fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/posix_acl_xattr.h>
  45 #include <linux/uio.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62 #include "qgroup.h"
  63
  64 struct btrfs_iget_args {
  65         struct btrfs_key *location;
  66         struct btrfs_root *root;
  67 };
  68
  69 static const struct inode_operations btrfs_dir_inode_operations;
  70 static const struct inode_operations btrfs_symlink_inode_operations;
  71 static const struct inode_operations btrfs_dir_ro_inode_operations;
  72 static const struct inode_operations btrfs_special_inode_operations;
  73 static const struct inode_operations btrfs_file_inode_operations;
  74 static const struct address_space_operations btrfs_aops;
  75 static const struct address_space_operations btrfs_symlink_aops;
  76 static const struct file_operations btrfs_dir_file_operations;
  77 static struct extent_io_ops btrfs_extent_io_ops;
  78
  79 static struct kmem_cache *btrfs_inode_cachep;
  80 static struct kmem_cache *btrfs_delalloc_work_cachep;
  81 struct kmem_cache *btrfs_trans_handle_cachep;
  82 struct kmem_cache *btrfs_transaction_cachep;
  83 struct kmem_cache *btrfs_path_cachep;
  84 struct kmem_cache *btrfs_free_space_cachep;
  85
  86 #define S_SHIFT 12
  87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  88         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  89         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  90         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  91         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  92         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  93         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  94         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  95 };
  96
  97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  98 static int btrfs_truncate(struct inode *inode);
  99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
 100 static noinline int cow_file_range(struct inode *inode,
 101                                    struct page *locked_page,
 102                                    u64 start, u64 end, int *page_started,
 103                                    unsigned long *nr_written, int unlock);
 104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 105                                            u64 len, u64 orig_start,
 106                                            u64 block_start, u64 block_len,
 107                                            u64 orig_block_len, u64 ram_bytes,
 108                                            int type);
 109
 110 static int btrfs_dirty_inode(struct inode *inode);
 111
 112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 113 void btrfs_test_inode_set_ops(struct inode *inode)
 114 {
 115         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 116 }
 117 #endif
 118
 119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 120                                      struct inode *inode,  struct inode *dir,
 121                                      const struct qstr *qstr)
 122 {
 123         int err;
 124
 125         err = btrfs_init_acl(trans, inode, dir);
 126         if (!err)
 127                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 128         return err;
 129 }
 130
 131 /*
 132  * this does all the hard work for inserting an inline extent into
 133  * the btree.  The caller should have done a btrfs_drop_extents so that
 134  * no overlapping inline items exist in the btree
 135  */
 136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_path *path, int extent_inserted,
 138                                 struct btrfs_root *root, struct inode *inode,
 139                                 u64 start, size_t size, size_t compressed_size,
 140                                 int compress_type,
 141                                 struct page **compressed_pages)
 142 {
 143         struct extent_buffer *leaf;
 144         struct page *page = NULL;
 145         char *kaddr;
 146         unsigned long ptr;
 147         struct btrfs_file_extent_item *ei;
 148         int err = 0;
 149         int ret;
 150         size_t cur_size = size;
 151         unsigned long offset;
 152
 153         if (compressed_size && compressed_pages)
 154                 cur_size = compressed_size;
 155
 156         inode_add_bytes(inode, size);
 157
 158         if (!extent_inserted) {
 159                 struct btrfs_key key;
 160                 size_t datasize;
 161
 162                 key.objectid = btrfs_ino(inode);
 163                 key.offset = start;
 164                 key.type = BTRFS_EXTENT_DATA_KEY;
 165
 166                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 167                 path->leave_spinning = 1;
 168                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 169                                               datasize);
 170                 if (ret) {
 171                         err = ret;
 172                         goto fail;
 173                 }
 174         }
 175         leaf = path->nodes[0];
 176         ei = btrfs_item_ptr(leaf, path->slots[0],
 177                             struct btrfs_file_extent_item);
 178         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 179         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 180         btrfs_set_file_extent_encryption(leaf, ei, 0);
 181         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 182         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 183         ptr = btrfs_file_extent_inline_start(ei);
 184
 185         if (compress_type != BTRFS_COMPRESS_NONE) {
 186                 struct page *cpage;
 187                 int i = 0;
 188                 while (compressed_size > 0) {
 189                         cpage = compressed_pages[i];
 190                         cur_size = min_t(unsigned long, compressed_size,
 191                                        PAGE_CACHE_SIZE);
 192
 193                         kaddr = kmap_atomic(cpage);
 194                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 195                         kunmap_atomic(kaddr);
 196
 197                         i++;
 198                         ptr += cur_size;
 199                         compressed_size -= cur_size;
 200                 }
 201                 btrfs_set_file_extent_compression(leaf, ei,
 202                                                   compress_type);
 203         } else {
 204                 page = find_get_page(inode->i_mapping,
 205                                      start >> PAGE_CACHE_SHIFT);
 206                 btrfs_set_file_extent_compression(leaf, ei, 0);
 207                 kaddr = kmap_atomic(page);
 208                 offset = start & (PAGE_CACHE_SIZE - 1);
 209                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 210                 kunmap_atomic(kaddr);
 211                 page_cache_release(page);
 212         }
 213         btrfs_mark_buffer_dirty(leaf);
 214         btrfs_release_path(path);
 215
 216         /*
 217          * we're an inline extent, so nobody can
 218          * extend the file past i_size without locking
 219          * a page we already have locked.
 220          *
 221          * We must do any isize and inode updates
 222          * before we unlock the pages.  Otherwise we
 223          * could end up racing with unlink.
 224          */
 225         BTRFS_I(inode)->disk_i_size = inode->i_size;
 226         ret = btrfs_update_inode(trans, root, inode);
 227
 228         return ret;
 229 fail:
 230         return err;
 231 }
 232
 233
 234 /*
 235  * conditionally insert an inline extent into the file.  This
 236  * does the checks required to make sure the data is small enough
 237  * to fit as an inline extent.
 238  */
 239 static noinline int cow_file_range_inline(struct btrfs_root *root,
 240                                           struct inode *inode, u64 start,
 241                                           u64 end, size_t compressed_size,
 242                                           int compress_type,
 243                                           struct page **compressed_pages)
 244 {
 245         struct btrfs_trans_handle *trans;
 246         u64 isize = i_size_read(inode);
 247         u64 actual_end = min(end + 1, isize);
 248         u64 inline_len = actual_end - start;
 249         u64 aligned_end = ALIGN(end, root->sectorsize);
 250         u64 data_len = inline_len;
 251         int ret;
 252         struct btrfs_path *path;
 253         int extent_inserted = 0;
 254         u32 extent_item_size;
 255
 256         if (compressed_size)
 257                 data_len = compressed_size;
 258
 259         if (start > 0 ||
 260             actual_end > PAGE_CACHE_SIZE ||
 261             data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 262             (!compressed_size &&
 263             (actual_end & (root->sectorsize - 1)) == 0) ||
 264             end + 1 < isize ||
 265             data_len > root->fs_info->max_inline) {
 266                 return 1;
 267         }
 268
 269         path = btrfs_alloc_path();
 270         if (!path)
 271                 return -ENOMEM;
 272
 273         trans = btrfs_join_transaction(root);
 274         if (IS_ERR(trans)) {
 275                 btrfs_free_path(path);
 276                 return PTR_ERR(trans);
 277         }
 278         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 279
 280         if (compressed_size && compressed_pages)
 281                 extent_item_size = btrfs_file_extent_calc_inline_size(
 282                    compressed_size);
 283         else
 284                 extent_item_size = btrfs_file_extent_calc_inline_size(
 285                     inline_len);
 286
 287         ret = __btrfs_drop_extents(trans, root, inode, path,
 288                                    start, aligned_end, NULL,
 289                                    1, 1, extent_item_size, &extent_inserted);
 290         if (ret) {
 291                 btrfs_abort_transaction(trans, root, ret);
 292                 goto out;
 293         }
 294
 295         if (isize > actual_end)
 296                 inline_len = min_t(u64, isize, actual_end);
 297         ret = insert_inline_extent(trans, path, extent_inserted,
 298                                    root, inode, start,
 299                                    inline_len, compressed_size,
 300                                    compress_type, compressed_pages);
 301         if (ret && ret != -ENOSPC) {
 302                 btrfs_abort_transaction(trans, root, ret);
 303                 goto out;
 304         } else if (ret == -ENOSPC) {
 305                 ret = 1;
 306                 goto out;
 307         }
 308
 309         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 310         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 311         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 312 out:
 313         btrfs_free_path(path);
 314         btrfs_end_transaction(trans, root);
 315         return ret;
 316 }
 317
 318 struct async_extent {
 319         u64 start;
 320         u64 ram_size;
 321         u64 compressed_size;
 322         struct page **pages;
 323         unsigned long nr_pages;
 324         int compress_type;
 325         struct list_head list;
 326 };
 327
 328 struct async_cow {
 329         struct inode *inode;
 330         struct btrfs_root *root;
 331         struct page *locked_page;
 332         u64 start;
 333         u64 end;
 334         struct list_head extents;
 335         struct btrfs_work work;
 336 };
 337
 338 static noinline int add_async_extent(struct async_cow *cow,
 339                                      u64 start, u64 ram_size,
 340                                      u64 compressed_size,
 341                                      struct page **pages,
 342                                      unsigned long nr_pages,
 343                                      int compress_type)
 344 {
 345         struct async_extent *async_extent;
 346
 347         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 348         BUG_ON(!async_extent); /* -ENOMEM */
 349         async_extent->start = start;
 350         async_extent->ram_size = ram_size;
 351         async_extent->compressed_size = compressed_size;
 352         async_extent->pages = pages;
 353         async_extent->nr_pages = nr_pages;
 354         async_extent->compress_type = compress_type;
 355         list_add_tail(&async_extent->list, &cow->extents);
 356         return 0;
 357 }
 358
 359 static inline int inode_need_compress(struct inode *inode)
 360 {
 361         struct btrfs_root *root = BTRFS_I(inode)->root;
 362
 363         /* force compress */
 364         if (btrfs_test_opt(root, FORCE_COMPRESS))
 365                 return 1;
 366         /* bad compression ratios */
 367         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 368                 return 0;
 369         if (btrfs_test_opt(root, COMPRESS) ||
 370             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 371             BTRFS_I(inode)->force_compress)
 372                 return 1;
 373         return 0;
 374 }
 375
 376 /*
 377  * we create compressed extents in two phases.  The first
 378  * phase compresses a range of pages that have already been
 379  * locked (both pages and state bits are locked).
 380  *
 381  * This is done inside an ordered work queue, and the compression
 382  * is spread across many cpus.  The actual IO submission is step
 383  * two, and the ordered work queue takes care of making sure that
 384  * happens in the same order things were put onto the queue by
 385  * writepages and friends.
 386  *
 387  * If this code finds it can't get good compression, it puts an
 388  * entry onto the work queue to write the uncompressed bytes.  This
 389  * makes sure that both compressed inodes and uncompressed inodes
 390  * are written in the same order that the flusher thread sent them
 391  * down.
 392  */
 393 static noinline void compress_file_range(struct inode *inode,
 394                                         struct page *locked_page,
 395                                         u64 start, u64 end,
 396                                         struct async_cow *async_cow,
 397                                         int *num_added)
 398 {
 399         struct btrfs_root *root = BTRFS_I(inode)->root;
 400         u64 num_bytes;
 401         u64 blocksize = root->sectorsize;
 402         u64 actual_end;
 403         u64 isize = i_size_read(inode);
 404         int ret = 0;
 405         struct page **pages = NULL;
 406         unsigned long nr_pages;
 407         unsigned long nr_pages_ret = 0;
 408         unsigned long total_compressed = 0;
 409         unsigned long total_in = 0;
 410         unsigned long max_compressed = 128 * 1024;
 411         unsigned long max_uncompressed = 128 * 1024;
 412         int i;
 413         int will_compress;
 414         int compress_type = root->fs_info->compress_type;
 415         int redirty = 0;
 416
 417         /* if this is a small write inside eof, kick off a defrag */
 418         if ((end - start + 1) < 16 * 1024 &&
 419             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 420                 btrfs_add_inode_defrag(NULL, inode);
 421
 422         actual_end = min_t(u64, isize, end + 1);
 423 again:
 424         will_compress = 0;
 425         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 426         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 427
 428         /*
 429          * we don't want to send crud past the end of i_size through
 430          * compression, that's just a waste of CPU time.  So, if the
 431          * end of the file is before the start of our current
 432          * requested range of bytes, we bail out to the uncompressed
 433          * cleanup code that can deal with all of this.
 434          *
 435          * It isn't really the fastest way to fix things, but this is a
 436          * very uncommon corner.
 437          */
 438         if (actual_end <= start)
 439                 goto cleanup_and_bail_uncompressed;
 440
 441         total_compressed = actual_end - start;
 442
 443         /*
 444          * skip compression for a small file range(<=blocksize) that
 445          * isn't an inline extent, since it dosen't save disk space at all.
 446          */
 447         if (total_compressed <= blocksize &&
 448            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 449                 goto cleanup_and_bail_uncompressed;
 450
 451         /* we want to make sure that amount of ram required to uncompress
 452          * an extent is reasonable, so we limit the total size in ram
 453          * of a compressed extent to 128k.  This is a crucial number
 454          * because it also controls how easily we can spread reads across
 455          * cpus for decompression.
 456          *
 457          * We also want to make sure the amount of IO required to do
 458          * a random read is reasonably small, so we limit the size of
 459          * a compressed extent to 128k.
 460          */
 461         total_compressed = min(total_compressed, max_uncompressed);
 462         num_bytes = ALIGN(end - start + 1, blocksize);
 463         num_bytes = max(blocksize,  num_bytes);
 464         total_in = 0;
 465         ret = 0;
 466
 467         /*
 468          * we do compression for mount -o compress and when the
 469          * inode has not been flagged as nocompress.  This flag can
 470          * change at any time if we discover bad compression ratios.
 471          */
 472         if (inode_need_compress(inode)) {
 473                 WARN_ON(pages);
 474                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 475                 if (!pages) {
 476                         /* just bail out to the uncompressed code */
 477                         goto cont;
 478                 }
 479
 480                 if (BTRFS_I(inode)->force_compress)
 481                         compress_type = BTRFS_I(inode)->force_compress;
 482
 483                 /*
 484                  * we need to call clear_page_dirty_for_io on each
 485                  * page in the range.  Otherwise applications with the file
 486                  * mmap'd can wander in and change the page contents while
 487                  * we are compressing them.
 488                  *
 489                  * If the compression fails for any reason, we set the pages
 490                  * dirty again later on.
 491                  */
 492                 extent_range_clear_dirty_for_io(inode, start, end);
 493                 redirty = 1;
 494                 ret = btrfs_compress_pages(compress_type,
 495                                            inode->i_mapping, start,
 496                                            total_compressed, pages,
 497                                            nr_pages, &nr_pages_ret,
 498                                            &total_in,
 499                                            &total_compressed,
 500                                            max_compressed);
 501
 502                 if (!ret) {
 503                         unsigned long offset = total_compressed &
 504                                 (PAGE_CACHE_SIZE - 1);
 505                         struct page *page = pages[nr_pages_ret - 1];
 506                         char *kaddr;
 507
 508                         /* zero the tail end of the last page, we might be
 509                          * sending it down to disk
 510                          */
 511                         if (offset) {
 512                                 kaddr = kmap_atomic(page);
 513                                 memset(kaddr + offset, 0,
 514                                        PAGE_CACHE_SIZE - offset);
 515                                 kunmap_atomic(kaddr);
 516                         }
 517                         will_compress = 1;
 518                 }
 519         }
 520 cont:
 521         if (start == 0) {
 522                 /* lets try to make an inline extent */
 523                 if (ret || total_in < (actual_end - start)) {
 524                         /* we didn't compress the entire range, try
 525                          * to make an uncompressed inline extent.
 526                          */
 527                         ret = cow_file_range_inline(root, inode, start, end,
 528                                                     0, 0, NULL);
 529                 } else {
 530                         /* try making a compressed inline extent */
 531                         ret = cow_file_range_inline(root, inode, start, end,
 532                                                     total_compressed,
 533                                                     compress_type, pages);
 534                 }
 535                 if (ret <= 0) {
 536                         unsigned long clear_flags = EXTENT_DELALLOC |
 537                                 EXTENT_DEFRAG;
 538                         unsigned long page_error_op;
 539
 540                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 541                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 542
 543                         /*
 544                          * inline extent creation worked or returned error,
 545                          * we don't need to create any more async work items.
 546                          * Unlock and free up our temp pages.
 547                          */
 548                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 549                                                      clear_flags, PAGE_UNLOCK |
 550                                                      PAGE_CLEAR_DIRTY |
 551                                                      PAGE_SET_WRITEBACK |
 552                                                      page_error_op |
 553                                                      PAGE_END_WRITEBACK);
 554                         goto free_pages_out;
 555                 }
 556         }
 557
 558         if (will_compress) {
 559                 /*
 560                  * we aren't doing an inline extent round the compressed size
 561                  * up to a block size boundary so the allocator does sane
 562                  * things
 563                  */
 564                 total_compressed = ALIGN(total_compressed, blocksize);
 565
 566                 /*
 567                  * one last check to make sure the compression is really a
 568                  * win, compare the page count read with the blocks on disk
 569                  */
 570                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 571                 if (total_compressed >= total_in) {
 572                         will_compress = 0;
 573                 } else {
 574                         num_bytes = total_in;
 575                 }
 576         }
 577         if (!will_compress && pages) {
 578                 /*
 579                  * the compression code ran but failed to make things smaller,
 580                  * free any pages it allocated and our page pointer array
 581                  */
 582                 for (i = 0; i < nr_pages_ret; i++) {
 583                         WARN_ON(pages[i]->mapping);
 584                         page_cache_release(pages[i]);
 585                 }
 586                 kfree(pages);
 587                 pages = NULL;
 588                 total_compressed = 0;
 589                 nr_pages_ret = 0;
 590
 591                 /* flag the file so we don't compress in the future */
 592                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 593                     !(BTRFS_I(inode)->force_compress)) {
 594                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 595                 }
 596         }
 597         if (will_compress) {
 598                 *num_added += 1;
 599
 600                 /* the async work queues will take care of doing actual
 601                  * allocation on disk for these compressed pages,
 602                  * and will submit them to the elevator.
 603                  */
 604                 add_async_extent(async_cow, start, num_bytes,
 605                                  total_compressed, pages, nr_pages_ret,
 606                                  compress_type);
 607
 608                 if (start + num_bytes < end) {
 609                         start += num_bytes;
 610                         pages = NULL;
 611                         cond_resched();
 612                         goto again;
 613                 }
 614         } else {
 615 cleanup_and_bail_uncompressed:
 616                 /*
 617                  * No compression, but we still need to write the pages in
 618                  * the file we've been given so far.  redirty the locked
 619                  * page if it corresponds to our extent and set things up
 620                  * for the async work queue to run cow_file_range to do
 621                  * the normal delalloc dance
 622                  */
 623                 if (page_offset(locked_page) >= start &&
 624                     page_offset(locked_page) <= end) {
 625                         __set_page_dirty_nobuffers(locked_page);
 626                         /* unlocked later on in the async handlers */
 627                 }
 628                 if (redirty)
 629                         extent_range_redirty_for_io(inode, start, end);
 630                 add_async_extent(async_cow, start, end - start + 1,
 631                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 632                 *num_added += 1;
 633         }
 634
 635         return;
 636
 637 free_pages_out:
 638         for (i = 0; i < nr_pages_ret; i++) {
 639                 WARN_ON(pages[i]->mapping);
 640                 page_cache_release(pages[i]);
 641         }
 642         kfree(pages);
 643 }
 644
 645 static void free_async_extent_pages(struct async_extent *async_extent)
 646 {
 647         int i;
 648
 649         if (!async_extent->pages)
 650                 return;
 651
 652         for (i = 0; i < async_extent->nr_pages; i++) {
 653                 WARN_ON(async_extent->pages[i]->mapping);
 654                 page_cache_release(async_extent->pages[i]);
 655         }
 656         kfree(async_extent->pages);
 657         async_extent->nr_pages = 0;
 658         async_extent->pages = NULL;
 659 }
 660
 661 /*
 662  * phase two of compressed writeback.  This is the ordered portion
 663  * of the code, which only gets called in the order the work was
 664  * queued.  We walk all the async extents created by compress_file_range
 665  * and send them down to the disk.
 666  */
 667 static noinline void submit_compressed_extents(struct inode *inode,
 668                                               struct async_cow *async_cow)
 669 {
 670         struct async_extent *async_extent;
 671         u64 alloc_hint = 0;
 672         struct btrfs_key ins;
 673         struct extent_map *em;
 674         struct btrfs_root *root = BTRFS_I(inode)->root;
 675         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 676         struct extent_io_tree *io_tree;
 677         int ret = 0;
 678
 679 again:
 680         while (!list_empty(&async_cow->extents)) {
 681                 async_extent = list_entry(async_cow->extents.next,
 682                                           struct async_extent, list);
 683                 list_del(&async_extent->list);
 684
 685                 io_tree = &BTRFS_I(inode)->io_tree;
 686
 687 retry:
 688                 /* did the compression code fall back to uncompressed IO? */
 689                 if (!async_extent->pages) {
 690                         int page_started = 0;
 691                         unsigned long nr_written = 0;
 692
 693                         lock_extent(io_tree, async_extent->start,
 694                                          async_extent->start +
 695                                          async_extent->ram_size - 1);
 696
 697                         /* allocate blocks */
 698                         ret = cow_file_range(inode, async_cow->locked_page,
 699                                              async_extent->start,
 700                                              async_extent->start +
 701                                              async_extent->ram_size - 1,
 702                                              &page_started, &nr_written, 0);
 703
 704                         /* JDM XXX */
 705
 706                         /*
 707                          * if page_started, cow_file_range inserted an
 708                          * inline extent and took care of all the unlocking
 709                          * and IO for us.  Otherwise, we need to submit
 710                          * all those pages down to the drive.
 711                          */
 712                         if (!page_started && !ret)
 713                                 extent_write_locked_range(io_tree,
 714                                                   inode, async_extent->start,
 715                                                   async_extent->start +
 716                                                   async_extent->ram_size - 1,
 717                                                   btrfs_get_extent,
 718                                                   WB_SYNC_ALL);
 719                         else if (ret)
 720                                 unlock_page(async_cow->locked_page);
 721                         kfree(async_extent);
 722                         cond_resched();
 723                         continue;
 724                 }
 725
 726                 lock_extent(io_tree, async_extent->start,
 727                             async_extent->start + async_extent->ram_size - 1);
 728
 729                 ret = btrfs_reserve_extent(root,
 730                                            async_extent->compressed_size,
 731                                            async_extent->compressed_size,
 732                                            0, alloc_hint, &ins, 1, 1);
 733                 if (ret) {
 734                         free_async_extent_pages(async_extent);
 735
 736                         if (ret == -ENOSPC) {
 737                                 unlock_extent(io_tree, async_extent->start,
 738                                               async_extent->start +
 739                                               async_extent->ram_size - 1);
 740
 741                                 /*
 742                                  * we need to redirty the pages if we decide to
 743                                  * fallback to uncompressed IO, otherwise we
 744                                  * will not submit these pages down to lower
 745                                  * layers.
 746                                  */
 747                                 extent_range_redirty_for_io(inode,
 748                                                 async_extent->start,
 749                                                 async_extent->start +
 750                                                 async_extent->ram_size - 1);
 751
 752                                 goto retry;
 753                         }
 754                         goto out_free;
 755                 }
 756                 /*
 757                  * here we're doing allocation and writeback of the
 758                  * compressed pages
 759                  */
 760                 btrfs_drop_extent_cache(inode, async_extent->start,
 761                                         async_extent->start +
 762                                         async_extent->ram_size - 1, 0);
 763
 764                 em = alloc_extent_map();
 765                 if (!em) {
 766                         ret = -ENOMEM;
 767                         goto out_free_reserve;
 768                 }
 769                 em->start = async_extent->start;
 770                 em->len = async_extent->ram_size;
 771                 em->orig_start = em->start;
 772                 em->mod_start = em->start;
 773                 em->mod_len = em->len;
 774
 775                 em->block_start = ins.objectid;
 776                 em->block_len = ins.offset;
 777                 em->orig_block_len = ins.offset;
 778                 em->ram_bytes = async_extent->ram_size;
 779                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 780                 em->compress_type = async_extent->compress_type;
 781                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 782                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 783                 em->generation = -1;
 784
 785                 while (1) {
 786                         write_lock(&em_tree->lock);
 787                         ret = add_extent_mapping(em_tree, em, 1);
 788                         write_unlock(&em_tree->lock);
 789                         if (ret != -EEXIST) {
 790                                 free_extent_map(em);
 791                                 break;
 792                         }
 793                         btrfs_drop_extent_cache(inode, async_extent->start,
 794                                                 async_extent->start +
 795                                                 async_extent->ram_size - 1, 0);
 796                 }
 797
 798                 if (ret)
 799                         goto out_free_reserve;
 800
 801                 ret = btrfs_add_ordered_extent_compress(inode,
 802                                                 async_extent->start,
 803                                                 ins.objectid,
 804                                                 async_extent->ram_size,
 805                                                 ins.offset,
 806                                                 BTRFS_ORDERED_COMPRESSED,
 807                                                 async_extent->compress_type);
 808                 if (ret) {
 809                         btrfs_drop_extent_cache(inode, async_extent->start,
 810                                                 async_extent->start +
 811                                                 async_extent->ram_size - 1, 0);
 812                         goto out_free_reserve;
 813                 }
 814
 815                 /*
 816                  * clear dirty, set writeback and unlock the pages.
 817                  */
 818                 extent_clear_unlock_delalloc(inode, async_extent->start,
 819                                 async_extent->start +
 820                                 async_extent->ram_size - 1,
 821                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 822                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 823                                 PAGE_SET_WRITEBACK);
 824                 ret = btrfs_submit_compressed_write(inode,
 825                                     async_extent->start,
 826                                     async_extent->ram_size,
 827                                     ins.objectid,
 828                                     ins.offset, async_extent->pages,
 829                                     async_extent->nr_pages);
 830                 if (ret) {
 831                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 832                         struct page *p = async_extent->pages[0];
 833                         const u64 start = async_extent->start;
 834                         const u64 end = start + async_extent->ram_size - 1;
 835
 836                         p->mapping = inode->i_mapping;
 837                         tree->ops->writepage_end_io_hook(p, start, end,
 838                                                          NULL, 0);
 839                         p->mapping = NULL;
 840                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
 841                                                      PAGE_END_WRITEBACK |
 842                                                      PAGE_SET_ERROR);
 843                         free_async_extent_pages(async_extent);
 844                 }
 845                 alloc_hint = ins.objectid + ins.offset;
 846                 kfree(async_extent);
 847                 cond_resched();
 848         }
 849         return;
 850 out_free_reserve:
 851         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 852 out_free:
 853         extent_clear_unlock_delalloc(inode, async_extent->start,
 854                                      async_extent->start +
 855                                      async_extent->ram_size - 1,
 856                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 857                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 858                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 859                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 860                                      PAGE_SET_ERROR);
 861         free_async_extent_pages(async_extent);
 862         kfree(async_extent);
 863         goto again;
 864 }
 865
 866 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 867                                       u64 num_bytes)
 868 {
 869         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870         struct extent_map *em;
 871         u64 alloc_hint = 0;
 872
 873         read_lock(&em_tree->lock);
 874         em = search_extent_mapping(em_tree, start, num_bytes);
 875         if (em) {
 876                 /*
 877                  * if block start isn't an actual block number then find the
 878                  * first block in this inode and use that as a hint.  If that
 879                  * block is also bogus then just don't worry about it.
 880                  */
 881                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 882                         free_extent_map(em);
 883                         em = search_extent_mapping(em_tree, 0, 0);
 884                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 885                                 alloc_hint = em->block_start;
 886                         if (em)
 887                                 free_extent_map(em);
 888                 } else {
 889                         alloc_hint = em->block_start;
 890                         free_extent_map(em);
 891                 }
 892         }
 893         read_unlock(&em_tree->lock);
 894
 895         return alloc_hint;
 896 }
 897
 898 /*
 899  * when extent_io.c finds a delayed allocation range in the file,
 900  * the call backs end up in this code.  The basic idea is to
 901  * allocate extents on disk for the range, and create ordered data structs
 902  * in ram to track those extents.
 903  *
 904  * locked_page is the page that writepage had locked already.  We use
 905  * it to make sure we don't do extra locks or unlocks.
 906  *
 907  * *page_started is set to one if we unlock locked_page and do everything
 908  * required to start IO on it.  It may be clean and already done with
 909  * IO when we return.
 910  */
 911 static noinline int cow_file_range(struct inode *inode,
 912                                    struct page *locked_page,
 913                                    u64 start, u64 end, int *page_started,
 914                                    unsigned long *nr_written,
 915                                    int unlock)
 916 {
 917         struct btrfs_root *root = BTRFS_I(inode)->root;
 918         u64 alloc_hint = 0;
 919         u64 num_bytes;
 920         unsigned long ram_size;
 921         u64 disk_num_bytes;
 922         u64 cur_alloc_size;
 923         u64 blocksize = root->sectorsize;
 924         struct btrfs_key ins;
 925         struct extent_map *em;
 926         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 927         int ret = 0;
 928
 929         if (btrfs_is_free_space_inode(inode)) {
 930                 WARN_ON_ONCE(1);
 931                 ret = -EINVAL;
 932                 goto out_unlock;
 933         }
 934
 935         num_bytes = ALIGN(end - start + 1, blocksize);
 936         num_bytes = max(blocksize,  num_bytes);
 937         disk_num_bytes = num_bytes;
 938
 939         /* if this is a small write inside eof, kick off defrag */
 940         if (num_bytes < 64 * 1024 &&
 941             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 942                 btrfs_add_inode_defrag(NULL, inode);
 943
 944         if (start == 0) {
 945                 /* lets try to make an inline extent */
 946                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 947                                             NULL);
 948                 if (ret == 0) {
 949                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 950                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 951                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 952                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 953                                      PAGE_END_WRITEBACK);
 954
 955                         *nr_written = *nr_written +
 956                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 957                         *page_started = 1;
 958                         goto out;
 959                 } else if (ret < 0) {
 960                         goto out_unlock;
 961                 }
 962         }
 963
 964         BUG_ON(disk_num_bytes >
 965                btrfs_super_total_bytes(root->fs_info->super_copy));
 966
 967         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 968         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 969
 970         while (disk_num_bytes > 0) {
 971                 unsigned long op;
 972
 973                 cur_alloc_size = disk_num_bytes;
 974                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 975                                            root->sectorsize, 0, alloc_hint,
 976                                            &ins, 1, 1);
 977                 if (ret < 0)
 978                         goto out_unlock;
 979
 980                 em = alloc_extent_map();
 981                 if (!em) {
 982                         ret = -ENOMEM;
 983                         goto out_reserve;
 984                 }
 985                 em->start = start;
 986                 em->orig_start = em->start;
 987                 ram_size = ins.offset;
 988                 em->len = ins.offset;
 989                 em->mod_start = em->start;
 990                 em->mod_len = em->len;
 991
 992                 em->block_start = ins.objectid;
 993                 em->block_len = ins.offset;
 994                 em->orig_block_len = ins.offset;
 995                 em->ram_bytes = ram_size;
 996                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 997                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 998                 em->generation = -1;
 999
1000                 while (1) {
1001                         write_lock(&em_tree->lock);
1002                         ret = add_extent_mapping(em_tree, em, 1);
1003                         write_unlock(&em_tree->lock);
1004                         if (ret != -EEXIST) {
1005                                 free_extent_map(em);
1006                                 break;
1007                         }
1008                         btrfs_drop_extent_cache(inode, start,
1009                                                 start + ram_size - 1, 0);
1010                 }
1011                 if (ret)
1012                         goto out_reserve;
1013
1014                 cur_alloc_size = ins.offset;
1015                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1016                                                ram_size, cur_alloc_size, 0);
1017                 if (ret)
1018                         goto out_drop_extent_cache;
1019
1020                 if (root->root_key.objectid ==
1021                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1022                         ret = btrfs_reloc_clone_csums(inode, start,
1023                                                       cur_alloc_size);
1024                         if (ret)
1025                                 goto out_drop_extent_cache;
1026                 }
1027
1028                 if (disk_num_bytes < cur_alloc_size)
1029                         break;
1030
1031                 /* we're not doing compressed IO, don't unlock the first
1032                  * page (which the caller expects to stay locked), don't
1033                  * clear any dirty bits and don't set any writeback bits
1034                  *
1035                  * Do set the Private2 bit so we know this page was properly
1036                  * setup for writepage
1037                  */
1038                 op = unlock ? PAGE_UNLOCK : 0;
1039                 op |= PAGE_SET_PRIVATE2;
1040
1041                 extent_clear_unlock_delalloc(inode, start,
1042                                              start + ram_size - 1, locked_page,
1043                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1044                                              op);
1045                 disk_num_bytes -= cur_alloc_size;
1046                 num_bytes -= cur_alloc_size;
1047                 alloc_hint = ins.objectid + ins.offset;
1048                 start += cur_alloc_size;
1049         }
1050 out:
1051         return ret;
1052
1053 out_drop_extent_cache:
1054         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1055 out_reserve:
1056         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1057 out_unlock:
1058         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1059                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1060                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1061                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1062                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1063         goto out;
1064 }
1065
1066 /*
1067  * work queue call back to started compression on a file and pages
1068  */
1069 static noinline void async_cow_start(struct btrfs_work *work)
1070 {
1071         struct async_cow *async_cow;
1072         int num_added = 0;
1073         async_cow = container_of(work, struct async_cow, work);
1074
1075         compress_file_range(async_cow->inode, async_cow->locked_page,
1076                             async_cow->start, async_cow->end, async_cow,
1077                             &num_added);
1078         if (num_added == 0) {
1079                 btrfs_add_delayed_iput(async_cow->inode);
1080                 async_cow->inode = NULL;
1081         }
1082 }
1083
1084 /*
1085  * work queue call back to submit previously compressed pages
1086  */
1087 static noinline void async_cow_submit(struct btrfs_work *work)
1088 {
1089         struct async_cow *async_cow;
1090         struct btrfs_root *root;
1091         unsigned long nr_pages;
1092
1093         async_cow = container_of(work, struct async_cow, work);
1094
1095         root = async_cow->root;
1096         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1097                 PAGE_CACHE_SHIFT;
1098
1099         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1100             5 * 1024 * 1024 &&
1101             waitqueue_active(&root->fs_info->async_submit_wait))
1102                 wake_up(&root->fs_info->async_submit_wait);
1103
1104         if (async_cow->inode)
1105                 submit_compressed_extents(async_cow->inode, async_cow);
1106 }
1107
1108 static noinline void async_cow_free(struct btrfs_work *work)
1109 {
1110         struct async_cow *async_cow;
1111         async_cow = container_of(work, struct async_cow, work);
1112         if (async_cow->inode)
1113                 btrfs_add_delayed_iput(async_cow->inode);
1114         kfree(async_cow);
1115 }
1116
1117 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1118                                 u64 start, u64 end, int *page_started,
1119                                 unsigned long *nr_written)
1120 {
1121         struct async_cow *async_cow;
1122         struct btrfs_root *root = BTRFS_I(inode)->root;
1123         unsigned long nr_pages;
1124         u64 cur_end;
1125         int limit = 10 * 1024 * 1024;
1126
1127         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1128                          1, 0, NULL, GFP_NOFS);
1129         while (start < end) {
1130                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1131                 BUG_ON(!async_cow); /* -ENOMEM */
1132                 async_cow->inode = igrab(inode);
1133                 async_cow->root = root;
1134                 async_cow->locked_page = locked_page;
1135                 async_cow->start = start;
1136
1137                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1138                     !btrfs_test_opt(root, FORCE_COMPRESS))
1139                         cur_end = end;
1140                 else
1141                         cur_end = min(end, start + 512 * 1024 - 1);
1142
1143                 async_cow->end = cur_end;
1144                 INIT_LIST_HEAD(&async_cow->extents);
1145
1146                 btrfs_init_work(&async_cow->work,
1147                                 btrfs_delalloc_helper,
1148                                 async_cow_start, async_cow_submit,
1149                                 async_cow_free);
1150
1151                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1152                         PAGE_CACHE_SHIFT;
1153                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1154
1155                 btrfs_queue_work(root->fs_info->delalloc_workers,
1156                                  &async_cow->work);
1157
1158                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1159                         wait_event(root->fs_info->async_submit_wait,
1160                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1161                             limit));
1162                 }
1163
1164                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1165                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1166                         wait_event(root->fs_info->async_submit_wait,
1167                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1168                            0));
1169                 }
1170
1171                 *nr_written += nr_pages;
1172                 start = cur_end + 1;
1173         }
1174         *page_started = 1;
1175         return 0;
1176 }
1177
1178 static noinline int csum_exist_in_range(struct btrfs_root *root,
1179                                         u64 bytenr, u64 num_bytes)
1180 {
1181         int ret;
1182         struct btrfs_ordered_sum *sums;
1183         LIST_HEAD(list);
1184
1185         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1186                                        bytenr + num_bytes - 1, &list, 0);
1187         if (ret == 0 && list_empty(&list))
1188                 return 0;
1189
1190         while (!list_empty(&list)) {
1191                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1192                 list_del(&sums->list);
1193                 kfree(sums);
1194         }
1195         return 1;
1196 }
1197
1198 /*
1199  * when nowcow writeback call back.  This checks for snapshots or COW copies
1200  * of the extents that exist in the file, and COWs the file as required.
1201  *
1202  * If no cow copies or snapshots exist, we write directly to the existing
1203  * blocks on disk
1204  */
1205 static noinline int run_delalloc_nocow(struct inode *inode,
1206                                        struct page *locked_page,
1207                               u64 start, u64 end, int *page_started, int force,
1208                               unsigned long *nr_written)
1209 {
1210         struct btrfs_root *root = BTRFS_I(inode)->root;
1211         struct btrfs_trans_handle *trans;
1212         struct extent_buffer *leaf;
1213         struct btrfs_path *path;
1214         struct btrfs_file_extent_item *fi;
1215         struct btrfs_key found_key;
1216         u64 cow_start;
1217         u64 cur_offset;
1218         u64 extent_end;
1219         u64 extent_offset;
1220         u64 disk_bytenr;
1221         u64 num_bytes;
1222         u64 disk_num_bytes;
1223         u64 ram_bytes;
1224         int extent_type;
1225         int ret, err;
1226         int type;
1227         int nocow;
1228         int check_prev = 1;
1229         bool nolock;
1230         u64 ino = btrfs_ino(inode);
1231
1232         path = btrfs_alloc_path();
1233         if (!path) {
1234                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1235                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1236                                              EXTENT_DO_ACCOUNTING |
1237                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1238                                              PAGE_CLEAR_DIRTY |
1239                                              PAGE_SET_WRITEBACK |
1240                                              PAGE_END_WRITEBACK);
1241                 return -ENOMEM;
1242         }
1243
1244         nolock = btrfs_is_free_space_inode(inode);
1245
1246         if (nolock)
1247                 trans = btrfs_join_transaction_nolock(root);
1248         else
1249                 trans = btrfs_join_transaction(root);
1250
1251         if (IS_ERR(trans)) {
1252                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1253                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1254                                              EXTENT_DO_ACCOUNTING |
1255                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1256                                              PAGE_CLEAR_DIRTY |
1257                                              PAGE_SET_WRITEBACK |
1258                                              PAGE_END_WRITEBACK);
1259                 btrfs_free_path(path);
1260                 return PTR_ERR(trans);
1261         }
1262
1263         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1264
1265         cow_start = (u64)-1;
1266         cur_offset = start;
1267         while (1) {
1268                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1269                                                cur_offset, 0);
1270                 if (ret < 0)
1271                         goto error;
1272                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1273                         leaf = path->nodes[0];
1274                         btrfs_item_key_to_cpu(leaf, &found_key,
1275                                               path->slots[0] - 1);
1276                         if (found_key.objectid == ino &&
1277                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1278                                 path->slots[0]--;
1279                 }
1280                 check_prev = 0;
1281 next_slot:
1282                 leaf = path->nodes[0];
1283                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1284                         ret = btrfs_next_leaf(root, path);
1285                         if (ret < 0)
1286                                 goto error;
1287                         if (ret > 0)
1288                                 break;
1289                         leaf = path->nodes[0];
1290                 }
1291
1292                 nocow = 0;
1293                 disk_bytenr = 0;
1294                 num_bytes = 0;
1295                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1296
1297                 if (found_key.objectid > ino ||
1298                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1299                     found_key.offset > end)
1300                         break;
1301
1302                 if (found_key.offset > cur_offset) {
1303                         extent_end = found_key.offset;
1304                         extent_type = 0;
1305                         goto out_check;
1306                 }
1307
1308                 fi = btrfs_item_ptr(leaf, path->slots[0],
1309                                     struct btrfs_file_extent_item);
1310                 extent_type = btrfs_file_extent_type(leaf, fi);
1311
1312                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1313                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1314                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1315                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1316                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1317                         extent_end = found_key.offset +
1318                                 btrfs_file_extent_num_bytes(leaf, fi);
1319                         disk_num_bytes =
1320                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1321                         if (extent_end <= start) {
1322                                 path->slots[0]++;
1323                                 goto next_slot;
1324                         }
1325                         if (disk_bytenr == 0)
1326                                 goto out_check;
1327                         if (btrfs_file_extent_compression(leaf, fi) ||
1328                             btrfs_file_extent_encryption(leaf, fi) ||
1329                             btrfs_file_extent_other_encoding(leaf, fi))
1330                                 goto out_check;
1331                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1332                                 goto out_check;
1333                         if (btrfs_extent_readonly(root, disk_bytenr))
1334                                 goto out_check;
1335                         if (btrfs_cross_ref_exist(trans, root, ino,
1336                                                   found_key.offset -
1337                                                   extent_offset, disk_bytenr))
1338                                 goto out_check;
1339                         disk_bytenr += extent_offset;
1340                         disk_bytenr += cur_offset - found_key.offset;
1341                         num_bytes = min(end + 1, extent_end) - cur_offset;
1342                         /*
1343                          * if there are pending snapshots for this root,
1344                          * we fall into common COW way.
1345                          */
1346                         if (!nolock) {
1347                                 err = btrfs_start_write_no_snapshoting(root);
1348                                 if (!err)
1349                                         goto out_check;
1350                         }
1351                         /*
1352                          * force cow if csum exists in the range.
1353                          * this ensure that csum for a given extent are
1354                          * either valid or do not exist.
1355                          */
1356                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1357                                 goto out_check;
1358                         nocow = 1;
1359                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1360                         extent_end = found_key.offset +
1361                                 btrfs_file_extent_inline_len(leaf,
1362                                                      path->slots[0], fi);
1363                         extent_end = ALIGN(extent_end, root->sectorsize);
1364                 } else {
1365                         BUG_ON(1);
1366                 }
1367 out_check:
1368                 if (extent_end <= start) {
1369                         path->slots[0]++;
1370                         if (!nolock && nocow)
1371                                 btrfs_end_write_no_snapshoting(root);
1372                         goto next_slot;
1373                 }
1374                 if (!nocow) {
1375                         if (cow_start == (u64)-1)
1376                                 cow_start = cur_offset;
1377                         cur_offset = extent_end;
1378                         if (cur_offset > end)
1379                                 break;
1380                         path->slots[0]++;
1381                         goto next_slot;
1382                 }
1383
1384                 btrfs_release_path(path);
1385                 if (cow_start != (u64)-1) {
1386                         ret = cow_file_range(inode, locked_page,
1387                                              cow_start, found_key.offset - 1,
1388                                              page_started, nr_written, 1);
1389                         if (ret) {
1390                                 if (!nolock && nocow)
1391                                         btrfs_end_write_no_snapshoting(root);
1392                                 goto error;
1393                         }
1394                         cow_start = (u64)-1;
1395                 }
1396
1397                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1398                         struct extent_map *em;
1399                         struct extent_map_tree *em_tree;
1400                         em_tree = &BTRFS_I(inode)->extent_tree;
1401                         em = alloc_extent_map();
1402                         BUG_ON(!em); /* -ENOMEM */
1403                         em->start = cur_offset;
1404                         em->orig_start = found_key.offset - extent_offset;
1405                         em->len = num_bytes;
1406                         em->block_len = num_bytes;
1407                         em->block_start = disk_bytenr;
1408                         em->orig_block_len = disk_num_bytes;
1409                         em->ram_bytes = ram_bytes;
1410                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1411                         em->mod_start = em->start;
1412                         em->mod_len = em->len;
1413                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1414                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1415                         em->generation = -1;
1416                         while (1) {
1417                                 write_lock(&em_tree->lock);
1418                                 ret = add_extent_mapping(em_tree, em, 1);
1419                                 write_unlock(&em_tree->lock);
1420                                 if (ret != -EEXIST) {
1421                                         free_extent_map(em);
1422                                         break;
1423                                 }
1424                                 btrfs_drop_extent_cache(inode, em->start,
1425                                                 em->start + em->len - 1, 0);
1426                         }
1427                         type = BTRFS_ORDERED_PREALLOC;
1428                 } else {
1429                         type = BTRFS_ORDERED_NOCOW;
1430                 }
1431
1432                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1433                                                num_bytes, num_bytes, type);
1434                 BUG_ON(ret); /* -ENOMEM */
1435
1436                 if (root->root_key.objectid ==
1437                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1438                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1439                                                       num_bytes);
1440                         if (ret) {
1441                                 if (!nolock && nocow)
1442                                         btrfs_end_write_no_snapshoting(root);
1443                                 goto error;
1444                         }
1445                 }
1446
1447                 extent_clear_unlock_delalloc(inode, cur_offset,
1448                                              cur_offset + num_bytes - 1,
1449                                              locked_page, EXTENT_LOCKED |
1450                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1451                                              PAGE_SET_PRIVATE2);
1452                 if (!nolock && nocow)
1453                         btrfs_end_write_no_snapshoting(root);
1454                 cur_offset = extent_end;
1455                 if (cur_offset > end)
1456                         break;
1457         }
1458         btrfs_release_path(path);
1459
1460         if (cur_offset <= end && cow_start == (u64)-1) {
1461                 cow_start = cur_offset;
1462                 cur_offset = end;
1463         }
1464
1465         if (cow_start != (u64)-1) {
1466                 ret = cow_file_range(inode, locked_page, cow_start, end,
1467                                      page_started, nr_written, 1);
1468                 if (ret)
1469                         goto error;
1470         }
1471
1472 error:
1473         err = btrfs_end_transaction(trans, root);
1474         if (!ret)
1475                 ret = err;
1476
1477         if (ret && cur_offset < end)
1478                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1479                                              locked_page, EXTENT_LOCKED |
1480                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1481                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1482                                              PAGE_CLEAR_DIRTY |
1483                                              PAGE_SET_WRITEBACK |
1484                                              PAGE_END_WRITEBACK);
1485         btrfs_free_path(path);
1486         return ret;
1487 }
1488
1489 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1490 {
1491
1492         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1493             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1494                 return 0;
1495
1496         /*
1497          * @defrag_bytes is a hint value, no spinlock held here,
1498          * if is not zero, it means the file is defragging.
1499          * Force cow if given extent needs to be defragged.
1500          */
1501         if (BTRFS_I(inode)->defrag_bytes &&
1502             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1503                            EXTENT_DEFRAG, 0, NULL))
1504                 return 1;
1505
1506         return 0;
1507 }
1508
1509 /*
1510  * extent_io.c call back to do delayed allocation processing
1511  */
1512 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1513                               u64 start, u64 end, int *page_started,
1514                               unsigned long *nr_written)
1515 {
1516         int ret;
1517         int force_cow = need_force_cow(inode, start, end);
1518
1519         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1520                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1521                                          page_started, 1, nr_written);
1522         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1523                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1524                                          page_started, 0, nr_written);
1525         } else if (!inode_need_compress(inode)) {
1526                 ret = cow_file_range(inode, locked_page, start, end,
1527                                       page_started, nr_written, 1);
1528         } else {
1529                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1530                         &BTRFS_I(inode)->runtime_flags);
1531                 ret = cow_file_range_async(inode, locked_page, start, end,
1532                                            page_started, nr_written);
1533         }
1534         return ret;
1535 }
1536
1537 static void btrfs_split_extent_hook(struct inode *inode,
1538                                     struct extent_state *orig, u64 split)
1539 {
1540         u64 size;
1541
1542         /* not delalloc, ignore it */
1543         if (!(orig->state & EXTENT_DELALLOC))
1544                 return;
1545
1546         size = orig->end - orig->start + 1;
1547         if (size > BTRFS_MAX_EXTENT_SIZE) {
1548                 u64 num_extents;
1549                 u64 new_size;
1550
1551                 /*
1552                  * See the explanation in btrfs_merge_extent_hook, the same
1553                  * applies here, just in reverse.
1554                  */
1555                 new_size = orig->end - split + 1;
1556                 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1557                                         BTRFS_MAX_EXTENT_SIZE);
1558                 new_size = split - orig->start;
1559                 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1560                                         BTRFS_MAX_EXTENT_SIZE);
1561                 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1562                               BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1563                         return;
1564         }
1565
1566         spin_lock(&BTRFS_I(inode)->lock);
1567         BTRFS_I(inode)->outstanding_extents++;
1568         spin_unlock(&BTRFS_I(inode)->lock);
1569 }
1570
1571 /*
1572  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1573  * extents so we can keep track of new extents that are just merged onto old
1574  * extents, such as when we are doing sequential writes, so we can properly
1575  * account for the metadata space we'll need.
1576  */
1577 static void btrfs_merge_extent_hook(struct inode *inode,
1578                                     struct extent_state *new,
1579                                     struct extent_state *other)
1580 {
1581         u64 new_size, old_size;
1582         u64 num_extents;
1583
1584         /* not delalloc, ignore it */
1585         if (!(other->state & EXTENT_DELALLOC))
1586                 return;
1587
1588         if (new->start > other->start)
1589                 new_size = new->end - other->start + 1;
1590         else
1591                 new_size = other->end - new->start + 1;
1592
1593         /* we're not bigger than the max, unreserve the space and go */
1594         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1595                 spin_lock(&BTRFS_I(inode)->lock);
1596                 BTRFS_I(inode)->outstanding_extents--;
1597                 spin_unlock(&BTRFS_I(inode)->lock);
1598                 return;
1599         }
1600
1601         /*
1602          * We have to add up either side to figure out how many extents were
1603          * accounted for before we merged into one big extent.  If the number of
1604          * extents we accounted for is <= the amount we need for the new range
1605          * then we can return, otherwise drop.  Think of it like this
1606          *
1607          * [ 4k][MAX_SIZE]
1608          *
1609          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1610          * need 2 outstanding extents, on one side we have 1 and the other side
1611          * we have 1 so they are == and we can return.  But in this case
1612          *
1613          * [MAX_SIZE+4k][MAX_SIZE+4k]
1614          *
1615          * Each range on their own accounts for 2 extents, but merged together
1616          * they are only 3 extents worth of accounting, so we need to drop in
1617          * this case.
1618          */
1619         old_size = other->end - other->start + 1;
1620         num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1621                                 BTRFS_MAX_EXTENT_SIZE);
1622         old_size = new->end - new->start + 1;
1623         num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1624                                  BTRFS_MAX_EXTENT_SIZE);
1625
1626         if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1627                       BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1628                 return;
1629
1630         spin_lock(&BTRFS_I(inode)->lock);
1631         BTRFS_I(inode)->outstanding_extents--;
1632         spin_unlock(&BTRFS_I(inode)->lock);
1633 }
1634
1635 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1636                                       struct inode *inode)
1637 {
1638         spin_lock(&root->delalloc_lock);
1639         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1640                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1641                               &root->delalloc_inodes);
1642                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1643                         &BTRFS_I(inode)->runtime_flags);
1644                 root->nr_delalloc_inodes++;
1645                 if (root->nr_delalloc_inodes == 1) {
1646                         spin_lock(&root->fs_info->delalloc_root_lock);
1647                         BUG_ON(!list_empty(&root->delalloc_root));
1648                         list_add_tail(&root->delalloc_root,
1649                                       &root->fs_info->delalloc_roots);
1650                         spin_unlock(&root->fs_info->delalloc_root_lock);
1651                 }
1652         }
1653         spin_unlock(&root->delalloc_lock);
1654 }
1655
1656 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1657                                      struct inode *inode)
1658 {
1659         spin_lock(&root->delalloc_lock);
1660         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1661                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1662                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1663                           &BTRFS_I(inode)->runtime_flags);
1664                 root->nr_delalloc_inodes--;
1665                 if (!root->nr_delalloc_inodes) {
1666                         spin_lock(&root->fs_info->delalloc_root_lock);
1667                         BUG_ON(list_empty(&root->delalloc_root));
1668                         list_del_init(&root->delalloc_root);
1669                         spin_unlock(&root->fs_info->delalloc_root_lock);
1670                 }
1671         }
1672         spin_unlock(&root->delalloc_lock);
1673 }
1674
1675 /*
1676  * extent_io.c set_bit_hook, used to track delayed allocation
1677  * bytes in this file, and to maintain the list of inodes that
1678  * have pending delalloc work to be done.
1679  */
1680 static void btrfs_set_bit_hook(struct inode *inode,
1681                                struct extent_state *state, unsigned *bits)
1682 {
1683
1684         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1685                 WARN_ON(1);
1686         /*
1687          * set_bit and clear bit hooks normally require _irqsave/restore
1688          * but in this case, we are only testing for the DELALLOC
1689          * bit, which is only set or cleared with irqs on
1690          */
1691         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1692                 struct btrfs_root *root = BTRFS_I(inode)->root;
1693                 u64 len = state->end + 1 - state->start;
1694                 bool do_list = !btrfs_is_free_space_inode(inode);
1695
1696                 if (*bits & EXTENT_FIRST_DELALLOC) {
1697                         *bits &= ~EXTENT_FIRST_DELALLOC;
1698                 } else {
1699                         spin_lock(&BTRFS_I(inode)->lock);
1700                         BTRFS_I(inode)->outstanding_extents++;
1701                         spin_unlock(&BTRFS_I(inode)->lock);
1702                 }
1703
1704                 /* For sanity tests */
1705                 if (btrfs_test_is_dummy_root(root))
1706                         return;
1707
1708                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1709                                      root->fs_info->delalloc_batch);
1710                 spin_lock(&BTRFS_I(inode)->lock);
1711                 BTRFS_I(inode)->delalloc_bytes += len;
1712                 if (*bits & EXTENT_DEFRAG)
1713                         BTRFS_I(inode)->defrag_bytes += len;
1714                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1715                                          &BTRFS_I(inode)->runtime_flags))
1716                         btrfs_add_delalloc_inodes(root, inode);
1717                 spin_unlock(&BTRFS_I(inode)->lock);
1718         }
1719 }
1720
1721 /*
1722  * extent_io.c clear_bit_hook, see set_bit_hook for why
1723  */
1724 static void btrfs_clear_bit_hook(struct inode *inode,
1725                                  struct extent_state *state,
1726                                  unsigned *bits)
1727 {
1728         u64 len = state->end + 1 - state->start;
1729         u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1730                                     BTRFS_MAX_EXTENT_SIZE);
1731
1732         spin_lock(&BTRFS_I(inode)->lock);
1733         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1734                 BTRFS_I(inode)->defrag_bytes -= len;
1735         spin_unlock(&BTRFS_I(inode)->lock);
1736
1737         /*
1738          * set_bit and clear bit hooks normally require _irqsave/restore
1739          * but in this case, we are only testing for the DELALLOC
1740          * bit, which is only set or cleared with irqs on
1741          */
1742         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1743                 struct btrfs_root *root = BTRFS_I(inode)->root;
1744                 bool do_list = !btrfs_is_free_space_inode(inode);
1745
1746                 if (*bits & EXTENT_FIRST_DELALLOC) {
1747                         *bits &= ~EXTENT_FIRST_DELALLOC;
1748                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1749                         spin_lock(&BTRFS_I(inode)->lock);
1750                         BTRFS_I(inode)->outstanding_extents -= num_extents;
1751                         spin_unlock(&BTRFS_I(inode)->lock);
1752                 }
1753
1754                 /*
1755                  * We don't reserve metadata space for space cache inodes so we
1756                  * don't need to call dellalloc_release_metadata if there is an
1757                  * error.
1758                  */
1759                 if (*bits & EXTENT_DO_ACCOUNTING &&
1760                     root != root->fs_info->tree_root)
1761                         btrfs_delalloc_release_metadata(inode, len);
1762
1763                 /* For sanity tests. */
1764                 if (btrfs_test_is_dummy_root(root))
1765                         return;
1766
1767                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1768                     && do_list && !(state->state & EXTENT_NORESERVE))
1769                         btrfs_free_reserved_data_space(inode, len);
1770
1771                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1772                                      root->fs_info->delalloc_batch);
1773                 spin_lock(&BTRFS_I(inode)->lock);
1774                 BTRFS_I(inode)->delalloc_bytes -= len;
1775                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1776                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1777                              &BTRFS_I(inode)->runtime_flags))
1778                         btrfs_del_delalloc_inode(root, inode);
1779                 spin_unlock(&BTRFS_I(inode)->lock);
1780         }
1781 }
1782
1783 /*
1784  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1785  * we don't create bios that span stripes or chunks
1786  */
1787 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1788                          size_t size, struct bio *bio,
1789                          unsigned long bio_flags)
1790 {
1791         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1792         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1793         u64 length = 0;
1794         u64 map_length;
1795         int ret;
1796
1797         if (bio_flags & EXTENT_BIO_COMPRESSED)
1798                 return 0;
1799
1800         length = bio->bi_iter.bi_size;
1801         map_length = length;
1802         ret = btrfs_map_block(root->fs_info, rw, logical,
1803                               &map_length, NULL, 0);
1804         /* Will always return 0 with map_multi == NULL */
1805         BUG_ON(ret < 0);
1806         if (map_length < length + size)
1807                 return 1;
1808         return 0;
1809 }
1810
1811 /*
1812  * in order to insert checksums into the metadata in large chunks,
1813  * we wait until bio submission time.   All the pages in the bio are
1814  * checksummed and sums are attached onto the ordered extent record.
1815  *
1816  * At IO completion time the cums attached on the ordered extent record
1817  * are inserted into the btree
1818  */
1819 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1820                                     struct bio *bio, int mirror_num,
1821                                     unsigned long bio_flags,
1822                                     u64 bio_offset)
1823 {
1824         struct btrfs_root *root = BTRFS_I(inode)->root;
1825         int ret = 0;
1826
1827         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1828         BUG_ON(ret); /* -ENOMEM */
1829         return 0;
1830 }
1831
1832 /*
1833  * in order to insert checksums into the metadata in large chunks,
1834  * we wait until bio submission time.   All the pages in the bio are
1835  * checksummed and sums are attached onto the ordered extent record.
1836  *
1837  * At IO completion time the cums attached on the ordered extent record
1838  * are inserted into the btree
1839  */
1840 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1841                           int mirror_num, unsigned long bio_flags,
1842                           u64 bio_offset)
1843 {
1844         struct btrfs_root *root = BTRFS_I(inode)->root;
1845         int ret;
1846
1847         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1848         if (ret)
1849                 bio_endio(bio, ret);
1850         return ret;
1851 }
1852
1853 /*
1854  * extent_io.c submission hook. This does the right thing for csum calculation
1855  * on write, or reading the csums from the tree before a read
1856  */
1857 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1858                           int mirror_num, unsigned long bio_flags,
1859                           u64 bio_offset)
1860 {
1861         struct btrfs_root *root = BTRFS_I(inode)->root;
1862         int ret = 0;
1863         int skip_sum;
1864         int metadata = 0;
1865         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1866
1867         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1868
1869         if (btrfs_is_free_space_inode(inode))
1870                 metadata = 2;
1871
1872         if (!(rw & REQ_WRITE)) {
1873                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1874                 if (ret)
1875                         goto out;
1876
1877                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1878                         ret = btrfs_submit_compressed_read(inode, bio,
1879                                                            mirror_num,
1880                                                            bio_flags);
1881                         goto out;
1882                 } else if (!skip_sum) {
1883                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1884                         if (ret)
1885                                 goto out;
1886                 }
1887                 goto mapit;
1888         } else if (async && !skip_sum) {
1889                 /* csum items have already been cloned */
1890                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1891                         goto mapit;
1892                 /* we're doing a write, do the async checksumming */
1893                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1894                                    inode, rw, bio, mirror_num,
1895                                    bio_flags, bio_offset,
1896                                    __btrfs_submit_bio_start,
1897                                    __btrfs_submit_bio_done);
1898                 goto out;
1899         } else if (!skip_sum) {
1900                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1901                 if (ret)
1902                         goto out;
1903         }
1904
1905 mapit:
1906         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1907
1908 out:
1909         if (ret < 0)
1910                 bio_endio(bio, ret);
1911         return ret;
1912 }
1913
1914 /*
1915  * given a list of ordered sums record them in the inode.  This happens
1916  * at IO completion time based on sums calculated at bio submission time.
1917  */
1918 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1919                              struct inode *inode, u64 file_offset,
1920                              struct list_head *list)
1921 {
1922         struct btrfs_ordered_sum *sum;
1923
1924         list_for_each_entry(sum, list, list) {
1925                 trans->adding_csums = 1;
1926                 btrfs_csum_file_blocks(trans,
1927                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1928                 trans->adding_csums = 0;
1929         }
1930         return 0;
1931 }
1932
1933 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1934                               struct extent_state **cached_state)
1935 {
1936         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1937         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1938                                    cached_state, GFP_NOFS);
1939 }
1940
1941 /* see btrfs_writepage_start_hook for details on why this is required */
1942 struct btrfs_writepage_fixup {
1943         struct page *page;
1944         struct btrfs_work work;
1945 };
1946
1947 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1948 {
1949         struct btrfs_writepage_fixup *fixup;
1950         struct btrfs_ordered_extent *ordered;
1951         struct extent_state *cached_state = NULL;
1952         struct page *page;
1953         struct inode *inode;
1954         u64 page_start;
1955         u64 page_end;
1956         int ret;
1957
1958         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1959         page = fixup->page;
1960 again:
1961         lock_page(page);
1962         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1963                 ClearPageChecked(page);
1964                 goto out_page;
1965         }
1966
1967         inode = page->mapping->host;
1968         page_start = page_offset(page);
1969         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1970
1971         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1972                          &cached_state);
1973
1974         /* already ordered? We're done */
1975         if (PagePrivate2(page))
1976                 goto out;
1977
1978         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1979         if (ordered) {
1980                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1981                                      page_end, &cached_state, GFP_NOFS);
1982                 unlock_page(page);
1983                 btrfs_start_ordered_extent(inode, ordered, 1);
1984                 btrfs_put_ordered_extent(ordered);
1985                 goto again;
1986         }
1987
1988         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1989         if (ret) {
1990                 mapping_set_error(page->mapping, ret);
1991                 end_extent_writepage(page, ret, page_start, page_end);
1992                 ClearPageChecked(page);
1993                 goto out;
1994          }
1995
1996         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1997         ClearPageChecked(page);
1998         set_page_dirty(page);
1999 out:
2000         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2001                              &cached_state, GFP_NOFS);
2002 out_page:
2003         unlock_page(page);
2004         page_cache_release(page);
2005         kfree(fixup);
2006 }
2007
2008 /*
2009  * There are a few paths in the higher layers of the kernel that directly
2010  * set the page dirty bit without asking the filesystem if it is a
2011  * good idea.  This causes problems because we want to make sure COW
2012  * properly happens and the data=ordered rules are followed.
2013  *
2014  * In our case any range that doesn't have the ORDERED bit set
2015  * hasn't been properly setup for IO.  We kick off an async process
2016  * to fix it up.  The async helper will wait for ordered extents, set
2017  * the delalloc bit and make it safe to write the page.
2018  */
2019 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2020 {
2021         struct inode *inode = page->mapping->host;
2022         struct btrfs_writepage_fixup *fixup;
2023         struct btrfs_root *root = BTRFS_I(inode)->root;
2024
2025         /* this page is properly in the ordered list */
2026         if (TestClearPagePrivate2(page))
2027                 return 0;
2028
2029         if (PageChecked(page))
2030                 return -EAGAIN;
2031
2032         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2033         if (!fixup)
2034                 return -EAGAIN;
2035
2036         SetPageChecked(page);
2037         page_cache_get(page);
2038         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2039                         btrfs_writepage_fixup_worker, NULL, NULL);
2040         fixup->page = page;
2041         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2042         return -EBUSY;
2043 }
2044
2045 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2046                                        struct inode *inode, u64 file_pos,
2047                                        u64 disk_bytenr, u64 disk_num_bytes,
2048                                        u64 num_bytes, u64 ram_bytes,
2049                                        u8 compression, u8 encryption,
2050                                        u16 other_encoding, int extent_type)
2051 {
2052         struct btrfs_root *root = BTRFS_I(inode)->root;
2053         struct btrfs_file_extent_item *fi;
2054         struct btrfs_path *path;
2055         struct extent_buffer *leaf;
2056         struct btrfs_key ins;
2057         int extent_inserted = 0;
2058         int ret;
2059
2060         path = btrfs_alloc_path();
2061         if (!path)
2062                 return -ENOMEM;
2063
2064         /*
2065          * we may be replacing one extent in the tree with another.
2066          * The new extent is pinned in the extent map, and we don't want
2067          * to drop it from the cache until it is completely in the btree.
2068          *
2069          * So, tell btrfs_drop_extents to leave this extent in the cache.
2070          * the caller is expected to unpin it and allow it to be merged
2071          * with the others.
2072          */
2073         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2074                                    file_pos + num_bytes, NULL, 0,
2075                                    1, sizeof(*fi), &extent_inserted);
2076         if (ret)
2077                 goto out;
2078
2079         if (!extent_inserted) {
2080                 ins.objectid = btrfs_ino(inode);
2081                 ins.offset = file_pos;
2082                 ins.type = BTRFS_EXTENT_DATA_KEY;
2083
2084                 path->leave_spinning = 1;
2085                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2086                                               sizeof(*fi));
2087                 if (ret)
2088                         goto out;
2089         }
2090         leaf = path->nodes[0];
2091         fi = btrfs_item_ptr(leaf, path->slots[0],
2092                             struct btrfs_file_extent_item);
2093         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2094         btrfs_set_file_extent_type(leaf, fi, extent_type);
2095         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2096         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2097         btrfs_set_file_extent_offset(leaf, fi, 0);
2098         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2099         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2100         btrfs_set_file_extent_compression(leaf, fi, compression);
2101         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2102         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2103
2104         btrfs_mark_buffer_dirty(leaf);
2105         btrfs_release_path(path);
2106
2107         inode_add_bytes(inode, num_bytes);
2108
2109         ins.objectid = disk_bytenr;
2110         ins.offset = disk_num_bytes;
2111         ins.type = BTRFS_EXTENT_ITEM_KEY;
2112         ret = btrfs_alloc_reserved_file_extent(trans, root,
2113                                         root->root_key.objectid,
2114                                         btrfs_ino(inode), file_pos, &ins);
2115 out:
2116         btrfs_free_path(path);
2117
2118         return ret;
2119 }
2120
2121 /* snapshot-aware defrag */
2122 struct sa_defrag_extent_backref {
2123         struct rb_node node;
2124         struct old_sa_defrag_extent *old;
2125         u64 root_id;
2126         u64 inum;
2127         u64 file_pos;
2128         u64 extent_offset;
2129         u64 num_bytes;
2130         u64 generation;
2131 };
2132
2133 struct old_sa_defrag_extent {
2134         struct list_head list;
2135         struct new_sa_defrag_extent *new;
2136
2137         u64 extent_offset;
2138         u64 bytenr;
2139         u64 offset;
2140         u64 len;
2141         int count;
2142 };
2143
2144 struct new_sa_defrag_extent {
2145         struct rb_root root;
2146         struct list_head head;
2147         struct btrfs_path *path;
2148         struct inode *inode;
2149         u64 file_pos;
2150         u64 len;
2151         u64 bytenr;
2152         u64 disk_len;
2153         u8 compress_type;
2154 };
2155
2156 static int backref_comp(struct sa_defrag_extent_backref *b1,
2157                         struct sa_defrag_extent_backref *b2)
2158 {
2159         if (b1->root_id < b2->root_id)
2160                 return -1;
2161         else if (b1->root_id > b2->root_id)
2162                 return 1;
2163
2164         if (b1->inum < b2->inum)
2165                 return -1;
2166         else if (b1->inum > b2->inum)
2167                 return 1;
2168
2169         if (b1->file_pos < b2->file_pos)
2170                 return -1;
2171         else if (b1->file_pos > b2->file_pos)
2172                 return 1;
2173
2174         /*
2175          * [------------------------------] ===> (a range of space)
2176          *     |<--->|   |<---->| =============> (fs/file tree A)
2177          * |<---------------------------->| ===> (fs/file tree B)
2178          *
2179          * A range of space can refer to two file extents in one tree while
2180          * refer to only one file extent in another tree.
2181          *
2182          * So we may process a disk offset more than one time(two extents in A)
2183          * and locate at the same extent(one extent in B), then insert two same
2184          * backrefs(both refer to the extent in B).
2185          */
2186         return 0;
2187 }
2188
2189 static void backref_insert(struct rb_root *root,
2190                            struct sa_defrag_extent_backref *backref)
2191 {
2192         struct rb_node **p = &root->rb_node;
2193         struct rb_node *parent = NULL;
2194         struct sa_defrag_extent_backref *entry;
2195         int ret;
2196
2197         while (*p) {
2198                 parent = *p;
2199                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2200
2201                 ret = backref_comp(backref, entry);
2202                 if (ret < 0)
2203                         p = &(*p)->rb_left;
2204                 else
2205                         p = &(*p)->rb_right;
2206         }
2207
2208         rb_link_node(&backref->node, parent, p);
2209         rb_insert_color(&backref->node, root);
2210 }
2211
2212 /*
2213  * Note the backref might has changed, and in this case we just return 0.
2214  */
2215 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2216                                        void *ctx)
2217 {
2218         struct btrfs_file_extent_item *extent;
2219         struct btrfs_fs_info *fs_info;
2220         struct old_sa_defrag_extent *old = ctx;
2221         struct new_sa_defrag_extent *new = old->new;
2222         struct btrfs_path *path = new->path;
2223         struct btrfs_key key;
2224         struct btrfs_root *root;
2225         struct sa_defrag_extent_backref *backref;
2226         struct extent_buffer *leaf;
2227         struct inode *inode = new->inode;
2228         int slot;
2229         int ret;
2230         u64 extent_offset;
2231         u64 num_bytes;
2232
2233         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2234             inum == btrfs_ino(inode))
2235                 return 0;
2236
2237         key.objectid = root_id;
2238         key.type = BTRFS_ROOT_ITEM_KEY;
2239         key.offset = (u64)-1;
2240
2241         fs_info = BTRFS_I(inode)->root->fs_info;
2242         root = btrfs_read_fs_root_no_name(fs_info, &key);
2243         if (IS_ERR(root)) {
2244                 if (PTR_ERR(root) == -ENOENT)
2245                         return 0;
2246                 WARN_ON(1);
2247                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2248                          inum, offset, root_id);
2249                 return PTR_ERR(root);
2250         }
2251
2252         key.objectid = inum;
2253         key.type = BTRFS_EXTENT_DATA_KEY;
2254         if (offset > (u64)-1 << 32)
2255                 key.offset = 0;
2256         else
2257                 key.offset = offset;
2258
2259         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2260         if (WARN_ON(ret < 0))
2261                 return ret;
2262         ret = 0;
2263
2264         while (1) {
2265                 cond_resched();
2266
2267                 leaf = path->nodes[0];
2268                 slot = path->slots[0];
2269
2270                 if (slot >= btrfs_header_nritems(leaf)) {
2271                         ret = btrfs_next_leaf(root, path);
2272                         if (ret < 0) {
2273                                 goto out;
2274                         } else if (ret > 0) {
2275                                 ret = 0;
2276                                 goto out;
2277                         }
2278                         continue;
2279                 }
2280
2281                 path->slots[0]++;
2282
2283                 btrfs_item_key_to_cpu(leaf, &key, slot);
2284
2285                 if (key.objectid > inum)
2286                         goto out;
2287
2288                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2289                         continue;
2290
2291                 extent = btrfs_item_ptr(leaf, slot,
2292                                         struct btrfs_file_extent_item);
2293
2294                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2295                         continue;
2296
2297                 /*
2298                  * 'offset' refers to the exact key.offset,
2299                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2300                  * (key.offset - extent_offset).
2301                  */
2302                 if (key.offset != offset)
2303                         continue;
2304
2305                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2306                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2307
2308                 if (extent_offset >= old->extent_offset + old->offset +
2309                     old->len || extent_offset + num_bytes <=
2310                     old->extent_offset + old->offset)
2311                         continue;
2312                 break;
2313         }
2314
2315         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2316         if (!backref) {
2317                 ret = -ENOENT;
2318                 goto out;
2319         }
2320
2321         backref->root_id = root_id;
2322         backref->inum = inum;
2323         backref->file_pos = offset;
2324         backref->num_bytes = num_bytes;
2325         backref->extent_offset = extent_offset;
2326         backref->generation = btrfs_file_extent_generation(leaf, extent);
2327         backref->old = old;
2328         backref_insert(&new->root, backref);
2329         old->count++;
2330 out:
2331         btrfs_release_path(path);
2332         WARN_ON(ret);
2333         return ret;
2334 }
2335
2336 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2337                                    struct new_sa_defrag_extent *new)
2338 {
2339         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2340         struct old_sa_defrag_extent *old, *tmp;
2341         int ret;
2342
2343         new->path = path;
2344
2345         list_for_each_entry_safe(old, tmp, &new->head, list) {
2346                 ret = iterate_inodes_from_logical(old->bytenr +
2347                                                   old->extent_offset, fs_info,
2348                                                   path, record_one_backref,
2349                                                   old);
2350                 if (ret < 0 && ret != -ENOENT)
2351                         return false;
2352
2353                 /* no backref to be processed for this extent */
2354                 if (!old->count) {
2355                         list_del(&old->list);
2356                         kfree(old);
2357                 }
2358         }
2359
2360         if (list_empty(&new->head))
2361                 return false;
2362
2363         return true;
2364 }
2365
2366 static int relink_is_mergable(struct extent_buffer *leaf,
2367                               struct btrfs_file_extent_item *fi,
2368                               struct new_sa_defrag_extent *new)
2369 {
2370         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2371                 return 0;
2372
2373         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2374                 return 0;
2375
2376         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2377                 return 0;
2378
2379         if (btrfs_file_extent_encryption(leaf, fi) ||
2380             btrfs_file_extent_other_encoding(leaf, fi))
2381                 return 0;
2382
2383         return 1;
2384 }
2385
2386 /*
2387  * Note the backref might has changed, and in this case we just return 0.
2388  */
2389 static noinline int relink_extent_backref(struct btrfs_path *path,
2390                                  struct sa_defrag_extent_backref *prev,
2391                                  struct sa_defrag_extent_backref *backref)
2392 {
2393         struct btrfs_file_extent_item *extent;
2394         struct btrfs_file_extent_item *item;
2395         struct btrfs_ordered_extent *ordered;
2396         struct btrfs_trans_handle *trans;
2397         struct btrfs_fs_info *fs_info;
2398         struct btrfs_root *root;
2399         struct btrfs_key key;
2400         struct extent_buffer *leaf;
2401         struct old_sa_defrag_extent *old = backref->old;
2402         struct new_sa_defrag_extent *new = old->new;
2403         struct inode *src_inode = new->inode;
2404         struct inode *inode;
2405         struct extent_state *cached = NULL;
2406         int ret = 0;
2407         u64 start;
2408         u64 len;
2409         u64 lock_start;
2410         u64 lock_end;
2411         bool merge = false;
2412         int index;
2413
2414         if (prev && prev->root_id == backref->root_id &&
2415             prev->inum == backref->inum &&
2416             prev->file_pos + prev->num_bytes == backref->file_pos)
2417                 merge = true;
2418
2419         /* step 1: get root */
2420         key.objectid = backref->root_id;
2421         key.type = BTRFS_ROOT_ITEM_KEY;
2422         key.offset = (u64)-1;
2423
2424         fs_info = BTRFS_I(src_inode)->root->fs_info;
2425         index = srcu_read_lock(&fs_info->subvol_srcu);
2426
2427         root = btrfs_read_fs_root_no_name(fs_info, &key);
2428         if (IS_ERR(root)) {
2429                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2430                 if (PTR_ERR(root) == -ENOENT)
2431                         return 0;
2432                 return PTR_ERR(root);
2433         }
2434
2435         if (btrfs_root_readonly(root)) {
2436                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2437                 return 0;
2438         }
2439
2440         /* step 2: get inode */
2441         key.objectid = backref->inum;
2442         key.type = BTRFS_INODE_ITEM_KEY;
2443         key.offset = 0;
2444
2445         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2446         if (IS_ERR(inode)) {
2447                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2448                 return 0;
2449         }
2450
2451         srcu_read_unlock(&fs_info->subvol_srcu, index);
2452
2453         /* step 3: relink backref */
2454         lock_start = backref->file_pos;
2455         lock_end = backref->file_pos + backref->num_bytes - 1;
2456         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2457                          0, &cached);
2458
2459         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2460         if (ordered) {
2461                 btrfs_put_ordered_extent(ordered);
2462                 goto out_unlock;
2463         }
2464
2465         trans = btrfs_join_transaction(root);
2466         if (IS_ERR(trans)) {
2467                 ret = PTR_ERR(trans);
2468                 goto out_unlock;
2469         }
2470
2471         key.objectid = backref->inum;
2472         key.type = BTRFS_EXTENT_DATA_KEY;
2473         key.offset = backref->file_pos;
2474
2475         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2476         if (ret < 0) {
2477                 goto out_free_path;
2478         } else if (ret > 0) {
2479                 ret = 0;
2480                 goto out_free_path;
2481         }
2482
2483         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2484                                 struct btrfs_file_extent_item);
2485
2486         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2487             backref->generation)
2488                 goto out_free_path;
2489
2490         btrfs_release_path(path);
2491
2492         start = backref->file_pos;
2493         if (backref->extent_offset < old->extent_offset + old->offset)
2494                 start += old->extent_offset + old->offset -
2495                          backref->extent_offset;
2496
2497         len = min(backref->extent_offset + backref->num_bytes,
2498                   old->extent_offset + old->offset + old->len);
2499         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2500
2501         ret = btrfs_drop_extents(trans, root, inode, start,
2502                                  start + len, 1);
2503         if (ret)
2504                 goto out_free_path;
2505 again:
2506         key.objectid = btrfs_ino(inode);
2507         key.type = BTRFS_EXTENT_DATA_KEY;
2508         key.offset = start;
2509
2510         path->leave_spinning = 1;
2511         if (merge) {
2512                 struct btrfs_file_extent_item *fi;
2513                 u64 extent_len;
2514                 struct btrfs_key found_key;
2515
2516                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2517                 if (ret < 0)
2518                         goto out_free_path;
2519
2520                 path->slots[0]--;
2521                 leaf = path->nodes[0];
2522                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2523
2524                 fi = btrfs_item_ptr(leaf, path->slots[0],
2525                                     struct btrfs_file_extent_item);
2526                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2527
2528                 if (extent_len + found_key.offset == start &&
2529                     relink_is_mergable(leaf, fi, new)) {
2530                         btrfs_set_file_extent_num_bytes(leaf, fi,
2531                                                         extent_len + len);
2532                         btrfs_mark_buffer_dirty(leaf);
2533                         inode_add_bytes(inode, len);
2534
2535                         ret = 1;
2536                         goto out_free_path;
2537                 } else {
2538                         merge = false;
2539                         btrfs_release_path(path);
2540                         goto again;
2541                 }
2542         }
2543
2544         ret = btrfs_insert_empty_item(trans, root, path, &key,
2545                                         sizeof(*extent));
2546         if (ret) {
2547                 btrfs_abort_transaction(trans, root, ret);
2548                 goto out_free_path;
2549         }
2550
2551         leaf = path->nodes[0];
2552         item = btrfs_item_ptr(leaf, path->slots[0],
2553                                 struct btrfs_file_extent_item);
2554         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2555         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2556         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2557         btrfs_set_file_extent_num_bytes(leaf, item, len);
2558         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2559         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2560         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2561         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2562         btrfs_set_file_extent_encryption(leaf, item, 0);
2563         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2564
2565         btrfs_mark_buffer_dirty(leaf);
2566         inode_add_bytes(inode, len);
2567         btrfs_release_path(path);
2568
2569         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2570                         new->disk_len, 0,
2571                         backref->root_id, backref->inum,
2572                         new->file_pos, 0);      /* start - extent_offset */
2573         if (ret) {
2574                 btrfs_abort_transaction(trans, root, ret);
2575                 goto out_free_path;
2576         }
2577
2578         ret = 1;
2579 out_free_path:
2580         btrfs_release_path(path);
2581         path->leave_spinning = 0;
2582         btrfs_end_transaction(trans, root);
2583 out_unlock:
2584         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2585                              &cached, GFP_NOFS);
2586         iput(inode);
2587         return ret;
2588 }
2589
2590 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2591 {
2592         struct old_sa_defrag_extent *old, *tmp;
2593
2594         if (!new)
2595                 return;
2596
2597         list_for_each_entry_safe(old, tmp, &new->head, list) {
2598                 list_del(&old->list);
2599                 kfree(old);
2600         }
2601         kfree(new);
2602 }
2603
2604 static void relink_file_extents(struct new_sa_defrag_extent *new)
2605 {
2606         struct btrfs_path *path;
2607         struct sa_defrag_extent_backref *backref;
2608         struct sa_defrag_extent_backref *prev = NULL;
2609         struct inode *inode;
2610         struct btrfs_root *root;
2611         struct rb_node *node;
2612         int ret;
2613
2614         inode = new->inode;
2615         root = BTRFS_I(inode)->root;
2616
2617         path = btrfs_alloc_path();
2618         if (!path)
2619                 return;
2620
2621         if (!record_extent_backrefs(path, new)) {
2622                 btrfs_free_path(path);
2623                 goto out;
2624         }
2625         btrfs_release_path(path);
2626
2627         while (1) {
2628                 node = rb_first(&new->root);
2629                 if (!node)
2630                         break;
2631                 rb_erase(node, &new->root);
2632
2633                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2634
2635                 ret = relink_extent_backref(path, prev, backref);
2636                 WARN_ON(ret < 0);
2637
2638                 kfree(prev);
2639
2640                 if (ret == 1)
2641                         prev = backref;
2642                 else
2643                         prev = NULL;
2644                 cond_resched();
2645         }
2646         kfree(prev);
2647
2648         btrfs_free_path(path);
2649 out:
2650         free_sa_defrag_extent(new);
2651
2652         atomic_dec(&root->fs_info->defrag_running);
2653         wake_up(&root->fs_info->transaction_wait);
2654 }
2655
2656 static struct new_sa_defrag_extent *
2657 record_old_file_extents(struct inode *inode,
2658                         struct btrfs_ordered_extent *ordered)
2659 {
2660         struct btrfs_root *root = BTRFS_I(inode)->root;
2661         struct btrfs_path *path;
2662         struct btrfs_key key;
2663         struct old_sa_defrag_extent *old;
2664         struct new_sa_defrag_extent *new;
2665         int ret;
2666
2667         new = kmalloc(sizeof(*new), GFP_NOFS);
2668         if (!new)
2669                 return NULL;
2670
2671         new->inode = inode;
2672         new->file_pos = ordered->file_offset;
2673         new->len = ordered->len;
2674         new->bytenr = ordered->start;
2675         new->disk_len = ordered->disk_len;
2676         new->compress_type = ordered->compress_type;
2677         new->root = RB_ROOT;
2678         INIT_LIST_HEAD(&new->head);
2679
2680         path = btrfs_alloc_path();
2681         if (!path)
2682                 goto out_kfree;
2683
2684         key.objectid = btrfs_ino(inode);
2685         key.type = BTRFS_EXTENT_DATA_KEY;
2686         key.offset = new->file_pos;
2687
2688         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2689         if (ret < 0)
2690                 goto out_free_path;
2691         if (ret > 0 && path->slots[0] > 0)
2692                 path->slots[0]--;
2693
2694         /* find out all the old extents for the file range */
2695         while (1) {
2696                 struct btrfs_file_extent_item *extent;
2697                 struct extent_buffer *l;
2698                 int slot;
2699                 u64 num_bytes;
2700                 u64 offset;
2701                 u64 end;
2702                 u64 disk_bytenr;
2703                 u64 extent_offset;
2704
2705                 l = path->nodes[0];
2706                 slot = path->slots[0];
2707
2708                 if (slot >= btrfs_header_nritems(l)) {
2709                         ret = btrfs_next_leaf(root, path);
2710                         if (ret < 0)
2711                                 goto out_free_path;
2712                         else if (ret > 0)
2713                                 break;
2714                         continue;
2715                 }
2716
2717                 btrfs_item_key_to_cpu(l, &key, slot);
2718
2719                 if (key.objectid != btrfs_ino(inode))
2720                         break;
2721                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2722                         break;
2723                 if (key.offset >= new->file_pos + new->len)
2724                         break;
2725
2726                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2727
2728                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2729                 if (key.offset + num_bytes < new->file_pos)
2730                         goto next;
2731
2732                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2733                 if (!disk_bytenr)
2734                         goto next;
2735
2736                 extent_offset = btrfs_file_extent_offset(l, extent);
2737
2738                 old = kmalloc(sizeof(*old), GFP_NOFS);
2739                 if (!old)
2740                         goto out_free_path;
2741
2742                 offset = max(new->file_pos, key.offset);
2743                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2744
2745                 old->bytenr = disk_bytenr;
2746                 old->extent_offset = extent_offset;
2747                 old->offset = offset - key.offset;
2748                 old->len = end - offset;
2749                 old->new = new;
2750                 old->count = 0;
2751                 list_add_tail(&old->list, &new->head);
2752 next:
2753                 path->slots[0]++;
2754                 cond_resched();
2755         }
2756
2757         btrfs_free_path(path);
2758         atomic_inc(&root->fs_info->defrag_running);
2759
2760         return new;
2761
2762 out_free_path:
2763         btrfs_free_path(path);
2764 out_kfree:
2765         free_sa_defrag_extent(new);
2766         return NULL;
2767 }
2768
2769 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2770                                          u64 start, u64 len)
2771 {
2772         struct btrfs_block_group_cache *cache;
2773
2774         cache = btrfs_lookup_block_group(root->fs_info, start);
2775         ASSERT(cache);
2776
2777         spin_lock(&cache->lock);
2778         cache->delalloc_bytes -= len;
2779         spin_unlock(&cache->lock);
2780
2781         btrfs_put_block_group(cache);
2782 }
2783
2784 /* as ordered data IO finishes, this gets called so we can finish
2785  * an ordered extent if the range of bytes in the file it covers are
2786  * fully written.
2787  */
2788 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2789 {
2790         struct inode *inode = ordered_extent->inode;
2791         struct btrfs_root *root = BTRFS_I(inode)->root;
2792         struct btrfs_trans_handle *trans = NULL;
2793         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2794         struct extent_state *cached_state = NULL;
2795         struct new_sa_defrag_extent *new = NULL;
2796         int compress_type = 0;
2797         int ret = 0;
2798         u64 logical_len = ordered_extent->len;
2799         bool nolock;
2800         bool truncated = false;
2801
2802         nolock = btrfs_is_free_space_inode(inode);
2803
2804         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2805                 ret = -EIO;
2806                 goto out;
2807         }
2808
2809         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2810                                      ordered_extent->file_offset +
2811                                      ordered_extent->len - 1);
2812
2813         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2814                 truncated = true;
2815                 logical_len = ordered_extent->truncated_len;
2816                 /* Truncated the entire extent, don't bother adding */
2817                 if (!logical_len)
2818                         goto out;
2819         }
2820
2821         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2822                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2823                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2824                 if (nolock)
2825                         trans = btrfs_join_transaction_nolock(root);
2826                 else
2827                         trans = btrfs_join_transaction(root);
2828                 if (IS_ERR(trans)) {
2829                         ret = PTR_ERR(trans);
2830                         trans = NULL;
2831                         goto out;
2832                 }
2833                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2834                 ret = btrfs_update_inode_fallback(trans, root, inode);
2835                 if (ret) /* -ENOMEM or corruption */
2836                         btrfs_abort_transaction(trans, root, ret);
2837                 goto out;
2838         }
2839
2840         lock_extent_bits(io_tree, ordered_extent->file_offset,
2841                          ordered_extent->file_offset + ordered_extent->len - 1,
2842                          0, &cached_state);
2843
2844         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2845                         ordered_extent->file_offset + ordered_extent->len - 1,
2846                         EXTENT_DEFRAG, 1, cached_state);
2847         if (ret) {
2848                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2849                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2850                         /* the inode is shared */
2851                         new = record_old_file_extents(inode, ordered_extent);
2852
2853                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2854                         ordered_extent->file_offset + ordered_extent->len - 1,
2855                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2856         }
2857
2858         if (nolock)
2859                 trans = btrfs_join_transaction_nolock(root);
2860         else
2861                 trans = btrfs_join_transaction(root);
2862         if (IS_ERR(trans)) {
2863                 ret = PTR_ERR(trans);
2864                 trans = NULL;
2865                 goto out_unlock;
2866         }
2867
2868         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2869
2870         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2871                 compress_type = ordered_extent->compress_type;
2872         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2873                 BUG_ON(compress_type);
2874                 ret = btrfs_mark_extent_written(trans, inode,
2875                                                 ordered_extent->file_offset,
2876                                                 ordered_extent->file_offset +
2877                                                 logical_len);
2878         } else {
2879                 BUG_ON(root == root->fs_info->tree_root);
2880                 ret = insert_reserved_file_extent(trans, inode,
2881                                                 ordered_extent->file_offset,
2882                                                 ordered_extent->start,
2883                                                 ordered_extent->disk_len,
2884                                                 logical_len, logical_len,
2885                                                 compress_type, 0, 0,
2886                                                 BTRFS_FILE_EXTENT_REG);
2887                 if (!ret)
2888                         btrfs_release_delalloc_bytes(root,
2889                                                      ordered_extent->start,
2890                                                      ordered_extent->disk_len);
2891         }
2892         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2893                            ordered_extent->file_offset, ordered_extent->len,
2894                            trans->transid);
2895         if (ret < 0) {
2896                 btrfs_abort_transaction(trans, root, ret);
2897                 goto out_unlock;
2898         }
2899
2900         add_pending_csums(trans, inode, ordered_extent->file_offset,
2901                           &ordered_extent->list);
2902
2903         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2904         ret = btrfs_update_inode_fallback(trans, root, inode);
2905         if (ret) { /* -ENOMEM or corruption */
2906                 btrfs_abort_transaction(trans, root, ret);
2907                 goto out_unlock;
2908         }
2909         ret = 0;
2910 out_unlock:
2911         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2912                              ordered_extent->file_offset +
2913                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2914 out:
2915         if (root != root->fs_info->tree_root)
2916                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2917         if (trans)
2918                 btrfs_end_transaction(trans, root);
2919
2920         if (ret || truncated) {
2921                 u64 start, end;
2922
2923                 if (truncated)
2924                         start = ordered_extent->file_offset + logical_len;
2925                 else
2926                         start = ordered_extent->file_offset;
2927                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2928                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2929
2930                 /* Drop the cache for the part of the extent we didn't write. */
2931                 btrfs_drop_extent_cache(inode, start, end, 0);
2932
2933                 /*
2934                  * If the ordered extent had an IOERR or something else went
2935                  * wrong we need to return the space for this ordered extent
2936                  * back to the allocator.  We only free the extent in the
2937                  * truncated case if we didn't write out the extent at all.
2938                  */
2939                 if ((ret || !logical_len) &&
2940                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2941                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2942                         btrfs_free_reserved_extent(root, ordered_extent->start,
2943                                                    ordered_extent->disk_len, 1);
2944         }
2945
2946
2947         /*
2948          * This needs to be done to make sure anybody waiting knows we are done
2949          * updating everything for this ordered extent.
2950          */
2951         btrfs_remove_ordered_extent(inode, ordered_extent);
2952
2953         /* for snapshot-aware defrag */
2954         if (new) {
2955                 if (ret) {
2956                         free_sa_defrag_extent(new);
2957                         atomic_dec(&root->fs_info->defrag_running);
2958                 } else {
2959                         relink_file_extents(new);
2960                 }
2961         }
2962
2963         /* once for us */
2964         btrfs_put_ordered_extent(ordered_extent);
2965         /* once for the tree */
2966         btrfs_put_ordered_extent(ordered_extent);
2967
2968         return ret;
2969 }
2970
2971 static void finish_ordered_fn(struct btrfs_work *work)
2972 {
2973         struct btrfs_ordered_extent *ordered_extent;
2974         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2975         btrfs_finish_ordered_io(ordered_extent);
2976 }
2977
2978 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2979                                 struct extent_state *state, int uptodate)
2980 {
2981         struct inode *inode = page->mapping->host;
2982         struct btrfs_root *root = BTRFS_I(inode)->root;
2983         struct btrfs_ordered_extent *ordered_extent = NULL;
2984         struct btrfs_workqueue *wq;
2985         btrfs_work_func_t func;
2986
2987         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2988
2989         ClearPagePrivate2(page);
2990         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2991                                             end - start + 1, uptodate))
2992                 return 0;
2993
2994         if (btrfs_is_free_space_inode(inode)) {
2995                 wq = root->fs_info->endio_freespace_worker;
2996                 func = btrfs_freespace_write_helper;
2997         } else {
2998                 wq = root->fs_info->endio_write_workers;
2999                 func = btrfs_endio_write_helper;
3000         }
3001
3002         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3003                         NULL);
3004         btrfs_queue_work(wq, &ordered_extent->work);
3005
3006         return 0;
3007 }
3008
3009 static int __readpage_endio_check(struct inode *inode,
3010                                   struct btrfs_io_bio *io_bio,
3011                                   int icsum, struct page *page,
3012                                   int pgoff, u64 start, size_t len)
3013 {
3014         char *kaddr;
3015         u32 csum_expected;
3016         u32 csum = ~(u32)0;
3017         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3018                                       DEFAULT_RATELIMIT_BURST);
3019
3020         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3021
3022         kaddr = kmap_atomic(page);
3023         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3024         btrfs_csum_final(csum, (char *)&csum);
3025         if (csum != csum_expected)
3026                 goto zeroit;
3027
3028         kunmap_atomic(kaddr);
3029         return 0;
3030 zeroit:
3031         if (__ratelimit(&_rs))
3032                 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3033                            "csum failed ino %llu off %llu csum %u expected csum %u",
3034                            btrfs_ino(inode), start, csum, csum_expected);
3035         memset(kaddr + pgoff, 1, len);
3036         flush_dcache_page(page);
3037         kunmap_atomic(kaddr);
3038         if (csum_expected == 0)
3039                 return 0;
3040         return -EIO;
3041 }
3042
3043 /*
3044  * when reads are done, we need to check csums to verify the data is correct
3045  * if there's a match, we allow the bio to finish.  If not, the code in
3046  * extent_io.c will try to find good copies for us.
3047  */
3048 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3049                                       u64 phy_offset, struct page *page,
3050                                       u64 start, u64 end, int mirror)
3051 {
3052         size_t offset = start - page_offset(page);
3053         struct inode *inode = page->mapping->host;
3054         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3055         struct btrfs_root *root = BTRFS_I(inode)->root;
3056
3057         if (PageChecked(page)) {
3058                 ClearPageChecked(page);
3059                 return 0;
3060         }
3061
3062         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3063                 return 0;
3064
3065         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3066             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3067                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3068                                   GFP_NOFS);
3069                 return 0;
3070         }
3071
3072         phy_offset >>= inode->i_sb->s_blocksize_bits;
3073         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3074                                       start, (size_t)(end - start + 1));
3075 }
3076
3077 struct delayed_iput {
3078         struct list_head list;
3079         struct inode *inode;
3080 };
3081
3082 /* JDM: If this is fs-wide, why can't we add a pointer to
3083  * btrfs_inode instead and avoid the allocation? */
3084 void btrfs_add_delayed_iput(struct inode *inode)
3085 {
3086         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3087         struct delayed_iput *delayed;
3088
3089         if (atomic_add_unless(&inode->i_count, -1, 1))
3090                 return;
3091
3092         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3093         delayed->inode = inode;
3094
3095         spin_lock(&fs_info->delayed_iput_lock);
3096         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3097         spin_unlock(&fs_info->delayed_iput_lock);
3098 }
3099
3100 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3101 {
3102         LIST_HEAD(list);
3103         struct btrfs_fs_info *fs_info = root->fs_info;
3104         struct delayed_iput *delayed;
3105         int empty;
3106
3107         spin_lock(&fs_info->delayed_iput_lock);
3108         empty = list_empty(&fs_info->delayed_iputs);
3109         spin_unlock(&fs_info->delayed_iput_lock);
3110         if (empty)
3111                 return;
3112
3113         down_read(&fs_info->delayed_iput_sem);
3114
3115         spin_lock(&fs_info->delayed_iput_lock);
3116         list_splice_init(&fs_info->delayed_iputs, &list);
3117         spin_unlock(&fs_info->delayed_iput_lock);
3118
3119         while (!list_empty(&list)) {
3120                 delayed = list_entry(list.next, struct delayed_iput, list);
3121                 list_del(&delayed->list);
3122                 iput(delayed->inode);
3123                 kfree(delayed);
3124         }
3125
3126         up_read(&root->fs_info->delayed_iput_sem);
3127 }
3128
3129 /*
3130  * This is called in transaction commit time. If there are no orphan
3131  * files in the subvolume, it removes orphan item and frees block_rsv
3132  * structure.
3133  */
3134 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3135                               struct btrfs_root *root)
3136 {
3137         struct btrfs_block_rsv *block_rsv;
3138         int ret;
3139
3140         if (atomic_read(&root->orphan_inodes) ||
3141             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3142                 return;
3143
3144         spin_lock(&root->orphan_lock);
3145         if (atomic_read(&root->orphan_inodes)) {
3146                 spin_unlock(&root->orphan_lock);
3147                 return;
3148         }
3149
3150         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3151                 spin_unlock(&root->orphan_lock);
3152                 return;
3153         }
3154
3155         block_rsv = root->orphan_block_rsv;
3156         root->orphan_block_rsv = NULL;
3157         spin_unlock(&root->orphan_lock);
3158
3159         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3160             btrfs_root_refs(&root->root_item) > 0) {
3161                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3162                                             root->root_key.objectid);
3163                 if (ret)
3164                         btrfs_abort_transaction(trans, root, ret);
3165                 else
3166                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3167                                   &root->state);
3168         }
3169
3170         if (block_rsv) {
3171                 WARN_ON(block_rsv->size > 0);
3172                 btrfs_free_block_rsv(root, block_rsv);
3173         }
3174 }
3175
3176 /*
3177  * This creates an orphan entry for the given inode in case something goes
3178  * wrong in the middle of an unlink/truncate.
3179  *
3180  * NOTE: caller of this function should reserve 5 units of metadata for
3181  *       this function.
3182  */
3183 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3184 {
3185         struct btrfs_root *root = BTRFS_I(inode)->root;
3186         struct btrfs_block_rsv *block_rsv = NULL;
3187         int reserve = 0;
3188         int insert = 0;
3189         int ret;
3190
3191         if (!root->orphan_block_rsv) {
3192                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3193                 if (!block_rsv)
3194                         return -ENOMEM;
3195         }
3196
3197         spin_lock(&root->orphan_lock);
3198         if (!root->orphan_block_rsv) {
3199                 root->orphan_block_rsv = block_rsv;
3200         } else if (block_rsv) {
3201                 btrfs_free_block_rsv(root, block_rsv);
3202                 block_rsv = NULL;
3203         }
3204
3205         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3206                               &BTRFS_I(inode)->runtime_flags)) {
3207 #if 0
3208                 /*
3209                  * For proper ENOSPC handling, we should do orphan
3210                  * cleanup when mounting. But this introduces backward
3211                  * compatibility issue.
3212                  */
3213                 if (!xchg(&root->orphan_item_inserted, 1))
3214                         insert = 2;
3215                 else
3216                         insert = 1;
3217 #endif
3218                 insert = 1;
3219                 atomic_inc(&root->orphan_inodes);
3220         }
3221
3222         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3223                               &BTRFS_I(inode)->runtime_flags))
3224                 reserve = 1;
3225         spin_unlock(&root->orphan_lock);
3226
3227         /* grab metadata reservation from transaction handle */
3228         if (reserve) {
3229                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3230                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3231         }
3232
3233         /* insert an orphan item to track this unlinked/truncated file */
3234         if (insert >= 1) {
3235                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3236                 if (ret) {
3237                         atomic_dec(&root->orphan_inodes);
3238                         if (reserve) {
3239                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3240                                           &BTRFS_I(inode)->runtime_flags);
3241                                 btrfs_orphan_release_metadata(inode);
3242                         }
3243                         if (ret != -EEXIST) {
3244                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3245                                           &BTRFS_I(inode)->runtime_flags);
3246                                 btrfs_abort_transaction(trans, root, ret);
3247                                 return ret;
3248                         }
3249                 }
3250                 ret = 0;
3251         }
3252
3253         /* insert an orphan item to track subvolume contains orphan files */
3254         if (insert >= 2) {
3255                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3256                                                root->root_key.objectid);
3257                 if (ret && ret != -EEXIST) {
3258                         btrfs_abort_transaction(trans, root, ret);
3259                         return ret;
3260                 }
3261         }
3262         return 0;
3263 }
3264
3265 /*
3266  * We have done the truncate/delete so we can go ahead and remove the orphan
3267  * item for this particular inode.
3268  */
3269 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3270                             struct inode *inode)
3271 {
3272         struct btrfs_root *root = BTRFS_I(inode)->root;
3273         int delete_item = 0;
3274         int release_rsv = 0;
3275         int ret = 0;
3276
3277         spin_lock(&root->orphan_lock);
3278         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3279                                &BTRFS_I(inode)->runtime_flags))
3280                 delete_item = 1;
3281
3282         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3283                                &BTRFS_I(inode)->runtime_flags))
3284                 release_rsv = 1;
3285         spin_unlock(&root->orphan_lock);
3286
3287         if (delete_item) {
3288                 atomic_dec(&root->orphan_inodes);
3289                 if (trans)
3290                         ret = btrfs_del_orphan_item(trans, root,
3291                                                     btrfs_ino(inode));
3292         }
3293
3294         if (release_rsv)
3295                 btrfs_orphan_release_metadata(inode);
3296
3297         return ret;
3298 }
3299
3300 /*
3301  * this cleans up any orphans that may be left on the list from the last use
3302  * of this root.
3303  */
3304 int btrfs_orphan_cleanup(struct btrfs_root *root)
3305 {
3306         struct btrfs_path *path;
3307         struct extent_buffer *leaf;
3308         struct btrfs_key key, found_key;
3309         struct btrfs_trans_handle *trans;
3310         struct inode *inode;
3311         u64 last_objectid = 0;
3312         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3313
3314         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3315                 return 0;
3316
3317         path = btrfs_alloc_path();
3318         if (!path) {
3319                 ret = -ENOMEM;
3320                 goto out;
3321         }
3322         path->reada = -1;
3323
3324         key.objectid = BTRFS_ORPHAN_OBJECTID;
3325         key.type = BTRFS_ORPHAN_ITEM_KEY;
3326         key.offset = (u64)-1;
3327
3328         while (1) {
3329                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3330                 if (ret < 0)
3331                         goto out;
3332
3333                 /*
3334                  * if ret == 0 means we found what we were searching for, which
3335                  * is weird, but possible, so only screw with path if we didn't
3336                  * find the key and see if we have stuff that matches
3337                  */
3338                 if (ret > 0) {
3339                         ret = 0;
3340                         if (path->slots[0] == 0)
3341                                 break;
3342                         path->slots[0]--;
3343                 }
3344
3345                 /* pull out the item */
3346                 leaf = path->nodes[0];
3347                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3348
3349                 /* make sure the item matches what we want */
3350                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3351                         break;
3352                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3353                         break;
3354
3355                 /* release the path since we're done with it */
3356                 btrfs_release_path(path);
3357
3358                 /*
3359                  * this is where we are basically btrfs_lookup, without the
3360                  * crossing root thing.  we store the inode number in the
3361                  * offset of the orphan item.
3362                  */
3363
3364                 if (found_key.offset == last_objectid) {
3365                         btrfs_err(root->fs_info,
3366                                 "Error removing orphan entry, stopping orphan cleanup");
3367                         ret = -EINVAL;
3368                         goto out;
3369                 }
3370
3371                 last_objectid = found_key.offset;
3372
3373                 found_key.objectid = found_key.offset;
3374                 found_key.type = BTRFS_INODE_ITEM_KEY;
3375                 found_key.offset = 0;
3376                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3377                 ret = PTR_ERR_OR_ZERO(inode);
3378                 if (ret && ret != -ESTALE)
3379                         goto out;
3380
3381                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3382                         struct btrfs_root *dead_root;
3383                         struct btrfs_fs_info *fs_info = root->fs_info;
3384                         int is_dead_root = 0;
3385
3386                         /*
3387                          * this is an orphan in the tree root. Currently these
3388                          * could come from 2 sources:
3389                          *  a) a snapshot deletion in progress
3390                          *  b) a free space cache inode
3391                          * We need to distinguish those two, as the snapshot
3392                          * orphan must not get deleted.
3393                          * find_dead_roots already ran before us, so if this
3394                          * is a snapshot deletion, we should find the root
3395                          * in the dead_roots list
3396                          */
3397                         spin_lock(&fs_info->trans_lock);
3398                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3399                                             root_list) {
3400                                 if (dead_root->root_key.objectid ==
3401                                     found_key.objectid) {
3402                                         is_dead_root = 1;
3403                                         break;
3404                                 }
3405                         }
3406                         spin_unlock(&fs_info->trans_lock);
3407                         if (is_dead_root) {
3408                                 /* prevent this orphan from being found again */
3409                                 key.offset = found_key.objectid - 1;
3410                                 continue;
3411                         }
3412                 }
3413                 /*
3414                  * Inode is already gone but the orphan item is still there,
3415                  * kill the orphan item.
3416                  */
3417                 if (ret == -ESTALE) {
3418                         trans = btrfs_start_transaction(root, 1);
3419                         if (IS_ERR(trans)) {
3420                                 ret = PTR_ERR(trans);
3421                                 goto out;
3422                         }
3423                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3424                                 found_key.objectid);
3425                         ret = btrfs_del_orphan_item(trans, root,
3426                                                     found_key.objectid);
3427                         btrfs_end_transaction(trans, root);
3428                         if (ret)
3429                                 goto out;
3430                         continue;
3431                 }
3432
3433                 /*
3434                  * add this inode to the orphan list so btrfs_orphan_del does
3435                  * the proper thing when we hit it
3436                  */
3437                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3438                         &BTRFS_I(inode)->runtime_flags);
3439                 atomic_inc(&root->orphan_inodes);
3440
3441                 /* if we have links, this was a truncate, lets do that */
3442                 if (inode->i_nlink) {
3443                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3444                                 iput(inode);
3445                                 continue;
3446                         }
3447                         nr_truncate++;
3448
3449                         /* 1 for the orphan item deletion. */
3450                         trans = btrfs_start_transaction(root, 1);
3451                         if (IS_ERR(trans)) {
3452                                 iput(inode);
3453                                 ret = PTR_ERR(trans);
3454                                 goto out;
3455                         }
3456                         ret = btrfs_orphan_add(trans, inode);
3457                         btrfs_end_transaction(trans, root);
3458                         if (ret) {
3459                                 iput(inode);
3460                                 goto out;
3461                         }
3462
3463                         ret = btrfs_truncate(inode);
3464                         if (ret)
3465                                 btrfs_orphan_del(NULL, inode);
3466                 } else {
3467                         nr_unlink++;
3468                 }
3469
3470                 /* this will do delete_inode and everything for us */
3471                 iput(inode);
3472                 if (ret)
3473                         goto out;
3474         }
3475         /* release the path since we're done with it */
3476         btrfs_release_path(path);
3477
3478         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3479
3480         if (root->orphan_block_rsv)
3481                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3482                                         (u64)-1);
3483
3484         if (root->orphan_block_rsv ||
3485             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3486                 trans = btrfs_join_transaction(root);
3487                 if (!IS_ERR(trans))
3488                         btrfs_end_transaction(trans, root);
3489         }
3490
3491         if (nr_unlink)
3492                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3493         if (nr_truncate)
3494                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3495
3496 out:
3497         if (ret)
3498                 btrfs_err(root->fs_info,
3499                         "could not do orphan cleanup %d", ret);
3500         btrfs_free_path(path);
3501         return ret;
3502 }
3503
3504 /*
3505  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3506  * don't find any xattrs, we know there can't be any acls.
3507  *
3508  * slot is the slot the inode is in, objectid is the objectid of the inode
3509  */
3510 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3511                                           int slot, u64 objectid,
3512                                           int *first_xattr_slot)
3513 {
3514         u32 nritems = btrfs_header_nritems(leaf);
3515         struct btrfs_key found_key;
3516         static u64 xattr_access = 0;
3517         static u64 xattr_default = 0;
3518         int scanned = 0;
3519
3520         if (!xattr_access) {
3521                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3522                                         strlen(POSIX_ACL_XATTR_ACCESS));
3523                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3524                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3525         }
3526
3527         slot++;
3528         *first_xattr_slot = -1;
3529         while (slot < nritems) {
3530                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3531
3532                 /* we found a different objectid, there must not be acls */
3533                 if (found_key.objectid != objectid)
3534                         return 0;
3535
3536                 /* we found an xattr, assume we've got an acl */
3537                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3538                         if (*first_xattr_slot == -1)
3539                                 *first_xattr_slot = slot;
3540                         if (found_key.offset == xattr_access ||
3541                             found_key.offset == xattr_default)
3542                                 return 1;
3543                 }
3544
3545                 /*
3546                  * we found a key greater than an xattr key, there can't
3547                  * be any acls later on
3548                  */
3549                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3550                         return 0;
3551
3552                 slot++;
3553                 scanned++;
3554
3555                 /*
3556                  * it goes inode, inode backrefs, xattrs, extents,
3557                  * so if there are a ton of hard links to an inode there can
3558                  * be a lot of backrefs.  Don't waste time searching too hard,
3559                  * this is just an optimization
3560                  */
3561                 if (scanned >= 8)
3562                         break;
3563         }
3564         /* we hit the end of the leaf before we found an xattr or
3565          * something larger than an xattr.  We have to assume the inode
3566          * has acls
3567          */
3568         if (*first_xattr_slot == -1)
3569                 *first_xattr_slot = slot;
3570         return 1;
3571 }
3572
3573 /*
3574  * read an inode from the btree into the in-memory inode
3575  */
3576 static void btrfs_read_locked_inode(struct inode *inode)
3577 {
3578         struct btrfs_path *path;
3579         struct extent_buffer *leaf;
3580         struct btrfs_inode_item *inode_item;
3581         struct btrfs_root *root = BTRFS_I(inode)->root;
3582         struct btrfs_key location;
3583         unsigned long ptr;
3584         int maybe_acls;
3585         u32 rdev;
3586         int ret;
3587         bool filled = false;
3588         int first_xattr_slot;
3589
3590         ret = btrfs_fill_inode(inode, &rdev);
3591         if (!ret)
3592                 filled = true;
3593
3594         path = btrfs_alloc_path();
3595         if (!path)
3596                 goto make_bad;
3597
3598         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3599
3600         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3601         if (ret)
3602                 goto make_bad;
3603
3604         leaf = path->nodes[0];
3605
3606         if (filled)
3607                 goto cache_index;
3608
3609         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3610                                     struct btrfs_inode_item);
3611         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3612         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3613         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3614         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3615         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3616
3617         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3619
3620         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3622
3623         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3625
3626         BTRFS_I(inode)->i_otime.tv_sec =
3627                 btrfs_timespec_sec(leaf, &inode_item->otime);
3628         BTRFS_I(inode)->i_otime.tv_nsec =
3629                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3630
3631         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3634
3635         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3636         inode->i_generation = BTRFS_I(inode)->generation;
3637         inode->i_rdev = 0;
3638         rdev = btrfs_inode_rdev(leaf, inode_item);
3639
3640         BTRFS_I(inode)->index_cnt = (u64)-1;
3641         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3642
3643 cache_index:
3644         /*
3645          * If we were modified in the current generation and evicted from memory
3646          * and then re-read we need to do a full sync since we don't have any
3647          * idea about which extents were modified before we were evicted from
3648          * cache.
3649          *
3650          * This is required for both inode re-read from disk and delayed inode
3651          * in delayed_nodes_tree.
3652          */
3653         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3654                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3655                         &BTRFS_I(inode)->runtime_flags);
3656
3657         path->slots[0]++;
3658         if (inode->i_nlink != 1 ||
3659             path->slots[0] >= btrfs_header_nritems(leaf))
3660                 goto cache_acl;
3661
3662         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3663         if (location.objectid != btrfs_ino(inode))
3664                 goto cache_acl;
3665
3666         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3667         if (location.type == BTRFS_INODE_REF_KEY) {
3668                 struct btrfs_inode_ref *ref;
3669
3670                 ref = (struct btrfs_inode_ref *)ptr;
3671                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3672         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3673                 struct btrfs_inode_extref *extref;
3674
3675                 extref = (struct btrfs_inode_extref *)ptr;
3676                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3677                                                                      extref);
3678         }
3679 cache_acl:
3680         /*
3681          * try to precache a NULL acl entry for files that don't have
3682          * any xattrs or acls
3683          */
3684         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3685                                            btrfs_ino(inode), &first_xattr_slot);
3686         if (first_xattr_slot != -1) {
3687                 path->slots[0] = first_xattr_slot;
3688                 ret = btrfs_load_inode_props(inode, path);
3689                 if (ret)
3690                         btrfs_err(root->fs_info,
3691                                   "error loading props for ino %llu (root %llu): %d",
3692                                   btrfs_ino(inode),
3693                                   root->root_key.objectid, ret);
3694         }
3695         btrfs_free_path(path);
3696
3697         if (!maybe_acls)
3698                 cache_no_acl(inode);
3699
3700         switch (inode->i_mode & S_IFMT) {
3701         case S_IFREG:
3702                 inode->i_mapping->a_ops = &btrfs_aops;
3703                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3704                 inode->i_fop = &btrfs_file_operations;
3705                 inode->i_op = &btrfs_file_inode_operations;
3706                 break;
3707         case S_IFDIR:
3708                 inode->i_fop = &btrfs_dir_file_operations;
3709                 if (root == root->fs_info->tree_root)
3710                         inode->i_op = &btrfs_dir_ro_inode_operations;
3711                 else
3712                         inode->i_op = &btrfs_dir_inode_operations;
3713                 break;
3714         case S_IFLNK:
3715                 inode->i_op = &btrfs_symlink_inode_operations;
3716                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3717                 break;
3718         default:
3719                 inode->i_op = &btrfs_special_inode_operations;
3720                 init_special_inode(inode, inode->i_mode, rdev);
3721                 break;
3722         }
3723
3724         btrfs_update_iflags(inode);
3725         return;
3726
3727 make_bad:
3728         btrfs_free_path(path);
3729         make_bad_inode(inode);
3730 }
3731
3732 /*
3733  * given a leaf and an inode, copy the inode fields into the leaf
3734  */
3735 static void fill_inode_item(struct btrfs_trans_handle *trans,
3736                             struct extent_buffer *leaf,
3737                             struct btrfs_inode_item *item,
3738                             struct inode *inode)
3739 {
3740         struct btrfs_map_token token;
3741
3742         btrfs_init_map_token(&token);
3743
3744         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3745         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3746         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3747                                    &token);
3748         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3749         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3750
3751         btrfs_set_token_timespec_sec(leaf, &item->atime,
3752                                      inode->i_atime.tv_sec, &token);
3753         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3754                                       inode->i_atime.tv_nsec, &token);
3755
3756         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3757                                      inode->i_mtime.tv_sec, &token);
3758         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3759                                       inode->i_mtime.tv_nsec, &token);
3760
3761         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3762                                      inode->i_ctime.tv_sec, &token);
3763         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3764                                       inode->i_ctime.tv_nsec, &token);
3765
3766         btrfs_set_token_timespec_sec(leaf, &item->otime,
3767                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3768         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3769                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3770
3771         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3772                                      &token);
3773         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3774                                          &token);
3775         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3776         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3777         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3778         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3779         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3780 }
3781
3782 /*
3783  * copy everything in the in-memory inode into the btree.
3784  */
3785 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3786                                 struct btrfs_root *root, struct inode *inode)
3787 {
3788         struct btrfs_inode_item *inode_item;
3789         struct btrfs_path *path;
3790         struct extent_buffer *leaf;
3791         int ret;
3792
3793         path = btrfs_alloc_path();
3794         if (!path)
3795                 return -ENOMEM;
3796
3797         path->leave_spinning = 1;
3798         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3799                                  1);
3800         if (ret) {
3801                 if (ret > 0)
3802                         ret = -ENOENT;
3803                 goto failed;
3804         }
3805
3806         leaf = path->nodes[0];
3807         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3808                                     struct btrfs_inode_item);
3809
3810         fill_inode_item(trans, leaf, inode_item, inode);
3811         btrfs_mark_buffer_dirty(leaf);
3812         btrfs_set_inode_last_trans(trans, inode);
3813         ret = 0;
3814 failed:
3815         btrfs_free_path(path);
3816         return ret;
3817 }
3818
3819 /*
3820  * copy everything in the in-memory inode into the btree.
3821  */
3822 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3823                                 struct btrfs_root *root, struct inode *inode)
3824 {
3825         int ret;
3826
3827         /*
3828          * If the inode is a free space inode, we can deadlock during commit
3829          * if we put it into the delayed code.
3830          *
3831          * The data relocation inode should also be directly updated
3832          * without delay
3833          */
3834         if (!btrfs_is_free_space_inode(inode)
3835             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3836             && !root->fs_info->log_root_recovering) {
3837                 btrfs_update_root_times(trans, root);
3838
3839                 ret = btrfs_delayed_update_inode(trans, root, inode);
3840                 if (!ret)
3841                         btrfs_set_inode_last_trans(trans, inode);
3842                 return ret;
3843         }
3844
3845         return btrfs_update_inode_item(trans, root, inode);
3846 }
3847
3848 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3849                                          struct btrfs_root *root,
3850                                          struct inode *inode)
3851 {
3852         int ret;
3853
3854         ret = btrfs_update_inode(trans, root, inode);
3855         if (ret == -ENOSPC)
3856                 return btrfs_update_inode_item(trans, root, inode);
3857         return ret;
3858 }
3859
3860 /*
3861  * unlink helper that gets used here in inode.c and in the tree logging
3862  * recovery code.  It remove a link in a directory with a given name, and
3863  * also drops the back refs in the inode to the directory
3864  */
3865 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3866                                 struct btrfs_root *root,
3867                                 struct inode *dir, struct inode *inode,
3868                                 const char *name, int name_len)
3869 {
3870         struct btrfs_path *path;
3871         int ret = 0;
3872         struct extent_buffer *leaf;
3873         struct btrfs_dir_item *di;
3874         struct btrfs_key key;
3875         u64 index;
3876         u64 ino = btrfs_ino(inode);
3877         u64 dir_ino = btrfs_ino(dir);
3878
3879         path = btrfs_alloc_path();
3880         if (!path) {
3881                 ret = -ENOMEM;
3882                 goto out;
3883         }
3884
3885         path->leave_spinning = 1;
3886         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3887                                     name, name_len, -1);
3888         if (IS_ERR(di)) {
3889                 ret = PTR_ERR(di);
3890                 goto err;
3891         }
3892         if (!di) {
3893                 ret = -ENOENT;
3894                 goto err;
3895         }
3896         leaf = path->nodes[0];
3897         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3898         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3899         if (ret)
3900                 goto err;
3901         btrfs_release_path(path);
3902
3903         /*
3904          * If we don't have dir index, we have to get it by looking up
3905          * the inode ref, since we get the inode ref, remove it directly,
3906          * it is unnecessary to do delayed deletion.
3907          *
3908          * But if we have dir index, needn't search inode ref to get it.
3909          * Since the inode ref is close to the inode item, it is better
3910          * that we delay to delete it, and just do this deletion when
3911          * we update the inode item.
3912          */
3913         if (BTRFS_I(inode)->dir_index) {
3914                 ret = btrfs_delayed_delete_inode_ref(inode);
3915                 if (!ret) {
3916                         index = BTRFS_I(inode)->dir_index;
3917                         goto skip_backref;
3918                 }
3919         }
3920
3921         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3922                                   dir_ino, &index);
3923         if (ret) {
3924                 btrfs_info(root->fs_info,
3925                         "failed to delete reference to %.*s, inode %llu parent %llu",
3926                         name_len, name, ino, dir_ino);
3927                 btrfs_abort_transaction(trans, root, ret);
3928                 goto err;
3929         }
3930 skip_backref:
3931         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3932         if (ret) {
3933                 btrfs_abort_transaction(trans, root, ret);
3934                 goto err;
3935         }
3936
3937         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3938                                          inode, dir_ino);
3939         if (ret != 0 && ret != -ENOENT) {
3940                 btrfs_abort_transaction(trans, root, ret);
3941                 goto err;
3942         }
3943
3944         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3945                                            dir, index);
3946         if (ret == -ENOENT)
3947                 ret = 0;
3948         else if (ret)
3949                 btrfs_abort_transaction(trans, root, ret);
3950 err:
3951         btrfs_free_path(path);
3952         if (ret)
3953                 goto out;
3954
3955         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3956         inode_inc_iversion(inode);
3957         inode_inc_iversion(dir);
3958         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3959         ret = btrfs_update_inode(trans, root, dir);
3960 out:
3961         return ret;
3962 }
3963
3964 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3965                        struct btrfs_root *root,
3966                        struct inode *dir, struct inode *inode,
3967                        const char *name, int name_len)
3968 {
3969         int ret;
3970         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3971         if (!ret) {
3972                 drop_nlink(inode);
3973                 ret = btrfs_update_inode(trans, root, inode);
3974         }
3975         return ret;
3976 }
3977
3978 /*
3979  * helper to start transaction for unlink and rmdir.
3980  *
3981  * unlink and rmdir are special in btrfs, they do not always free space, so
3982  * if we cannot make our reservations the normal way try and see if there is
3983  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3984  * allow the unlink to occur.
3985  */
3986 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3987 {
3988         struct btrfs_trans_handle *trans;
3989         struct btrfs_root *root = BTRFS_I(dir)->root;
3990         int ret;
3991
3992         /*
3993          * 1 for the possible orphan item
3994          * 1 for the dir item
3995          * 1 for the dir index
3996          * 1 for the inode ref
3997          * 1 for the inode
3998          */
3999         trans = btrfs_start_transaction(root, 5);
4000         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
4001                 return trans;
4002
4003         if (PTR_ERR(trans) == -ENOSPC) {
4004                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4005
4006                 trans = btrfs_start_transaction(root, 0);
4007                 if (IS_ERR(trans))
4008                         return trans;
4009                 ret = btrfs_cond_migrate_bytes(root->fs_info,
4010                                                &root->fs_info->trans_block_rsv,
4011                                                num_bytes, 5);
4012                 if (ret) {
4013                         btrfs_end_transaction(trans, root);
4014                         return ERR_PTR(ret);
4015                 }
4016                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4017                 trans->bytes_reserved = num_bytes;
4018         }
4019         return trans;
4020 }
4021
4022 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4023 {
4024         struct btrfs_root *root = BTRFS_I(dir)->root;
4025         struct btrfs_trans_handle *trans;
4026         struct inode *inode = d_inode(dentry);
4027         int ret;
4028
4029         trans = __unlink_start_trans(dir);
4030         if (IS_ERR(trans))
4031                 return PTR_ERR(trans);
4032
4033         btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4034
4035         ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4036                                  dentry->d_name.name, dentry->d_name.len);
4037         if (ret)
4038                 goto out;
4039
4040         if (inode->i_nlink == 0) {
4041                 ret = btrfs_orphan_add(trans, inode);
4042                 if (ret)
4043                         goto out;
4044         }
4045
4046 out:
4047         btrfs_end_transaction(trans, root);
4048         btrfs_btree_balance_dirty(root);
4049         return ret;
4050 }
4051
4052 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4053                         struct btrfs_root *root,
4054                         struct inode *dir, u64 objectid,
4055                         const char *name, int name_len)
4056 {
4057         struct btrfs_path *path;
4058         struct extent_buffer *leaf;
4059         struct btrfs_dir_item *di;
4060         struct btrfs_key key;
4061         u64 index;
4062         int ret;
4063         u64 dir_ino = btrfs_ino(dir);
4064
4065         path = btrfs_alloc_path();
4066         if (!path)
4067                 return -ENOMEM;
4068
4069         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4070                                    name, name_len, -1);
4071         if (IS_ERR_OR_NULL(di)) {
4072                 if (!di)
4073                         ret = -ENOENT;
4074                 else
4075                         ret = PTR_ERR(di);
4076                 goto out;
4077         }
4078
4079         leaf = path->nodes[0];
4080         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4081         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4082         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4083         if (ret) {
4084                 btrfs_abort_transaction(trans, root, ret);
4085                 goto out;
4086         }
4087         btrfs_release_path(path);
4088
4089         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4090                                  objectid, root->root_key.objectid,
4091                                  dir_ino, &index, name, name_len);
4092         if (ret < 0) {
4093                 if (ret != -ENOENT) {
4094                         btrfs_abort_transaction(trans, root, ret);
4095                         goto out;
4096                 }
4097                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4098                                                  name, name_len);
4099                 if (IS_ERR_OR_NULL(di)) {
4100                         if (!di)
4101                                 ret = -ENOENT;
4102                         else
4103                                 ret = PTR_ERR(di);
4104                         btrfs_abort_transaction(trans, root, ret);
4105                         goto out;
4106                 }
4107
4108                 leaf = path->nodes[0];
4109                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4110                 btrfs_release_path(path);
4111                 index = key.offset;
4112         }
4113         btrfs_release_path(path);
4114
4115         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4116         if (ret) {
4117                 btrfs_abort_transaction(trans, root, ret);
4118                 goto out;
4119         }
4120
4121         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4122         inode_inc_iversion(dir);
4123         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4124         ret = btrfs_update_inode_fallback(trans, root, dir);
4125         if (ret)
4126                 btrfs_abort_transaction(trans, root, ret);
4127 out:
4128         btrfs_free_path(path);
4129         return ret;
4130 }
4131
4132 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4133 {
4134         struct inode *inode = d_inode(dentry);
4135         int err = 0;
4136         struct btrfs_root *root = BTRFS_I(dir)->root;
4137         struct btrfs_trans_handle *trans;
4138
4139         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4140                 return -ENOTEMPTY;
4141         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4142                 return -EPERM;
4143
4144         trans = __unlink_start_trans(dir);
4145         if (IS_ERR(trans))
4146                 return PTR_ERR(trans);
4147
4148         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4149                 err = btrfs_unlink_subvol(trans, root, dir,
4150                                           BTRFS_I(inode)->location.objectid,
4151                                           dentry->d_name.name,
4152                                           dentry->d_name.len);
4153                 goto out;
4154         }
4155
4156         err = btrfs_orphan_add(trans, inode);
4157         if (err)
4158                 goto out;
4159
4160         /* now the directory is empty */
4161         err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4162                                  dentry->d_name.name, dentry->d_name.len);
4163         if (!err)
4164                 btrfs_i_size_write(inode, 0);
4165 out:
4166         btrfs_end_transaction(trans, root);
4167         btrfs_btree_balance_dirty(root);
4168
4169         return err;
4170 }
4171
4172 static int truncate_space_check(struct btrfs_trans_handle *trans,
4173                                 struct btrfs_root *root,
4174                                 u64 bytes_deleted)
4175 {
4176         int ret;
4177
4178         bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4179         ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4180                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4181         if (!ret)
4182                 trans->bytes_reserved += bytes_deleted;
4183         return ret;
4184
4185 }
4186
4187 /*
4188  * this can truncate away extent items, csum items and directory items.
4189  * It starts at a high offset and removes keys until it can't find
4190  * any higher than new_size
4191  *
4192  * csum items that cross the new i_size are truncated to the new size
4193  * as well.
4194  *
4195  * min_type is the minimum key type to truncate down to.  If set to 0, this
4196  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4197  */
4198 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4199                                struct btrfs_root *root,
4200                                struct inode *inode,
4201                                u64 new_size, u32 min_type)
4202 {
4203         struct btrfs_path *path;
4204         struct extent_buffer *leaf;
4205         struct btrfs_file_extent_item *fi;
4206         struct btrfs_key key;
4207         struct btrfs_key found_key;
4208         u64 extent_start = 0;
4209         u64 extent_num_bytes = 0;
4210         u64 extent_offset = 0;
4211         u64 item_end = 0;
4212         u64 last_size = (u64)-1;
4213         u32 found_type = (u8)-1;
4214         int found_extent;
4215         int del_item;
4216         int pending_del_nr = 0;
4217         int pending_del_slot = 0;
4218         int extent_type = -1;
4219         int ret;
4220         int err = 0;
4221         u64 ino = btrfs_ino(inode);
4222         u64 bytes_deleted = 0;
4223         bool be_nice = 0;
4224         bool should_throttle = 0;
4225         bool should_end = 0;
4226
4227         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4228
4229         /*
4230          * for non-free space inodes and ref cows, we want to back off from
4231          * time to time
4232          */
4233         if (!btrfs_is_free_space_inode(inode) &&
4234             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4235                 be_nice = 1;
4236
4237         path = btrfs_alloc_path();
4238         if (!path)
4239                 return -ENOMEM;
4240         path->reada = -1;
4241
4242         /*
4243          * We want to drop from the next block forward in case this new size is
4244          * not block aligned since we will be keeping the last block of the
4245          * extent just the way it is.
4246          */
4247         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4248             root == root->fs_info->tree_root)
4249                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4250                                         root->sectorsize), (u64)-1, 0);
4251
4252         /*
4253          * This function is also used to drop the items in the log tree before
4254          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4255          * it is used to drop the loged items. So we shouldn't kill the delayed
4256          * items.
4257          */
4258         if (min_type == 0 && root == BTRFS_I(inode)->root)
4259                 btrfs_kill_delayed_inode_items(inode);
4260
4261         key.objectid = ino;
4262         key.offset = (u64)-1;
4263         key.type = (u8)-1;
4264
4265 search_again:
4266         /*
4267          * with a 16K leaf size and 128MB extents, you can actually queue
4268          * up a huge file in a single leaf.  Most of the time that
4269          * bytes_deleted is > 0, it will be huge by the time we get here
4270          */
4271         if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4272                 if (btrfs_should_end_transaction(trans, root)) {
4273                         err = -EAGAIN;
4274                         goto error;
4275                 }
4276         }
4277
4278
4279         path->leave_spinning = 1;
4280         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4281         if (ret < 0) {
4282                 err = ret;
4283                 goto out;
4284         }
4285
4286         if (ret > 0) {
4287                 /* there are no items in the tree for us to truncate, we're
4288                  * done
4289                  */
4290                 if (path->slots[0] == 0)
4291                         goto out;
4292                 path->slots[0]--;
4293         }
4294
4295         while (1) {
4296                 fi = NULL;
4297                 leaf = path->nodes[0];
4298                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4299                 found_type = found_key.type;
4300
4301                 if (found_key.objectid != ino)
4302                         break;
4303
4304                 if (found_type < min_type)
4305                         break;
4306
4307                 item_end = found_key.offset;
4308                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4309                         fi = btrfs_item_ptr(leaf, path->slots[0],
4310                                             struct btrfs_file_extent_item);
4311                         extent_type = btrfs_file_extent_type(leaf, fi);
4312                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4313                                 item_end +=
4314                                     btrfs_file_extent_num_bytes(leaf, fi);
4315                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4316                                 item_end += btrfs_file_extent_inline_len(leaf,
4317                                                          path->slots[0], fi);
4318                         }
4319                         item_end--;
4320                 }
4321                 if (found_type > min_type) {
4322                         del_item = 1;
4323                 } else {
4324                         if (item_end < new_size)
4325                                 break;
4326                         if (found_key.offset >= new_size)
4327                                 del_item = 1;
4328                         else
4329                                 del_item = 0;
4330                 }
4331                 found_extent = 0;
4332                 /* FIXME, shrink the extent if the ref count is only 1 */
4333                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4334                         goto delete;
4335
4336                 if (del_item)
4337                         last_size = found_key.offset;
4338                 else
4339                         last_size = new_size;
4340
4341                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4342                         u64 num_dec;
4343                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4344                         if (!del_item) {
4345                                 u64 orig_num_bytes =
4346                                         btrfs_file_extent_num_bytes(leaf, fi);
4347                                 extent_num_bytes = ALIGN(new_size -
4348                                                 found_key.offset,
4349                                                 root->sectorsize);
4350                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4351                                                          extent_num_bytes);
4352                                 num_dec = (orig_num_bytes -
4353                                            extent_num_bytes);
4354                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4355                                              &root->state) &&
4356                                     extent_start != 0)
4357                                         inode_sub_bytes(inode, num_dec);
4358                                 btrfs_mark_buffer_dirty(leaf);
4359                         } else {
4360                                 extent_num_bytes =
4361                                         btrfs_file_extent_disk_num_bytes(leaf,
4362                                                                          fi);
4363                                 extent_offset = found_key.offset -
4364                                         btrfs_file_extent_offset(leaf, fi);
4365
4366                                 /* FIXME blocksize != 4096 */
4367                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4368                                 if (extent_start != 0) {
4369                                         found_extent = 1;
4370                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4371                                                      &root->state))
4372                                                 inode_sub_bytes(inode, num_dec);
4373                                 }
4374                         }
4375                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4376                         /*
4377                          * we can't truncate inline items that have had
4378                          * special encodings
4379                          */
4380                         if (!del_item &&
4381                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4382                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4383                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4384                                 u32 size = new_size - found_key.offset;
4385
4386                                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4387                                         inode_sub_bytes(inode, item_end + 1 -
4388                                                         new_size);
4389
4390                                 /*
4391                                  * update the ram bytes to properly reflect
4392                                  * the new size of our item
4393                                  */
4394                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4395                                 size =
4396                                     btrfs_file_extent_calc_inline_size(size);
4397                                 btrfs_truncate_item(root, path, size, 1);
4398                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4399                                             &root->state)) {
4400                                 inode_sub_bytes(inode, item_end + 1 -
4401                                                 found_key.offset);
4402                         }
4403                 }
4404 delete:
4405                 if (del_item) {
4406                         if (!pending_del_nr) {
4407                                 /* no pending yet, add ourselves */
4408                                 pending_del_slot = path->slots[0];
4409                                 pending_del_nr = 1;
4410                         } else if (pending_del_nr &&
4411                                    path->slots[0] + 1 == pending_del_slot) {
4412                                 /* hop on the pending chunk */
4413                                 pending_del_nr++;
4414                                 pending_del_slot = path->slots[0];
4415                         } else {
4416                                 BUG();
4417                         }
4418                 } else {
4419                         break;
4420                 }
4421                 should_throttle = 0;
4422
4423                 if (found_extent &&
4424                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4425                      root == root->fs_info->tree_root)) {
4426                         btrfs_set_path_blocking(path);
4427                         bytes_deleted += extent_num_bytes;
4428                         ret = btrfs_free_extent(trans, root, extent_start,
4429                                                 extent_num_bytes, 0,
4430                                                 btrfs_header_owner(leaf),
4431                                                 ino, extent_offset, 0);
4432                         BUG_ON(ret);
4433                         if (btrfs_should_throttle_delayed_refs(trans, root))
4434                                 btrfs_async_run_delayed_refs(root,
4435                                         trans->delayed_ref_updates * 2, 0);
4436                         if (be_nice) {
4437                                 if (truncate_space_check(trans, root,
4438                                                          extent_num_bytes)) {
4439                                         should_end = 1;
4440                                 }
4441                                 if (btrfs_should_throttle_delayed_refs(trans,
4442                                                                        root)) {
4443                                         should_throttle = 1;
4444                                 }
4445                         }
4446                 }
4447
4448                 if (found_type == BTRFS_INODE_ITEM_KEY)
4449                         break;
4450
4451                 if (path->slots[0] == 0 ||
4452                     path->slots[0] != pending_del_slot ||
4453                     should_throttle || should_end) {
4454                         if (pending_del_nr) {
4455                                 ret = btrfs_del_items(trans, root, path,
4456                                                 pending_del_slot,
4457                                                 pending_del_nr);
4458                                 if (ret) {
4459                                         btrfs_abort_transaction(trans,
4460                                                                 root, ret);
4461                                         goto error;
4462                                 }
4463                                 pending_del_nr = 0;
4464                         }
4465                         btrfs_release_path(path);
4466                         if (should_throttle) {
4467                                 unsigned long updates = trans->delayed_ref_updates;
4468                                 if (updates) {
4469                                         trans->delayed_ref_updates = 0;
4470                                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4471                                         if (ret && !err)
4472                                                 err = ret;
4473                                 }
4474                         }
4475                         /*
4476                          * if we failed to refill our space rsv, bail out
4477                          * and let the transaction restart
4478                          */
4479                         if (should_end) {
4480                                 err = -EAGAIN;
4481                                 goto error;
4482                         }
4483                         goto search_again;
4484                 } else {
4485                         path->slots[0]--;
4486                 }
4487         }
4488 out:
4489         if (pending_del_nr) {
4490                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4491                                       pending_del_nr);
4492                 if (ret)
4493                         btrfs_abort_transaction(trans, root, ret);
4494         }
4495 error:
4496         if (last_size != (u64)-1 &&
4497             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4498                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4499
4500         btrfs_free_path(path);
4501
4502         if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4503                 unsigned long updates = trans->delayed_ref_updates;
4504                 if (updates) {
4505                         trans->delayed_ref_updates = 0;
4506                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4507                         if (ret && !err)
4508                                 err = ret;
4509                 }
4510         }
4511         return err;
4512 }
4513
4514 /*
4515  * btrfs_truncate_page - read, zero a chunk and write a page
4516  * @inode - inode that we're zeroing
4517  * @from - the offset to start zeroing
4518  * @len - the length to zero, 0 to zero the entire range respective to the
4519  *      offset
4520  * @front - zero up to the offset instead of from the offset on
4521  *
4522  * This will find the page for the "from" offset and cow the page and zero the
4523  * part we want to zero.  This is used with truncate and hole punching.
4524  */
4525 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4526                         int front)
4527 {
4528         struct address_space *mapping = inode->i_mapping;
4529         struct btrfs_root *root = BTRFS_I(inode)->root;
4530         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4531         struct btrfs_ordered_extent *ordered;
4532         struct extent_state *cached_state = NULL;
4533         char *kaddr;
4534         u32 blocksize = root->sectorsize;
4535         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4536         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4537         struct page *page;
4538         gfp_t mask = btrfs_alloc_write_mask(mapping);
4539         int ret = 0;
4540         u64 page_start;
4541         u64 page_end;
4542
4543         if ((offset & (blocksize - 1)) == 0 &&
4544             (!len || ((len & (blocksize - 1)) == 0)))
4545                 goto out;
4546         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4547         if (ret)
4548                 goto out;
4549
4550 again:
4551         page = find_or_create_page(mapping, index, mask);
4552         if (!page) {
4553                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4554                 ret = -ENOMEM;
4555                 goto out;
4556         }
4557
4558         page_start = page_offset(page);
4559         page_end = page_start + PAGE_CACHE_SIZE - 1;
4560
4561         if (!PageUptodate(page)) {
4562                 ret = btrfs_readpage(NULL, page);
4563                 lock_page(page);
4564                 if (page->mapping != mapping) {
4565                         unlock_page(page);
4566                         page_cache_release(page);
4567                         goto again;
4568                 }
4569                 if (!PageUptodate(page)) {
4570                         ret = -EIO;
4571                         goto out_unlock;
4572                 }
4573         }
4574         wait_on_page_writeback(page);
4575
4576         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4577         set_page_extent_mapped(page);
4578
4579         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4580         if (ordered) {
4581                 unlock_extent_cached(io_tree, page_start, page_end,
4582                                      &cached_state, GFP_NOFS);
4583                 unlock_page(page);
4584                 page_cache_release(page);
4585                 btrfs_start_ordered_extent(inode, ordered, 1);
4586                 btrfs_put_ordered_extent(ordered);
4587                 goto again;
4588         }
4589
4590         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4591                           EXTENT_DIRTY | EXTENT_DELALLOC |
4592                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4593                           0, 0, &cached_state, GFP_NOFS);
4594
4595         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4596                                         &cached_state);
4597         if (ret) {
4598                 unlock_extent_cached(io_tree, page_start, page_end,
4599                                      &cached_state, GFP_NOFS);
4600                 goto out_unlock;
4601         }
4602
4603         if (offset != PAGE_CACHE_SIZE) {
4604                 if (!len)
4605                         len = PAGE_CACHE_SIZE - offset;
4606                 kaddr = kmap(page);
4607                 if (front)
4608                         memset(kaddr, 0, offset);
4609                 else
4610                         memset(kaddr + offset, 0, len);
4611                 flush_dcache_page(page);
4612                 kunmap(page);
4613         }
4614         ClearPageChecked(page);
4615         set_page_dirty(page);
4616         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4617                              GFP_NOFS);
4618
4619 out_unlock:
4620         if (ret)
4621                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4622         unlock_page(page);
4623         page_cache_release(page);
4624 out:
4625         return ret;
4626 }
4627
4628 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4629                              u64 offset, u64 len)
4630 {
4631         struct btrfs_trans_handle *trans;
4632         int ret;
4633
4634         /*
4635          * Still need to make sure the inode looks like it's been updated so
4636          * that any holes get logged if we fsync.
4637          */
4638         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4639                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4640                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4641                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4642                 return 0;
4643         }
4644
4645         /*
4646          * 1 - for the one we're dropping
4647          * 1 - for the one we're adding
4648          * 1 - for updating the inode.
4649          */
4650         trans = btrfs_start_transaction(root, 3);
4651         if (IS_ERR(trans))
4652                 return PTR_ERR(trans);
4653
4654         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4655         if (ret) {
4656                 btrfs_abort_transaction(trans, root, ret);
4657                 btrfs_end_transaction(trans, root);
4658                 return ret;
4659         }
4660
4661         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4662                                        0, 0, len, 0, len, 0, 0, 0);
4663         if (ret)
4664                 btrfs_abort_transaction(trans, root, ret);
4665         else
4666                 btrfs_update_inode(trans, root, inode);
4667         btrfs_end_transaction(trans, root);
4668         return ret;
4669 }
4670
4671 /*
4672  * This function puts in dummy file extents for the area we're creating a hole
4673  * for.  So if we are truncating this file to a larger size we need to insert
4674  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4675  * the range between oldsize and size
4676  */
4677 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4678 {
4679         struct btrfs_root *root = BTRFS_I(inode)->root;
4680         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4681         struct extent_map *em = NULL;
4682         struct extent_state *cached_state = NULL;
4683         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4684         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4685         u64 block_end = ALIGN(size, root->sectorsize);
4686         u64 last_byte;
4687         u64 cur_offset;
4688         u64 hole_size;
4689         int err = 0;
4690
4691         /*
4692          * If our size started in the middle of a page we need to zero out the
4693          * rest of the page before we expand the i_size, otherwise we could
4694          * expose stale data.
4695          */
4696         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4697         if (err)
4698                 return err;
4699
4700         if (size <= hole_start)
4701                 return 0;
4702
4703         while (1) {
4704                 struct btrfs_ordered_extent *ordered;
4705
4706                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4707                                  &cached_state);
4708                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4709                                                      block_end - hole_start);
4710                 if (!ordered)
4711                         break;
4712                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4713                                      &cached_state, GFP_NOFS);
4714                 btrfs_start_ordered_extent(inode, ordered, 1);
4715                 btrfs_put_ordered_extent(ordered);
4716         }
4717
4718         cur_offset = hole_start;
4719         while (1) {
4720                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4721                                 block_end - cur_offset, 0);
4722                 if (IS_ERR(em)) {
4723                         err = PTR_ERR(em);
4724                         em = NULL;
4725                         break;
4726                 }
4727                 last_byte = min(extent_map_end(em), block_end);
4728                 last_byte = ALIGN(last_byte , root->sectorsize);
4729                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4730                         struct extent_map *hole_em;
4731                         hole_size = last_byte - cur_offset;
4732
4733                         err = maybe_insert_hole(root, inode, cur_offset,
4734                                                 hole_size);
4735                         if (err)
4736                                 break;
4737                         btrfs_drop_extent_cache(inode, cur_offset,
4738                                                 cur_offset + hole_size - 1, 0);
4739                         hole_em = alloc_extent_map();
4740                         if (!hole_em) {
4741                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4742                                         &BTRFS_I(inode)->runtime_flags);
4743                                 goto next;
4744                         }
4745                         hole_em->start = cur_offset;
4746                         hole_em->len = hole_size;
4747                         hole_em->orig_start = cur_offset;
4748
4749                         hole_em->block_start = EXTENT_MAP_HOLE;
4750                         hole_em->block_len = 0;
4751                         hole_em->orig_block_len = 0;
4752                         hole_em->ram_bytes = hole_size;
4753                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4754                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4755                         hole_em->generation = root->fs_info->generation;
4756
4757                         while (1) {
4758                                 write_lock(&em_tree->lock);
4759                                 err = add_extent_mapping(em_tree, hole_em, 1);
4760                                 write_unlock(&em_tree->lock);
4761                                 if (err != -EEXIST)
4762                                         break;
4763                                 btrfs_drop_extent_cache(inode, cur_offset,
4764                                                         cur_offset +
4765                                                         hole_size - 1, 0);
4766                         }
4767                         free_extent_map(hole_em);
4768                 }
4769 next:
4770                 free_extent_map(em);
4771                 em = NULL;
4772                 cur_offset = last_byte;
4773                 if (cur_offset >= block_end)
4774                         break;
4775         }
4776         free_extent_map(em);
4777         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4778                              GFP_NOFS);
4779         return err;
4780 }
4781
4782 static int wait_snapshoting_atomic_t(atomic_t *a)
4783 {
4784         schedule();
4785         return 0;
4786 }
4787
4788 static void wait_for_snapshot_creation(struct btrfs_root *root)
4789 {
4790         while (true) {
4791                 int ret;
4792
4793                 ret = btrfs_start_write_no_snapshoting(root);
4794                 if (ret)
4795                         break;
4796                 wait_on_atomic_t(&root->will_be_snapshoted,
4797                                  wait_snapshoting_atomic_t,
4798                                  TASK_UNINTERRUPTIBLE);
4799         }
4800 }
4801
4802 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4803 {
4804         struct btrfs_root *root = BTRFS_I(inode)->root;
4805         struct btrfs_trans_handle *trans;
4806         loff_t oldsize = i_size_read(inode);
4807         loff_t newsize = attr->ia_size;
4808         int mask = attr->ia_valid;
4809         int ret;
4810
4811         /*
4812          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4813          * special case where we need to update the times despite not having
4814          * these flags set.  For all other operations the VFS set these flags
4815          * explicitly if it wants a timestamp update.
4816          */
4817         if (newsize != oldsize) {
4818                 inode_inc_iversion(inode);
4819                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4820                         inode->i_ctime = inode->i_mtime =
4821                                 current_fs_time(inode->i_sb);
4822         }
4823
4824         if (newsize > oldsize) {
4825                 truncate_pagecache(inode, newsize);
4826                 /*
4827                  * Don't do an expanding truncate while snapshoting is ongoing.
4828                  * This is to ensure the snapshot captures a fully consistent
4829                  * state of this file - if the snapshot captures this expanding
4830                  * truncation, it must capture all writes that happened before
4831                  * this truncation.
4832                  */
4833                 wait_for_snapshot_creation(root);
4834                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4835                 if (ret) {
4836                         btrfs_end_write_no_snapshoting(root);
4837                         return ret;
4838                 }
4839
4840                 trans = btrfs_start_transaction(root, 1);
4841                 if (IS_ERR(trans)) {
4842                         btrfs_end_write_no_snapshoting(root);
4843                         return PTR_ERR(trans);
4844                 }
4845
4846                 i_size_write(inode, newsize);
4847                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4848                 ret = btrfs_update_inode(trans, root, inode);
4849                 btrfs_end_write_no_snapshoting(root);
4850                 btrfs_end_transaction(trans, root);
4851         } else {
4852
4853                 /*
4854                  * We're truncating a file that used to have good data down to
4855                  * zero. Make sure it gets into the ordered flush list so that
4856                  * any new writes get down to disk quickly.
4857                  */
4858                 if (newsize == 0)
4859                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4860                                 &BTRFS_I(inode)->runtime_flags);
4861
4862                 /*
4863                  * 1 for the orphan item we're going to add
4864                  * 1 for the orphan item deletion.
4865                  */
4866                 trans = btrfs_start_transaction(root, 2);
4867                 if (IS_ERR(trans))
4868                         return PTR_ERR(trans);
4869
4870                 /*
4871                  * We need to do this in case we fail at _any_ point during the
4872                  * actual truncate.  Once we do the truncate_setsize we could
4873                  * invalidate pages which forces any outstanding ordered io to
4874                  * be instantly completed which will give us extents that need
4875                  * to be truncated.  If we fail to get an orphan inode down we
4876                  * could have left over extents that were never meant to live,
4877                  * so we need to garuntee from this point on that everything
4878                  * will be consistent.
4879                  */
4880                 ret = btrfs_orphan_add(trans, inode);
4881                 btrfs_end_transaction(trans, root);
4882                 if (ret)
4883                         return ret;
4884
4885                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4886                 truncate_setsize(inode, newsize);
4887
4888                 /* Disable nonlocked read DIO to avoid the end less truncate */
4889                 btrfs_inode_block_unlocked_dio(inode);
4890                 inode_dio_wait(inode);
4891                 btrfs_inode_resume_unlocked_dio(inode);
4892
4893                 ret = btrfs_truncate(inode);
4894                 if (ret && inode->i_nlink) {
4895                         int err;
4896
4897                         /*
4898                          * failed to truncate, disk_i_size is only adjusted down
4899                          * as we remove extents, so it should represent the true
4900                          * size of the inode, so reset the in memory size and
4901                          * delete our orphan entry.
4902                          */
4903                         trans = btrfs_join_transaction(root);
4904                         if (IS_ERR(trans)) {
4905                                 btrfs_orphan_del(NULL, inode);
4906                                 return ret;
4907                         }
4908                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4909                         err = btrfs_orphan_del(trans, inode);
4910                         if (err)
4911                                 btrfs_abort_transaction(trans, root, err);
4912                         btrfs_end_transaction(trans, root);
4913                 }
4914         }
4915
4916         return ret;
4917 }
4918
4919 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4920 {
4921         struct inode *inode = d_inode(dentry);
4922         struct btrfs_root *root = BTRFS_I(inode)->root;
4923         int err;
4924
4925         if (btrfs_root_readonly(root))
4926                 return -EROFS;
4927
4928         err = inode_change_ok(inode, attr);
4929         if (err)
4930                 return err;
4931
4932         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4933                 err = btrfs_setsize(inode, attr);
4934                 if (err)
4935                         return err;
4936         }
4937
4938         if (attr->ia_valid) {
4939                 setattr_copy(inode, attr);
4940                 inode_inc_iversion(inode);
4941                 err = btrfs_dirty_inode(inode);
4942
4943                 if (!err && attr->ia_valid & ATTR_MODE)
4944                         err = posix_acl_chmod(inode, inode->i_mode);
4945         }
4946
4947         return err;
4948 }
4949
4950 /*
4951  * While truncating the inode pages during eviction, we get the VFS calling
4952  * btrfs_invalidatepage() against each page of the inode. This is slow because
4953  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4954  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4955  * extent_state structures over and over, wasting lots of time.
4956  *
4957  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4958  * those expensive operations on a per page basis and do only the ordered io
4959  * finishing, while we release here the extent_map and extent_state structures,
4960  * without the excessive merging and splitting.
4961  */
4962 static void evict_inode_truncate_pages(struct inode *inode)
4963 {
4964         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4965         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4966         struct rb_node *node;
4967
4968         ASSERT(inode->i_state & I_FREEING);
4969         truncate_inode_pages_final(&inode->i_data);
4970
4971         write_lock(&map_tree->lock);
4972         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4973                 struct extent_map *em;
4974
4975                 node = rb_first(&map_tree->map);
4976                 em = rb_entry(node, struct extent_map, rb_node);
4977                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4978                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4979                 remove_extent_mapping(map_tree, em);
4980                 free_extent_map(em);
4981                 if (need_resched()) {
4982                         write_unlock(&map_tree->lock);
4983                         cond_resched();
4984                         write_lock(&map_tree->lock);
4985                 }
4986         }
4987         write_unlock(&map_tree->lock);
4988
4989         spin_lock(&io_tree->lock);
4990         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4991                 struct extent_state *state;
4992                 struct extent_state *cached_state = NULL;
4993
4994                 node = rb_first(&io_tree->state);
4995                 state = rb_entry(node, struct extent_state, rb_node);
4996                 atomic_inc(&state->refs);
4997                 spin_unlock(&io_tree->lock);
4998
4999                 lock_extent_bits(io_tree, state->start, state->end,
5000                                  0, &cached_state);
5001                 clear_extent_bit(io_tree, state->start, state->end,
5002                                  EXTENT_LOCKED | EXTENT_DIRTY |
5003                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5004                                  EXTENT_DEFRAG, 1, 1,
5005                                  &cached_state, GFP_NOFS);
5006                 free_extent_state(state);
5007
5008                 cond_resched();
5009                 spin_lock(&io_tree->lock);
5010         }
5011         spin_unlock(&io_tree->lock);
5012 }
5013
5014 void btrfs_evict_inode(struct inode *inode)
5015 {
5016         struct btrfs_trans_handle *trans;
5017         struct btrfs_root *root = BTRFS_I(inode)->root;
5018         struct btrfs_block_rsv *rsv, *global_rsv;
5019         int steal_from_global = 0;
5020         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5021         int ret;
5022
5023         trace_btrfs_inode_evict(inode);
5024
5025         evict_inode_truncate_pages(inode);
5026
5027         if (inode->i_nlink &&
5028             ((btrfs_root_refs(&root->root_item) != 0 &&
5029               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5030              btrfs_is_free_space_inode(inode)))
5031                 goto no_delete;
5032
5033         if (is_bad_inode(inode)) {
5034                 btrfs_orphan_del(NULL, inode);
5035                 goto no_delete;
5036         }
5037         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5038         btrfs_wait_ordered_range(inode, 0, (u64)-1);
5039
5040         btrfs_free_io_failure_record(inode, 0, (u64)-1);
5041
5042         if (root->fs_info->log_root_recovering) {
5043                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5044                                  &BTRFS_I(inode)->runtime_flags));
5045                 goto no_delete;
5046         }
5047
5048         if (inode->i_nlink > 0) {
5049                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5050                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5051                 goto no_delete;
5052         }
5053
5054         ret = btrfs_commit_inode_delayed_inode(inode);
5055         if (ret) {
5056                 btrfs_orphan_del(NULL, inode);
5057                 goto no_delete;
5058         }
5059
5060         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5061         if (!rsv) {
5062                 btrfs_orphan_del(NULL, inode);
5063                 goto no_delete;
5064         }
5065         rsv->size = min_size;
5066         rsv->failfast = 1;
5067         global_rsv = &root->fs_info->global_block_rsv;
5068
5069         btrfs_i_size_write(inode, 0);
5070
5071         /*
5072          * This is a bit simpler than btrfs_truncate since we've already
5073          * reserved our space for our orphan item in the unlink, so we just
5074          * need to reserve some slack space in case we add bytes and update
5075          * inode item when doing the truncate.
5076          */
5077         while (1) {
5078                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5079                                              BTRFS_RESERVE_FLUSH_LIMIT);
5080
5081                 /*
5082                  * Try and steal from the global reserve since we will
5083                  * likely not use this space anyway, we want to try as
5084                  * hard as possible to get this to work.
5085                  */
5086                 if (ret)
5087                         steal_from_global++;
5088                 else
5089                         steal_from_global = 0;
5090                 ret = 0;
5091
5092                 /*
5093                  * steal_from_global == 0: we reserved stuff, hooray!
5094                  * steal_from_global == 1: we didn't reserve stuff, boo!
5095                  * steal_from_global == 2: we've committed, still not a lot of
5096                  * room but maybe we'll have room in the global reserve this
5097                  * time.
5098                  * steal_from_global == 3: abandon all hope!
5099                  */
5100                 if (steal_from_global > 2) {
5101                         btrfs_warn(root->fs_info,
5102                                 "Could not get space for a delete, will truncate on mount %d",
5103                                 ret);
5104                         btrfs_orphan_del(NULL, inode);
5105                         btrfs_free_block_rsv(root, rsv);
5106                         goto no_delete;
5107                 }
5108
5109                 trans = btrfs_join_transaction(root);
5110                 if (IS_ERR(trans)) {
5111                         btrfs_orphan_del(NULL, inode);
5112                         btrfs_free_block_rsv(root, rsv);
5113                         goto no_delete;
5114                 }
5115
5116                 /*
5117                  * We can't just steal from the global reserve, we need tomake
5118                  * sure there is room to do it, if not we need to commit and try
5119                  * again.
5120                  */
5121                 if (steal_from_global) {
5122                         if (!btrfs_check_space_for_delayed_refs(trans, root))
5123                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5124                                                               min_size);
5125                         else
5126                                 ret = -ENOSPC;
5127                 }
5128
5129                 /*
5130                  * Couldn't steal from the global reserve, we have too much
5131                  * pending stuff built up, commit the transaction and try it
5132                  * again.
5133                  */
5134                 if (ret) {
5135                         ret = btrfs_commit_transaction(trans, root);
5136                         if (ret) {
5137                                 btrfs_orphan_del(NULL, inode);
5138                                 btrfs_free_block_rsv(root, rsv);
5139                                 goto no_delete;
5140                         }
5141                         continue;
5142                 } else {
5143                         steal_from_global = 0;
5144                 }
5145
5146                 trans->block_rsv = rsv;
5147
5148                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5149                 if (ret != -ENOSPC && ret != -EAGAIN)
5150                         break;
5151
5152                 trans->block_rsv = &root->fs_info->trans_block_rsv;
5153                 btrfs_end_transaction(trans, root);
5154                 trans = NULL;
5155                 btrfs_btree_balance_dirty(root);
5156         }
5157
5158         btrfs_free_block_rsv(root, rsv);
5159
5160         /*
5161          * Errors here aren't a big deal, it just means we leave orphan items
5162          * in the tree.  They will be cleaned up on the next mount.
5163          */
5164         if (ret == 0) {
5165                 trans->block_rsv = root->orphan_block_rsv;
5166                 btrfs_orphan_del(trans, inode);
5167         } else {
5168                 btrfs_orphan_del(NULL, inode);
5169         }
5170
5171         trans->block_rsv = &root->fs_info->trans_block_rsv;
5172         if (!(root == root->fs_info->tree_root ||
5173               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5174                 btrfs_return_ino(root, btrfs_ino(inode));
5175
5176         btrfs_end_transaction(trans, root);
5177         btrfs_btree_balance_dirty(root);
5178 no_delete:
5179         btrfs_remove_delayed_node(inode);
5180         clear_inode(inode);
5181         return;
5182 }
5183
5184 /*
5185  * this returns the key found in the dir entry in the location pointer.
5186  * If no dir entries were found, location->objectid is 0.
5187  */
5188 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5189                                struct btrfs_key *location)
5190 {
5191         const char *name = dentry->d_name.name;
5192         int namelen = dentry->d_name.len;
5193         struct btrfs_dir_item *di;
5194         struct btrfs_path *path;
5195         struct btrfs_root *root = BTRFS_I(dir)->root;
5196         int ret = 0;
5197
5198         path = btrfs_alloc_path();
5199         if (!path)
5200                 return -ENOMEM;
5201
5202         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5203                                     namelen, 0);
5204         if (IS_ERR(di))
5205                 ret = PTR_ERR(di);
5206
5207         if (IS_ERR_OR_NULL(di))
5208                 goto out_err;
5209
5210         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5211 out:
5212         btrfs_free_path(path);
5213         return ret;
5214 out_err:
5215         location->objectid = 0;
5216         goto out;
5217 }
5218
5219 /*
5220  * when we hit a tree root in a directory, the btrfs part of the inode
5221  * needs to be changed to reflect the root directory of the tree root.  This
5222  * is kind of like crossing a mount point.
5223  */
5224 static int fixup_tree_root_location(struct btrfs_root *root,
5225                                     struct inode *dir,
5226                                     struct dentry *dentry,
5227                                     struct btrfs_key *location,
5228                                     struct btrfs_root **sub_root)
5229 {
5230         struct btrfs_path *path;
5231         struct btrfs_root *new_root;
5232         struct btrfs_root_ref *ref;
5233         struct extent_buffer *leaf;
5234         struct btrfs_key key;
5235         int ret;
5236         int err = 0;
5237
5238         path = btrfs_alloc_path();
5239         if (!path) {
5240                 err = -ENOMEM;
5241                 goto out;
5242         }
5243
5244         err = -ENOENT;
5245         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5246         key.type = BTRFS_ROOT_REF_KEY;
5247         key.offset = location->objectid;
5248
5249         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5250                                 0, 0);
5251         if (ret) {
5252                 if (ret < 0)
5253                         err = ret;
5254                 goto out;
5255         }
5256
5257         leaf = path->nodes[0];
5258         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5259         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5260             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5261                 goto out;
5262
5263         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5264                                    (unsigned long)(ref + 1),
5265                                    dentry->d_name.len);
5266         if (ret)
5267                 goto out;
5268
5269         btrfs_release_path(path);
5270
5271         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5272         if (IS_ERR(new_root)) {
5273                 err = PTR_ERR(new_root);
5274                 goto out;
5275         }
5276
5277         *sub_root = new_root;
5278         location->objectid = btrfs_root_dirid(&new_root->root_item);
5279         location->type = BTRFS_INODE_ITEM_KEY;
5280         location->offset = 0;
5281         err = 0;
5282 out:
5283         btrfs_free_path(path);
5284         return err;
5285 }
5286
5287 static void inode_tree_add(struct inode *inode)
5288 {
5289         struct btrfs_root *root = BTRFS_I(inode)->root;
5290         struct btrfs_inode *entry;
5291         struct rb_node **p;
5292         struct rb_node *parent;
5293         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5294         u64 ino = btrfs_ino(inode);
5295
5296         if (inode_unhashed(inode))
5297                 return;
5298         parent = NULL;
5299         spin_lock(&root->inode_lock);
5300         p = &root->inode_tree.rb_node;
5301         while (*p) {
5302                 parent = *p;
5303                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5304
5305                 if (ino < btrfs_ino(&entry->vfs_inode))
5306                         p = &parent->rb_left;
5307                 else if (ino > btrfs_ino(&entry->vfs_inode))
5308                         p = &parent->rb_right;
5309                 else {
5310                         WARN_ON(!(entry->vfs_inode.i_state &
5311                                   (I_WILL_FREE | I_FREEING)));
5312                         rb_replace_node(parent, new, &root->inode_tree);
5313                         RB_CLEAR_NODE(parent);
5314                         spin_unlock(&root->inode_lock);
5315                         return;
5316                 }
5317         }
5318         rb_link_node(new, parent, p);
5319         rb_insert_color(new, &root->inode_tree);
5320         spin_unlock(&root->inode_lock);
5321 }
5322
5323 static void inode_tree_del(struct inode *inode)
5324 {
5325         struct btrfs_root *root = BTRFS_I(inode)->root;
5326         int empty = 0;
5327
5328         spin_lock(&root->inode_lock);
5329         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5330                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5331                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5332                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5333         }
5334         spin_unlock(&root->inode_lock);
5335
5336         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5337                 synchronize_srcu(&root->fs_info->subvol_srcu);
5338                 spin_lock(&root->inode_lock);
5339                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5340                 spin_unlock(&root->inode_lock);
5341                 if (empty)
5342                         btrfs_add_dead_root(root);
5343         }
5344 }
5345
5346 void btrfs_invalidate_inodes(struct btrfs_root *root)
5347 {
5348         struct rb_node *node;
5349         struct rb_node *prev;
5350         struct btrfs_inode *entry;
5351         struct inode *inode;
5352         u64 objectid = 0;
5353
5354         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5355                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5356
5357         spin_lock(&root->inode_lock);
5358 again:
5359         node = root->inode_tree.rb_node;
5360         prev = NULL;
5361         while (node) {
5362                 prev = node;
5363                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5364
5365                 if (objectid < btrfs_ino(&entry->vfs_inode))
5366                         node = node->rb_left;
5367                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5368                         node = node->rb_right;
5369                 else
5370                         break;
5371         }
5372         if (!node) {
5373                 while (prev) {
5374                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5375                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5376                                 node = prev;
5377                                 break;
5378                         }
5379                         prev = rb_next(prev);
5380                 }
5381         }
5382         while (node) {
5383                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5384                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5385                 inode = igrab(&entry->vfs_inode);
5386                 if (inode) {
5387                         spin_unlock(&root->inode_lock);
5388                         if (atomic_read(&inode->i_count) > 1)
5389                                 d_prune_aliases(inode);
5390                         /*
5391                          * btrfs_drop_inode will have it removed from
5392                          * the inode cache when its usage count
5393                          * hits zero.
5394                          */
5395                         iput(inode);
5396                         cond_resched();
5397                         spin_lock(&root->inode_lock);
5398                         goto again;
5399                 }
5400
5401                 if (cond_resched_lock(&root->inode_lock))
5402                         goto again;
5403
5404                 node = rb_next(node);
5405         }
5406         spin_unlock(&root->inode_lock);
5407 }
5408
5409 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5410 {
5411         struct btrfs_iget_args *args = p;
5412         inode->i_ino = args->location->objectid;
5413         memcpy(&BTRFS_I(inode)->location, args->location,
5414                sizeof(*args->location));
5415         BTRFS_I(inode)->root = args->root;
5416         return 0;
5417 }
5418
5419 static int btrfs_find_actor(struct inode *inode, void *opaque)
5420 {
5421         struct btrfs_iget_args *args = opaque;
5422         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5423                 args->root == BTRFS_I(inode)->root;
5424 }
5425
5426 static struct inode *btrfs_iget_locked(struct super_block *s,
5427                                        struct btrfs_key *location,
5428                                        struct btrfs_root *root)
5429 {
5430         struct inode *inode;
5431         struct btrfs_iget_args args;
5432         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5433
5434         args.location = location;
5435         args.root = root;
5436
5437         inode = iget5_locked(s, hashval, btrfs_find_actor,
5438                              btrfs_init_locked_inode,
5439                              (void *)&args);
5440         return inode;
5441 }
5442
5443 /* Get an inode object given its location and corresponding root.
5444  * Returns in *is_new if the inode was read from disk
5445  */
5446 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5447                          struct btrfs_root *root, int *new)
5448 {
5449         struct inode *inode;
5450
5451         inode = btrfs_iget_locked(s, location, root);
5452         if (!inode)
5453                 return ERR_PTR(-ENOMEM);
5454
5455         if (inode->i_state & I_NEW) {
5456                 btrfs_read_locked_inode(inode);
5457                 if (!is_bad_inode(inode)) {
5458                         inode_tree_add(inode);
5459                         unlock_new_inode(inode);
5460                         if (new)
5461                                 *new = 1;
5462                 } else {
5463                         unlock_new_inode(inode);
5464                         iput(inode);
5465                         inode = ERR_PTR(-ESTALE);
5466                 }
5467         }
5468
5469         return inode;
5470 }
5471
5472 static struct inode *new_simple_dir(struct super_block *s,
5473                                     struct btrfs_key *key,
5474                                     struct btrfs_root *root)
5475 {
5476         struct inode *inode = new_inode(s);
5477
5478         if (!inode)
5479                 return ERR_PTR(-ENOMEM);
5480
5481         BTRFS_I(inode)->root = root;
5482         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5483         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5484
5485         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5486         inode->i_op = &btrfs_dir_ro_inode_operations;
5487         inode->i_fop = &simple_dir_operations;
5488         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5489         inode->i_mtime = CURRENT_TIME;
5490         inode->i_atime = inode->i_mtime;
5491         inode->i_ctime = inode->i_mtime;
5492         BTRFS_I(inode)->i_otime = inode->i_mtime;
5493
5494         return inode;
5495 }
5496
5497 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5498 {
5499         struct inode *inode;
5500         struct btrfs_root *root = BTRFS_I(dir)->root;
5501         struct btrfs_root *sub_root = root;
5502         struct btrfs_key location;
5503         int index;
5504         int ret = 0;
5505
5506         if (dentry->d_name.len > BTRFS_NAME_LEN)
5507                 return ERR_PTR(-ENAMETOOLONG);
5508
5509         ret = btrfs_inode_by_name(dir, dentry, &location);
5510         if (ret < 0)
5511                 return ERR_PTR(ret);
5512
5513         if (location.objectid == 0)
5514                 return ERR_PTR(-ENOENT);
5515
5516         if (location.type == BTRFS_INODE_ITEM_KEY) {
5517                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5518                 return inode;
5519         }
5520
5521         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5522
5523         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5524         ret = fixup_tree_root_location(root, dir, dentry,
5525                                        &location, &sub_root);
5526         if (ret < 0) {
5527                 if (ret != -ENOENT)
5528                         inode = ERR_PTR(ret);
5529                 else
5530                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5531         } else {
5532                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5533         }
5534         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5535
5536         if (!IS_ERR(inode) && root != sub_root) {
5537                 down_read(&root->fs_info->cleanup_work_sem);
5538                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5539                         ret = btrfs_orphan_cleanup(sub_root);
5540                 up_read(&root->fs_info->cleanup_work_sem);
5541                 if (ret) {
5542                         iput(inode);
5543                         inode = ERR_PTR(ret);
5544                 }
5545         }
5546
5547         return inode;
5548 }
5549
5550 static int btrfs_dentry_delete(const struct dentry *dentry)
5551 {
5552         struct btrfs_root *root;
5553         struct inode *inode = d_inode(dentry);
5554
5555         if (!inode && !IS_ROOT(dentry))
5556                 inode = d_inode(dentry->d_parent);
5557
5558         if (inode) {
5559                 root = BTRFS_I(inode)->root;
5560                 if (btrfs_root_refs(&root->root_item) == 0)
5561                         return 1;
5562
5563                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5564                         return 1;
5565         }
5566         return 0;
5567 }
5568
5569 static void btrfs_dentry_release(struct dentry *dentry)
5570 {
5571         kfree(dentry->d_fsdata);
5572 }
5573
5574 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5575                                    unsigned int flags)
5576 {
5577         struct inode *inode;
5578
5579         inode = btrfs_lookup_dentry(dir, dentry);
5580         if (IS_ERR(inode)) {
5581                 if (PTR_ERR(inode) == -ENOENT)
5582                         inode = NULL;
5583                 else
5584                         return ERR_CAST(inode);
5585         }
5586
5587         return d_splice_alias(inode, dentry);
5588 }
5589
5590 unsigned char btrfs_filetype_table[] = {
5591         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5592 };
5593
5594 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5595 {
5596         struct inode *inode = file_inode(file);
5597         struct btrfs_root *root = BTRFS_I(inode)->root;
5598         struct btrfs_item *item;
5599         struct btrfs_dir_item *di;
5600         struct btrfs_key key;
5601         struct btrfs_key found_key;
5602         struct btrfs_path *path;
5603         struct list_head ins_list;
5604         struct list_head del_list;
5605         int ret;
5606         struct extent_buffer *leaf;
5607         int slot;
5608         unsigned char d_type;
5609         int over = 0;
5610         u32 di_cur;
5611         u32 di_total;
5612         u32 di_len;
5613         int key_type = BTRFS_DIR_INDEX_KEY;
5614         char tmp_name[32];
5615         char *name_ptr;
5616         int name_len;
5617         int is_curr = 0;        /* ctx->pos points to the current index? */
5618
5619         /* FIXME, use a real flag for deciding about the key type */
5620         if (root->fs_info->tree_root == root)
5621                 key_type = BTRFS_DIR_ITEM_KEY;
5622
5623         if (!dir_emit_dots(file, ctx))
5624                 return 0;
5625
5626         path = btrfs_alloc_path();
5627         if (!path)
5628                 return -ENOMEM;
5629
5630         path->reada = 1;
5631
5632         if (key_type == BTRFS_DIR_INDEX_KEY) {
5633                 INIT_LIST_HEAD(&ins_list);
5634                 INIT_LIST_HEAD(&del_list);
5635                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5636         }
5637
5638         key.type = key_type;
5639         key.offset = ctx->pos;
5640         key.objectid = btrfs_ino(inode);
5641
5642         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5643         if (ret < 0)
5644                 goto err;
5645
5646         while (1) {
5647                 leaf = path->nodes[0];
5648                 slot = path->slots[0];
5649                 if (slot >= btrfs_header_nritems(leaf)) {
5650                         ret = btrfs_next_leaf(root, path);
5651                         if (ret < 0)
5652                                 goto err;
5653                         else if (ret > 0)
5654                                 break;
5655                         continue;
5656                 }
5657
5658                 item = btrfs_item_nr(slot);
5659                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5660
5661                 if (found_key.objectid != key.objectid)
5662                         break;
5663                 if (found_key.type != key_type)
5664                         break;
5665                 if (found_key.offset < ctx->pos)
5666                         goto next;
5667                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5668                     btrfs_should_delete_dir_index(&del_list,
5669                                                   found_key.offset))
5670                         goto next;
5671
5672                 ctx->pos = found_key.offset;
5673                 is_curr = 1;
5674
5675                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5676                 di_cur = 0;
5677                 di_total = btrfs_item_size(leaf, item);
5678
5679                 while (di_cur < di_total) {
5680                         struct btrfs_key location;
5681
5682                         if (verify_dir_item(root, leaf, di))
5683                                 break;
5684
5685                         name_len = btrfs_dir_name_len(leaf, di);
5686                         if (name_len <= sizeof(tmp_name)) {
5687                                 name_ptr = tmp_name;
5688                         } else {
5689                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5690                                 if (!name_ptr) {
5691                                         ret = -ENOMEM;
5692                                         goto err;
5693                                 }
5694                         }
5695                         read_extent_buffer(leaf, name_ptr,
5696                                            (unsigned long)(di + 1), name_len);
5697
5698                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5699                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5700
5701
5702                         /* is this a reference to our own snapshot? If so
5703                          * skip it.
5704                          *
5705                          * In contrast to old kernels, we insert the snapshot's
5706                          * dir item and dir index after it has been created, so
5707                          * we won't find a reference to our own snapshot. We
5708                          * still keep the following code for backward
5709                          * compatibility.
5710                          */
5711                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5712                             location.objectid == root->root_key.objectid) {
5713                                 over = 0;
5714                                 goto skip;
5715                         }
5716                         over = !dir_emit(ctx, name_ptr, name_len,
5717                                        location.objectid, d_type);
5718
5719 skip:
5720                         if (name_ptr != tmp_name)
5721                                 kfree(name_ptr);
5722
5723                         if (over)
5724                                 goto nopos;
5725                         di_len = btrfs_dir_name_len(leaf, di) +
5726                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5727                         di_cur += di_len;
5728                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5729                 }
5730 next:
5731                 path->slots[0]++;
5732         }
5733
5734         if (key_type == BTRFS_DIR_INDEX_KEY) {
5735                 if (is_curr)
5736                         ctx->pos++;
5737                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5738                 if (ret)
5739                         goto nopos;
5740         }
5741
5742         /* Reached end of directory/root. Bump pos past the last item. */
5743         ctx->pos++;
5744
5745         /*
5746          * Stop new entries from being returned after we return the last
5747          * entry.
5748          *
5749          * New directory entries are assigned a strictly increasing
5750          * offset.  This means that new entries created during readdir
5751          * are *guaranteed* to be seen in the future by that readdir.
5752          * This has broken buggy programs which operate on names as
5753          * they're returned by readdir.  Until we re-use freed offsets
5754          * we have this hack to stop new entries from being returned
5755          * under the assumption that they'll never reach this huge
5756          * offset.
5757          *
5758          * This is being careful not to overflow 32bit loff_t unless the
5759          * last entry requires it because doing so has broken 32bit apps
5760          * in the past.
5761          */
5762         if (key_type == BTRFS_DIR_INDEX_KEY) {
5763                 if (ctx->pos >= INT_MAX)
5764                         ctx->pos = LLONG_MAX;
5765                 else
5766                         ctx->pos = INT_MAX;
5767         }
5768 nopos:
5769         ret = 0;
5770 err:
5771         if (key_type == BTRFS_DIR_INDEX_KEY)
5772                 btrfs_put_delayed_items(&ins_list, &del_list);
5773         btrfs_free_path(path);
5774         return ret;
5775 }
5776
5777 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5778 {
5779         struct btrfs_root *root = BTRFS_I(inode)->root;
5780         struct btrfs_trans_handle *trans;
5781         int ret = 0;
5782         bool nolock = false;
5783
5784         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5785                 return 0;
5786
5787         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5788                 nolock = true;
5789
5790         if (wbc->sync_mode == WB_SYNC_ALL) {
5791                 if (nolock)
5792                         trans = btrfs_join_transaction_nolock(root);
5793                 else
5794                         trans = btrfs_join_transaction(root);
5795                 if (IS_ERR(trans))
5796                         return PTR_ERR(trans);
5797                 ret = btrfs_commit_transaction(trans, root);
5798         }
5799         return ret;
5800 }
5801
5802 /*
5803  * This is somewhat expensive, updating the tree every time the
5804  * inode changes.  But, it is most likely to find the inode in cache.
5805  * FIXME, needs more benchmarking...there are no reasons other than performance
5806  * to keep or drop this code.
5807  */
5808 static int btrfs_dirty_inode(struct inode *inode)
5809 {
5810         struct btrfs_root *root = BTRFS_I(inode)->root;
5811         struct btrfs_trans_handle *trans;
5812         int ret;
5813
5814         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5815                 return 0;
5816
5817         trans = btrfs_join_transaction(root);
5818         if (IS_ERR(trans))
5819                 return PTR_ERR(trans);
5820
5821         ret = btrfs_update_inode(trans, root, inode);
5822         if (ret && ret == -ENOSPC) {
5823                 /* whoops, lets try again with the full transaction */
5824                 btrfs_end_transaction(trans, root);
5825                 trans = btrfs_start_transaction(root, 1);
5826                 if (IS_ERR(trans))
5827                         return PTR_ERR(trans);
5828
5829                 ret = btrfs_update_inode(trans, root, inode);
5830         }
5831         btrfs_end_transaction(trans, root);
5832         if (BTRFS_I(inode)->delayed_node)
5833                 btrfs_balance_delayed_items(root);
5834
5835         return ret;
5836 }
5837
5838 /*
5839  * This is a copy of file_update_time.  We need this so we can return error on
5840  * ENOSPC for updating the inode in the case of file write and mmap writes.
5841  */
5842 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5843                              int flags)
5844 {
5845         struct btrfs_root *root = BTRFS_I(inode)->root;
5846
5847         if (btrfs_root_readonly(root))
5848                 return -EROFS;
5849
5850         if (flags & S_VERSION)
5851                 inode_inc_iversion(inode);
5852         if (flags & S_CTIME)
5853                 inode->i_ctime = *now;
5854         if (flags & S_MTIME)
5855                 inode->i_mtime = *now;
5856         if (flags & S_ATIME)
5857                 inode->i_atime = *now;
5858         return btrfs_dirty_inode(inode);
5859 }
5860
5861 /*
5862  * find the highest existing sequence number in a directory
5863  * and then set the in-memory index_cnt variable to reflect
5864  * free sequence numbers
5865  */
5866 static int btrfs_set_inode_index_count(struct inode *inode)
5867 {
5868         struct btrfs_root *root = BTRFS_I(inode)->root;
5869         struct btrfs_key key, found_key;
5870         struct btrfs_path *path;
5871         struct extent_buffer *leaf;
5872         int ret;
5873
5874         key.objectid = btrfs_ino(inode);
5875         key.type = BTRFS_DIR_INDEX_KEY;
5876         key.offset = (u64)-1;
5877
5878         path = btrfs_alloc_path();
5879         if (!path)
5880                 return -ENOMEM;
5881
5882         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5883         if (ret < 0)
5884                 goto out;
5885         /* FIXME: we should be able to handle this */
5886         if (ret == 0)
5887                 goto out;
5888         ret = 0;
5889
5890         /*
5891          * MAGIC NUMBER EXPLANATION:
5892          * since we search a directory based on f_pos we have to start at 2
5893          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5894          * else has to start at 2
5895          */
5896         if (path->slots[0] == 0) {
5897                 BTRFS_I(inode)->index_cnt = 2;
5898                 goto out;
5899         }
5900
5901         path->slots[0]--;
5902
5903         leaf = path->nodes[0];
5904         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5905
5906         if (found_key.objectid != btrfs_ino(inode) ||
5907             found_key.type != BTRFS_DIR_INDEX_KEY) {
5908                 BTRFS_I(inode)->index_cnt = 2;
5909                 goto out;
5910         }
5911
5912         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5913 out:
5914         btrfs_free_path(path);
5915         return ret;
5916 }
5917
5918 /*
5919  * helper to find a free sequence number in a given directory.  This current
5920  * code is very simple, later versions will do smarter things in the btree
5921  */
5922 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5923 {
5924         int ret = 0;
5925
5926         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5927                 ret = btrfs_inode_delayed_dir_index_count(dir);
5928                 if (ret) {
5929                         ret = btrfs_set_inode_index_count(dir);
5930                         if (ret)
5931                                 return ret;
5932                 }
5933         }
5934
5935         *index = BTRFS_I(dir)->index_cnt;
5936         BTRFS_I(dir)->index_cnt++;
5937
5938         return ret;
5939 }
5940
5941 static int btrfs_insert_inode_locked(struct inode *inode)
5942 {
5943         struct btrfs_iget_args args;
5944         args.location = &BTRFS_I(inode)->location;
5945         args.root = BTRFS_I(inode)->root;
5946
5947         return insert_inode_locked4(inode,
5948                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5949                    btrfs_find_actor, &args);
5950 }
5951
5952 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5953                                      struct btrfs_root *root,
5954                                      struct inode *dir,
5955                                      const char *name, int name_len,
5956                                      u64 ref_objectid, u64 objectid,
5957                                      umode_t mode, u64 *index)
5958 {
5959         struct inode *inode;
5960         struct btrfs_inode_item *inode_item;
5961         struct btrfs_key *location;
5962         struct btrfs_path *path;
5963         struct btrfs_inode_ref *ref;
5964         struct btrfs_key key[2];
5965         u32 sizes[2];
5966         int nitems = name ? 2 : 1;
5967         unsigned long ptr;
5968         int ret;
5969
5970         path = btrfs_alloc_path();
5971         if (!path)
5972                 return ERR_PTR(-ENOMEM);
5973
5974         inode = new_inode(root->fs_info->sb);
5975         if (!inode) {
5976                 btrfs_free_path(path);
5977                 return ERR_PTR(-ENOMEM);
5978         }
5979
5980         /*
5981          * O_TMPFILE, set link count to 0, so that after this point,
5982          * we fill in an inode item with the correct link count.
5983          */
5984         if (!name)
5985                 set_nlink(inode, 0);
5986
5987         /*
5988          * we have to initialize this early, so we can reclaim the inode
5989          * number if we fail afterwards in this function.
5990          */
5991         inode->i_ino = objectid;
5992
5993         if (dir && name) {
5994                 trace_btrfs_inode_request(dir);
5995
5996                 ret = btrfs_set_inode_index(dir, index);
5997                 if (ret) {
5998                         btrfs_free_path(path);
5999                         iput(inode);
6000                         return ERR_PTR(ret);
6001                 }
6002         } else if (dir) {
6003                 *index = 0;
6004         }
6005         /*
6006          * index_cnt is ignored for everything but a dir,
6007          * btrfs_get_inode_index_count has an explanation for the magic
6008          * number
6009          */
6010         BTRFS_I(inode)->index_cnt = 2;
6011         BTRFS_I(inode)->dir_index = *index;
6012         BTRFS_I(inode)->root = root;
6013         BTRFS_I(inode)->generation = trans->transid;
6014         inode->i_generation = BTRFS_I(inode)->generation;
6015
6016         /*
6017          * We could have gotten an inode number from somebody who was fsynced
6018          * and then removed in this same transaction, so let's just set full
6019          * sync since it will be a full sync anyway and this will blow away the
6020          * old info in the log.
6021          */
6022         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6023
6024         key[0].objectid = objectid;
6025         key[0].type = BTRFS_INODE_ITEM_KEY;
6026         key[0].offset = 0;
6027
6028         sizes[0] = sizeof(struct btrfs_inode_item);
6029
6030         if (name) {
6031                 /*
6032                  * Start new inodes with an inode_ref. This is slightly more
6033                  * efficient for small numbers of hard links since they will
6034                  * be packed into one item. Extended refs will kick in if we
6035                  * add more hard links than can fit in the ref item.
6036                  */
6037                 key[1].objectid = objectid;
6038                 key[1].type = BTRFS_INODE_REF_KEY;
6039                 key[1].offset = ref_objectid;
6040
6041                 sizes[1] = name_len + sizeof(*ref);
6042         }
6043
6044         location = &BTRFS_I(inode)->location;
6045         location->objectid = objectid;
6046         location->offset = 0;
6047         location->type = BTRFS_INODE_ITEM_KEY;
6048
6049         ret = btrfs_insert_inode_locked(inode);
6050         if (ret < 0)
6051                 goto fail;
6052
6053         path->leave_spinning = 1;
6054         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6055         if (ret != 0)
6056                 goto fail_unlock;
6057
6058         inode_init_owner(inode, dir, mode);
6059         inode_set_bytes(inode, 0);
6060
6061         inode->i_mtime = CURRENT_TIME;
6062         inode->i_atime = inode->i_mtime;
6063         inode->i_ctime = inode->i_mtime;
6064         BTRFS_I(inode)->i_otime = inode->i_mtime;
6065
6066         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6067                                   struct btrfs_inode_item);
6068         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6069                              sizeof(*inode_item));
6070         fill_inode_item(trans, path->nodes[0], inode_item, inode);
6071
6072         if (name) {
6073                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6074                                      struct btrfs_inode_ref);
6075                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6076                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6077                 ptr = (unsigned long)(ref + 1);
6078                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6079         }
6080
6081         btrfs_mark_buffer_dirty(path->nodes[0]);
6082         btrfs_free_path(path);
6083
6084         btrfs_inherit_iflags(inode, dir);
6085
6086         if (S_ISREG(mode)) {
6087                 if (btrfs_test_opt(root, NODATASUM))
6088                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6089                 if (btrfs_test_opt(root, NODATACOW))
6090                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6091                                 BTRFS_INODE_NODATASUM;
6092         }
6093
6094         inode_tree_add(inode);
6095
6096         trace_btrfs_inode_new(inode);
6097         btrfs_set_inode_last_trans(trans, inode);
6098
6099         btrfs_update_root_times(trans, root);
6100
6101         ret = btrfs_inode_inherit_props(trans, inode, dir);
6102         if (ret)
6103                 btrfs_err(root->fs_info,
6104                           "error inheriting props for ino %llu (root %llu): %d",
6105                           btrfs_ino(inode), root->root_key.objectid, ret);
6106
6107         return inode;
6108
6109 fail_unlock:
6110         unlock_new_inode(inode);
6111 fail:
6112         if (dir && name)
6113                 BTRFS_I(dir)->index_cnt--;
6114         btrfs_free_path(path);
6115         iput(inode);
6116         return ERR_PTR(ret);
6117 }
6118
6119 static inline u8 btrfs_inode_type(struct inode *inode)
6120 {
6121         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6122 }
6123
6124 /*
6125  * utility function to add 'inode' into 'parent_inode' with
6126  * a give name and a given sequence number.
6127  * if 'add_backref' is true, also insert a backref from the
6128  * inode to the parent directory.
6129  */
6130 int btrfs_add_link(struct btrfs_trans_handle *trans,
6131                    struct inode *parent_inode, struct inode *inode,
6132                    const char *name, int name_len, int add_backref, u64 index)
6133 {
6134         int ret = 0;
6135         struct btrfs_key key;
6136         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6137         u64 ino = btrfs_ino(inode);
6138         u64 parent_ino = btrfs_ino(parent_inode);
6139
6140         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6141                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6142         } else {
6143                 key.objectid = ino;
6144                 key.type = BTRFS_INODE_ITEM_KEY;
6145                 key.offset = 0;
6146         }
6147
6148         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6149                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6150                                          key.objectid, root->root_key.objectid,
6151                                          parent_ino, index, name, name_len);
6152         } else if (add_backref) {
6153                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6154                                              parent_ino, index);
6155         }
6156
6157         /* Nothing to clean up yet */
6158         if (ret)
6159                 return ret;
6160
6161         ret = btrfs_insert_dir_item(trans, root, name, name_len,
6162                                     parent_inode, &key,
6163                                     btrfs_inode_type(inode), index);
6164         if (ret == -EEXIST || ret == -EOVERFLOW)
6165                 goto fail_dir_item;
6166         else if (ret) {
6167                 btrfs_abort_transaction(trans, root, ret);
6168                 return ret;
6169         }
6170
6171         btrfs_i_size_write(parent_inode, parent_inode->i_size +
6172                            name_len * 2);
6173         inode_inc_iversion(parent_inode);
6174         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6175         ret = btrfs_update_inode(trans, root, parent_inode);
6176         if (ret)
6177                 btrfs_abort_transaction(trans, root, ret);
6178         return ret;
6179
6180 fail_dir_item:
6181         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6182                 u64 local_index;
6183                 int err;
6184                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6185                                  key.objectid, root->root_key.objectid,
6186                                  parent_ino, &local_index, name, name_len);
6187
6188         } else if (add_backref) {
6189                 u64 local_index;
6190                 int err;
6191
6192                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6193                                           ino, parent_ino, &local_index);
6194         }
6195         return ret;
6196 }
6197
6198 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6199                             struct inode *dir, struct dentry *dentry,
6200                             struct inode *inode, int backref, u64 index)
6201 {
6202         int err = btrfs_add_link(trans, dir, inode,
6203                                  dentry->d_name.name, dentry->d_name.len,
6204                                  backref, index);
6205         if (err > 0)
6206                 err = -EEXIST;
6207         return err;
6208 }
6209
6210 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6211                         umode_t mode, dev_t rdev)
6212 {
6213         struct btrfs_trans_handle *trans;
6214         struct btrfs_root *root = BTRFS_I(dir)->root;
6215         struct inode *inode = NULL;
6216         int err;
6217         int drop_inode = 0;
6218         u64 objectid;
6219         u64 index = 0;
6220
6221         if (!new_valid_dev(rdev))
6222                 return -EINVAL;
6223
6224         /*
6225          * 2 for inode item and ref
6226          * 2 for dir items
6227          * 1 for xattr if selinux is on
6228          */
6229         trans = btrfs_start_transaction(root, 5);
6230         if (IS_ERR(trans))
6231                 return PTR_ERR(trans);
6232
6233         err = btrfs_find_free_ino(root, &objectid);
6234         if (err)
6235                 goto out_unlock;
6236
6237         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6238                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6239                                 mode, &index);
6240         if (IS_ERR(inode)) {
6241                 err = PTR_ERR(inode);
6242                 goto out_unlock;
6243         }
6244
6245         /*
6246         * If the active LSM wants to access the inode during
6247         * d_instantiate it needs these. Smack checks to see
6248         * if the filesystem supports xattrs by looking at the
6249         * ops vector.
6250         */
6251         inode->i_op = &btrfs_special_inode_operations;
6252         init_special_inode(inode, inode->i_mode, rdev);
6253
6254         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6255         if (err)
6256                 goto out_unlock_inode;
6257
6258         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6259         if (err) {
6260                 goto out_unlock_inode;
6261         } else {
6262                 btrfs_update_inode(trans, root, inode);
6263                 unlock_new_inode(inode);
6264                 d_instantiate(dentry, inode);
6265         }
6266
6267 out_unlock:
6268         btrfs_end_transaction(trans, root);
6269         btrfs_balance_delayed_items(root);
6270         btrfs_btree_balance_dirty(root);
6271         if (drop_inode) {
6272                 inode_dec_link_count(inode);
6273                 iput(inode);
6274         }
6275         return err;
6276
6277 out_unlock_inode:
6278         drop_inode = 1;
6279         unlock_new_inode(inode);
6280         goto out_unlock;
6281
6282 }
6283
6284 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6285                         umode_t mode, bool excl)
6286 {
6287         struct btrfs_trans_handle *trans;
6288         struct btrfs_root *root = BTRFS_I(dir)->root;
6289         struct inode *inode = NULL;
6290         int drop_inode_on_err = 0;
6291         int err;
6292         u64 objectid;
6293         u64 index = 0;
6294
6295         /*
6296          * 2 for inode item and ref
6297          * 2 for dir items
6298          * 1 for xattr if selinux is on
6299          */
6300         trans = btrfs_start_transaction(root, 5);
6301         if (IS_ERR(trans))
6302                 return PTR_ERR(trans);
6303
6304         err = btrfs_find_free_ino(root, &objectid);
6305         if (err)
6306                 goto out_unlock;
6307
6308         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6309                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6310                                 mode, &index);
6311         if (IS_ERR(inode)) {
6312                 err = PTR_ERR(inode);
6313                 goto out_unlock;
6314         }
6315         drop_inode_on_err = 1;
6316         /*
6317         * If the active LSM wants to access the inode during
6318         * d_instantiate it needs these. Smack checks to see
6319         * if the filesystem supports xattrs by looking at the
6320         * ops vector.
6321         */
6322         inode->i_fop = &btrfs_file_operations;
6323         inode->i_op = &btrfs_file_inode_operations;
6324         inode->i_mapping->a_ops = &btrfs_aops;
6325
6326         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6327         if (err)
6328                 goto out_unlock_inode;
6329
6330         err = btrfs_update_inode(trans, root, inode);
6331         if (err)
6332                 goto out_unlock_inode;
6333
6334         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6335         if (err)
6336                 goto out_unlock_inode;
6337
6338         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6339         unlock_new_inode(inode);
6340         d_instantiate(dentry, inode);
6341
6342 out_unlock:
6343         btrfs_end_transaction(trans, root);
6344         if (err && drop_inode_on_err) {
6345                 inode_dec_link_count(inode);
6346                 iput(inode);
6347         }
6348         btrfs_balance_delayed_items(root);
6349         btrfs_btree_balance_dirty(root);
6350         return err;
6351
6352 out_unlock_inode:
6353         unlock_new_inode(inode);
6354         goto out_unlock;
6355
6356 }
6357
6358 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6359                       struct dentry *dentry)
6360 {
6361         struct btrfs_trans_handle *trans;
6362         struct btrfs_root *root = BTRFS_I(dir)->root;
6363         struct inode *inode = d_inode(old_dentry);
6364         u64 index;
6365         int err;
6366         int drop_inode = 0;
6367
6368         /* do not allow sys_link's with other subvols of the same device */
6369         if (root->objectid != BTRFS_I(inode)->root->objectid)
6370                 return -EXDEV;
6371
6372         if (inode->i_nlink >= BTRFS_LINK_MAX)
6373                 return -EMLINK;
6374
6375         err = btrfs_set_inode_index(dir, &index);
6376         if (err)
6377                 goto fail;
6378
6379         /*
6380          * 2 items for inode and inode ref
6381          * 2 items for dir items
6382          * 1 item for parent inode
6383          */
6384         trans = btrfs_start_transaction(root, 5);
6385         if (IS_ERR(trans)) {
6386                 err = PTR_ERR(trans);
6387                 goto fail;
6388         }
6389
6390         /* There are several dir indexes for this inode, clear the cache. */
6391         BTRFS_I(inode)->dir_index = 0ULL;
6392         inc_nlink(inode);
6393         inode_inc_iversion(inode);
6394         inode->i_ctime = CURRENT_TIME;
6395         ihold(inode);
6396         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6397
6398         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6399
6400         if (err) {
6401                 drop_inode = 1;
6402         } else {
6403                 struct dentry *parent = dentry->d_parent;
6404                 err = btrfs_update_inode(trans, root, inode);
6405                 if (err)
6406                         goto fail;
6407                 if (inode->i_nlink == 1) {
6408                         /*
6409                          * If new hard link count is 1, it's a file created
6410                          * with open(2) O_TMPFILE flag.
6411                          */
6412                         err = btrfs_orphan_del(trans, inode);
6413                         if (err)
6414                                 goto fail;
6415                 }
6416                 d_instantiate(dentry, inode);
6417                 btrfs_log_new_name(trans, inode, NULL, parent);
6418         }
6419
6420         btrfs_end_transaction(trans, root);
6421         btrfs_balance_delayed_items(root);
6422 fail:
6423         if (drop_inode) {
6424                 inode_dec_link_count(inode);
6425                 iput(inode);
6426         }
6427         btrfs_btree_balance_dirty(root);
6428         return err;
6429 }
6430
6431 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6432 {
6433         struct inode *inode = NULL;
6434         struct btrfs_trans_handle *trans;
6435         struct btrfs_root *root = BTRFS_I(dir)->root;
6436         int err = 0;
6437         int drop_on_err = 0;
6438         u64 objectid = 0;
6439         u64 index = 0;
6440
6441         /*
6442          * 2 items for inode and ref
6443          * 2 items for dir items
6444          * 1 for xattr if selinux is on
6445          */
6446         trans = btrfs_start_transaction(root, 5);
6447         if (IS_ERR(trans))
6448                 return PTR_ERR(trans);
6449
6450         err = btrfs_find_free_ino(root, &objectid);
6451         if (err)
6452                 goto out_fail;
6453
6454         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6455                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6456                                 S_IFDIR | mode, &index);
6457         if (IS_ERR(inode)) {
6458                 err = PTR_ERR(inode);
6459                 goto out_fail;
6460         }
6461
6462         drop_on_err = 1;
6463         /* these must be set before we unlock the inode */
6464         inode->i_op = &btrfs_dir_inode_operations;
6465         inode->i_fop = &btrfs_dir_file_operations;
6466
6467         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6468         if (err)
6469                 goto out_fail_inode;
6470
6471         btrfs_i_size_write(inode, 0);
6472         err = btrfs_update_inode(trans, root, inode);
6473         if (err)
6474                 goto out_fail_inode;
6475
6476         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6477                              dentry->d_name.len, 0, index);
6478         if (err)
6479                 goto out_fail_inode;
6480
6481         d_instantiate(dentry, inode);
6482         /*
6483          * mkdir is special.  We're unlocking after we call d_instantiate
6484          * to avoid a race with nfsd calling d_instantiate.
6485          */
6486         unlock_new_inode(inode);
6487         drop_on_err = 0;
6488
6489 out_fail:
6490         btrfs_end_transaction(trans, root);
6491         if (drop_on_err) {
6492                 inode_dec_link_count(inode);
6493                 iput(inode);
6494         }
6495         btrfs_balance_delayed_items(root);
6496         btrfs_btree_balance_dirty(root);
6497         return err;
6498
6499 out_fail_inode:
6500         unlock_new_inode(inode);
6501         goto out_fail;
6502 }
6503
6504 /* Find next extent map of a given extent map, caller needs to ensure locks */
6505 static struct extent_map *next_extent_map(struct extent_map *em)
6506 {
6507         struct rb_node *next;
6508
6509         next = rb_next(&em->rb_node);
6510         if (!next)
6511                 return NULL;
6512         return container_of(next, struct extent_map, rb_node);
6513 }
6514
6515 static struct extent_map *prev_extent_map(struct extent_map *em)
6516 {
6517         struct rb_node *prev;
6518
6519         prev = rb_prev(&em->rb_node);
6520         if (!prev)
6521                 return NULL;
6522         return container_of(prev, struct extent_map, rb_node);
6523 }
6524
6525 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6526  * the existing extent is the nearest extent to map_start,
6527  * and an extent that you want to insert, deal with overlap and insert
6528  * the best fitted new extent into the tree.
6529  */
6530 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6531                                 struct extent_map *existing,
6532                                 struct extent_map *em,
6533                                 u64 map_start)
6534 {
6535         struct extent_map *prev;
6536         struct extent_map *next;
6537         u64 start;
6538         u64 end;
6539         u64 start_diff;
6540
6541         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6542
6543         if (existing->start > map_start) {
6544                 next = existing;
6545                 prev = prev_extent_map(next);
6546         } else {
6547                 prev = existing;
6548                 next = next_extent_map(prev);
6549         }
6550
6551         start = prev ? extent_map_end(prev) : em->start;
6552         start = max_t(u64, start, em->start);
6553         end = next ? next->start : extent_map_end(em);
6554         end = min_t(u64, end, extent_map_end(em));
6555         start_diff = start - em->start;
6556         em->start = start;
6557         em->len = end - start;
6558         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6559             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6560                 em->block_start += start_diff;
6561                 em->block_len -= start_diff;
6562         }
6563         return add_extent_mapping(em_tree, em, 0);
6564 }
6565
6566 static noinline int uncompress_inline(struct btrfs_path *path,
6567                                       struct inode *inode, struct page *page,
6568                                       size_t pg_offset, u64 extent_offset,
6569                                       struct btrfs_file_extent_item *item)
6570 {
6571         int ret;
6572         struct extent_buffer *leaf = path->nodes[0];
6573         char *tmp;
6574         size_t max_size;
6575         unsigned long inline_size;
6576         unsigned long ptr;
6577         int compress_type;
6578
6579         WARN_ON(pg_offset != 0);
6580         compress_type = btrfs_file_extent_compression(leaf, item);
6581         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6582         inline_size = btrfs_file_extent_inline_item_len(leaf,
6583                                         btrfs_item_nr(path->slots[0]));
6584         tmp = kmalloc(inline_size, GFP_NOFS);
6585         if (!tmp)
6586                 return -ENOMEM;
6587         ptr = btrfs_file_extent_inline_start(item);
6588
6589         read_extent_buffer(leaf, tmp, ptr, inline_size);
6590
6591         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6592         ret = btrfs_decompress(compress_type, tmp, page,
6593                                extent_offset, inline_size, max_size);
6594         kfree(tmp);
6595         return ret;
6596 }
6597
6598 /*
6599  * a bit scary, this does extent mapping from logical file offset to the disk.
6600  * the ugly parts come from merging extents from the disk with the in-ram
6601  * representation.  This gets more complex because of the data=ordered code,
6602  * where the in-ram extents might be locked pending data=ordered completion.
6603  *
6604  * This also copies inline extents directly into the page.
6605  */
6606
6607 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6608                                     size_t pg_offset, u64 start, u64 len,
6609                                     int create)
6610 {
6611         int ret;
6612         int err = 0;
6613         u64 extent_start = 0;
6614         u64 extent_end = 0;
6615         u64 objectid = btrfs_ino(inode);
6616         u32 found_type;
6617         struct btrfs_path *path = NULL;
6618         struct btrfs_root *root = BTRFS_I(inode)->root;
6619         struct btrfs_file_extent_item *item;
6620         struct extent_buffer *leaf;
6621         struct btrfs_key found_key;
6622         struct extent_map *em = NULL;
6623         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6624         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6625         struct btrfs_trans_handle *trans = NULL;
6626         const bool new_inline = !page || create;
6627
6628 again:
6629         read_lock(&em_tree->lock);
6630         em = lookup_extent_mapping(em_tree, start, len);
6631         if (em)
6632                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6633         read_unlock(&em_tree->lock);
6634
6635         if (em) {
6636                 if (em->start > start || em->start + em->len <= start)
6637                         free_extent_map(em);
6638                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6639                         free_extent_map(em);
6640                 else
6641                         goto out;
6642         }
6643         em = alloc_extent_map();
6644         if (!em) {
6645                 err = -ENOMEM;
6646                 goto out;
6647         }
6648         em->bdev = root->fs_info->fs_devices->latest_bdev;
6649         em->start = EXTENT_MAP_HOLE;
6650         em->orig_start = EXTENT_MAP_HOLE;
6651         em->len = (u64)-1;
6652         em->block_len = (u64)-1;
6653
6654         if (!path) {
6655                 path = btrfs_alloc_path();
6656                 if (!path) {
6657                         err = -ENOMEM;
6658                         goto out;
6659                 }
6660                 /*
6661                  * Chances are we'll be called again, so go ahead and do
6662                  * readahead
6663                  */
6664                 path->reada = 1;
6665         }
6666
6667         ret = btrfs_lookup_file_extent(trans, root, path,
6668                                        objectid, start, trans != NULL);
6669         if (ret < 0) {
6670                 err = ret;
6671                 goto out;
6672         }
6673
6674         if (ret != 0) {
6675                 if (path->slots[0] == 0)
6676                         goto not_found;
6677                 path->slots[0]--;
6678         }
6679
6680         leaf = path->nodes[0];
6681         item = btrfs_item_ptr(leaf, path->slots[0],
6682                               struct btrfs_file_extent_item);
6683         /* are we inside the extent that was found? */
6684         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6685         found_type = found_key.type;
6686         if (found_key.objectid != objectid ||
6687             found_type != BTRFS_EXTENT_DATA_KEY) {
6688                 /*
6689                  * If we backup past the first extent we want to move forward
6690                  * and see if there is an extent in front of us, otherwise we'll
6691                  * say there is a hole for our whole search range which can
6692                  * cause problems.
6693                  */
6694                 extent_end = start;
6695                 goto next;
6696         }
6697
6698         found_type = btrfs_file_extent_type(leaf, item);
6699         extent_start = found_key.offset;
6700         if (found_type == BTRFS_FILE_EXTENT_REG ||
6701             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6702                 extent_end = extent_start +
6703                        btrfs_file_extent_num_bytes(leaf, item);
6704         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6705                 size_t size;
6706                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6707                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6708         }
6709 next:
6710         if (start >= extent_end) {
6711                 path->slots[0]++;
6712                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6713                         ret = btrfs_next_leaf(root, path);
6714                         if (ret < 0) {
6715                                 err = ret;
6716                                 goto out;
6717                         }
6718                         if (ret > 0)
6719                                 goto not_found;
6720                         leaf = path->nodes[0];
6721                 }
6722                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6723                 if (found_key.objectid != objectid ||
6724                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6725                         goto not_found;
6726                 if (start + len <= found_key.offset)
6727                         goto not_found;
6728                 if (start > found_key.offset)
6729                         goto next;
6730                 em->start = start;
6731                 em->orig_start = start;
6732                 em->len = found_key.offset - start;
6733                 goto not_found_em;
6734         }
6735
6736         btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6737
6738         if (found_type == BTRFS_FILE_EXTENT_REG ||
6739             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6740                 goto insert;
6741         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6742                 unsigned long ptr;
6743                 char *map;
6744                 size_t size;
6745                 size_t extent_offset;
6746                 size_t copy_size;
6747
6748                 if (new_inline)
6749                         goto out;
6750
6751                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6752                 extent_offset = page_offset(page) + pg_offset - extent_start;
6753                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6754                                 size - extent_offset);
6755                 em->start = extent_start + extent_offset;
6756                 em->len = ALIGN(copy_size, root->sectorsize);
6757                 em->orig_block_len = em->len;
6758                 em->orig_start = em->start;
6759                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6760                 if (create == 0 && !PageUptodate(page)) {
6761                         if (btrfs_file_extent_compression(leaf, item) !=
6762                             BTRFS_COMPRESS_NONE) {
6763                                 ret = uncompress_inline(path, inode, page,
6764                                                         pg_offset,
6765                                                         extent_offset, item);
6766                                 if (ret) {
6767                                         err = ret;
6768                                         goto out;
6769                                 }
6770                         } else {
6771                                 map = kmap(page);
6772                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6773                                                    copy_size);
6774                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6775                                         memset(map + pg_offset + copy_size, 0,
6776                                                PAGE_CACHE_SIZE - pg_offset -
6777                                                copy_size);
6778                                 }
6779                                 kunmap(page);
6780                         }
6781                         flush_dcache_page(page);
6782                 } else if (create && PageUptodate(page)) {
6783                         BUG();
6784                         if (!trans) {
6785                                 kunmap(page);
6786                                 free_extent_map(em);
6787                                 em = NULL;
6788
6789                                 btrfs_release_path(path);
6790                                 trans = btrfs_join_transaction(root);
6791
6792                                 if (IS_ERR(trans))
6793                                         return ERR_CAST(trans);
6794                                 goto again;
6795                         }
6796                         map = kmap(page);
6797                         write_extent_buffer(leaf, map + pg_offset, ptr,
6798                                             copy_size);
6799                         kunmap(page);
6800                         btrfs_mark_buffer_dirty(leaf);
6801                 }
6802                 set_extent_uptodate(io_tree, em->start,
6803                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6804                 goto insert;
6805         }
6806 not_found:
6807         em->start = start;
6808         em->orig_start = start;
6809         em->len = len;
6810 not_found_em:
6811         em->block_start = EXTENT_MAP_HOLE;
6812         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6813 insert:
6814         btrfs_release_path(path);
6815         if (em->start > start || extent_map_end(em) <= start) {
6816                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6817                         em->start, em->len, start, len);
6818                 err = -EIO;
6819                 goto out;
6820         }
6821
6822         err = 0;
6823         write_lock(&em_tree->lock);
6824         ret = add_extent_mapping(em_tree, em, 0);
6825         /* it is possible that someone inserted the extent into the tree
6826          * while we had the lock dropped.  It is also possible that
6827          * an overlapping map exists in the tree
6828          */
6829         if (ret == -EEXIST) {
6830                 struct extent_map *existing;
6831
6832                 ret = 0;
6833
6834                 existing = search_extent_mapping(em_tree, start, len);
6835                 /*
6836                  * existing will always be non-NULL, since there must be
6837                  * extent causing the -EEXIST.
6838                  */
6839                 if (start >= extent_map_end(existing) ||
6840                     start <= existing->start) {
6841                         /*
6842                          * The existing extent map is the one nearest to
6843                          * the [start, start + len) range which overlaps
6844                          */
6845                         err = merge_extent_mapping(em_tree, existing,
6846                                                    em, start);
6847                         free_extent_map(existing);
6848                         if (err) {
6849                                 free_extent_map(em);
6850                                 em = NULL;
6851                         }
6852                 } else {
6853                         free_extent_map(em);
6854                         em = existing;
6855                         err = 0;
6856                 }
6857         }
6858         write_unlock(&em_tree->lock);
6859 out:
6860
6861         trace_btrfs_get_extent(root, em);
6862
6863         if (path)
6864                 btrfs_free_path(path);
6865         if (trans) {
6866                 ret = btrfs_end_transaction(trans, root);
6867                 if (!err)
6868                         err = ret;
6869         }
6870         if (err) {
6871                 free_extent_map(em);
6872                 return ERR_PTR(err);
6873         }
6874         BUG_ON(!em); /* Error is always set */
6875         return em;
6876 }
6877
6878 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6879                                            size_t pg_offset, u64 start, u64 len,
6880                                            int create)
6881 {
6882         struct extent_map *em;
6883         struct extent_map *hole_em = NULL;
6884         u64 range_start = start;
6885         u64 end;
6886         u64 found;
6887         u64 found_end;
6888         int err = 0;
6889
6890         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6891         if (IS_ERR(em))
6892                 return em;
6893         if (em) {
6894                 /*
6895                  * if our em maps to
6896                  * -  a hole or
6897                  * -  a pre-alloc extent,
6898                  * there might actually be delalloc bytes behind it.
6899                  */
6900                 if (em->block_start != EXTENT_MAP_HOLE &&
6901                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6902                         return em;
6903                 else
6904                         hole_em = em;
6905         }
6906
6907         /* check to see if we've wrapped (len == -1 or similar) */
6908         end = start + len;
6909         if (end < start)
6910                 end = (u64)-1;
6911         else
6912                 end -= 1;
6913
6914         em = NULL;
6915
6916         /* ok, we didn't find anything, lets look for delalloc */
6917         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6918                                  end, len, EXTENT_DELALLOC, 1);
6919         found_end = range_start + found;
6920         if (found_end < range_start)
6921                 found_end = (u64)-1;
6922
6923         /*
6924          * we didn't find anything useful, return
6925          * the original results from get_extent()
6926          */
6927         if (range_start > end || found_end <= start) {
6928                 em = hole_em;
6929                 hole_em = NULL;
6930                 goto out;
6931         }
6932
6933         /* adjust the range_start to make sure it doesn't
6934          * go backwards from the start they passed in
6935          */
6936         range_start = max(start, range_start);
6937         found = found_end - range_start;
6938
6939         if (found > 0) {
6940                 u64 hole_start = start;
6941                 u64 hole_len = len;
6942
6943                 em = alloc_extent_map();
6944                 if (!em) {
6945                         err = -ENOMEM;
6946                         goto out;
6947                 }
6948                 /*
6949                  * when btrfs_get_extent can't find anything it
6950                  * returns one huge hole
6951                  *
6952                  * make sure what it found really fits our range, and
6953                  * adjust to make sure it is based on the start from
6954                  * the caller
6955                  */
6956                 if (hole_em) {
6957                         u64 calc_end = extent_map_end(hole_em);
6958
6959                         if (calc_end <= start || (hole_em->start > end)) {
6960                                 free_extent_map(hole_em);
6961                                 hole_em = NULL;
6962                         } else {
6963                                 hole_start = max(hole_em->start, start);
6964                                 hole_len = calc_end - hole_start;
6965                         }
6966                 }
6967                 em->bdev = NULL;
6968                 if (hole_em && range_start > hole_start) {
6969                         /* our hole starts before our delalloc, so we
6970                          * have to return just the parts of the hole
6971                          * that go until  the delalloc starts
6972                          */
6973                         em->len = min(hole_len,
6974                                       range_start - hole_start);
6975                         em->start = hole_start;
6976                         em->orig_start = hole_start;
6977                         /*
6978                          * don't adjust block start at all,
6979                          * it is fixed at EXTENT_MAP_HOLE
6980                          */
6981                         em->block_start = hole_em->block_start;
6982                         em->block_len = hole_len;
6983                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6984                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6985                 } else {
6986                         em->start = range_start;
6987                         em->len = found;
6988                         em->orig_start = range_start;
6989                         em->block_start = EXTENT_MAP_DELALLOC;
6990                         em->block_len = found;
6991                 }
6992         } else if (hole_em) {
6993                 return hole_em;
6994         }
6995 out:
6996
6997         free_extent_map(hole_em);
6998         if (err) {
6999                 free_extent_map(em);
7000                 return ERR_PTR(err);
7001         }
7002         return em;
7003 }
7004
7005 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7006                                                   u64 start, u64 len)
7007 {
7008         struct btrfs_root *root = BTRFS_I(inode)->root;
7009         struct extent_map *em;
7010         struct btrfs_key ins;
7011         u64 alloc_hint;
7012         int ret;
7013
7014         alloc_hint = get_extent_allocation_hint(inode, start, len);
7015         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7016                                    alloc_hint, &ins, 1, 1);
7017         if (ret)
7018                 return ERR_PTR(ret);
7019
7020         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7021                               ins.offset, ins.offset, ins.offset, 0);
7022         if (IS_ERR(em)) {
7023                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7024                 return em;
7025         }
7026
7027         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7028                                            ins.offset, ins.offset, 0);
7029         if (ret) {
7030                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7031                 free_extent_map(em);
7032                 return ERR_PTR(ret);
7033         }
7034
7035         return em;
7036 }
7037
7038 /*
7039  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7040  * block must be cow'd
7041  */
7042 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7043                               u64 *orig_start, u64 *orig_block_len,
7044                               u64 *ram_bytes)
7045 {
7046         struct btrfs_trans_handle *trans;
7047         struct btrfs_path *path;
7048         int ret;
7049         struct extent_buffer *leaf;
7050         struct btrfs_root *root = BTRFS_I(inode)->root;
7051         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7052         struct btrfs_file_extent_item *fi;
7053         struct btrfs_key key;
7054         u64 disk_bytenr;
7055         u64 backref_offset;
7056         u64 extent_end;
7057         u64 num_bytes;
7058         int slot;
7059         int found_type;
7060         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7061
7062         path = btrfs_alloc_path();
7063         if (!path)
7064                 return -ENOMEM;
7065
7066         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7067                                        offset, 0);
7068         if (ret < 0)
7069                 goto out;
7070
7071         slot = path->slots[0];
7072         if (ret == 1) {
7073                 if (slot == 0) {
7074                         /* can't find the item, must cow */
7075                         ret = 0;
7076                         goto out;
7077                 }
7078                 slot--;
7079         }
7080         ret = 0;
7081         leaf = path->nodes[0];
7082         btrfs_item_key_to_cpu(leaf, &key, slot);
7083         if (key.objectid != btrfs_ino(inode) ||
7084             key.type != BTRFS_EXTENT_DATA_KEY) {
7085                 /* not our file or wrong item type, must cow */
7086                 goto out;
7087         }
7088
7089         if (key.offset > offset) {
7090                 /* Wrong offset, must cow */
7091                 goto out;
7092         }
7093
7094         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7095         found_type = btrfs_file_extent_type(leaf, fi);
7096         if (found_type != BTRFS_FILE_EXTENT_REG &&
7097             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7098                 /* not a regular extent, must cow */
7099                 goto out;
7100         }
7101
7102         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7103                 goto out;
7104
7105         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7106         if (extent_end <= offset)
7107                 goto out;
7108
7109         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7110         if (disk_bytenr == 0)
7111                 goto out;
7112
7113         if (btrfs_file_extent_compression(leaf, fi) ||
7114             btrfs_file_extent_encryption(leaf, fi) ||
7115             btrfs_file_extent_other_encoding(leaf, fi))
7116                 goto out;
7117
7118         backref_offset = btrfs_file_extent_offset(leaf, fi);
7119
7120         if (orig_start) {
7121                 *orig_start = key.offset - backref_offset;
7122                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7123                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7124         }
7125
7126         if (btrfs_extent_readonly(root, disk_bytenr))
7127                 goto out;
7128
7129         num_bytes = min(offset + *len, extent_end) - offset;
7130         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7131                 u64 range_end;
7132
7133                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7134                 ret = test_range_bit(io_tree, offset, range_end,
7135                                      EXTENT_DELALLOC, 0, NULL);
7136                 if (ret) {
7137                         ret = -EAGAIN;
7138                         goto out;
7139                 }
7140         }
7141
7142         btrfs_release_path(path);
7143
7144         /*
7145          * look for other files referencing this extent, if we
7146          * find any we must cow
7147          */
7148         trans = btrfs_join_transaction(root);
7149         if (IS_ERR(trans)) {
7150                 ret = 0;
7151                 goto out;
7152         }
7153
7154         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7155                                     key.offset - backref_offset, disk_bytenr);
7156         btrfs_end_transaction(trans, root);
7157         if (ret) {
7158                 ret = 0;
7159                 goto out;
7160         }
7161
7162         /*
7163          * adjust disk_bytenr and num_bytes to cover just the bytes
7164          * in this extent we are about to write.  If there
7165          * are any csums in that range we have to cow in order
7166          * to keep the csums correct
7167          */
7168         disk_bytenr += backref_offset;
7169         disk_bytenr += offset - key.offset;
7170         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7171                                 goto out;
7172         /*
7173          * all of the above have passed, it is safe to overwrite this extent
7174          * without cow
7175          */
7176         *len = num_bytes;
7177         ret = 1;
7178 out:
7179         btrfs_free_path(path);
7180         return ret;
7181 }
7182
7183 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7184 {
7185         struct radix_tree_root *root = &inode->i_mapping->page_tree;
7186         int found = false;
7187         void **pagep = NULL;
7188         struct page *page = NULL;
7189         int start_idx;
7190         int end_idx;
7191
7192         start_idx = start >> PAGE_CACHE_SHIFT;
7193
7194         /*
7195          * end is the last byte in the last page.  end == start is legal
7196          */
7197         end_idx = end >> PAGE_CACHE_SHIFT;
7198
7199         rcu_read_lock();
7200
7201         /* Most of the code in this while loop is lifted from
7202          * find_get_page.  It's been modified to begin searching from a
7203          * page and return just the first page found in that range.  If the
7204          * found idx is less than or equal to the end idx then we know that
7205          * a page exists.  If no pages are found or if those pages are
7206          * outside of the range then we're fine (yay!) */
7207         while (page == NULL &&
7208                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7209                 page = radix_tree_deref_slot(pagep);
7210                 if (unlikely(!page))
7211                         break;
7212
7213                 if (radix_tree_exception(page)) {
7214                         if (radix_tree_deref_retry(page)) {
7215                                 page = NULL;
7216                                 continue;
7217                         }
7218                         /*
7219                          * Otherwise, shmem/tmpfs must be storing a swap entry
7220                          * here as an exceptional entry: so return it without
7221                          * attempting to raise page count.
7222                          */
7223                         page = NULL;
7224                         break; /* TODO: Is this relevant for this use case? */
7225                 }
7226
7227                 if (!page_cache_get_speculative(page)) {
7228                         page = NULL;
7229                         continue;
7230                 }
7231
7232                 /*
7233                  * Has the page moved?
7234                  * This is part of the lockless pagecache protocol. See
7235                  * include/linux/pagemap.h for details.
7236                  */
7237                 if (unlikely(page != *pagep)) {
7238                         page_cache_release(page);
7239                         page = NULL;
7240                 }
7241         }
7242
7243         if (page) {
7244                 if (page->index <= end_idx)
7245                         found = true;
7246                 page_cache_release(page);
7247         }
7248
7249         rcu_read_unlock();
7250         return found;
7251 }
7252
7253 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7254                               struct extent_state **cached_state, int writing)
7255 {
7256         struct btrfs_ordered_extent *ordered;
7257         int ret = 0;
7258
7259         while (1) {
7260                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7261                                  0, cached_state);
7262                 /*
7263                  * We're concerned with the entire range that we're going to be
7264                  * doing DIO to, so we need to make sure theres no ordered
7265                  * extents in this range.
7266                  */
7267                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7268                                                      lockend - lockstart + 1);
7269
7270                 /*
7271                  * We need to make sure there are no buffered pages in this
7272                  * range either, we could have raced between the invalidate in
7273                  * generic_file_direct_write and locking the extent.  The
7274                  * invalidate needs to happen so that reads after a write do not
7275                  * get stale data.
7276                  */
7277                 if (!ordered &&
7278                     (!writing ||
7279                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7280                         break;
7281
7282                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7283                                      cached_state, GFP_NOFS);
7284
7285                 if (ordered) {
7286                         btrfs_start_ordered_extent(inode, ordered, 1);
7287                         btrfs_put_ordered_extent(ordered);
7288                 } else {
7289                         /* Screw you mmap */
7290                         ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7291                         if (ret)
7292                                 break;
7293                         ret = filemap_fdatawait_range(inode->i_mapping,
7294                                                       lockstart,
7295                                                       lockend);
7296                         if (ret)
7297                                 break;
7298
7299                         /*
7300                          * If we found a page that couldn't be invalidated just
7301                          * fall back to buffered.
7302                          */
7303                         ret = invalidate_inode_pages2_range(inode->i_mapping,
7304                                         lockstart >> PAGE_CACHE_SHIFT,
7305                                         lockend >> PAGE_CACHE_SHIFT);
7306                         if (ret)
7307                                 break;
7308                 }
7309
7310                 cond_resched();
7311         }
7312
7313         return ret;
7314 }
7315
7316 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7317                                            u64 len, u64 orig_start,
7318                                            u64 block_start, u64 block_len,
7319                                            u64 orig_block_len, u64 ram_bytes,
7320                                            int type)
7321 {
7322         struct extent_map_tree *em_tree;
7323         struct extent_map *em;
7324         struct btrfs_root *root = BTRFS_I(inode)->root;
7325         int ret;
7326
7327         em_tree = &BTRFS_I(inode)->extent_tree;
7328         em = alloc_extent_map();
7329         if (!em)
7330                 return ERR_PTR(-ENOMEM);
7331
7332         em->start = start;
7333         em->orig_start = orig_start;
7334         em->mod_start = start;
7335         em->mod_len = len;
7336         em->len = len;
7337         em->block_len = block_len;
7338         em->block_start = block_start;
7339         em->bdev = root->fs_info->fs_devices->latest_bdev;
7340         em->orig_block_len = orig_block_len;
7341         em->ram_bytes = ram_bytes;
7342         em->generation = -1;
7343         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7344         if (type == BTRFS_ORDERED_PREALLOC)
7345                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7346
7347         do {
7348                 btrfs_drop_extent_cache(inode, em->start,
7349                                 em->start + em->len - 1, 0);
7350                 write_lock(&em_tree->lock);
7351                 ret = add_extent_mapping(em_tree, em, 1);
7352                 write_unlock(&em_tree->lock);
7353         } while (ret == -EEXIST);
7354
7355         if (ret) {
7356                 free_extent_map(em);
7357                 return ERR_PTR(ret);
7358         }
7359
7360         return em;
7361 }
7362
7363
7364 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7365                                    struct buffer_head *bh_result, int create)
7366 {
7367         struct extent_map *em;
7368         struct btrfs_root *root = BTRFS_I(inode)->root;
7369         struct extent_state *cached_state = NULL;
7370         u64 start = iblock << inode->i_blkbits;
7371         u64 lockstart, lockend;
7372         u64 len = bh_result->b_size;
7373         u64 *outstanding_extents = NULL;
7374         int unlock_bits = EXTENT_LOCKED;
7375         int ret = 0;
7376
7377         if (create)
7378                 unlock_bits |= EXTENT_DIRTY;
7379         else
7380                 len = min_t(u64, len, root->sectorsize);
7381
7382         lockstart = start;
7383         lockend = start + len - 1;
7384
7385         if (current->journal_info) {
7386                 /*
7387                  * Need to pull our outstanding extents and set journal_info to NULL so
7388                  * that anything that needs to check if there's a transction doesn't get
7389                  * confused.
7390                  */
7391                 outstanding_extents = current->journal_info;
7392                 current->journal_info = NULL;
7393         }
7394
7395         /*
7396          * If this errors out it's because we couldn't invalidate pagecache for
7397          * this range and we need to fallback to buffered.
7398          */
7399         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7400                 return -ENOTBLK;
7401
7402         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7403         if (IS_ERR(em)) {
7404                 ret = PTR_ERR(em);
7405                 goto unlock_err;
7406         }
7407
7408         /*
7409          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7410          * io.  INLINE is special, and we could probably kludge it in here, but
7411          * it's still buffered so for safety lets just fall back to the generic
7412          * buffered path.
7413          *
7414          * For COMPRESSED we _have_ to read the entire extent in so we can
7415          * decompress it, so there will be buffering required no matter what we
7416          * do, so go ahead and fallback to buffered.
7417          *
7418          * We return -ENOTBLK because thats what makes DIO go ahead and go back
7419          * to buffered IO.  Don't blame me, this is the price we pay for using
7420          * the generic code.
7421          */
7422         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7423             em->block_start == EXTENT_MAP_INLINE) {
7424                 free_extent_map(em);
7425                 ret = -ENOTBLK;
7426                 goto unlock_err;
7427         }
7428
7429         /* Just a good old fashioned hole, return */
7430         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7431                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7432                 free_extent_map(em);
7433                 goto unlock_err;
7434         }
7435
7436         /*
7437          * We don't allocate a new extent in the following cases
7438          *
7439          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7440          * existing extent.
7441          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7442          * just use the extent.
7443          *
7444          */
7445         if (!create) {
7446                 len = min(len, em->len - (start - em->start));
7447                 lockstart = start + len;
7448                 goto unlock;
7449         }
7450
7451         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7452             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7453              em->block_start != EXTENT_MAP_HOLE)) {
7454                 int type;
7455                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7456
7457                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7458                         type = BTRFS_ORDERED_PREALLOC;
7459                 else
7460                         type = BTRFS_ORDERED_NOCOW;
7461                 len = min(len, em->len - (start - em->start));
7462                 block_start = em->block_start + (start - em->start);
7463
7464                 if (can_nocow_extent(inode, start, &len, &orig_start,
7465                                      &orig_block_len, &ram_bytes) == 1) {
7466                         if (type == BTRFS_ORDERED_PREALLOC) {
7467                                 free_extent_map(em);
7468                                 em = create_pinned_em(inode, start, len,
7469                                                        orig_start,
7470                                                        block_start, len,
7471                                                        orig_block_len,
7472                                                        ram_bytes, type);
7473                                 if (IS_ERR(em)) {
7474                                         ret = PTR_ERR(em);
7475                                         goto unlock_err;
7476                                 }
7477                         }
7478
7479                         ret = btrfs_add_ordered_extent_dio(inode, start,
7480                                            block_start, len, len, type);
7481                         if (ret) {
7482                                 free_extent_map(em);
7483                                 goto unlock_err;
7484                         }
7485                         goto unlock;
7486                 }
7487         }
7488
7489         /*
7490          * this will cow the extent, reset the len in case we changed
7491          * it above
7492          */
7493         len = bh_result->b_size;
7494         free_extent_map(em);
7495         em = btrfs_new_extent_direct(inode, start, len);
7496         if (IS_ERR(em)) {
7497                 ret = PTR_ERR(em);
7498                 goto unlock_err;
7499         }
7500         len = min(len, em->len - (start - em->start));
7501 unlock:
7502         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7503                 inode->i_blkbits;
7504         bh_result->b_size = len;
7505         bh_result->b_bdev = em->bdev;
7506         set_buffer_mapped(bh_result);
7507         if (create) {
7508                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7509                         set_buffer_new(bh_result);
7510
7511                 /*
7512                  * Need to update the i_size under the extent lock so buffered
7513                  * readers will get the updated i_size when we unlock.
7514                  */
7515                 if (start + len > i_size_read(inode))
7516                         i_size_write(inode, start + len);
7517
7518                 /*
7519                  * If we have an outstanding_extents count still set then we're
7520                  * within our reservation, otherwise we need to adjust our inode
7521                  * counter appropriately.
7522                  */
7523                 if (*outstanding_extents) {
7524                         (*outstanding_extents)--;
7525                 } else {
7526                         spin_lock(&BTRFS_I(inode)->lock);
7527                         BTRFS_I(inode)->outstanding_extents++;
7528                         spin_unlock(&BTRFS_I(inode)->lock);
7529                 }
7530
7531                 current->journal_info = outstanding_extents;
7532                 btrfs_free_reserved_data_space(inode, len);
7533         }
7534
7535         /*
7536          * In the case of write we need to clear and unlock the entire range,
7537          * in the case of read we need to unlock only the end area that we
7538          * aren't using if there is any left over space.
7539          */
7540         if (lockstart < lockend) {
7541                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7542                                  lockend, unlock_bits, 1, 0,
7543                                  &cached_state, GFP_NOFS);
7544         } else {
7545                 free_extent_state(cached_state);
7546         }
7547
7548         free_extent_map(em);
7549
7550         return 0;
7551
7552 unlock_err:
7553         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7554                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7555         if (outstanding_extents)
7556                 current->journal_info = outstanding_extents;
7557         return ret;
7558 }
7559
7560 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7561                                         int rw, int mirror_num)
7562 {
7563         struct btrfs_root *root = BTRFS_I(inode)->root;
7564         int ret;
7565
7566         BUG_ON(rw & REQ_WRITE);
7567
7568         bio_get(bio);
7569
7570         ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7571                                   BTRFS_WQ_ENDIO_DIO_REPAIR);
7572         if (ret)
7573                 goto err;
7574
7575         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7576 err:
7577         bio_put(bio);
7578         return ret;
7579 }
7580
7581 static int btrfs_check_dio_repairable(struct inode *inode,
7582                                       struct bio *failed_bio,
7583                                       struct io_failure_record *failrec,
7584                                       int failed_mirror)
7585 {
7586         int num_copies;
7587
7588         num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7589                                       failrec->logical, failrec->len);
7590         if (num_copies == 1) {
7591                 /*
7592                  * we only have a single copy of the data, so don't bother with
7593                  * all the retry and error correction code that follows. no
7594                  * matter what the error is, it is very likely to persist.
7595                  */
7596                 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7597                          num_copies, failrec->this_mirror, failed_mirror);
7598                 return 0;
7599         }
7600
7601         failrec->failed_mirror = failed_mirror;
7602         failrec->this_mirror++;
7603         if (failrec->this_mirror == failed_mirror)
7604                 failrec->this_mirror++;
7605
7606         if (failrec->this_mirror > num_copies) {
7607                 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7608                          num_copies, failrec->this_mirror, failed_mirror);
7609                 return 0;
7610         }
7611
7612         return 1;
7613 }
7614
7615 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7616                           struct page *page, u64 start, u64 end,
7617                           int failed_mirror, bio_end_io_t *repair_endio,
7618                           void *repair_arg)
7619 {
7620         struct io_failure_record *failrec;
7621         struct bio *bio;
7622         int isector;
7623         int read_mode;
7624         int ret;
7625
7626         BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7627
7628         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7629         if (ret)
7630                 return ret;
7631
7632         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7633                                          failed_mirror);
7634         if (!ret) {
7635                 free_io_failure(inode, failrec);
7636                 return -EIO;
7637         }
7638
7639         if (failed_bio->bi_vcnt > 1)
7640                 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7641         else
7642                 read_mode = READ_SYNC;
7643
7644         isector = start - btrfs_io_bio(failed_bio)->logical;
7645         isector >>= inode->i_sb->s_blocksize_bits;
7646         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7647                                       0, isector, repair_endio, repair_arg);
7648         if (!bio) {
7649                 free_io_failure(inode, failrec);
7650                 return -EIO;
7651         }
7652
7653         btrfs_debug(BTRFS_I(inode)->root->fs_info,
7654                     "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7655                     read_mode, failrec->this_mirror, failrec->in_validation);
7656
7657         ret = submit_dio_repair_bio(inode, bio, read_mode,
7658                                     failrec->this_mirror);
7659         if (ret) {
7660                 free_io_failure(inode, failrec);
7661                 bio_put(bio);
7662         }
7663
7664         return ret;
7665 }
7666
7667 struct btrfs_retry_complete {
7668         struct completion done;
7669         struct inode *inode;
7670         u64 start;
7671         int uptodate;
7672 };
7673
7674 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7675 {
7676         struct btrfs_retry_complete *done = bio->bi_private;
7677         struct bio_vec *bvec;
7678         int i;
7679
7680         if (err)
7681                 goto end;
7682
7683         done->uptodate = 1;
7684         bio_for_each_segment_all(bvec, bio, i)
7685                 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7686 end:
7687         complete(&done->done);
7688         bio_put(bio);
7689 }
7690
7691 static int __btrfs_correct_data_nocsum(struct inode *inode,
7692                                        struct btrfs_io_bio *io_bio)
7693 {
7694         struct bio_vec *bvec;
7695         struct btrfs_retry_complete done;
7696         u64 start;
7697         int i;
7698         int ret;
7699
7700         start = io_bio->logical;
7701         done.inode = inode;
7702
7703         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7704 try_again:
7705                 done.uptodate = 0;
7706                 done.start = start;
7707                 init_completion(&done.done);
7708
7709                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7710                                      start + bvec->bv_len - 1,
7711                                      io_bio->mirror_num,
7712                                      btrfs_retry_endio_nocsum, &done);
7713                 if (ret)
7714                         return ret;
7715
7716                 wait_for_completion(&done.done);
7717
7718                 if (!done.uptodate) {
7719                         /* We might have another mirror, so try again */
7720                         goto try_again;
7721                 }
7722
7723                 start += bvec->bv_len;
7724         }
7725
7726         return 0;
7727 }
7728
7729 static void btrfs_retry_endio(struct bio *bio, int err)
7730 {
7731         struct btrfs_retry_complete *done = bio->bi_private;
7732         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7733         struct bio_vec *bvec;
7734         int uptodate;
7735         int ret;
7736         int i;
7737
7738         if (err)
7739                 goto end;
7740
7741         uptodate = 1;
7742         bio_for_each_segment_all(bvec, bio, i) {
7743                 ret = __readpage_endio_check(done->inode, io_bio, i,
7744                                              bvec->bv_page, 0,
7745                                              done->start, bvec->bv_len);
7746                 if (!ret)
7747                         clean_io_failure(done->inode, done->start,
7748                                          bvec->bv_page, 0);
7749                 else
7750                         uptodate = 0;
7751         }
7752
7753         done->uptodate = uptodate;
7754 end:
7755         complete(&done->done);
7756         bio_put(bio);
7757 }
7758
7759 static int __btrfs_subio_endio_read(struct inode *inode,
7760                                     struct btrfs_io_bio *io_bio, int err)
7761 {
7762         struct bio_vec *bvec;
7763         struct btrfs_retry_complete done;
7764         u64 start;
7765         u64 offset = 0;
7766         int i;
7767         int ret;
7768
7769         err = 0;
7770         start = io_bio->logical;
7771         done.inode = inode;
7772
7773         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7774                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7775                                              0, start, bvec->bv_len);
7776                 if (likely(!ret))
7777                         goto next;
7778 try_again:
7779                 done.uptodate = 0;
7780                 done.start = start;
7781                 init_completion(&done.done);
7782
7783                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7784                                      start + bvec->bv_len - 1,
7785                                      io_bio->mirror_num,
7786                                      btrfs_retry_endio, &done);
7787                 if (ret) {
7788                         err = ret;
7789                         goto next;
7790                 }
7791
7792                 wait_for_completion(&done.done);
7793
7794                 if (!done.uptodate) {
7795                         /* We might have another mirror, so try again */
7796                         goto try_again;
7797                 }
7798 next:
7799                 offset += bvec->bv_len;
7800                 start += bvec->bv_len;
7801         }
7802
7803         return err;
7804 }
7805
7806 static int btrfs_subio_endio_read(struct inode *inode,
7807                                   struct btrfs_io_bio *io_bio, int err)
7808 {
7809         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7810
7811         if (skip_csum) {
7812                 if (unlikely(err))
7813                         return __btrfs_correct_data_nocsum(inode, io_bio);
7814                 else
7815                         return 0;
7816         } else {
7817                 return __btrfs_subio_endio_read(inode, io_bio, err);
7818         }
7819 }
7820
7821 static void btrfs_endio_direct_read(struct bio *bio, int err)
7822 {
7823         struct btrfs_dio_private *dip = bio->bi_private;
7824         struct inode *inode = dip->inode;
7825         struct bio *dio_bio;
7826         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7827
7828         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7829                 err = btrfs_subio_endio_read(inode, io_bio, err);
7830
7831         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7832                       dip->logical_offset + dip->bytes - 1);
7833         dio_bio = dip->dio_bio;
7834
7835         kfree(dip);
7836
7837         /* If we had a csum failure make sure to clear the uptodate flag */
7838         if (err)
7839                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7840         dio_end_io(dio_bio, err);
7841
7842         if (io_bio->end_io)
7843                 io_bio->end_io(io_bio, err);
7844         bio_put(bio);
7845 }
7846
7847 static void btrfs_endio_direct_write(struct bio *bio, int err)
7848 {
7849         struct btrfs_dio_private *dip = bio->bi_private;
7850         struct inode *inode = dip->inode;
7851         struct btrfs_root *root = BTRFS_I(inode)->root;
7852         struct btrfs_ordered_extent *ordered = NULL;
7853         u64 ordered_offset = dip->logical_offset;
7854         u64 ordered_bytes = dip->bytes;
7855         struct bio *dio_bio;
7856         int ret;
7857
7858         if (err)
7859                 goto out_done;
7860 again:
7861         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7862                                                    &ordered_offset,
7863                                                    ordered_bytes, !err);
7864         if (!ret)
7865                 goto out_test;
7866
7867         btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7868                         finish_ordered_fn, NULL, NULL);
7869         btrfs_queue_work(root->fs_info->endio_write_workers,
7870                          &ordered->work);
7871 out_test:
7872         /*
7873          * our bio might span multiple ordered extents.  If we haven't
7874          * completed the accounting for the whole dio, go back and try again
7875          */
7876         if (ordered_offset < dip->logical_offset + dip->bytes) {
7877                 ordered_bytes = dip->logical_offset + dip->bytes -
7878                         ordered_offset;
7879                 ordered = NULL;
7880                 goto again;
7881         }
7882 out_done:
7883         dio_bio = dip->dio_bio;
7884
7885         kfree(dip);
7886
7887         /* If we had an error make sure to clear the uptodate flag */
7888         if (err)
7889                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7890         dio_end_io(dio_bio, err);
7891         bio_put(bio);
7892 }
7893
7894 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7895                                     struct bio *bio, int mirror_num,
7896                                     unsigned long bio_flags, u64 offset)
7897 {
7898         int ret;
7899         struct btrfs_root *root = BTRFS_I(inode)->root;
7900         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7901         BUG_ON(ret); /* -ENOMEM */
7902         return 0;
7903 }
7904
7905 static void btrfs_end_dio_bio(struct bio *bio, int err)
7906 {
7907         struct btrfs_dio_private *dip = bio->bi_private;
7908
7909         if (err)
7910                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7911                            "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7912                            btrfs_ino(dip->inode), bio->bi_rw,
7913                            (unsigned long long)bio->bi_iter.bi_sector,
7914                            bio->bi_iter.bi_size, err);
7915
7916         if (dip->subio_endio)
7917                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7918
7919         if (err) {
7920                 dip->errors = 1;
7921
7922                 /*
7923                  * before atomic variable goto zero, we must make sure
7924                  * dip->errors is perceived to be set.
7925                  */
7926                 smp_mb__before_atomic();
7927         }
7928
7929         /* if there are more bios still pending for this dio, just exit */
7930         if (!atomic_dec_and_test(&dip->pending_bios))
7931                 goto out;
7932
7933         if (dip->errors) {
7934                 bio_io_error(dip->orig_bio);
7935         } else {
7936                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7937                 bio_endio(dip->orig_bio, 0);
7938         }
7939 out:
7940         bio_put(bio);
7941 }
7942
7943 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7944                                        u64 first_sector, gfp_t gfp_flags)
7945 {
7946         int nr_vecs = bio_get_nr_vecs(bdev);
7947         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7948 }
7949
7950 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7951                                                  struct inode *inode,
7952                                                  struct btrfs_dio_private *dip,
7953                                                  struct bio *bio,
7954                                                  u64 file_offset)
7955 {
7956         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7957         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7958         int ret;
7959
7960         /*
7961          * We load all the csum data we need when we submit
7962          * the first bio to reduce the csum tree search and
7963          * contention.
7964          */
7965         if (dip->logical_offset == file_offset) {
7966                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7967                                                 file_offset);
7968                 if (ret)
7969                         return ret;
7970         }
7971
7972         if (bio == dip->orig_bio)
7973                 return 0;
7974
7975         file_offset -= dip->logical_offset;
7976         file_offset >>= inode->i_sb->s_blocksize_bits;
7977         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7978
7979         return 0;
7980 }
7981
7982 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7983                                          int rw, u64 file_offset, int skip_sum,
7984                                          int async_submit)
7985 {
7986         struct btrfs_dio_private *dip = bio->bi_private;
7987         int write = rw & REQ_WRITE;
7988         struct btrfs_root *root = BTRFS_I(inode)->root;
7989         int ret;
7990
7991         if (async_submit)
7992                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7993
7994         bio_get(bio);
7995
7996         if (!write) {
7997                 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7998                                 BTRFS_WQ_ENDIO_DATA);
7999                 if (ret)
8000                         goto err;
8001         }
8002
8003         if (skip_sum)
8004                 goto map;
8005
8006         if (write && async_submit) {
8007                 ret = btrfs_wq_submit_bio(root->fs_info,
8008                                    inode, rw, bio, 0, 0,
8009                                    file_offset,
8010                                    __btrfs_submit_bio_start_direct_io,
8011                                    __btrfs_submit_bio_done);
8012                 goto err;
8013         } else if (write) {
8014                 /*
8015                  * If we aren't doing async submit, calculate the csum of the
8016                  * bio now.
8017                  */
8018                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8019                 if (ret)
8020                         goto err;
8021         } else {
8022                 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8023                                                      file_offset);
8024                 if (ret)
8025                         goto err;
8026         }
8027 map:
8028         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8029 err:
8030         bio_put(bio);
8031         return ret;
8032 }
8033
8034 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8035                                     int skip_sum)
8036 {
8037         struct inode *inode = dip->inode;
8038         struct btrfs_root *root = BTRFS_I(inode)->root;
8039         struct bio *bio;
8040         struct bio *orig_bio = dip->orig_bio;
8041         struct bio_vec *bvec = orig_bio->bi_io_vec;
8042         u64 start_sector = orig_bio->bi_iter.bi_sector;
8043         u64 file_offset = dip->logical_offset;
8044         u64 submit_len = 0;
8045         u64 map_length;
8046         int nr_pages = 0;
8047         int ret;
8048         int async_submit = 0;
8049
8050         map_length = orig_bio->bi_iter.bi_size;
8051         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8052                               &map_length, NULL, 0);
8053         if (ret)
8054                 return -EIO;
8055
8056         if (map_length >= orig_bio->bi_iter.bi_size) {
8057                 bio = orig_bio;
8058                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8059                 goto submit;
8060         }
8061
8062         /* async crcs make it difficult to collect full stripe writes. */
8063         if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8064                 async_submit = 0;
8065         else
8066                 async_submit = 1;
8067
8068         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8069         if (!bio)
8070                 return -ENOMEM;
8071
8072         bio->bi_private = dip;
8073         bio->bi_end_io = btrfs_end_dio_bio;
8074         btrfs_io_bio(bio)->logical = file_offset;
8075         atomic_inc(&dip->pending_bios);
8076
8077         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8078                 if (map_length < submit_len + bvec->bv_len ||
8079                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8080                                  bvec->bv_offset) < bvec->bv_len) {
8081                         /*
8082                          * inc the count before we submit the bio so
8083                          * we know the end IO handler won't happen before
8084                          * we inc the count. Otherwise, the dip might get freed
8085                          * before we're done setting it up
8086                          */
8087                         atomic_inc(&dip->pending_bios);
8088                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
8089                                                      file_offset, skip_sum,
8090                                                      async_submit);
8091                         if (ret) {
8092                                 bio_put(bio);
8093                                 atomic_dec(&dip->pending_bios);
8094                                 goto out_err;
8095                         }
8096
8097                         start_sector += submit_len >> 9;
8098                         file_offset += submit_len;
8099
8100                         submit_len = 0;
8101                         nr_pages = 0;
8102
8103                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8104                                                   start_sector, GFP_NOFS);
8105                         if (!bio)
8106                                 goto out_err;
8107                         bio->bi_private = dip;
8108                         bio->bi_end_io = btrfs_end_dio_bio;
8109                         btrfs_io_bio(bio)->logical = file_offset;
8110
8111                         map_length = orig_bio->bi_iter.bi_size;
8112                         ret = btrfs_map_block(root->fs_info, rw,
8113                                               start_sector << 9,
8114                                               &map_length, NULL, 0);
8115                         if (ret) {
8116                                 bio_put(bio);
8117                                 goto out_err;
8118                         }
8119                 } else {
8120                         submit_len += bvec->bv_len;
8121                         nr_pages++;
8122                         bvec++;
8123                 }
8124         }
8125
8126 submit:
8127         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8128                                      async_submit);
8129         if (!ret)
8130                 return 0;
8131
8132         bio_put(bio);
8133 out_err:
8134         dip->errors = 1;
8135         /*
8136          * before atomic variable goto zero, we must
8137          * make sure dip->errors is perceived to be set.
8138          */
8139         smp_mb__before_atomic();
8140         if (atomic_dec_and_test(&dip->pending_bios))
8141                 bio_io_error(dip->orig_bio);
8142
8143         /* bio_end_io() will handle error, so we needn't return it */
8144         return 0;
8145 }
8146
8147 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8148                                 struct inode *inode, loff_t file_offset)
8149 {
8150         struct btrfs_root *root = BTRFS_I(inode)->root;
8151         struct btrfs_dio_private *dip;
8152         struct bio *io_bio;
8153         struct btrfs_io_bio *btrfs_bio;
8154         int skip_sum;
8155         int write = rw & REQ_WRITE;
8156         int ret = 0;
8157
8158         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8159
8160         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8161         if (!io_bio) {
8162                 ret = -ENOMEM;
8163                 goto free_ordered;
8164         }
8165
8166         dip = kzalloc(sizeof(*dip), GFP_NOFS);
8167         if (!dip) {
8168                 ret = -ENOMEM;
8169                 goto free_io_bio;
8170         }
8171
8172         dip->private = dio_bio->bi_private;
8173         dip->inode = inode;
8174         dip->logical_offset = file_offset;
8175         dip->bytes = dio_bio->bi_iter.bi_size;
8176         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8177         io_bio->bi_private = dip;
8178         dip->orig_bio = io_bio;
8179         dip->dio_bio = dio_bio;
8180         atomic_set(&dip->pending_bios, 0);
8181         btrfs_bio = btrfs_io_bio(io_bio);
8182         btrfs_bio->logical = file_offset;
8183
8184         if (write) {
8185                 io_bio->bi_end_io = btrfs_endio_direct_write;
8186         } else {
8187                 io_bio->bi_end_io = btrfs_endio_direct_read;
8188                 dip->subio_endio = btrfs_subio_endio_read;
8189         }
8190
8191         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8192         if (!ret)
8193                 return;
8194
8195         if (btrfs_bio->end_io)
8196                 btrfs_bio->end_io(btrfs_bio, ret);
8197 free_io_bio:
8198         bio_put(io_bio);
8199
8200 free_ordered:
8201         /*
8202          * If this is a write, we need to clean up the reserved space and kill
8203          * the ordered extent.
8204          */
8205         if (write) {
8206                 struct btrfs_ordered_extent *ordered;
8207                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
8208                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
8209                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
8210                         btrfs_free_reserved_extent(root, ordered->start,
8211                                                    ordered->disk_len, 1);
8212                 btrfs_put_ordered_extent(ordered);
8213                 btrfs_put_ordered_extent(ordered);
8214         }
8215         bio_endio(dio_bio, ret);
8216 }
8217
8218 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8219                         const struct iov_iter *iter, loff_t offset)
8220 {
8221         int seg;
8222         int i;
8223         unsigned blocksize_mask = root->sectorsize - 1;
8224         ssize_t retval = -EINVAL;
8225
8226         if (offset & blocksize_mask)
8227                 goto out;
8228
8229         if (iov_iter_alignment(iter) & blocksize_mask)
8230                 goto out;
8231
8232         /* If this is a write we don't need to check anymore */
8233         if (iov_iter_rw(iter) == WRITE)
8234                 return 0;
8235         /*
8236          * Check to make sure we don't have duplicate iov_base's in this
8237          * iovec, if so return EINVAL, otherwise we'll get csum errors
8238          * when reading back.
8239          */
8240         for (seg = 0; seg < iter->nr_segs; seg++) {
8241                 for (i = seg + 1; i < iter->nr_segs; i++) {
8242                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8243                                 goto out;
8244                 }
8245         }
8246         retval = 0;
8247 out:
8248         return retval;
8249 }
8250
8251 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8252                                loff_t offset)
8253 {
8254         struct file *file = iocb->ki_filp;
8255         struct inode *inode = file->f_mapping->host;
8256         u64 outstanding_extents = 0;
8257         size_t count = 0;
8258         int flags = 0;
8259         bool wakeup = true;
8260         bool relock = false;
8261         ssize_t ret;
8262
8263         if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8264                 return 0;
8265
8266         inode_dio_begin(inode);
8267         smp_mb__after_atomic();
8268
8269         /*
8270          * The generic stuff only does filemap_write_and_wait_range, which
8271          * isn't enough if we've written compressed pages to this area, so
8272          * we need to flush the dirty pages again to make absolutely sure
8273          * that any outstanding dirty pages are on disk.
8274          */
8275         count = iov_iter_count(iter);
8276         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8277                      &BTRFS_I(inode)->runtime_flags))
8278                 filemap_fdatawrite_range(inode->i_mapping, offset,
8279                                          offset + count - 1);
8280
8281         if (iov_iter_rw(iter) == WRITE) {
8282                 /*
8283                  * If the write DIO is beyond the EOF, we need update
8284                  * the isize, but it is protected by i_mutex. So we can
8285                  * not unlock the i_mutex at this case.
8286                  */
8287                 if (offset + count <= inode->i_size) {
8288                         mutex_unlock(&inode->i_mutex);
8289                         relock = true;
8290                 }
8291                 ret = btrfs_delalloc_reserve_space(inode, count);
8292                 if (ret)
8293                         goto out;
8294                 outstanding_extents = div64_u64(count +
8295                                                 BTRFS_MAX_EXTENT_SIZE - 1,
8296                                                 BTRFS_MAX_EXTENT_SIZE);
8297
8298                 /*
8299                  * We need to know how many extents we reserved so that we can
8300                  * do the accounting properly if we go over the number we
8301                  * originally calculated.  Abuse current->journal_info for this.
8302                  */
8303                 current->journal_info = &outstanding_extents;
8304         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8305                                      &BTRFS_I(inode)->runtime_flags)) {
8306                 inode_dio_end(inode);
8307                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8308                 wakeup = false;
8309         }
8310
8311         ret = __blockdev_direct_IO(iocb, inode,
8312                                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8313                                    iter, offset, btrfs_get_blocks_direct, NULL,
8314                                    btrfs_submit_direct, flags);
8315         if (iov_iter_rw(iter) == WRITE) {
8316                 current->journal_info = NULL;
8317                 if (ret < 0 && ret != -EIOCBQUEUED)
8318                         btrfs_delalloc_release_space(inode, count);
8319                 else if (ret >= 0 && (size_t)ret < count)
8320                         btrfs_delalloc_release_space(inode,
8321                                                      count - (size_t)ret);
8322         }
8323 out:
8324         if (wakeup)
8325                 inode_dio_end(inode);
8326         if (relock)
8327                 mutex_lock(&inode->i_mutex);
8328
8329         return ret;
8330 }
8331
8332 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8333
8334 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8335                 __u64 start, __u64 len)
8336 {
8337         int     ret;
8338
8339         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8340         if (ret)
8341                 return ret;
8342
8343         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8344 }
8345
8346 int btrfs_readpage(struct file *file, struct page *page)
8347 {
8348         struct extent_io_tree *tree;
8349         tree = &BTRFS_I(page->mapping->host)->io_tree;
8350         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8351 }
8352
8353 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8354 {
8355         struct extent_io_tree *tree;
8356
8357
8358         if (current->flags & PF_MEMALLOC) {
8359                 redirty_page_for_writepage(wbc, page);
8360                 unlock_page(page);
8361                 return 0;
8362         }
8363         tree = &BTRFS_I(page->mapping->host)->io_tree;
8364         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8365 }
8366
8367 static int btrfs_writepages(struct address_space *mapping,
8368                             struct writeback_control *wbc)
8369 {
8370         struct extent_io_tree *tree;
8371
8372         tree = &BTRFS_I(mapping->host)->io_tree;
8373         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8374 }
8375
8376 static int
8377 btrfs_readpages(struct file *file, struct address_space *mapping,
8378                 struct list_head *pages, unsigned nr_pages)
8379 {
8380         struct extent_io_tree *tree;
8381         tree = &BTRFS_I(mapping->host)->io_tree;
8382         return extent_readpages(tree, mapping, pages, nr_pages,
8383                                 btrfs_get_extent);
8384 }
8385 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8386 {
8387         struct extent_io_tree *tree;
8388         struct extent_map_tree *map;
8389         int ret;
8390
8391         tree = &BTRFS_I(page->mapping->host)->io_tree;
8392         map = &BTRFS_I(page->mapping->host)->extent_tree;
8393         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8394         if (ret == 1) {
8395                 ClearPagePrivate(page);
8396                 set_page_private(page, 0);
8397                 page_cache_release(page);
8398         }
8399         return ret;
8400 }
8401
8402 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8403 {
8404         if (PageWriteback(page) || PageDirty(page))
8405                 return 0;
8406         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8407 }
8408
8409 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8410                                  unsigned int length)
8411 {
8412         struct inode *inode = page->mapping->host;
8413         struct extent_io_tree *tree;
8414         struct btrfs_ordered_extent *ordered;
8415         struct extent_state *cached_state = NULL;
8416         u64 page_start = page_offset(page);
8417         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8418         int inode_evicting = inode->i_state & I_FREEING;
8419
8420         /*
8421          * we have the page locked, so new writeback can't start,
8422          * and the dirty bit won't be cleared while we are here.
8423          *
8424          * Wait for IO on this page so that we can safely clear
8425          * the PagePrivate2 bit and do ordered accounting
8426          */
8427         wait_on_page_writeback(page);
8428
8429         tree = &BTRFS_I(inode)->io_tree;
8430         if (offset) {
8431                 btrfs_releasepage(page, GFP_NOFS);
8432                 return;
8433         }
8434
8435         if (!inode_evicting)
8436                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8437         ordered = btrfs_lookup_ordered_extent(inode, page_start);
8438         if (ordered) {
8439                 /*
8440                  * IO on this page will never be started, so we need
8441                  * to account for any ordered extents now
8442                  */
8443                 if (!inode_evicting)
8444                         clear_extent_bit(tree, page_start, page_end,
8445                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8446                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8447                                          EXTENT_DEFRAG, 1, 0, &cached_state,
8448                                          GFP_NOFS);
8449                 /*
8450                  * whoever cleared the private bit is responsible
8451                  * for the finish_ordered_io
8452                  */
8453                 if (TestClearPagePrivate2(page)) {
8454                         struct btrfs_ordered_inode_tree *tree;
8455                         u64 new_len;
8456
8457                         tree = &BTRFS_I(inode)->ordered_tree;
8458
8459                         spin_lock_irq(&tree->lock);
8460                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8461                         new_len = page_start - ordered->file_offset;
8462                         if (new_len < ordered->truncated_len)
8463                                 ordered->truncated_len = new_len;
8464                         spin_unlock_irq(&tree->lock);
8465
8466                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8467                                                            page_start,
8468                                                            PAGE_CACHE_SIZE, 1))
8469                                 btrfs_finish_ordered_io(ordered);
8470                 }
8471                 btrfs_put_ordered_extent(ordered);
8472                 if (!inode_evicting) {
8473                         cached_state = NULL;
8474                         lock_extent_bits(tree, page_start, page_end, 0,
8475                                          &cached_state);
8476                 }
8477         }
8478
8479         if (!inode_evicting) {
8480                 clear_extent_bit(tree, page_start, page_end,
8481                                  EXTENT_LOCKED | EXTENT_DIRTY |
8482                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8483                                  EXTENT_DEFRAG, 1, 1,
8484                                  &cached_state, GFP_NOFS);
8485
8486                 __btrfs_releasepage(page, GFP_NOFS);
8487         }
8488
8489         ClearPageChecked(page);
8490         if (PagePrivate(page)) {
8491                 ClearPagePrivate(page);
8492                 set_page_private(page, 0);
8493                 page_cache_release(page);
8494         }
8495 }
8496
8497 /*
8498  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8499  * called from a page fault handler when a page is first dirtied. Hence we must
8500  * be careful to check for EOF conditions here. We set the page up correctly
8501  * for a written page which means we get ENOSPC checking when writing into
8502  * holes and correct delalloc and unwritten extent mapping on filesystems that
8503  * support these features.
8504  *
8505  * We are not allowed to take the i_mutex here so we have to play games to
8506  * protect against truncate races as the page could now be beyond EOF.  Because
8507  * vmtruncate() writes the inode size before removing pages, once we have the
8508  * page lock we can determine safely if the page is beyond EOF. If it is not
8509  * beyond EOF, then the page is guaranteed safe against truncation until we
8510  * unlock the page.
8511  */
8512 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8513 {
8514         struct page *page = vmf->page;
8515         struct inode *inode = file_inode(vma->vm_file);
8516         struct btrfs_root *root = BTRFS_I(inode)->root;
8517         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8518         struct btrfs_ordered_extent *ordered;
8519         struct extent_state *cached_state = NULL;
8520         char *kaddr;
8521         unsigned long zero_start;
8522         loff_t size;
8523         int ret;
8524         int reserved = 0;
8525         u64 page_start;
8526         u64 page_end;
8527
8528         sb_start_pagefault(inode->i_sb);
8529         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8530         if (!ret) {
8531                 ret = file_update_time(vma->vm_file);
8532                 reserved = 1;
8533         }
8534         if (ret) {
8535                 if (ret == -ENOMEM)
8536                         ret = VM_FAULT_OOM;
8537                 else /* -ENOSPC, -EIO, etc */
8538                         ret = VM_FAULT_SIGBUS;
8539                 if (reserved)
8540                         goto out;
8541                 goto out_noreserve;
8542         }
8543
8544         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8545 again:
8546         lock_page(page);
8547         size = i_size_read(inode);
8548         page_start = page_offset(page);
8549         page_end = page_start + PAGE_CACHE_SIZE - 1;
8550
8551         if ((page->mapping != inode->i_mapping) ||
8552             (page_start >= size)) {
8553                 /* page got truncated out from underneath us */
8554                 goto out_unlock;
8555         }
8556         wait_on_page_writeback(page);
8557
8558         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8559         set_page_extent_mapped(page);
8560
8561         /*
8562          * we can't set the delalloc bits if there are pending ordered
8563          * extents.  Drop our locks and wait for them to finish
8564          */
8565         ordered = btrfs_lookup_ordered_extent(inode, page_start);
8566         if (ordered) {
8567                 unlock_extent_cached(io_tree, page_start, page_end,
8568                                      &cached_state, GFP_NOFS);
8569                 unlock_page(page);
8570                 btrfs_start_ordered_extent(inode, ordered, 1);
8571                 btrfs_put_ordered_extent(ordered);
8572                 goto again;
8573         }
8574
8575         /*
8576          * XXX - page_mkwrite gets called every time the page is dirtied, even
8577          * if it was already dirty, so for space accounting reasons we need to
8578          * clear any delalloc bits for the range we are fixing to save.  There
8579          * is probably a better way to do this, but for now keep consistent with
8580          * prepare_pages in the normal write path.
8581          */
8582         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8583                           EXTENT_DIRTY | EXTENT_DELALLOC |
8584                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8585                           0, 0, &cached_state, GFP_NOFS);
8586
8587         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8588                                         &cached_state);
8589         if (ret) {
8590                 unlock_extent_cached(io_tree, page_start, page_end,
8591                                      &cached_state, GFP_NOFS);
8592                 ret = VM_FAULT_SIGBUS;
8593                 goto out_unlock;
8594         }
8595         ret = 0;
8596
8597         /* page is wholly or partially inside EOF */
8598         if (page_start + PAGE_CACHE_SIZE > size)
8599                 zero_start = size & ~PAGE_CACHE_MASK;
8600         else
8601                 zero_start = PAGE_CACHE_SIZE;
8602
8603         if (zero_start != PAGE_CACHE_SIZE) {
8604                 kaddr = kmap(page);
8605                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8606                 flush_dcache_page(page);
8607                 kunmap(page);
8608         }
8609         ClearPageChecked(page);
8610         set_page_dirty(page);
8611         SetPageUptodate(page);
8612
8613         BTRFS_I(inode)->last_trans = root->fs_info->generation;
8614         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8615         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8616
8617         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8618
8619 out_unlock:
8620         if (!ret) {
8621                 sb_end_pagefault(inode->i_sb);
8622                 return VM_FAULT_LOCKED;
8623         }
8624         unlock_page(page);
8625 out:
8626         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8627 out_noreserve:
8628         sb_end_pagefault(inode->i_sb);
8629         return ret;
8630 }
8631
8632 static int btrfs_truncate(struct inode *inode)
8633 {
8634         struct btrfs_root *root = BTRFS_I(inode)->root;
8635         struct btrfs_block_rsv *rsv;
8636         int ret = 0;
8637         int err = 0;
8638         struct btrfs_trans_handle *trans;
8639         u64 mask = root->sectorsize - 1;
8640         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8641
8642         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8643                                        (u64)-1);
8644         if (ret)
8645                 return ret;
8646
8647         /*
8648          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
8649          * 3 things going on here
8650          *
8651          * 1) We need to reserve space for our orphan item and the space to
8652          * delete our orphan item.  Lord knows we don't want to have a dangling
8653          * orphan item because we didn't reserve space to remove it.
8654          *
8655          * 2) We need to reserve space to update our inode.
8656          *
8657          * 3) We need to have something to cache all the space that is going to
8658          * be free'd up by the truncate operation, but also have some slack
8659          * space reserved in case it uses space during the truncate (thank you
8660          * very much snapshotting).
8661          *
8662          * And we need these to all be seperate.  The fact is we can use alot of
8663          * space doing the truncate, and we have no earthly idea how much space
8664          * we will use, so we need the truncate reservation to be seperate so it
8665          * doesn't end up using space reserved for updating the inode or
8666          * removing the orphan item.  We also need to be able to stop the
8667          * transaction and start a new one, which means we need to be able to
8668          * update the inode several times, and we have no idea of knowing how
8669          * many times that will be, so we can't just reserve 1 item for the
8670          * entirety of the opration, so that has to be done seperately as well.
8671          * Then there is the orphan item, which does indeed need to be held on
8672          * to for the whole operation, and we need nobody to touch this reserved
8673          * space except the orphan code.
8674          *
8675          * So that leaves us with
8676          *
8677          * 1) root->orphan_block_rsv - for the orphan deletion.
8678          * 2) rsv - for the truncate reservation, which we will steal from the
8679          * transaction reservation.
8680          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8681          * updating the inode.
8682          */
8683         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8684         if (!rsv)
8685                 return -ENOMEM;
8686         rsv->size = min_size;
8687         rsv->failfast = 1;
8688
8689         /*
8690          * 1 for the truncate slack space
8691          * 1 for updating the inode.
8692          */
8693         trans = btrfs_start_transaction(root, 2);
8694         if (IS_ERR(trans)) {
8695                 err = PTR_ERR(trans);
8696                 goto out;
8697         }
8698
8699         /* Migrate the slack space for the truncate to our reserve */
8700         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8701                                       min_size);
8702         BUG_ON(ret);
8703
8704         /*
8705          * So if we truncate and then write and fsync we normally would just
8706          * write the extents that changed, which is a problem if we need to
8707          * first truncate that entire inode.  So set this flag so we write out
8708          * all of the extents in the inode to the sync log so we're completely
8709          * safe.
8710          */
8711         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8712         trans->block_rsv = rsv;
8713
8714         while (1) {
8715                 ret = btrfs_truncate_inode_items(trans, root, inode,
8716                                                  inode->i_size,
8717                                                  BTRFS_EXTENT_DATA_KEY);
8718                 if (ret != -ENOSPC && ret != -EAGAIN) {
8719                         err = ret;
8720                         break;
8721                 }
8722
8723                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8724                 ret = btrfs_update_inode(trans, root, inode);
8725                 if (ret) {
8726                         err = ret;
8727                         break;
8728                 }
8729
8730                 btrfs_end_transaction(trans, root);
8731                 btrfs_btree_balance_dirty(root);
8732
8733                 trans = btrfs_start_transaction(root, 2);
8734                 if (IS_ERR(trans)) {
8735                         ret = err = PTR_ERR(trans);
8736                         trans = NULL;
8737                         break;
8738                 }
8739
8740                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8741                                               rsv, min_size);
8742                 BUG_ON(ret);    /* shouldn't happen */
8743                 trans->block_rsv = rsv;
8744         }
8745
8746         if (ret == 0 && inode->i_nlink > 0) {
8747                 trans->block_rsv = root->orphan_block_rsv;
8748                 ret = btrfs_orphan_del(trans, inode);
8749                 if (ret)
8750                         err = ret;
8751         }
8752
8753         if (trans) {
8754                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8755                 ret = btrfs_update_inode(trans, root, inode);
8756                 if (ret && !err)
8757                         err = ret;
8758
8759                 ret = btrfs_end_transaction(trans, root);
8760                 btrfs_btree_balance_dirty(root);
8761         }
8762
8763 out:
8764         btrfs_free_block_rsv(root, rsv);
8765
8766         if (ret && !err)
8767                 err = ret;
8768
8769         return err;
8770 }
8771
8772 /*
8773  * create a new subvolume directory/inode (helper for the ioctl).
8774  */
8775 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8776                              struct btrfs_root *new_root,
8777                              struct btrfs_root *parent_root,
8778                              u64 new_dirid)
8779 {
8780         struct inode *inode;
8781         int err;
8782         u64 index = 0;
8783
8784         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8785                                 new_dirid, new_dirid,
8786                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
8787                                 &index);
8788         if (IS_ERR(inode))
8789                 return PTR_ERR(inode);
8790         inode->i_op = &btrfs_dir_inode_operations;
8791         inode->i_fop = &btrfs_dir_file_operations;
8792
8793         set_nlink(inode, 1);
8794         btrfs_i_size_write(inode, 0);
8795         unlock_new_inode(inode);
8796
8797         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8798         if (err)
8799                 btrfs_err(new_root->fs_info,
8800                           "error inheriting subvolume %llu properties: %d",
8801                           new_root->root_key.objectid, err);
8802
8803         err = btrfs_update_inode(trans, new_root, inode);
8804
8805         iput(inode);
8806         return err;
8807 }
8808
8809 struct inode *btrfs_alloc_inode(struct super_block *sb)
8810 {
8811         struct btrfs_inode *ei;
8812         struct inode *inode;
8813
8814         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8815         if (!ei)
8816                 return NULL;
8817
8818         ei->root = NULL;
8819         ei->generation = 0;
8820         ei->last_trans = 0;
8821         ei->last_sub_trans = 0;
8822         ei->logged_trans = 0;
8823         ei->delalloc_bytes = 0;
8824         ei->defrag_bytes = 0;
8825         ei->disk_i_size = 0;
8826         ei->flags = 0;
8827         ei->csum_bytes = 0;
8828         ei->index_cnt = (u64)-1;
8829         ei->dir_index = 0;
8830         ei->last_unlink_trans = 0;
8831         ei->last_log_commit = 0;
8832
8833         spin_lock_init(&ei->lock);
8834         ei->outstanding_extents = 0;
8835         ei->reserved_extents = 0;
8836
8837         ei->runtime_flags = 0;
8838         ei->force_compress = BTRFS_COMPRESS_NONE;
8839
8840         ei->delayed_node = NULL;
8841
8842         ei->i_otime.tv_sec = 0;
8843         ei->i_otime.tv_nsec = 0;
8844
8845         inode = &ei->vfs_inode;
8846         extent_map_tree_init(&ei->extent_tree);
8847         extent_io_tree_init(&ei->io_tree, &inode->i_data);
8848         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8849         ei->io_tree.track_uptodate = 1;
8850         ei->io_failure_tree.track_uptodate = 1;
8851         atomic_set(&ei->sync_writers, 0);
8852         mutex_init(&ei->log_mutex);
8853         mutex_init(&ei->delalloc_mutex);
8854         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8855         INIT_LIST_HEAD(&ei->delalloc_inodes);
8856         RB_CLEAR_NODE(&ei->rb_node);
8857
8858         return inode;
8859 }
8860
8861 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8862 void btrfs_test_destroy_inode(struct inode *inode)
8863 {
8864         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8865         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8866 }
8867 #endif
8868
8869 static void btrfs_i_callback(struct rcu_head *head)
8870 {
8871         struct inode *inode = container_of(head, struct inode, i_rcu);
8872         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8873 }
8874
8875 void btrfs_destroy_inode(struct inode *inode)
8876 {
8877         struct btrfs_ordered_extent *ordered;
8878         struct btrfs_root *root = BTRFS_I(inode)->root;
8879
8880         WARN_ON(!hlist_empty(&inode->i_dentry));
8881         WARN_ON(inode->i_data.nrpages);
8882         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8883         WARN_ON(BTRFS_I(inode)->reserved_extents);
8884         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8885         WARN_ON(BTRFS_I(inode)->csum_bytes);
8886         WARN_ON(BTRFS_I(inode)->defrag_bytes);
8887
8888         /*
8889          * This can happen where we create an inode, but somebody else also
8890          * created the same inode and we need to destroy the one we already
8891          * created.
8892          */
8893         if (!root)
8894                 goto free;
8895
8896         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8897                      &BTRFS_I(inode)->runtime_flags)) {
8898                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8899                         btrfs_ino(inode));
8900                 atomic_dec(&root->orphan_inodes);
8901         }
8902
8903         while (1) {
8904                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8905                 if (!ordered)
8906                         break;
8907                 else {
8908                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8909                                 ordered->file_offset, ordered->len);
8910                         btrfs_remove_ordered_extent(inode, ordered);
8911                         btrfs_put_ordered_extent(ordered);
8912                         btrfs_put_ordered_extent(ordered);
8913                 }
8914         }
8915         inode_tree_del(inode);
8916         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8917 free:
8918         call_rcu(&inode->i_rcu, btrfs_i_callback);
8919 }
8920
8921 int btrfs_drop_inode(struct inode *inode)
8922 {
8923         struct btrfs_root *root = BTRFS_I(inode)->root;
8924
8925         if (root == NULL)
8926                 return 1;
8927
8928         /* the snap/subvol tree is on deleting */
8929         if (btrfs_root_refs(&root->root_item) == 0)
8930                 return 1;
8931         else
8932                 return generic_drop_inode(inode);
8933 }
8934
8935 static void init_once(void *foo)
8936 {
8937         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8938
8939         inode_init_once(&ei->vfs_inode);
8940 }
8941
8942 void btrfs_destroy_cachep(void)
8943 {
8944         /*
8945          * Make sure all delayed rcu free inodes are flushed before we
8946          * destroy cache.
8947          */
8948         rcu_barrier();
8949         if (btrfs_inode_cachep)
8950                 kmem_cache_destroy(btrfs_inode_cachep);
8951         if (btrfs_trans_handle_cachep)
8952                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8953         if (btrfs_transaction_cachep)
8954                 kmem_cache_destroy(btrfs_transaction_cachep);
8955         if (btrfs_path_cachep)
8956                 kmem_cache_destroy(btrfs_path_cachep);
8957         if (btrfs_free_space_cachep)
8958                 kmem_cache_destroy(btrfs_free_space_cachep);
8959         if (btrfs_delalloc_work_cachep)
8960                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8961 }
8962
8963 int btrfs_init_cachep(void)
8964 {
8965         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8966                         sizeof(struct btrfs_inode), 0,
8967                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8968         if (!btrfs_inode_cachep)
8969                 goto fail;
8970
8971         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8972                         sizeof(struct btrfs_trans_handle), 0,
8973                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8974         if (!btrfs_trans_handle_cachep)
8975                 goto fail;
8976
8977         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8978                         sizeof(struct btrfs_transaction), 0,
8979                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8980         if (!btrfs_transaction_cachep)
8981                 goto fail;
8982
8983         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8984                         sizeof(struct btrfs_path), 0,
8985                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8986         if (!btrfs_path_cachep)
8987                 goto fail;
8988
8989         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8990                         sizeof(struct btrfs_free_space), 0,
8991                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8992         if (!btrfs_free_space_cachep)
8993                 goto fail;
8994
8995         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8996                         sizeof(struct btrfs_delalloc_work), 0,
8997                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8998                         NULL);
8999         if (!btrfs_delalloc_work_cachep)
9000                 goto fail;
9001
9002         return 0;
9003 fail:
9004         btrfs_destroy_cachep();
9005         return -ENOMEM;
9006 }
9007
9008 static int btrfs_getattr(struct vfsmount *mnt,
9009                          struct dentry *dentry, struct kstat *stat)
9010 {
9011         u64 delalloc_bytes;
9012         struct inode *inode = d_inode(dentry);
9013         u32 blocksize = inode->i_sb->s_blocksize;
9014
9015         generic_fillattr(inode, stat);
9016         stat->dev = BTRFS_I(inode)->root->anon_dev;
9017         stat->blksize = PAGE_CACHE_SIZE;
9018
9019         spin_lock(&BTRFS_I(inode)->lock);
9020         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9021         spin_unlock(&BTRFS_I(inode)->lock);
9022         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9023                         ALIGN(delalloc_bytes, blocksize)) >> 9;
9024         return 0;
9025 }
9026
9027 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9028                            struct inode *new_dir, struct dentry *new_dentry)
9029 {
9030         struct btrfs_trans_handle *trans;
9031         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9032         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9033         struct inode *new_inode = d_inode(new_dentry);
9034         struct inode *old_inode = d_inode(old_dentry);
9035         struct timespec ctime = CURRENT_TIME;
9036         u64 index = 0;
9037         u64 root_objectid;
9038         int ret;
9039         u64 old_ino = btrfs_ino(old_inode);
9040
9041         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9042                 return -EPERM;
9043
9044         /* we only allow rename subvolume link between subvolumes */
9045         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9046                 return -EXDEV;
9047
9048         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9049             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9050                 return -ENOTEMPTY;
9051
9052         if (S_ISDIR(old_inode->i_mode) && new_inode &&
9053             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9054                 return -ENOTEMPTY;
9055
9056
9057         /* check for collisions, even if the  name isn't there */
9058         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9059                              new_dentry->d_name.name,
9060                              new_dentry->d_name.len);
9061
9062         if (ret) {
9063                 if (ret == -EEXIST) {
9064                         /* we shouldn't get
9065                          * eexist without a new_inode */
9066                         if (WARN_ON(!new_inode)) {
9067                                 return ret;
9068                         }
9069                 } else {
9070                         /* maybe -EOVERFLOW */
9071                         return ret;
9072                 }
9073         }
9074         ret = 0;
9075
9076         /*
9077          * we're using rename to replace one file with another.  Start IO on it
9078          * now so  we don't add too much work to the end of the transaction
9079          */
9080         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9081                 filemap_flush(old_inode->i_mapping);
9082
9083         /* close the racy window with snapshot create/destroy ioctl */
9084         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9085                 down_read(&root->fs_info->subvol_sem);
9086         /*
9087          * We want to reserve the absolute worst case amount of items.  So if
9088          * both inodes are subvols and we need to unlink them then that would
9089          * require 4 item modifications, but if they are both normal inodes it
9090          * would require 5 item modifications, so we'll assume their normal
9091          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9092          * should cover the worst case number of items we'll modify.
9093          */
9094         trans = btrfs_start_transaction(root, 11);
9095         if (IS_ERR(trans)) {
9096                 ret = PTR_ERR(trans);
9097                 goto out_notrans;
9098         }
9099
9100         if (dest != root)
9101                 btrfs_record_root_in_trans(trans, dest);
9102
9103         ret = btrfs_set_inode_index(new_dir, &index);
9104         if (ret)
9105                 goto out_fail;
9106
9107         BTRFS_I(old_inode)->dir_index = 0ULL;
9108         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9109                 /* force full log commit if subvolume involved. */
9110                 btrfs_set_log_full_commit(root->fs_info, trans);
9111         } else {
9112                 ret = btrfs_insert_inode_ref(trans, dest,
9113                                              new_dentry->d_name.name,
9114                                              new_dentry->d_name.len,
9115                                              old_ino,
9116                                              btrfs_ino(new_dir), index);
9117                 if (ret)
9118                         goto out_fail;
9119                 /*
9120                  * this is an ugly little race, but the rename is required
9121                  * to make sure that if we crash, the inode is either at the
9122                  * old name or the new one.  pinning the log transaction lets
9123                  * us make sure we don't allow a log commit to come in after
9124                  * we unlink the name but before we add the new name back in.
9125                  */
9126                 btrfs_pin_log_trans(root);
9127         }
9128
9129         inode_inc_iversion(old_dir);
9130         inode_inc_iversion(new_dir);
9131         inode_inc_iversion(old_inode);
9132         old_dir->i_ctime = old_dir->i_mtime = ctime;
9133         new_dir->i_ctime = new_dir->i_mtime = ctime;
9134         old_inode->i_ctime = ctime;
9135
9136         if (old_dentry->d_parent != new_dentry->d_parent)
9137                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9138
9139         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9140                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9141                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9142                                         old_dentry->d_name.name,
9143                                         old_dentry->d_name.len);
9144         } else {
9145                 ret = __btrfs_unlink_inode(trans, root, old_dir,
9146                                         d_inode(old_dentry),
9147                                         old_dentry->d_name.name,
9148                                         old_dentry->d_name.len);
9149                 if (!ret)
9150                         ret = btrfs_update_inode(trans, root, old_inode);
9151         }
9152         if (ret) {
9153                 btrfs_abort_transaction(trans, root, ret);
9154                 goto out_fail;
9155         }
9156
9157         if (new_inode) {
9158                 inode_inc_iversion(new_inode);
9159                 new_inode->i_ctime = CURRENT_TIME;
9160                 if (unlikely(btrfs_ino(new_inode) ==
9161                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9162                         root_objectid = BTRFS_I(new_inode)->location.objectid;
9163                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
9164                                                 root_objectid,
9165                                                 new_dentry->d_name.name,
9166                                                 new_dentry->d_name.len);
9167                         BUG_ON(new_inode->i_nlink == 0);
9168                 } else {
9169                         ret = btrfs_unlink_inode(trans, dest, new_dir,
9170                                                  d_inode(new_dentry),
9171                                                  new_dentry->d_name.name,
9172                                                  new_dentry->d_name.len);
9173                 }
9174                 if (!ret && new_inode->i_nlink == 0)
9175                         ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9176                 if (ret) {
9177                         btrfs_abort_transaction(trans, root, ret);
9178                         goto out_fail;
9179                 }
9180         }
9181
9182         ret = btrfs_add_link(trans, new_dir, old_inode,
9183                              new_dentry->d_name.name,
9184                              new_dentry->d_name.len, 0, index);
9185         if (ret) {
9186                 btrfs_abort_transaction(trans, root, ret);
9187                 goto out_fail;
9188         }
9189
9190         if (old_inode->i_nlink == 1)
9191                 BTRFS_I(old_inode)->dir_index = index;
9192
9193         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9194                 struct dentry *parent = new_dentry->d_parent;
9195                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9196                 btrfs_end_log_trans(root);
9197         }
9198 out_fail:
9199         btrfs_end_transaction(trans, root);
9200 out_notrans:
9201         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9202                 up_read(&root->fs_info->subvol_sem);
9203
9204         return ret;
9205 }
9206
9207 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9208                          struct inode *new_dir, struct dentry *new_dentry,
9209                          unsigned int flags)
9210 {
9211         if (flags & ~RENAME_NOREPLACE)
9212                 return -EINVAL;
9213
9214         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9215 }
9216
9217 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9218 {
9219         struct btrfs_delalloc_work *delalloc_work;
9220         struct inode *inode;
9221
9222         delalloc_work = container_of(work, struct btrfs_delalloc_work,
9223                                      work);
9224         inode = delalloc_work->inode;
9225         if (delalloc_work->wait) {
9226                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9227         } else {
9228                 filemap_flush(inode->i_mapping);
9229                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9230                              &BTRFS_I(inode)->runtime_flags))
9231                         filemap_flush(inode->i_mapping);
9232         }
9233
9234         if (delalloc_work->delay_iput)
9235                 btrfs_add_delayed_iput(inode);
9236         else
9237                 iput(inode);
9238         complete(&delalloc_work->completion);
9239 }
9240
9241 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9242                                                     int wait, int delay_iput)
9243 {
9244         struct btrfs_delalloc_work *work;
9245
9246         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9247         if (!work)
9248                 return NULL;
9249
9250         init_completion(&work->completion);
9251         INIT_LIST_HEAD(&work->list);
9252         work->inode = inode;
9253         work->wait = wait;
9254         work->delay_iput = delay_iput;
9255         WARN_ON_ONCE(!inode);
9256         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9257                         btrfs_run_delalloc_work, NULL, NULL);
9258
9259         return work;
9260 }
9261
9262 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9263 {
9264         wait_for_completion(&work->completion);
9265         kmem_cache_free(btrfs_delalloc_work_cachep, work);
9266 }
9267
9268 /*
9269  * some fairly slow code that needs optimization. This walks the list
9270  * of all the inodes with pending delalloc and forces them to disk.
9271  */
9272 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9273                                    int nr)
9274 {
9275         struct btrfs_inode *binode;
9276         struct inode *inode;
9277         struct btrfs_delalloc_work *work, *next;
9278         struct list_head works;
9279         struct list_head splice;
9280         int ret = 0;
9281
9282         INIT_LIST_HEAD(&works);
9283         INIT_LIST_HEAD(&splice);
9284
9285         mutex_lock(&root->delalloc_mutex);
9286         spin_lock(&root->delalloc_lock);
9287         list_splice_init(&root->delalloc_inodes, &splice);
9288         while (!list_empty(&splice)) {
9289                 binode = list_entry(splice.next, struct btrfs_inode,
9290                                     delalloc_inodes);
9291
9292                 list_move_tail(&binode->delalloc_inodes,
9293                                &root->delalloc_inodes);
9294                 inode = igrab(&binode->vfs_inode);
9295                 if (!inode) {
9296                         cond_resched_lock(&root->delalloc_lock);
9297                         continue;
9298                 }
9299                 spin_unlock(&root->delalloc_lock);
9300
9301                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9302                 if (!work) {
9303                         if (delay_iput)
9304                                 btrfs_add_delayed_iput(inode);
9305                         else
9306                                 iput(inode);
9307                         ret = -ENOMEM;
9308                         goto out;
9309                 }
9310                 list_add_tail(&work->list, &works);
9311                 btrfs_queue_work(root->fs_info->flush_workers,
9312                                  &work->work);
9313                 ret++;
9314                 if (nr != -1 && ret >= nr)
9315                         goto out;
9316                 cond_resched();
9317                 spin_lock(&root->delalloc_lock);
9318         }
9319         spin_unlock(&root->delalloc_lock);
9320
9321 out:
9322         list_for_each_entry_safe(work, next, &works, list) {
9323                 list_del_init(&work->list);
9324                 btrfs_wait_and_free_delalloc_work(work);
9325         }
9326
9327         if (!list_empty_careful(&splice)) {
9328                 spin_lock(&root->delalloc_lock);
9329                 list_splice_tail(&splice, &root->delalloc_inodes);
9330                 spin_unlock(&root->delalloc_lock);
9331         }
9332         mutex_unlock(&root->delalloc_mutex);
9333         return ret;
9334 }
9335
9336 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9337 {
9338         int ret;
9339
9340         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9341                 return -EROFS;
9342
9343         ret = __start_delalloc_inodes(root, delay_iput, -1);
9344         if (ret > 0)
9345                 ret = 0;
9346         /*
9347          * the filemap_flush will queue IO into the worker threads, but
9348          * we have to make sure the IO is actually started and that
9349          * ordered extents get created before we return
9350          */
9351         atomic_inc(&root->fs_info->async_submit_draining);
9352         while (atomic_read(&root->fs_info->nr_async_submits) ||
9353               atomic_read(&root->fs_info->async_delalloc_pages)) {
9354                 wait_event(root->fs_info->async_submit_wait,
9355                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9356                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9357         }
9358         atomic_dec(&root->fs_info->async_submit_draining);
9359         return ret;
9360 }
9361
9362 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9363                                int nr)
9364 {
9365         struct btrfs_root *root;
9366         struct list_head splice;
9367         int ret;
9368
9369         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9370                 return -EROFS;
9371
9372         INIT_LIST_HEAD(&splice);
9373
9374         mutex_lock(&fs_info->delalloc_root_mutex);
9375         spin_lock(&fs_info->delalloc_root_lock);
9376         list_splice_init(&fs_info->delalloc_roots, &splice);
9377         while (!list_empty(&splice) && nr) {
9378                 root = list_first_entry(&splice, struct btrfs_root,
9379                                         delalloc_root);
9380                 root = btrfs_grab_fs_root(root);
9381                 BUG_ON(!root);
9382                 list_move_tail(&root->delalloc_root,
9383                                &fs_info->delalloc_roots);
9384                 spin_unlock(&fs_info->delalloc_root_lock);
9385
9386                 ret = __start_delalloc_inodes(root, delay_iput, nr);
9387                 btrfs_put_fs_root(root);
9388                 if (ret < 0)
9389                         goto out;
9390
9391                 if (nr != -1) {
9392                         nr -= ret;
9393                         WARN_ON(nr < 0);
9394                 }
9395                 spin_lock(&fs_info->delalloc_root_lock);
9396         }
9397         spin_unlock(&fs_info->delalloc_root_lock);
9398
9399         ret = 0;
9400         atomic_inc(&fs_info->async_submit_draining);
9401         while (atomic_read(&fs_info->nr_async_submits) ||
9402               atomic_read(&fs_info->async_delalloc_pages)) {
9403                 wait_event(fs_info->async_submit_wait,
9404                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
9405                     atomic_read(&fs_info->async_delalloc_pages) == 0));
9406         }
9407         atomic_dec(&fs_info->async_submit_draining);
9408 out:
9409         if (!list_empty_careful(&splice)) {
9410                 spin_lock(&fs_info->delalloc_root_lock);
9411                 list_splice_tail(&splice, &fs_info->delalloc_roots);
9412                 spin_unlock(&fs_info->delalloc_root_lock);
9413         }
9414         mutex_unlock(&fs_info->delalloc_root_mutex);
9415         return ret;
9416 }
9417
9418 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9419                          const char *symname)
9420 {
9421         struct btrfs_trans_handle *trans;
9422         struct btrfs_root *root = BTRFS_I(dir)->root;
9423         struct btrfs_path *path;
9424         struct btrfs_key key;
9425         struct inode *inode = NULL;
9426         int err;
9427         int drop_inode = 0;
9428         u64 objectid;
9429         u64 index = 0;
9430         int name_len;
9431         int datasize;
9432         unsigned long ptr;
9433         struct btrfs_file_extent_item *ei;
9434         struct extent_buffer *leaf;
9435
9436         name_len = strlen(symname);
9437         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9438                 return -ENAMETOOLONG;
9439
9440         /*
9441          * 2 items for inode item and ref
9442          * 2 items for dir items
9443          * 1 item for xattr if selinux is on
9444          */
9445         trans = btrfs_start_transaction(root, 5);
9446         if (IS_ERR(trans))
9447                 return PTR_ERR(trans);
9448
9449         err = btrfs_find_free_ino(root, &objectid);
9450         if (err)
9451                 goto out_unlock;
9452
9453         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9454                                 dentry->d_name.len, btrfs_ino(dir), objectid,
9455                                 S_IFLNK|S_IRWXUGO, &index);
9456         if (IS_ERR(inode)) {
9457                 err = PTR_ERR(inode);
9458                 goto out_unlock;
9459         }
9460
9461         /*
9462         * If the active LSM wants to access the inode during
9463         * d_instantiate it needs these. Smack checks to see
9464         * if the filesystem supports xattrs by looking at the
9465         * ops vector.
9466         */
9467         inode->i_fop = &btrfs_file_operations;
9468         inode->i_op = &btrfs_file_inode_operations;
9469         inode->i_mapping->a_ops = &btrfs_aops;
9470         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9471
9472         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9473         if (err)
9474                 goto out_unlock_inode;
9475
9476         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9477         if (err)
9478                 goto out_unlock_inode;
9479
9480         path = btrfs_alloc_path();
9481         if (!path) {
9482                 err = -ENOMEM;
9483                 goto out_unlock_inode;
9484         }
9485         key.objectid = btrfs_ino(inode);
9486         key.offset = 0;
9487         key.type = BTRFS_EXTENT_DATA_KEY;
9488         datasize = btrfs_file_extent_calc_inline_size(name_len);
9489         err = btrfs_insert_empty_item(trans, root, path, &key,
9490                                       datasize);
9491         if (err) {
9492                 btrfs_free_path(path);
9493                 goto out_unlock_inode;
9494         }
9495         leaf = path->nodes[0];
9496         ei = btrfs_item_ptr(leaf, path->slots[0],
9497                             struct btrfs_file_extent_item);
9498         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9499         btrfs_set_file_extent_type(leaf, ei,
9500                                    BTRFS_FILE_EXTENT_INLINE);
9501         btrfs_set_file_extent_encryption(leaf, ei, 0);
9502         btrfs_set_file_extent_compression(leaf, ei, 0);
9503         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9504         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9505
9506         ptr = btrfs_file_extent_inline_start(ei);
9507         write_extent_buffer(leaf, symname, ptr, name_len);
9508         btrfs_mark_buffer_dirty(leaf);
9509         btrfs_free_path(path);
9510
9511         inode->i_op = &btrfs_symlink_inode_operations;
9512         inode->i_mapping->a_ops = &btrfs_symlink_aops;
9513         inode_set_bytes(inode, name_len);
9514         btrfs_i_size_write(inode, name_len);
9515         err = btrfs_update_inode(trans, root, inode);
9516         if (err) {
9517                 drop_inode = 1;
9518                 goto out_unlock_inode;
9519         }
9520
9521         unlock_new_inode(inode);
9522         d_instantiate(dentry, inode);
9523
9524 out_unlock:
9525         btrfs_end_transaction(trans, root);
9526         if (drop_inode) {
9527                 inode_dec_link_count(inode);
9528                 iput(inode);
9529         }
9530         btrfs_btree_balance_dirty(root);
9531         return err;
9532
9533 out_unlock_inode:
9534         drop_inode = 1;
9535         unlock_new_inode(inode);
9536         goto out_unlock;
9537 }
9538
9539 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9540                                        u64 start, u64 num_bytes, u64 min_size,
9541                                        loff_t actual_len, u64 *alloc_hint,
9542                                        struct btrfs_trans_handle *trans)
9543 {
9544         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9545         struct extent_map *em;
9546         struct btrfs_root *root = BTRFS_I(inode)->root;
9547         struct btrfs_key ins;
9548         u64 cur_offset = start;
9549         u64 i_size;
9550         u64 cur_bytes;
9551         int ret = 0;
9552         bool own_trans = true;
9553
9554         if (trans)
9555                 own_trans = false;
9556         while (num_bytes > 0) {
9557                 if (own_trans) {
9558                         trans = btrfs_start_transaction(root, 3);
9559                         if (IS_ERR(trans)) {
9560                                 ret = PTR_ERR(trans);
9561                                 break;
9562                         }
9563                 }
9564
9565                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9566                 cur_bytes = max(cur_bytes, min_size);
9567                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9568                                            *alloc_hint, &ins, 1, 0);
9569                 if (ret) {
9570                         if (own_trans)
9571                                 btrfs_end_transaction(trans, root);
9572                         break;
9573                 }
9574
9575                 ret = insert_reserved_file_extent(trans, inode,
9576                                                   cur_offset, ins.objectid,
9577                                                   ins.offset, ins.offset,
9578                                                   ins.offset, 0, 0, 0,
9579                                                   BTRFS_FILE_EXTENT_PREALLOC);
9580                 if (ret) {
9581                         btrfs_free_reserved_extent(root, ins.objectid,
9582                                                    ins.offset, 0);
9583                         btrfs_abort_transaction(trans, root, ret);
9584                         if (own_trans)
9585                                 btrfs_end_transaction(trans, root);
9586                         break;
9587                 }
9588
9589                 btrfs_drop_extent_cache(inode, cur_offset,
9590                                         cur_offset + ins.offset -1, 0);
9591
9592                 em = alloc_extent_map();
9593                 if (!em) {
9594                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9595                                 &BTRFS_I(inode)->runtime_flags);
9596                         goto next;
9597                 }
9598
9599                 em->start = cur_offset;
9600                 em->orig_start = cur_offset;
9601                 em->len = ins.offset;
9602                 em->block_start = ins.objectid;
9603                 em->block_len = ins.offset;
9604                 em->orig_block_len = ins.offset;
9605                 em->ram_bytes = ins.offset;
9606                 em->bdev = root->fs_info->fs_devices->latest_bdev;
9607                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9608                 em->generation = trans->transid;
9609
9610                 while (1) {
9611                         write_lock(&em_tree->lock);
9612                         ret = add_extent_mapping(em_tree, em, 1);
9613                         write_unlock(&em_tree->lock);
9614                         if (ret != -EEXIST)
9615                                 break;
9616                         btrfs_drop_extent_cache(inode, cur_offset,
9617                                                 cur_offset + ins.offset - 1,
9618                                                 0);
9619                 }
9620                 free_extent_map(em);
9621 next:
9622                 num_bytes -= ins.offset;
9623                 cur_offset += ins.offset;
9624                 *alloc_hint = ins.objectid + ins.offset;
9625
9626                 inode_inc_iversion(inode);
9627                 inode->i_ctime = CURRENT_TIME;
9628                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9629                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9630                     (actual_len > inode->i_size) &&
9631                     (cur_offset > inode->i_size)) {
9632                         if (cur_offset > actual_len)
9633                                 i_size = actual_len;
9634                         else
9635                                 i_size = cur_offset;
9636                         i_size_write(inode, i_size);
9637                         btrfs_ordered_update_i_size(inode, i_size, NULL);
9638                 }
9639
9640                 ret = btrfs_update_inode(trans, root, inode);
9641
9642                 if (ret) {
9643                         btrfs_abort_transaction(trans, root, ret);
9644                         if (own_trans)
9645                                 btrfs_end_transaction(trans, root);
9646                         break;
9647                 }
9648
9649                 if (own_trans)
9650                         btrfs_end_transaction(trans, root);
9651         }
9652         return ret;
9653 }
9654
9655 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9656                               u64 start, u64 num_bytes, u64 min_size,
9657                               loff_t actual_len, u64 *alloc_hint)
9658 {
9659         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9660                                            min_size, actual_len, alloc_hint,
9661                                            NULL);
9662 }
9663
9664 int btrfs_prealloc_file_range_trans(struct inode *inode,
9665                                     struct btrfs_trans_handle *trans, int mode,
9666                                     u64 start, u64 num_bytes, u64 min_size,
9667                                     loff_t actual_len, u64 *alloc_hint)
9668 {
9669         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9670                                            min_size, actual_len, alloc_hint, trans);
9671 }
9672
9673 static int btrfs_set_page_dirty(struct page *page)
9674 {
9675         return __set_page_dirty_nobuffers(page);
9676 }
9677
9678 static int btrfs_permission(struct inode *inode, int mask)
9679 {
9680         struct btrfs_root *root = BTRFS_I(inode)->root;
9681         umode_t mode = inode->i_mode;
9682
9683         if (mask & MAY_WRITE &&
9684             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9685                 if (btrfs_root_readonly(root))
9686                         return -EROFS;
9687                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9688                         return -EACCES;
9689         }
9690         return generic_permission(inode, mask);
9691 }
9692
9693 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9694 {
9695         struct btrfs_trans_handle *trans;
9696         struct btrfs_root *root = BTRFS_I(dir)->root;
9697         struct inode *inode = NULL;
9698         u64 objectid;
9699         u64 index;
9700         int ret = 0;
9701
9702         /*
9703          * 5 units required for adding orphan entry
9704          */
9705         trans = btrfs_start_transaction(root, 5);
9706         if (IS_ERR(trans))
9707                 return PTR_ERR(trans);
9708
9709         ret = btrfs_find_free_ino(root, &objectid);
9710         if (ret)
9711                 goto out;
9712
9713         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9714                                 btrfs_ino(dir), objectid, mode, &index);
9715         if (IS_ERR(inode)) {
9716                 ret = PTR_ERR(inode);
9717                 inode = NULL;
9718                 goto out;
9719         }
9720
9721         inode->i_fop = &btrfs_file_operations;
9722         inode->i_op = &btrfs_file_inode_operations;
9723
9724         inode->i_mapping->a_ops = &btrfs_aops;
9725         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9726
9727         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9728         if (ret)
9729                 goto out_inode;
9730
9731         ret = btrfs_update_inode(trans, root, inode);
9732         if (ret)
9733                 goto out_inode;
9734         ret = btrfs_orphan_add(trans, inode);
9735         if (ret)
9736                 goto out_inode;
9737
9738         /*
9739          * We set number of links to 0 in btrfs_new_inode(), and here we set
9740          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9741          * through:
9742          *
9743          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9744          */
9745         set_nlink(inode, 1);
9746         unlock_new_inode(inode);
9747         d_tmpfile(dentry, inode);
9748         mark_inode_dirty(inode);
9749
9750 out:
9751         btrfs_end_transaction(trans, root);
9752         if (ret)
9753                 iput(inode);
9754         btrfs_balance_delayed_items(root);
9755         btrfs_btree_balance_dirty(root);
9756         return ret;
9757
9758 out_inode:
9759         unlock_new_inode(inode);
9760         goto out;
9761
9762 }
9763
9764 /* Inspired by filemap_check_errors() */
9765 int btrfs_inode_check_errors(struct inode *inode)
9766 {
9767         int ret = 0;
9768
9769         if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9770             test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9771                 ret = -ENOSPC;
9772         if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9773             test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9774                 ret = -EIO;
9775
9776         return ret;
9777 }
9778
9779 static const struct inode_operations btrfs_dir_inode_operations = {
9780         .getattr        = btrfs_getattr,
9781         .lookup         = btrfs_lookup,
9782         .create         = btrfs_create,
9783         .unlink         = btrfs_unlink,
9784         .link           = btrfs_link,
9785         .mkdir          = btrfs_mkdir,
9786         .rmdir          = btrfs_rmdir,
9787         .rename2        = btrfs_rename2,
9788         .symlink        = btrfs_symlink,
9789         .setattr        = btrfs_setattr,
9790         .mknod          = btrfs_mknod,
9791         .setxattr       = btrfs_setxattr,
9792         .getxattr       = btrfs_getxattr,
9793         .listxattr      = btrfs_listxattr,
9794         .removexattr    = btrfs_removexattr,
9795         .permission     = btrfs_permission,
9796         .get_acl        = btrfs_get_acl,
9797         .set_acl        = btrfs_set_acl,
9798         .update_time    = btrfs_update_time,
9799         .tmpfile        = btrfs_tmpfile,
9800 };
9801 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9802         .lookup         = btrfs_lookup,
9803         .permission     = btrfs_permission,
9804         .get_acl        = btrfs_get_acl,
9805         .set_acl        = btrfs_set_acl,
9806         .update_time    = btrfs_update_time,
9807 };
9808
9809 static const struct file_operations btrfs_dir_file_operations = {
9810         .llseek         = generic_file_llseek,
9811         .read           = generic_read_dir,
9812         .iterate        = btrfs_real_readdir,
9813         .unlocked_ioctl = btrfs_ioctl,
9814 #ifdef CONFIG_COMPAT
9815         .compat_ioctl   = btrfs_ioctl,
9816 #endif
9817         .release        = btrfs_release_file,
9818         .fsync          = btrfs_sync_file,
9819 };
9820
9821 static struct extent_io_ops btrfs_extent_io_ops = {
9822         .fill_delalloc = run_delalloc_range,
9823         .submit_bio_hook = btrfs_submit_bio_hook,
9824         .merge_bio_hook = btrfs_merge_bio_hook,
9825         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9826         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9827         .writepage_start_hook = btrfs_writepage_start_hook,
9828         .set_bit_hook = btrfs_set_bit_hook,
9829         .clear_bit_hook = btrfs_clear_bit_hook,
9830         .merge_extent_hook = btrfs_merge_extent_hook,
9831         .split_extent_hook = btrfs_split_extent_hook,
9832 };
9833
9834 /*
9835  * btrfs doesn't support the bmap operation because swapfiles
9836  * use bmap to make a mapping of extents in the file.  They assume
9837  * these extents won't change over the life of the file and they
9838  * use the bmap result to do IO directly to the drive.
9839  *
9840  * the btrfs bmap call would return logical addresses that aren't
9841  * suitable for IO and they also will change frequently as COW
9842  * operations happen.  So, swapfile + btrfs == corruption.
9843  *
9844  * For now we're avoiding this by dropping bmap.
9845  */
9846 static const struct address_space_operations btrfs_aops = {
9847         .readpage       = btrfs_readpage,
9848         .writepage      = btrfs_writepage,
9849         .writepages     = btrfs_writepages,
9850         .readpages      = btrfs_readpages,
9851         .direct_IO      = btrfs_direct_IO,
9852         .invalidatepage = btrfs_invalidatepage,
9853         .releasepage    = btrfs_releasepage,
9854         .set_page_dirty = btrfs_set_page_dirty,
9855         .error_remove_page = generic_error_remove_page,
9856 };
9857
9858 static const struct address_space_operations btrfs_symlink_aops = {
9859         .readpage       = btrfs_readpage,
9860         .writepage      = btrfs_writepage,
9861         .invalidatepage = btrfs_invalidatepage,
9862         .releasepage    = btrfs_releasepage,
9863 };
9864
9865 static const struct inode_operations btrfs_file_inode_operations = {
9866         .getattr        = btrfs_getattr,
9867         .setattr        = btrfs_setattr,
9868         .setxattr       = btrfs_setxattr,
9869         .getxattr       = btrfs_getxattr,
9870         .listxattr      = btrfs_listxattr,
9871         .removexattr    = btrfs_removexattr,
9872         .permission     = btrfs_permission,
9873         .fiemap         = btrfs_fiemap,
9874         .get_acl        = btrfs_get_acl,
9875         .set_acl        = btrfs_set_acl,
9876         .update_time    = btrfs_update_time,
9877 };
9878 static const struct inode_operations btrfs_special_inode_operations = {
9879         .getattr        = btrfs_getattr,
9880         .setattr        = btrfs_setattr,
9881         .permission     = btrfs_permission,
9882         .setxattr       = btrfs_setxattr,
9883         .getxattr       = btrfs_getxattr,
9884         .listxattr      = btrfs_listxattr,
9885         .removexattr    = btrfs_removexattr,
9886         .get_acl        = btrfs_get_acl,
9887         .set_acl        = btrfs_set_acl,
9888         .update_time    = btrfs_update_time,
9889 };
9890 static const struct inode_operations btrfs_symlink_inode_operations = {
9891         .readlink       = generic_readlink,
9892         .follow_link    = page_follow_link_light,
9893         .put_link       = page_put_link,
9894         .getattr        = btrfs_getattr,
9895         .setattr        = btrfs_setattr,
9896         .permission     = btrfs_permission,
9897         .setxattr       = btrfs_setxattr,
9898         .getxattr       = btrfs_getxattr,
9899         .listxattr      = btrfs_listxattr,
9900         .removexattr    = btrfs_removexattr,
9901         .update_time    = btrfs_update_time,
9902 };
9903
9904 const struct dentry_operations btrfs_dentry_operations = {
9905         .d_delete       = btrfs_dentry_delete,
9906         .d_release      = btrfs_dentry_release,
9907 };