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 "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "print-tree.h"
  50 #include "ordered-data.h"
  51 #include "xattr.h"
  52 #include "tree-log.h"
  53 #include "volumes.h"
  54 #include "compression.h"
  55 #include "locking.h"
  56 #include "free-space-cache.h"
  57 #include "inode-map.h"
  58 #include "backref.h"
  59
  60 struct btrfs_iget_args {
  61         u64 ino;
  62         struct btrfs_root *root;
  63 };
  64
  65 static const struct inode_operations btrfs_dir_inode_operations;
  66 static const struct inode_operations btrfs_symlink_inode_operations;
  67 static const struct inode_operations btrfs_dir_ro_inode_operations;
  68 static const struct inode_operations btrfs_special_inode_operations;
  69 static const struct inode_operations btrfs_file_inode_operations;
  70 static const struct address_space_operations btrfs_aops;
  71 static const struct address_space_operations btrfs_symlink_aops;
  72 static const struct file_operations btrfs_dir_file_operations;
  73 static struct extent_io_ops btrfs_extent_io_ops;
  74
  75 static struct kmem_cache *btrfs_inode_cachep;
  76 static struct kmem_cache *btrfs_delalloc_work_cachep;
  77 struct kmem_cache *btrfs_trans_handle_cachep;
  78 struct kmem_cache *btrfs_transaction_cachep;
  79 struct kmem_cache *btrfs_path_cachep;
  80 struct kmem_cache *btrfs_free_space_cachep;
  81
  82 #define S_SHIFT 12
  83 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  84         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  85         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  86         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  87         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  88         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  89         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  90         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  91 };
  92
  93 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  94 static int btrfs_truncate(struct inode *inode);
  95 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  96 static noinline int cow_file_range(struct inode *inode,
  97                                    struct page *locked_page,
  98                                    u64 start, u64 end, int *page_started,
  99                                    unsigned long *nr_written, int unlock);
 100 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 101                                            u64 len, u64 orig_start,
 102                                            u64 block_start, u64 block_len,
 103                                            u64 orig_block_len, int type);
 104
 105 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 106                                      struct inode *inode,  struct inode *dir,
 107                                      const struct qstr *qstr)
 108 {
 109         int err;
 110
 111         err = btrfs_init_acl(trans, inode, dir);
 112         if (!err)
 113                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 114         return err;
 115 }
 116
 117 /*
 118  * this does all the hard work for inserting an inline extent into
 119  * the btree.  The caller should have done a btrfs_drop_extents so that
 120  * no overlapping inline items exist in the btree
 121  */
 122 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 123                                 struct btrfs_root *root, struct inode *inode,
 124                                 u64 start, size_t size, size_t compressed_size,
 125                                 int compress_type,
 126                                 struct page **compressed_pages)
 127 {
 128         struct btrfs_key key;
 129         struct btrfs_path *path;
 130         struct extent_buffer *leaf;
 131         struct page *page = NULL;
 132         char *kaddr;
 133         unsigned long ptr;
 134         struct btrfs_file_extent_item *ei;
 135         int err = 0;
 136         int ret;
 137         size_t cur_size = size;
 138         size_t datasize;
 139         unsigned long offset;
 140
 141         if (compressed_size && compressed_pages)
 142                 cur_size = compressed_size;
 143
 144         path = btrfs_alloc_path();
 145         if (!path)
 146                 return -ENOMEM;
 147
 148         path->leave_spinning = 1;
 149
 150         key.objectid = btrfs_ino(inode);
 151         key.offset = start;
 152         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 153         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 154
 155         inode_add_bytes(inode, size);
 156         ret = btrfs_insert_empty_item(trans, root, path, &key,
 157                                       datasize);
 158         if (ret) {
 159                 err = ret;
 160                 goto fail;
 161         }
 162         leaf = path->nodes[0];
 163         ei = btrfs_item_ptr(leaf, path->slots[0],
 164                             struct btrfs_file_extent_item);
 165         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 166         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 167         btrfs_set_file_extent_encryption(leaf, ei, 0);
 168         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 169         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 170         ptr = btrfs_file_extent_inline_start(ei);
 171
 172         if (compress_type != BTRFS_COMPRESS_NONE) {
 173                 struct page *cpage;
 174                 int i = 0;
 175                 while (compressed_size > 0) {
 176                         cpage = compressed_pages[i];
 177                         cur_size = min_t(unsigned long, compressed_size,
 178                                        PAGE_CACHE_SIZE);
 179
 180                         kaddr = kmap_atomic(cpage);
 181                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 182                         kunmap_atomic(kaddr);
 183
 184                         i++;
 185                         ptr += cur_size;
 186                         compressed_size -= cur_size;
 187                 }
 188                 btrfs_set_file_extent_compression(leaf, ei,
 189                                                   compress_type);
 190         } else {
 191                 page = find_get_page(inode->i_mapping,
 192                                      start >> PAGE_CACHE_SHIFT);
 193                 btrfs_set_file_extent_compression(leaf, ei, 0);
 194                 kaddr = kmap_atomic(page);
 195                 offset = start & (PAGE_CACHE_SIZE - 1);
 196                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 197                 kunmap_atomic(kaddr);
 198                 page_cache_release(page);
 199         }
 200         btrfs_mark_buffer_dirty(leaf);
 201         btrfs_free_path(path);
 202
 203         /*
 204          * we're an inline extent, so nobody can
 205          * extend the file past i_size without locking
 206          * a page we already have locked.
 207          *
 208          * We must do any isize and inode updates
 209          * before we unlock the pages.  Otherwise we
 210          * could end up racing with unlink.
 211          */
 212         BTRFS_I(inode)->disk_i_size = inode->i_size;
 213         ret = btrfs_update_inode(trans, root, inode);
 214
 215         return ret;
 216 fail:
 217         btrfs_free_path(path);
 218         return err;
 219 }
 220
 221
 222 /*
 223  * conditionally insert an inline extent into the file.  This
 224  * does the checks required to make sure the data is small enough
 225  * to fit as an inline extent.
 226  */
 227 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 228                                  struct btrfs_root *root,
 229                                  struct inode *inode, u64 start, u64 end,
 230                                  size_t compressed_size, int compress_type,
 231                                  struct page **compressed_pages)
 232 {
 233         u64 isize = i_size_read(inode);
 234         u64 actual_end = min(end + 1, isize);
 235         u64 inline_len = actual_end - start;
 236         u64 aligned_end = ALIGN(end, root->sectorsize);
 237         u64 data_len = inline_len;
 238         int ret;
 239
 240         if (compressed_size)
 241                 data_len = compressed_size;
 242
 243         if (start > 0 ||
 244             actual_end >= PAGE_CACHE_SIZE ||
 245             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 246             (!compressed_size &&
 247             (actual_end & (root->sectorsize - 1)) == 0) ||
 248             end + 1 < isize ||
 249             data_len > root->fs_info->max_inline) {
 250                 return 1;
 251         }
 252
 253         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 254         if (ret)
 255                 return ret;
 256
 257         if (isize > actual_end)
 258                 inline_len = min_t(u64, isize, actual_end);
 259         ret = insert_inline_extent(trans, root, inode, start,
 260                                    inline_len, compressed_size,
 261                                    compress_type, compressed_pages);
 262         if (ret && ret != -ENOSPC) {
 263                 btrfs_abort_transaction(trans, root, ret);
 264                 return ret;
 265         } else if (ret == -ENOSPC) {
 266                 return 1;
 267         }
 268
 269         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 270         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 271         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 272         return 0;
 273 }
 274
 275 struct async_extent {
 276         u64 start;
 277         u64 ram_size;
 278         u64 compressed_size;
 279         struct page **pages;
 280         unsigned long nr_pages;
 281         int compress_type;
 282         struct list_head list;
 283 };
 284
 285 struct async_cow {
 286         struct inode *inode;
 287         struct btrfs_root *root;
 288         struct page *locked_page;
 289         u64 start;
 290         u64 end;
 291         struct list_head extents;
 292         struct btrfs_work work;
 293 };
 294
 295 static noinline int add_async_extent(struct async_cow *cow,
 296                                      u64 start, u64 ram_size,
 297                                      u64 compressed_size,
 298                                      struct page **pages,
 299                                      unsigned long nr_pages,
 300                                      int compress_type)
 301 {
 302         struct async_extent *async_extent;
 303
 304         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 305         BUG_ON(!async_extent); /* -ENOMEM */
 306         async_extent->start = start;
 307         async_extent->ram_size = ram_size;
 308         async_extent->compressed_size = compressed_size;
 309         async_extent->pages = pages;
 310         async_extent->nr_pages = nr_pages;
 311         async_extent->compress_type = compress_type;
 312         list_add_tail(&async_extent->list, &cow->extents);
 313         return 0;
 314 }
 315
 316 /*
 317  * we create compressed extents in two phases.  The first
 318  * phase compresses a range of pages that have already been
 319  * locked (both pages and state bits are locked).
 320  *
 321  * This is done inside an ordered work queue, and the compression
 322  * is spread across many cpus.  The actual IO submission is step
 323  * two, and the ordered work queue takes care of making sure that
 324  * happens in the same order things were put onto the queue by
 325  * writepages and friends.
 326  *
 327  * If this code finds it can't get good compression, it puts an
 328  * entry onto the work queue to write the uncompressed bytes.  This
 329  * makes sure that both compressed inodes and uncompressed inodes
 330  * are written in the same order that the flusher thread sent them
 331  * down.
 332  */
 333 static noinline int compress_file_range(struct inode *inode,
 334                                         struct page *locked_page,
 335                                         u64 start, u64 end,
 336                                         struct async_cow *async_cow,
 337                                         int *num_added)
 338 {
 339         struct btrfs_root *root = BTRFS_I(inode)->root;
 340         struct btrfs_trans_handle *trans;
 341         u64 num_bytes;
 342         u64 blocksize = root->sectorsize;
 343         u64 actual_end;
 344         u64 isize = i_size_read(inode);
 345         int ret = 0;
 346         struct page **pages = NULL;
 347         unsigned long nr_pages;
 348         unsigned long nr_pages_ret = 0;
 349         unsigned long total_compressed = 0;
 350         unsigned long total_in = 0;
 351         unsigned long max_compressed = 128 * 1024;
 352         unsigned long max_uncompressed = 128 * 1024;
 353         int i;
 354         int will_compress;
 355         int compress_type = root->fs_info->compress_type;
 356
 357         /* if this is a small write inside eof, kick off a defrag */
 358         if ((end - start + 1) < 16 * 1024 &&
 359             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 360                 btrfs_add_inode_defrag(NULL, inode);
 361
 362         actual_end = min_t(u64, isize, end + 1);
 363 again:
 364         will_compress = 0;
 365         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 366         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 367
 368         /*
 369          * we don't want to send crud past the end of i_size through
 370          * compression, that's just a waste of CPU time.  So, if the
 371          * end of the file is before the start of our current
 372          * requested range of bytes, we bail out to the uncompressed
 373          * cleanup code that can deal with all of this.
 374          *
 375          * It isn't really the fastest way to fix things, but this is a
 376          * very uncommon corner.
 377          */
 378         if (actual_end <= start)
 379                 goto cleanup_and_bail_uncompressed;
 380
 381         total_compressed = actual_end - start;
 382
 383         /* we want to make sure that amount of ram required to uncompress
 384          * an extent is reasonable, so we limit the total size in ram
 385          * of a compressed extent to 128k.  This is a crucial number
 386          * because it also controls how easily we can spread reads across
 387          * cpus for decompression.
 388          *
 389          * We also want to make sure the amount of IO required to do
 390          * a random read is reasonably small, so we limit the size of
 391          * a compressed extent to 128k.
 392          */
 393         total_compressed = min(total_compressed, max_uncompressed);
 394         num_bytes = ALIGN(end - start + 1, blocksize);
 395         num_bytes = max(blocksize,  num_bytes);
 396         total_in = 0;
 397         ret = 0;
 398
 399         /*
 400          * we do compression for mount -o compress and when the
 401          * inode has not been flagged as nocompress.  This flag can
 402          * change at any time if we discover bad compression ratios.
 403          */
 404         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 405             (btrfs_test_opt(root, COMPRESS) ||
 406              (BTRFS_I(inode)->force_compress) ||
 407              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 408                 WARN_ON(pages);
 409                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 410                 if (!pages) {
 411                         /* just bail out to the uncompressed code */
 412                         goto cont;
 413                 }
 414
 415                 if (BTRFS_I(inode)->force_compress)
 416                         compress_type = BTRFS_I(inode)->force_compress;
 417
 418                 ret = btrfs_compress_pages(compress_type,
 419                                            inode->i_mapping, start,
 420                                            total_compressed, pages,
 421                                            nr_pages, &nr_pages_ret,
 422                                            &total_in,
 423                                            &total_compressed,
 424                                            max_compressed);
 425
 426                 if (!ret) {
 427                         unsigned long offset = total_compressed &
 428                                 (PAGE_CACHE_SIZE - 1);
 429                         struct page *page = pages[nr_pages_ret - 1];
 430                         char *kaddr;
 431
 432                         /* zero the tail end of the last page, we might be
 433                          * sending it down to disk
 434                          */
 435                         if (offset) {
 436                                 kaddr = kmap_atomic(page);
 437                                 memset(kaddr + offset, 0,
 438                                        PAGE_CACHE_SIZE - offset);
 439                                 kunmap_atomic(kaddr);
 440                         }
 441                         will_compress = 1;
 442                 }
 443         }
 444 cont:
 445         if (start == 0) {
 446                 trans = btrfs_join_transaction(root);
 447                 if (IS_ERR(trans)) {
 448                         ret = PTR_ERR(trans);
 449                         trans = NULL;
 450                         goto cleanup_and_out;
 451                 }
 452                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 453
 454                 /* lets try to make an inline extent */
 455                 if (ret || total_in < (actual_end - start)) {
 456                         /* we didn't compress the entire range, try
 457                          * to make an uncompressed inline extent.
 458                          */
 459                         ret = cow_file_range_inline(trans, root, inode,
 460                                                     start, end, 0, 0, NULL);
 461                 } else {
 462                         /* try making a compressed inline extent */
 463                         ret = cow_file_range_inline(trans, root, inode,
 464                                                     start, end,
 465                                                     total_compressed,
 466                                                     compress_type, pages);
 467                 }
 468                 if (ret <= 0) {
 469                         /*
 470                          * inline extent creation worked or returned error,
 471                          * we don't need to create any more async work items.
 472                          * Unlock and free up our temp pages.
 473                          */
 474                         extent_clear_unlock_delalloc(inode,
 475                              &BTRFS_I(inode)->io_tree,
 476                              start, end, NULL,
 477                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 478                              EXTENT_CLEAR_DELALLOC |
 479                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 480
 481                         btrfs_end_transaction(trans, root);
 482                         goto free_pages_out;
 483                 }
 484                 btrfs_end_transaction(trans, root);
 485         }
 486
 487         if (will_compress) {
 488                 /*
 489                  * we aren't doing an inline extent round the compressed size
 490                  * up to a block size boundary so the allocator does sane
 491                  * things
 492                  */
 493                 total_compressed = ALIGN(total_compressed, blocksize);
 494
 495                 /*
 496                  * one last check to make sure the compression is really a
 497                  * win, compare the page count read with the blocks on disk
 498                  */
 499                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 500                 if (total_compressed >= total_in) {
 501                         will_compress = 0;
 502                 } else {
 503                         num_bytes = total_in;
 504                 }
 505         }
 506         if (!will_compress && pages) {
 507                 /*
 508                  * the compression code ran but failed to make things smaller,
 509                  * free any pages it allocated and our page pointer array
 510                  */
 511                 for (i = 0; i < nr_pages_ret; i++) {
 512                         WARN_ON(pages[i]->mapping);
 513                         page_cache_release(pages[i]);
 514                 }
 515                 kfree(pages);
 516                 pages = NULL;
 517                 total_compressed = 0;
 518                 nr_pages_ret = 0;
 519
 520                 /* flag the file so we don't compress in the future */
 521                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 522                     !(BTRFS_I(inode)->force_compress)) {
 523                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 524                 }
 525         }
 526         if (will_compress) {
 527                 *num_added += 1;
 528
 529                 /* the async work queues will take care of doing actual
 530                  * allocation on disk for these compressed pages,
 531                  * and will submit them to the elevator.
 532                  */
 533                 add_async_extent(async_cow, start, num_bytes,
 534                                  total_compressed, pages, nr_pages_ret,
 535                                  compress_type);
 536
 537                 if (start + num_bytes < end) {
 538                         start += num_bytes;
 539                         pages = NULL;
 540                         cond_resched();
 541                         goto again;
 542                 }
 543         } else {
 544 cleanup_and_bail_uncompressed:
 545                 /*
 546                  * No compression, but we still need to write the pages in
 547                  * the file we've been given so far.  redirty the locked
 548                  * page if it corresponds to our extent and set things up
 549                  * for the async work queue to run cow_file_range to do
 550                  * the normal delalloc dance
 551                  */
 552                 if (page_offset(locked_page) >= start &&
 553                     page_offset(locked_page) <= end) {
 554                         __set_page_dirty_nobuffers(locked_page);
 555                         /* unlocked later on in the async handlers */
 556                 }
 557                 add_async_extent(async_cow, start, end - start + 1,
 558                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 559                 *num_added += 1;
 560         }
 561
 562 out:
 563         return ret;
 564
 565 free_pages_out:
 566         for (i = 0; i < nr_pages_ret; i++) {
 567                 WARN_ON(pages[i]->mapping);
 568                 page_cache_release(pages[i]);
 569         }
 570         kfree(pages);
 571
 572         goto out;
 573
 574 cleanup_and_out:
 575         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 576                                      start, end, NULL,
 577                                      EXTENT_CLEAR_UNLOCK_PAGE |
 578                                      EXTENT_CLEAR_DIRTY |
 579                                      EXTENT_CLEAR_DELALLOC |
 580                                      EXTENT_SET_WRITEBACK |
 581                                      EXTENT_END_WRITEBACK);
 582         if (!trans || IS_ERR(trans))
 583                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 584         else
 585                 btrfs_abort_transaction(trans, root, ret);
 586         goto free_pages_out;
 587 }
 588
 589 /*
 590  * phase two of compressed writeback.  This is the ordered portion
 591  * of the code, which only gets called in the order the work was
 592  * queued.  We walk all the async extents created by compress_file_range
 593  * and send them down to the disk.
 594  */
 595 static noinline int submit_compressed_extents(struct inode *inode,
 596                                               struct async_cow *async_cow)
 597 {
 598         struct async_extent *async_extent;
 599         u64 alloc_hint = 0;
 600         struct btrfs_trans_handle *trans;
 601         struct btrfs_key ins;
 602         struct extent_map *em;
 603         struct btrfs_root *root = BTRFS_I(inode)->root;
 604         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 605         struct extent_io_tree *io_tree;
 606         int ret = 0;
 607
 608         if (list_empty(&async_cow->extents))
 609                 return 0;
 610
 611 again:
 612         while (!list_empty(&async_cow->extents)) {
 613                 async_extent = list_entry(async_cow->extents.next,
 614                                           struct async_extent, list);
 615                 list_del(&async_extent->list);
 616
 617                 io_tree = &BTRFS_I(inode)->io_tree;
 618
 619 retry:
 620                 /* did the compression code fall back to uncompressed IO? */
 621                 if (!async_extent->pages) {
 622                         int page_started = 0;
 623                         unsigned long nr_written = 0;
 624
 625                         lock_extent(io_tree, async_extent->start,
 626                                          async_extent->start +
 627                                          async_extent->ram_size - 1);
 628
 629                         /* allocate blocks */
 630                         ret = cow_file_range(inode, async_cow->locked_page,
 631                                              async_extent->start,
 632                                              async_extent->start +
 633                                              async_extent->ram_size - 1,
 634                                              &page_started, &nr_written, 0);
 635
 636                         /* JDM XXX */
 637
 638                         /*
 639                          * if page_started, cow_file_range inserted an
 640                          * inline extent and took care of all the unlocking
 641                          * and IO for us.  Otherwise, we need to submit
 642                          * all those pages down to the drive.
 643                          */
 644                         if (!page_started && !ret)
 645                                 extent_write_locked_range(io_tree,
 646                                                   inode, async_extent->start,
 647                                                   async_extent->start +
 648                                                   async_extent->ram_size - 1,
 649                                                   btrfs_get_extent,
 650                                                   WB_SYNC_ALL);
 651                         else if (ret)
 652                                 unlock_page(async_cow->locked_page);
 653                         kfree(async_extent);
 654                         cond_resched();
 655                         continue;
 656                 }
 657
 658                 lock_extent(io_tree, async_extent->start,
 659                             async_extent->start + async_extent->ram_size - 1);
 660
 661                 trans = btrfs_join_transaction(root);
 662                 if (IS_ERR(trans)) {
 663                         ret = PTR_ERR(trans);
 664                 } else {
 665                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 666                         ret = btrfs_reserve_extent(trans, root,
 667                                            async_extent->compressed_size,
 668                                            async_extent->compressed_size,
 669                                            0, alloc_hint, &ins, 1);
 670                         if (ret && ret != -ENOSPC)
 671                                 btrfs_abort_transaction(trans, root, ret);
 672                         btrfs_end_transaction(trans, root);
 673                 }
 674
 675                 if (ret) {
 676                         int i;
 677
 678                         for (i = 0; i < async_extent->nr_pages; i++) {
 679                                 WARN_ON(async_extent->pages[i]->mapping);
 680                                 page_cache_release(async_extent->pages[i]);
 681                         }
 682                         kfree(async_extent->pages);
 683                         async_extent->nr_pages = 0;
 684                         async_extent->pages = NULL;
 685
 686                         if (ret == -ENOSPC)
 687                                 goto retry;
 688                         goto out_free;
 689                 }
 690
 691                 /*
 692                  * here we're doing allocation and writeback of the
 693                  * compressed pages
 694                  */
 695                 btrfs_drop_extent_cache(inode, async_extent->start,
 696                                         async_extent->start +
 697                                         async_extent->ram_size - 1, 0);
 698
 699                 em = alloc_extent_map();
 700                 if (!em)
 701                         goto out_free_reserve;
 702                 em->start = async_extent->start;
 703                 em->len = async_extent->ram_size;
 704                 em->orig_start = em->start;
 705                 em->mod_start = em->start;
 706                 em->mod_len = em->len;
 707
 708                 em->block_start = ins.objectid;
 709                 em->block_len = ins.offset;
 710                 em->orig_block_len = ins.offset;
 711                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 712                 em->compress_type = async_extent->compress_type;
 713                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 714                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 715                 em->generation = -1;
 716
 717                 while (1) {
 718                         write_lock(&em_tree->lock);
 719                         ret = add_extent_mapping(em_tree, em);
 720                         if (!ret)
 721                                 list_move(&em->list,
 722                                           &em_tree->modified_extents);
 723                         write_unlock(&em_tree->lock);
 724                         if (ret != -EEXIST) {
 725                                 free_extent_map(em);
 726                                 break;
 727                         }
 728                         btrfs_drop_extent_cache(inode, async_extent->start,
 729                                                 async_extent->start +
 730                                                 async_extent->ram_size - 1, 0);
 731                 }
 732
 733                 if (ret)
 734                         goto out_free_reserve;
 735
 736                 ret = btrfs_add_ordered_extent_compress(inode,
 737                                                 async_extent->start,
 738                                                 ins.objectid,
 739                                                 async_extent->ram_size,
 740                                                 ins.offset,
 741                                                 BTRFS_ORDERED_COMPRESSED,
 742                                                 async_extent->compress_type);
 743                 if (ret)
 744                         goto out_free_reserve;
 745
 746                 /*
 747                  * clear dirty, set writeback and unlock the pages.
 748                  */
 749                 extent_clear_unlock_delalloc(inode,
 750                                 &BTRFS_I(inode)->io_tree,
 751                                 async_extent->start,
 752                                 async_extent->start +
 753                                 async_extent->ram_size - 1,
 754                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 755                                 EXTENT_CLEAR_UNLOCK |
 756                                 EXTENT_CLEAR_DELALLOC |
 757                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 758
 759                 ret = btrfs_submit_compressed_write(inode,
 760                                     async_extent->start,
 761                                     async_extent->ram_size,
 762                                     ins.objectid,
 763                                     ins.offset, async_extent->pages,
 764                                     async_extent->nr_pages);
 765                 alloc_hint = ins.objectid + ins.offset;
 766                 kfree(async_extent);
 767                 if (ret)
 768                         goto out;
 769                 cond_resched();
 770         }
 771         ret = 0;
 772 out:
 773         return ret;
 774 out_free_reserve:
 775         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 776 out_free:
 777         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 778                                      async_extent->start,
 779                                      async_extent->start +
 780                                      async_extent->ram_size - 1,
 781                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 782                                      EXTENT_CLEAR_UNLOCK |
 783                                      EXTENT_CLEAR_DELALLOC |
 784                                      EXTENT_CLEAR_DIRTY |
 785                                      EXTENT_SET_WRITEBACK |
 786                                      EXTENT_END_WRITEBACK);
 787         kfree(async_extent);
 788         goto again;
 789 }
 790
 791 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 792                                       u64 num_bytes)
 793 {
 794         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 795         struct extent_map *em;
 796         u64 alloc_hint = 0;
 797
 798         read_lock(&em_tree->lock);
 799         em = search_extent_mapping(em_tree, start, num_bytes);
 800         if (em) {
 801                 /*
 802                  * if block start isn't an actual block number then find the
 803                  * first block in this inode and use that as a hint.  If that
 804                  * block is also bogus then just don't worry about it.
 805                  */
 806                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 807                         free_extent_map(em);
 808                         em = search_extent_mapping(em_tree, 0, 0);
 809                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 810                                 alloc_hint = em->block_start;
 811                         if (em)
 812                                 free_extent_map(em);
 813                 } else {
 814                         alloc_hint = em->block_start;
 815                         free_extent_map(em);
 816                 }
 817         }
 818         read_unlock(&em_tree->lock);
 819
 820         return alloc_hint;
 821 }
 822
 823 /*
 824  * when extent_io.c finds a delayed allocation range in the file,
 825  * the call backs end up in this code.  The basic idea is to
 826  * allocate extents on disk for the range, and create ordered data structs
 827  * in ram to track those extents.
 828  *
 829  * locked_page is the page that writepage had locked already.  We use
 830  * it to make sure we don't do extra locks or unlocks.
 831  *
 832  * *page_started is set to one if we unlock locked_page and do everything
 833  * required to start IO on it.  It may be clean and already done with
 834  * IO when we return.
 835  */
 836 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 837                                      struct inode *inode,
 838                                      struct btrfs_root *root,
 839                                      struct page *locked_page,
 840                                      u64 start, u64 end, int *page_started,
 841                                      unsigned long *nr_written,
 842                                      int unlock)
 843 {
 844         u64 alloc_hint = 0;
 845         u64 num_bytes;
 846         unsigned long ram_size;
 847         u64 disk_num_bytes;
 848         u64 cur_alloc_size;
 849         u64 blocksize = root->sectorsize;
 850         struct btrfs_key ins;
 851         struct extent_map *em;
 852         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 853         int ret = 0;
 854
 855         BUG_ON(btrfs_is_free_space_inode(inode));
 856
 857         num_bytes = ALIGN(end - start + 1, blocksize);
 858         num_bytes = max(blocksize,  num_bytes);
 859         disk_num_bytes = num_bytes;
 860
 861         /* if this is a small write inside eof, kick off defrag */
 862         if (num_bytes < 64 * 1024 &&
 863             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 864                 btrfs_add_inode_defrag(trans, inode);
 865
 866         if (start == 0) {
 867                 /* lets try to make an inline extent */
 868                 ret = cow_file_range_inline(trans, root, inode,
 869                                             start, end, 0, 0, NULL);
 870                 if (ret == 0) {
 871                         extent_clear_unlock_delalloc(inode,
 872                                      &BTRFS_I(inode)->io_tree,
 873                                      start, end, NULL,
 874                                      EXTENT_CLEAR_UNLOCK_PAGE |
 875                                      EXTENT_CLEAR_UNLOCK |
 876                                      EXTENT_CLEAR_DELALLOC |
 877                                      EXTENT_CLEAR_DIRTY |
 878                                      EXTENT_SET_WRITEBACK |
 879                                      EXTENT_END_WRITEBACK);
 880
 881                         *nr_written = *nr_written +
 882                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 883                         *page_started = 1;
 884                         goto out;
 885                 } else if (ret < 0) {
 886                         btrfs_abort_transaction(trans, root, ret);
 887                         goto out_unlock;
 888                 }
 889         }
 890
 891         BUG_ON(disk_num_bytes >
 892                btrfs_super_total_bytes(root->fs_info->super_copy));
 893
 894         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 895         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 896
 897         while (disk_num_bytes > 0) {
 898                 unsigned long op;
 899
 900                 cur_alloc_size = disk_num_bytes;
 901                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 902                                            root->sectorsize, 0, alloc_hint,
 903                                            &ins, 1);
 904                 if (ret < 0) {
 905                         btrfs_abort_transaction(trans, root, ret);
 906                         goto out_unlock;
 907                 }
 908
 909                 em = alloc_extent_map();
 910                 BUG_ON(!em); /* -ENOMEM */
 911                 em->start = start;
 912                 em->orig_start = em->start;
 913                 ram_size = ins.offset;
 914                 em->len = ins.offset;
 915                 em->mod_start = em->start;
 916                 em->mod_len = em->len;
 917
 918                 em->block_start = ins.objectid;
 919                 em->block_len = ins.offset;
 920                 em->orig_block_len = ins.offset;
 921                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 922                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 923                 em->generation = -1;
 924
 925                 while (1) {
 926                         write_lock(&em_tree->lock);
 927                         ret = add_extent_mapping(em_tree, em);
 928                         if (!ret)
 929                                 list_move(&em->list,
 930                                           &em_tree->modified_extents);
 931                         write_unlock(&em_tree->lock);
 932                         if (ret != -EEXIST) {
 933                                 free_extent_map(em);
 934                                 break;
 935                         }
 936                         btrfs_drop_extent_cache(inode, start,
 937                                                 start + ram_size - 1, 0);
 938                 }
 939
 940                 cur_alloc_size = ins.offset;
 941                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 942                                                ram_size, cur_alloc_size, 0);
 943                 BUG_ON(ret); /* -ENOMEM */
 944
 945                 if (root->root_key.objectid ==
 946                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 947                         ret = btrfs_reloc_clone_csums(inode, start,
 948                                                       cur_alloc_size);
 949                         if (ret) {
 950                                 btrfs_abort_transaction(trans, root, ret);
 951                                 goto out_unlock;
 952                         }
 953                 }
 954
 955                 if (disk_num_bytes < cur_alloc_size)
 956                         break;
 957
 958                 /* we're not doing compressed IO, don't unlock the first
 959                  * page (which the caller expects to stay locked), don't
 960                  * clear any dirty bits and don't set any writeback bits
 961                  *
 962                  * Do set the Private2 bit so we know this page was properly
 963                  * setup for writepage
 964                  */
 965                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 966                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 967                         EXTENT_SET_PRIVATE2;
 968
 969                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 970                                              start, start + ram_size - 1,
 971                                              locked_page, op);
 972                 disk_num_bytes -= cur_alloc_size;
 973                 num_bytes -= cur_alloc_size;
 974                 alloc_hint = ins.objectid + ins.offset;
 975                 start += cur_alloc_size;
 976         }
 977 out:
 978         return ret;
 979
 980 out_unlock:
 981         extent_clear_unlock_delalloc(inode,
 982                      &BTRFS_I(inode)->io_tree,
 983                      start, end, locked_page,
 984                      EXTENT_CLEAR_UNLOCK_PAGE |
 985                      EXTENT_CLEAR_UNLOCK |
 986                      EXTENT_CLEAR_DELALLOC |
 987                      EXTENT_CLEAR_DIRTY |
 988                      EXTENT_SET_WRITEBACK |
 989                      EXTENT_END_WRITEBACK);
 990
 991         goto out;
 992 }
 993
 994 static noinline int cow_file_range(struct inode *inode,
 995                                    struct page *locked_page,
 996                                    u64 start, u64 end, int *page_started,
 997                                    unsigned long *nr_written,
 998                                    int unlock)
 999 {
1000         struct btrfs_trans_handle *trans;
1001         struct btrfs_root *root = BTRFS_I(inode)->root;
1002         int ret;
1003
1004         trans = btrfs_join_transaction(root);
1005         if (IS_ERR(trans)) {
1006                 extent_clear_unlock_delalloc(inode,
1007                              &BTRFS_I(inode)->io_tree,
1008                              start, end, locked_page,
1009                              EXTENT_CLEAR_UNLOCK_PAGE |
1010                              EXTENT_CLEAR_UNLOCK |
1011                              EXTENT_CLEAR_DELALLOC |
1012                              EXTENT_CLEAR_DIRTY |
1013                              EXTENT_SET_WRITEBACK |
1014                              EXTENT_END_WRITEBACK);
1015                 return PTR_ERR(trans);
1016         }
1017         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1018
1019         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1020                                page_started, nr_written, unlock);
1021
1022         btrfs_end_transaction(trans, root);
1023
1024         return ret;
1025 }
1026
1027 /*
1028  * work queue call back to started compression on a file and pages
1029  */
1030 static noinline void async_cow_start(struct btrfs_work *work)
1031 {
1032         struct async_cow *async_cow;
1033         int num_added = 0;
1034         async_cow = container_of(work, struct async_cow, work);
1035
1036         compress_file_range(async_cow->inode, async_cow->locked_page,
1037                             async_cow->start, async_cow->end, async_cow,
1038                             &num_added);
1039         if (num_added == 0) {
1040                 btrfs_add_delayed_iput(async_cow->inode);
1041                 async_cow->inode = NULL;
1042         }
1043 }
1044
1045 /*
1046  * work queue call back to submit previously compressed pages
1047  */
1048 static noinline void async_cow_submit(struct btrfs_work *work)
1049 {
1050         struct async_cow *async_cow;
1051         struct btrfs_root *root;
1052         unsigned long nr_pages;
1053
1054         async_cow = container_of(work, struct async_cow, work);
1055
1056         root = async_cow->root;
1057         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1058                 PAGE_CACHE_SHIFT;
1059
1060         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1061             5 * 1024 * 1024 &&
1062             waitqueue_active(&root->fs_info->async_submit_wait))
1063                 wake_up(&root->fs_info->async_submit_wait);
1064
1065         if (async_cow->inode)
1066                 submit_compressed_extents(async_cow->inode, async_cow);
1067 }
1068
1069 static noinline void async_cow_free(struct btrfs_work *work)
1070 {
1071         struct async_cow *async_cow;
1072         async_cow = container_of(work, struct async_cow, work);
1073         if (async_cow->inode)
1074                 btrfs_add_delayed_iput(async_cow->inode);
1075         kfree(async_cow);
1076 }
1077
1078 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1079                                 u64 start, u64 end, int *page_started,
1080                                 unsigned long *nr_written)
1081 {
1082         struct async_cow *async_cow;
1083         struct btrfs_root *root = BTRFS_I(inode)->root;
1084         unsigned long nr_pages;
1085         u64 cur_end;
1086         int limit = 10 * 1024 * 1024;
1087
1088         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1089                          1, 0, NULL, GFP_NOFS);
1090         while (start < end) {
1091                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1092                 BUG_ON(!async_cow); /* -ENOMEM */
1093                 async_cow->inode = igrab(inode);
1094                 async_cow->root = root;
1095                 async_cow->locked_page = locked_page;
1096                 async_cow->start = start;
1097
1098                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1099                         cur_end = end;
1100                 else
1101                         cur_end = min(end, start + 512 * 1024 - 1);
1102
1103                 async_cow->end = cur_end;
1104                 INIT_LIST_HEAD(&async_cow->extents);
1105
1106                 async_cow->work.func = async_cow_start;
1107                 async_cow->work.ordered_func = async_cow_submit;
1108                 async_cow->work.ordered_free = async_cow_free;
1109                 async_cow->work.flags = 0;
1110
1111                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1112                         PAGE_CACHE_SHIFT;
1113                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1114
1115                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1116                                    &async_cow->work);
1117
1118                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1119                         wait_event(root->fs_info->async_submit_wait,
1120                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1121                             limit));
1122                 }
1123
1124                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1125                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1126                         wait_event(root->fs_info->async_submit_wait,
1127                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1128                            0));
1129                 }
1130
1131                 *nr_written += nr_pages;
1132                 start = cur_end + 1;
1133         }
1134         *page_started = 1;
1135         return 0;
1136 }
1137
1138 static noinline int csum_exist_in_range(struct btrfs_root *root,
1139                                         u64 bytenr, u64 num_bytes)
1140 {
1141         int ret;
1142         struct btrfs_ordered_sum *sums;
1143         LIST_HEAD(list);
1144
1145         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1146                                        bytenr + num_bytes - 1, &list, 0);
1147         if (ret == 0 && list_empty(&list))
1148                 return 0;
1149
1150         while (!list_empty(&list)) {
1151                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1152                 list_del(&sums->list);
1153                 kfree(sums);
1154         }
1155         return 1;
1156 }
1157
1158 /*
1159  * when nowcow writeback call back.  This checks for snapshots or COW copies
1160  * of the extents that exist in the file, and COWs the file as required.
1161  *
1162  * If no cow copies or snapshots exist, we write directly to the existing
1163  * blocks on disk
1164  */
1165 static noinline int run_delalloc_nocow(struct inode *inode,
1166                                        struct page *locked_page,
1167                               u64 start, u64 end, int *page_started, int force,
1168                               unsigned long *nr_written)
1169 {
1170         struct btrfs_root *root = BTRFS_I(inode)->root;
1171         struct btrfs_trans_handle *trans;
1172         struct extent_buffer *leaf;
1173         struct btrfs_path *path;
1174         struct btrfs_file_extent_item *fi;
1175         struct btrfs_key found_key;
1176         u64 cow_start;
1177         u64 cur_offset;
1178         u64 extent_end;
1179         u64 extent_offset;
1180         u64 disk_bytenr;
1181         u64 num_bytes;
1182         u64 disk_num_bytes;
1183         int extent_type;
1184         int ret, err;
1185         int type;
1186         int nocow;
1187         int check_prev = 1;
1188         bool nolock;
1189         u64 ino = btrfs_ino(inode);
1190
1191         path = btrfs_alloc_path();
1192         if (!path) {
1193                 extent_clear_unlock_delalloc(inode,
1194                              &BTRFS_I(inode)->io_tree,
1195                              start, end, locked_page,
1196                              EXTENT_CLEAR_UNLOCK_PAGE |
1197                              EXTENT_CLEAR_UNLOCK |
1198                              EXTENT_CLEAR_DELALLOC |
1199                              EXTENT_CLEAR_DIRTY |
1200                              EXTENT_SET_WRITEBACK |
1201                              EXTENT_END_WRITEBACK);
1202                 return -ENOMEM;
1203         }
1204
1205         nolock = btrfs_is_free_space_inode(inode);
1206
1207         if (nolock)
1208                 trans = btrfs_join_transaction_nolock(root);
1209         else
1210                 trans = btrfs_join_transaction(root);
1211
1212         if (IS_ERR(trans)) {
1213                 extent_clear_unlock_delalloc(inode,
1214                              &BTRFS_I(inode)->io_tree,
1215                              start, end, locked_page,
1216                              EXTENT_CLEAR_UNLOCK_PAGE |
1217                              EXTENT_CLEAR_UNLOCK |
1218                              EXTENT_CLEAR_DELALLOC |
1219                              EXTENT_CLEAR_DIRTY |
1220                              EXTENT_SET_WRITEBACK |
1221                              EXTENT_END_WRITEBACK);
1222                 btrfs_free_path(path);
1223                 return PTR_ERR(trans);
1224         }
1225
1226         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1227
1228         cow_start = (u64)-1;
1229         cur_offset = start;
1230         while (1) {
1231                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1232                                                cur_offset, 0);
1233                 if (ret < 0) {
1234                         btrfs_abort_transaction(trans, root, ret);
1235                         goto error;
1236                 }
1237                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1238                         leaf = path->nodes[0];
1239                         btrfs_item_key_to_cpu(leaf, &found_key,
1240                                               path->slots[0] - 1);
1241                         if (found_key.objectid == ino &&
1242                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1243                                 path->slots[0]--;
1244                 }
1245                 check_prev = 0;
1246 next_slot:
1247                 leaf = path->nodes[0];
1248                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1249                         ret = btrfs_next_leaf(root, path);
1250                         if (ret < 0) {
1251                                 btrfs_abort_transaction(trans, root, ret);
1252                                 goto error;
1253                         }
1254                         if (ret > 0)
1255                                 break;
1256                         leaf = path->nodes[0];
1257                 }
1258
1259                 nocow = 0;
1260                 disk_bytenr = 0;
1261                 num_bytes = 0;
1262                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1263
1264                 if (found_key.objectid > ino ||
1265                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1266                     found_key.offset > end)
1267                         break;
1268
1269                 if (found_key.offset > cur_offset) {
1270                         extent_end = found_key.offset;
1271                         extent_type = 0;
1272                         goto out_check;
1273                 }
1274
1275                 fi = btrfs_item_ptr(leaf, path->slots[0],
1276                                     struct btrfs_file_extent_item);
1277                 extent_type = btrfs_file_extent_type(leaf, fi);
1278
1279                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1280                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1281                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1282                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1283                         extent_end = found_key.offset +
1284                                 btrfs_file_extent_num_bytes(leaf, fi);
1285                         disk_num_bytes =
1286                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1287                         if (extent_end <= start) {
1288                                 path->slots[0]++;
1289                                 goto next_slot;
1290                         }
1291                         if (disk_bytenr == 0)
1292                                 goto out_check;
1293                         if (btrfs_file_extent_compression(leaf, fi) ||
1294                             btrfs_file_extent_encryption(leaf, fi) ||
1295                             btrfs_file_extent_other_encoding(leaf, fi))
1296                                 goto out_check;
1297                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1298                                 goto out_check;
1299                         if (btrfs_extent_readonly(root, disk_bytenr))
1300                                 goto out_check;
1301                         if (btrfs_cross_ref_exist(trans, root, ino,
1302                                                   found_key.offset -
1303                                                   extent_offset, disk_bytenr))
1304                                 goto out_check;
1305                         disk_bytenr += extent_offset;
1306                         disk_bytenr += cur_offset - found_key.offset;
1307                         num_bytes = min(end + 1, extent_end) - cur_offset;
1308                         /*
1309                          * force cow if csum exists in the range.
1310                          * this ensure that csum for a given extent are
1311                          * either valid or do not exist.
1312                          */
1313                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1314                                 goto out_check;
1315                         nocow = 1;
1316                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1317                         extent_end = found_key.offset +
1318                                 btrfs_file_extent_inline_len(leaf, fi);
1319                         extent_end = ALIGN(extent_end, root->sectorsize);
1320                 } else {
1321                         BUG_ON(1);
1322                 }
1323 out_check:
1324                 if (extent_end <= start) {
1325                         path->slots[0]++;
1326                         goto next_slot;
1327                 }
1328                 if (!nocow) {
1329                         if (cow_start == (u64)-1)
1330                                 cow_start = cur_offset;
1331                         cur_offset = extent_end;
1332                         if (cur_offset > end)
1333                                 break;
1334                         path->slots[0]++;
1335                         goto next_slot;
1336                 }
1337
1338                 btrfs_release_path(path);
1339                 if (cow_start != (u64)-1) {
1340                         ret = __cow_file_range(trans, inode, root, locked_page,
1341                                                cow_start, found_key.offset - 1,
1342                                                page_started, nr_written, 1);
1343                         if (ret) {
1344                                 btrfs_abort_transaction(trans, root, ret);
1345                                 goto error;
1346                         }
1347                         cow_start = (u64)-1;
1348                 }
1349
1350                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1351                         struct extent_map *em;
1352                         struct extent_map_tree *em_tree;
1353                         em_tree = &BTRFS_I(inode)->extent_tree;
1354                         em = alloc_extent_map();
1355                         BUG_ON(!em); /* -ENOMEM */
1356                         em->start = cur_offset;
1357                         em->orig_start = found_key.offset - extent_offset;
1358                         em->len = num_bytes;
1359                         em->block_len = num_bytes;
1360                         em->block_start = disk_bytenr;
1361                         em->orig_block_len = disk_num_bytes;
1362                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1363                         em->mod_start = em->start;
1364                         em->mod_len = em->len;
1365                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1366                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1367                         em->generation = -1;
1368                         while (1) {
1369                                 write_lock(&em_tree->lock);
1370                                 ret = add_extent_mapping(em_tree, em);
1371                                 if (!ret)
1372                                         list_move(&em->list,
1373                                                   &em_tree->modified_extents);
1374                                 write_unlock(&em_tree->lock);
1375                                 if (ret != -EEXIST) {
1376                                         free_extent_map(em);
1377                                         break;
1378                                 }
1379                                 btrfs_drop_extent_cache(inode, em->start,
1380                                                 em->start + em->len - 1, 0);
1381                         }
1382                         type = BTRFS_ORDERED_PREALLOC;
1383                 } else {
1384                         type = BTRFS_ORDERED_NOCOW;
1385                 }
1386
1387                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1388                                                num_bytes, num_bytes, type);
1389                 BUG_ON(ret); /* -ENOMEM */
1390
1391                 if (root->root_key.objectid ==
1392                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1393                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1394                                                       num_bytes);
1395                         if (ret) {
1396                                 btrfs_abort_transaction(trans, root, ret);
1397                                 goto error;
1398                         }
1399                 }
1400
1401                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1402                                 cur_offset, cur_offset + num_bytes - 1,
1403                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1404                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1405                                 EXTENT_SET_PRIVATE2);
1406                 cur_offset = extent_end;
1407                 if (cur_offset > end)
1408                         break;
1409         }
1410         btrfs_release_path(path);
1411
1412         if (cur_offset <= end && cow_start == (u64)-1) {
1413                 cow_start = cur_offset;
1414                 cur_offset = end;
1415         }
1416
1417         if (cow_start != (u64)-1) {
1418                 ret = __cow_file_range(trans, inode, root, locked_page,
1419                                        cow_start, end,
1420                                        page_started, nr_written, 1);
1421                 if (ret) {
1422                         btrfs_abort_transaction(trans, root, ret);
1423                         goto error;
1424                 }
1425         }
1426
1427 error:
1428         err = btrfs_end_transaction(trans, root);
1429         if (!ret)
1430                 ret = err;
1431
1432         if (ret && cur_offset < end)
1433                 extent_clear_unlock_delalloc(inode,
1434                              &BTRFS_I(inode)->io_tree,
1435                              cur_offset, end, locked_page,
1436                              EXTENT_CLEAR_UNLOCK_PAGE |
1437                              EXTENT_CLEAR_UNLOCK |
1438                              EXTENT_CLEAR_DELALLOC |
1439                              EXTENT_CLEAR_DIRTY |
1440                              EXTENT_SET_WRITEBACK |
1441                              EXTENT_END_WRITEBACK);
1442
1443         btrfs_free_path(path);
1444         return ret;
1445 }
1446
1447 /*
1448  * extent_io.c call back to do delayed allocation processing
1449  */
1450 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1451                               u64 start, u64 end, int *page_started,
1452                               unsigned long *nr_written)
1453 {
1454         int ret;
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456
1457         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1458                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1459                                          page_started, 1, nr_written);
1460         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1461                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1462                                          page_started, 0, nr_written);
1463         } else if (!btrfs_test_opt(root, COMPRESS) &&
1464                    !(BTRFS_I(inode)->force_compress) &&
1465                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1466                 ret = cow_file_range(inode, locked_page, start, end,
1467                                       page_started, nr_written, 1);
1468         } else {
1469                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1470                         &BTRFS_I(inode)->runtime_flags);
1471                 ret = cow_file_range_async(inode, locked_page, start, end,
1472                                            page_started, nr_written);
1473         }
1474         return ret;
1475 }
1476
1477 static void btrfs_split_extent_hook(struct inode *inode,
1478                                     struct extent_state *orig, u64 split)
1479 {
1480         /* not delalloc, ignore it */
1481         if (!(orig->state & EXTENT_DELALLOC))
1482                 return;
1483
1484         spin_lock(&BTRFS_I(inode)->lock);
1485         BTRFS_I(inode)->outstanding_extents++;
1486         spin_unlock(&BTRFS_I(inode)->lock);
1487 }
1488
1489 /*
1490  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1491  * extents so we can keep track of new extents that are just merged onto old
1492  * extents, such as when we are doing sequential writes, so we can properly
1493  * account for the metadata space we'll need.
1494  */
1495 static void btrfs_merge_extent_hook(struct inode *inode,
1496                                     struct extent_state *new,
1497                                     struct extent_state *other)
1498 {
1499         /* not delalloc, ignore it */
1500         if (!(other->state & EXTENT_DELALLOC))
1501                 return;
1502
1503         spin_lock(&BTRFS_I(inode)->lock);
1504         BTRFS_I(inode)->outstanding_extents--;
1505         spin_unlock(&BTRFS_I(inode)->lock);
1506 }
1507
1508 /*
1509  * extent_io.c set_bit_hook, used to track delayed allocation
1510  * bytes in this file, and to maintain the list of inodes that
1511  * have pending delalloc work to be done.
1512  */
1513 static void btrfs_set_bit_hook(struct inode *inode,
1514                                struct extent_state *state, int *bits)
1515 {
1516
1517         /*
1518          * set_bit and clear bit hooks normally require _irqsave/restore
1519          * but in this case, we are only testing for the DELALLOC
1520          * bit, which is only set or cleared with irqs on
1521          */
1522         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1523                 struct btrfs_root *root = BTRFS_I(inode)->root;
1524                 u64 len = state->end + 1 - state->start;
1525                 bool do_list = !btrfs_is_free_space_inode(inode);
1526
1527                 if (*bits & EXTENT_FIRST_DELALLOC) {
1528                         *bits &= ~EXTENT_FIRST_DELALLOC;
1529                 } else {
1530                         spin_lock(&BTRFS_I(inode)->lock);
1531                         BTRFS_I(inode)->outstanding_extents++;
1532                         spin_unlock(&BTRFS_I(inode)->lock);
1533                 }
1534
1535                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1536                                      root->fs_info->delalloc_batch);
1537                 spin_lock(&BTRFS_I(inode)->lock);
1538                 BTRFS_I(inode)->delalloc_bytes += len;
1539                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1540                                          &BTRFS_I(inode)->runtime_flags)) {
1541                         spin_lock(&root->fs_info->delalloc_lock);
1542                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1543                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1544                                               &root->fs_info->delalloc_inodes);
1545                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1546                                         &BTRFS_I(inode)->runtime_flags);
1547                         }
1548                         spin_unlock(&root->fs_info->delalloc_lock);
1549                 }
1550                 spin_unlock(&BTRFS_I(inode)->lock);
1551         }
1552 }
1553
1554 /*
1555  * extent_io.c clear_bit_hook, see set_bit_hook for why
1556  */
1557 static void btrfs_clear_bit_hook(struct inode *inode,
1558                                  struct extent_state *state, int *bits)
1559 {
1560         /*
1561          * set_bit and clear bit hooks normally require _irqsave/restore
1562          * but in this case, we are only testing for the DELALLOC
1563          * bit, which is only set or cleared with irqs on
1564          */
1565         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1566                 struct btrfs_root *root = BTRFS_I(inode)->root;
1567                 u64 len = state->end + 1 - state->start;
1568                 bool do_list = !btrfs_is_free_space_inode(inode);
1569
1570                 if (*bits & EXTENT_FIRST_DELALLOC) {
1571                         *bits &= ~EXTENT_FIRST_DELALLOC;
1572                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1573                         spin_lock(&BTRFS_I(inode)->lock);
1574                         BTRFS_I(inode)->outstanding_extents--;
1575                         spin_unlock(&BTRFS_I(inode)->lock);
1576                 }
1577
1578                 if (*bits & EXTENT_DO_ACCOUNTING)
1579                         btrfs_delalloc_release_metadata(inode, len);
1580
1581                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1582                     && do_list)
1583                         btrfs_free_reserved_data_space(inode, len);
1584
1585                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1586                                      root->fs_info->delalloc_batch);
1587                 spin_lock(&BTRFS_I(inode)->lock);
1588                 BTRFS_I(inode)->delalloc_bytes -= len;
1589                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1590                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1591                              &BTRFS_I(inode)->runtime_flags)) {
1592                         spin_lock(&root->fs_info->delalloc_lock);
1593                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1594                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1595                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1596                                           &BTRFS_I(inode)->runtime_flags);
1597                         }
1598                         spin_unlock(&root->fs_info->delalloc_lock);
1599                 }
1600                 spin_unlock(&BTRFS_I(inode)->lock);
1601         }
1602 }
1603
1604 /*
1605  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1606  * we don't create bios that span stripes or chunks
1607  */
1608 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1609                          size_t size, struct bio *bio,
1610                          unsigned long bio_flags)
1611 {
1612         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1613         u64 logical = (u64)bio->bi_sector << 9;
1614         u64 length = 0;
1615         u64 map_length;
1616         int ret;
1617
1618         if (bio_flags & EXTENT_BIO_COMPRESSED)
1619                 return 0;
1620
1621         length = bio->bi_size;
1622         map_length = length;
1623         ret = btrfs_map_block(root->fs_info, rw, logical,
1624                               &map_length, NULL, 0);
1625         /* Will always return 0 with map_multi == NULL */
1626         BUG_ON(ret < 0);
1627         if (map_length < length + size)
1628                 return 1;
1629         return 0;
1630 }
1631
1632 /*
1633  * in order to insert checksums into the metadata in large chunks,
1634  * we wait until bio submission time.   All the pages in the bio are
1635  * checksummed and sums are attached onto the ordered extent record.
1636  *
1637  * At IO completion time the cums attached on the ordered extent record
1638  * are inserted into the btree
1639  */
1640 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1641                                     struct bio *bio, int mirror_num,
1642                                     unsigned long bio_flags,
1643                                     u64 bio_offset)
1644 {
1645         struct btrfs_root *root = BTRFS_I(inode)->root;
1646         int ret = 0;
1647
1648         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1649         BUG_ON(ret); /* -ENOMEM */
1650         return 0;
1651 }
1652
1653 /*
1654  * in order to insert checksums into the metadata in large chunks,
1655  * we wait until bio submission time.   All the pages in the bio are
1656  * checksummed and sums are attached onto the ordered extent record.
1657  *
1658  * At IO completion time the cums attached on the ordered extent record
1659  * are inserted into the btree
1660  */
1661 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1662                           int mirror_num, unsigned long bio_flags,
1663                           u64 bio_offset)
1664 {
1665         struct btrfs_root *root = BTRFS_I(inode)->root;
1666         int ret;
1667
1668         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1669         if (ret)
1670                 bio_endio(bio, ret);
1671         return ret;
1672 }
1673
1674 /*
1675  * extent_io.c submission hook. This does the right thing for csum calculation
1676  * on write, or reading the csums from the tree before a read
1677  */
1678 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1679                           int mirror_num, unsigned long bio_flags,
1680                           u64 bio_offset)
1681 {
1682         struct btrfs_root *root = BTRFS_I(inode)->root;
1683         int ret = 0;
1684         int skip_sum;
1685         int metadata = 0;
1686         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1687
1688         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1689
1690         if (btrfs_is_free_space_inode(inode))
1691                 metadata = 2;
1692
1693         if (!(rw & REQ_WRITE)) {
1694                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1695                 if (ret)
1696                         goto out;
1697
1698                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1699                         ret = btrfs_submit_compressed_read(inode, bio,
1700                                                            mirror_num,
1701                                                            bio_flags);
1702                         goto out;
1703                 } else if (!skip_sum) {
1704                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1705                         if (ret)
1706                                 goto out;
1707                 }
1708                 goto mapit;
1709         } else if (async && !skip_sum) {
1710                 /* csum items have already been cloned */
1711                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1712                         goto mapit;
1713                 /* we're doing a write, do the async checksumming */
1714                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1715                                    inode, rw, bio, mirror_num,
1716                                    bio_flags, bio_offset,
1717                                    __btrfs_submit_bio_start,
1718                                    __btrfs_submit_bio_done);
1719                 goto out;
1720         } else if (!skip_sum) {
1721                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1722                 if (ret)
1723                         goto out;
1724         }
1725
1726 mapit:
1727         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1728
1729 out:
1730         if (ret < 0)
1731                 bio_endio(bio, ret);
1732         return ret;
1733 }
1734
1735 /*
1736  * given a list of ordered sums record them in the inode.  This happens
1737  * at IO completion time based on sums calculated at bio submission time.
1738  */
1739 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1740                              struct inode *inode, u64 file_offset,
1741                              struct list_head *list)
1742 {
1743         struct btrfs_ordered_sum *sum;
1744
1745         list_for_each_entry(sum, list, list) {
1746                 btrfs_csum_file_blocks(trans,
1747                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1748         }
1749         return 0;
1750 }
1751
1752 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1753                               struct extent_state **cached_state)
1754 {
1755         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1756         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1757                                    cached_state, GFP_NOFS);
1758 }
1759
1760 /* see btrfs_writepage_start_hook for details on why this is required */
1761 struct btrfs_writepage_fixup {
1762         struct page *page;
1763         struct btrfs_work work;
1764 };
1765
1766 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1767 {
1768         struct btrfs_writepage_fixup *fixup;
1769         struct btrfs_ordered_extent *ordered;
1770         struct extent_state *cached_state = NULL;
1771         struct page *page;
1772         struct inode *inode;
1773         u64 page_start;
1774         u64 page_end;
1775         int ret;
1776
1777         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1778         page = fixup->page;
1779 again:
1780         lock_page(page);
1781         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1782                 ClearPageChecked(page);
1783                 goto out_page;
1784         }
1785
1786         inode = page->mapping->host;
1787         page_start = page_offset(page);
1788         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1789
1790         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1791                          &cached_state);
1792
1793         /* already ordered? We're done */
1794         if (PagePrivate2(page))
1795                 goto out;
1796
1797         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1798         if (ordered) {
1799                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1800                                      page_end, &cached_state, GFP_NOFS);
1801                 unlock_page(page);
1802                 btrfs_start_ordered_extent(inode, ordered, 1);
1803                 btrfs_put_ordered_extent(ordered);
1804                 goto again;
1805         }
1806
1807         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1808         if (ret) {
1809                 mapping_set_error(page->mapping, ret);
1810                 end_extent_writepage(page, ret, page_start, page_end);
1811                 ClearPageChecked(page);
1812                 goto out;
1813          }
1814
1815         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1816         ClearPageChecked(page);
1817         set_page_dirty(page);
1818 out:
1819         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1820                              &cached_state, GFP_NOFS);
1821 out_page:
1822         unlock_page(page);
1823         page_cache_release(page);
1824         kfree(fixup);
1825 }
1826
1827 /*
1828  * There are a few paths in the higher layers of the kernel that directly
1829  * set the page dirty bit without asking the filesystem if it is a
1830  * good idea.  This causes problems because we want to make sure COW
1831  * properly happens and the data=ordered rules are followed.
1832  *
1833  * In our case any range that doesn't have the ORDERED bit set
1834  * hasn't been properly setup for IO.  We kick off an async process
1835  * to fix it up.  The async helper will wait for ordered extents, set
1836  * the delalloc bit and make it safe to write the page.
1837  */
1838 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1839 {
1840         struct inode *inode = page->mapping->host;
1841         struct btrfs_writepage_fixup *fixup;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843
1844         /* this page is properly in the ordered list */
1845         if (TestClearPagePrivate2(page))
1846                 return 0;
1847
1848         if (PageChecked(page))
1849                 return -EAGAIN;
1850
1851         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1852         if (!fixup)
1853                 return -EAGAIN;
1854
1855         SetPageChecked(page);
1856         page_cache_get(page);
1857         fixup->work.func = btrfs_writepage_fixup_worker;
1858         fixup->page = page;
1859         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1860         return -EBUSY;
1861 }
1862
1863 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1864                                        struct inode *inode, u64 file_pos,
1865                                        u64 disk_bytenr, u64 disk_num_bytes,
1866                                        u64 num_bytes, u64 ram_bytes,
1867                                        u8 compression, u8 encryption,
1868                                        u16 other_encoding, int extent_type)
1869 {
1870         struct btrfs_root *root = BTRFS_I(inode)->root;
1871         struct btrfs_file_extent_item *fi;
1872         struct btrfs_path *path;
1873         struct extent_buffer *leaf;
1874         struct btrfs_key ins;
1875         int ret;
1876
1877         path = btrfs_alloc_path();
1878         if (!path)
1879                 return -ENOMEM;
1880
1881         path->leave_spinning = 1;
1882
1883         /*
1884          * we may be replacing one extent in the tree with another.
1885          * The new extent is pinned in the extent map, and we don't want
1886          * to drop it from the cache until it is completely in the btree.
1887          *
1888          * So, tell btrfs_drop_extents to leave this extent in the cache.
1889          * the caller is expected to unpin it and allow it to be merged
1890          * with the others.
1891          */
1892         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1893                                  file_pos + num_bytes, 0);
1894         if (ret)
1895                 goto out;
1896
1897         ins.objectid = btrfs_ino(inode);
1898         ins.offset = file_pos;
1899         ins.type = BTRFS_EXTENT_DATA_KEY;
1900         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1901         if (ret)
1902                 goto out;
1903         leaf = path->nodes[0];
1904         fi = btrfs_item_ptr(leaf, path->slots[0],
1905                             struct btrfs_file_extent_item);
1906         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1907         btrfs_set_file_extent_type(leaf, fi, extent_type);
1908         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1909         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1910         btrfs_set_file_extent_offset(leaf, fi, 0);
1911         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1912         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1913         btrfs_set_file_extent_compression(leaf, fi, compression);
1914         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1915         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1916
1917         btrfs_mark_buffer_dirty(leaf);
1918         btrfs_release_path(path);
1919
1920         inode_add_bytes(inode, num_bytes);
1921
1922         ins.objectid = disk_bytenr;
1923         ins.offset = disk_num_bytes;
1924         ins.type = BTRFS_EXTENT_ITEM_KEY;
1925         ret = btrfs_alloc_reserved_file_extent(trans, root,
1926                                         root->root_key.objectid,
1927                                         btrfs_ino(inode), file_pos, &ins);
1928 out:
1929         btrfs_free_path(path);
1930
1931         return ret;
1932 }
1933
1934 /* snapshot-aware defrag */
1935 struct sa_defrag_extent_backref {
1936         struct rb_node node;
1937         struct old_sa_defrag_extent *old;
1938         u64 root_id;
1939         u64 inum;
1940         u64 file_pos;
1941         u64 extent_offset;
1942         u64 num_bytes;
1943         u64 generation;
1944 };
1945
1946 struct old_sa_defrag_extent {
1947         struct list_head list;
1948         struct new_sa_defrag_extent *new;
1949
1950         u64 extent_offset;
1951         u64 bytenr;
1952         u64 offset;
1953         u64 len;
1954         int count;
1955 };
1956
1957 struct new_sa_defrag_extent {
1958         struct rb_root root;
1959         struct list_head head;
1960         struct btrfs_path *path;
1961         struct inode *inode;
1962         u64 file_pos;
1963         u64 len;
1964         u64 bytenr;
1965         u64 disk_len;
1966         u8 compress_type;
1967 };
1968
1969 static int backref_comp(struct sa_defrag_extent_backref *b1,
1970                         struct sa_defrag_extent_backref *b2)
1971 {
1972         if (b1->root_id < b2->root_id)
1973                 return -1;
1974         else if (b1->root_id > b2->root_id)
1975                 return 1;
1976
1977         if (b1->inum < b2->inum)
1978                 return -1;
1979         else if (b1->inum > b2->inum)
1980                 return 1;
1981
1982         if (b1->file_pos < b2->file_pos)
1983                 return -1;
1984         else if (b1->file_pos > b2->file_pos)
1985                 return 1;
1986
1987         /*
1988          * [------------------------------] ===> (a range of space)
1989          *     |<--->|   |<---->| =============> (fs/file tree A)
1990          * |<---------------------------->| ===> (fs/file tree B)
1991          *
1992          * A range of space can refer to two file extents in one tree while
1993          * refer to only one file extent in another tree.
1994          *
1995          * So we may process a disk offset more than one time(two extents in A)
1996          * and locate at the same extent(one extent in B), then insert two same
1997          * backrefs(both refer to the extent in B).
1998          */
1999         return 0;
2000 }
2001
2002 static void backref_insert(struct rb_root *root,
2003                            struct sa_defrag_extent_backref *backref)
2004 {
2005         struct rb_node **p = &root->rb_node;
2006         struct rb_node *parent = NULL;
2007         struct sa_defrag_extent_backref *entry;
2008         int ret;
2009
2010         while (*p) {
2011                 parent = *p;
2012                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2013
2014                 ret = backref_comp(backref, entry);
2015                 if (ret < 0)
2016                         p = &(*p)->rb_left;
2017                 else
2018                         p = &(*p)->rb_right;
2019         }
2020
2021         rb_link_node(&backref->node, parent, p);
2022         rb_insert_color(&backref->node, root);
2023 }
2024
2025 /*
2026  * Note the backref might has changed, and in this case we just return 0.
2027  */
2028 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2029                                        void *ctx)
2030 {
2031         struct btrfs_file_extent_item *extent;
2032         struct btrfs_fs_info *fs_info;
2033         struct old_sa_defrag_extent *old = ctx;
2034         struct new_sa_defrag_extent *new = old->new;
2035         struct btrfs_path *path = new->path;
2036         struct btrfs_key key;
2037         struct btrfs_root *root;
2038         struct sa_defrag_extent_backref *backref;
2039         struct extent_buffer *leaf;
2040         struct inode *inode = new->inode;
2041         int slot;
2042         int ret;
2043         u64 extent_offset;
2044         u64 num_bytes;
2045
2046         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2047             inum == btrfs_ino(inode))
2048                 return 0;
2049
2050         key.objectid = root_id;
2051         key.type = BTRFS_ROOT_ITEM_KEY;
2052         key.offset = (u64)-1;
2053
2054         fs_info = BTRFS_I(inode)->root->fs_info;
2055         root = btrfs_read_fs_root_no_name(fs_info, &key);
2056         if (IS_ERR(root)) {
2057                 if (PTR_ERR(root) == -ENOENT)
2058                         return 0;
2059                 WARN_ON(1);
2060                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2061                          inum, offset, root_id);
2062                 return PTR_ERR(root);
2063         }
2064
2065         key.objectid = inum;
2066         key.type = BTRFS_EXTENT_DATA_KEY;
2067         if (offset > (u64)-1 << 32)
2068                 key.offset = 0;
2069         else
2070                 key.offset = offset;
2071
2072         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2073         if (ret < 0) {
2074                 WARN_ON(1);
2075                 return ret;
2076         }
2077
2078         while (1) {
2079                 cond_resched();
2080
2081                 leaf = path->nodes[0];
2082                 slot = path->slots[0];
2083
2084                 if (slot >= btrfs_header_nritems(leaf)) {
2085                         ret = btrfs_next_leaf(root, path);
2086                         if (ret < 0) {
2087                                 goto out;
2088                         } else if (ret > 0) {
2089                                 ret = 0;
2090                                 goto out;
2091                         }
2092                         continue;
2093                 }
2094
2095                 path->slots[0]++;
2096
2097                 btrfs_item_key_to_cpu(leaf, &key, slot);
2098
2099                 if (key.objectid > inum)
2100                         goto out;
2101
2102                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2103                         continue;
2104
2105                 extent = btrfs_item_ptr(leaf, slot,
2106                                         struct btrfs_file_extent_item);
2107
2108                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2109                         continue;
2110
2111                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2112                 if (key.offset - extent_offset != offset)
2113                         continue;
2114
2115                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2116                 if (extent_offset >= old->extent_offset + old->offset +
2117                     old->len || extent_offset + num_bytes <=
2118                     old->extent_offset + old->offset)
2119                         continue;
2120
2121                 break;
2122         }
2123
2124         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2125         if (!backref) {
2126                 ret = -ENOENT;
2127                 goto out;
2128         }
2129
2130         backref->root_id = root_id;
2131         backref->inum = inum;
2132         backref->file_pos = offset + extent_offset;
2133         backref->num_bytes = num_bytes;
2134         backref->extent_offset = extent_offset;
2135         backref->generation = btrfs_file_extent_generation(leaf, extent);
2136         backref->old = old;
2137         backref_insert(&new->root, backref);
2138         old->count++;
2139 out:
2140         btrfs_release_path(path);
2141         WARN_ON(ret);
2142         return ret;
2143 }
2144
2145 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2146                                    struct new_sa_defrag_extent *new)
2147 {
2148         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2149         struct old_sa_defrag_extent *old, *tmp;
2150         int ret;
2151
2152         new->path = path;
2153
2154         list_for_each_entry_safe(old, tmp, &new->head, list) {
2155                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2156                                                   path, record_one_backref,
2157                                                   old);
2158                 BUG_ON(ret < 0 && ret != -ENOENT);
2159
2160                 /* no backref to be processed for this extent */
2161                 if (!old->count) {
2162                         list_del(&old->list);
2163                         kfree(old);
2164                 }
2165         }
2166
2167         if (list_empty(&new->head))
2168                 return false;
2169
2170         return true;
2171 }
2172
2173 static int relink_is_mergable(struct extent_buffer *leaf,
2174                               struct btrfs_file_extent_item *fi,
2175                               u64 disk_bytenr)
2176 {
2177         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2178                 return 0;
2179
2180         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2181                 return 0;
2182
2183         if (btrfs_file_extent_compression(leaf, fi) ||
2184             btrfs_file_extent_encryption(leaf, fi) ||
2185             btrfs_file_extent_other_encoding(leaf, fi))
2186                 return 0;
2187
2188         return 1;
2189 }
2190
2191 /*
2192  * Note the backref might has changed, and in this case we just return 0.
2193  */
2194 static noinline int relink_extent_backref(struct btrfs_path *path,
2195                                  struct sa_defrag_extent_backref *prev,
2196                                  struct sa_defrag_extent_backref *backref)
2197 {
2198         struct btrfs_file_extent_item *extent;
2199         struct btrfs_file_extent_item *item;
2200         struct btrfs_ordered_extent *ordered;
2201         struct btrfs_trans_handle *trans;
2202         struct btrfs_fs_info *fs_info;
2203         struct btrfs_root *root;
2204         struct btrfs_key key;
2205         struct extent_buffer *leaf;
2206         struct old_sa_defrag_extent *old = backref->old;
2207         struct new_sa_defrag_extent *new = old->new;
2208         struct inode *src_inode = new->inode;
2209         struct inode *inode;
2210         struct extent_state *cached = NULL;
2211         int ret = 0;
2212         u64 start;
2213         u64 len;
2214         u64 lock_start;
2215         u64 lock_end;
2216         bool merge = false;
2217         int index;
2218
2219         if (prev && prev->root_id == backref->root_id &&
2220             prev->inum == backref->inum &&
2221             prev->file_pos + prev->num_bytes == backref->file_pos)
2222                 merge = true;
2223
2224         /* step 1: get root */
2225         key.objectid = backref->root_id;
2226         key.type = BTRFS_ROOT_ITEM_KEY;
2227         key.offset = (u64)-1;
2228
2229         fs_info = BTRFS_I(src_inode)->root->fs_info;
2230         index = srcu_read_lock(&fs_info->subvol_srcu);
2231
2232         root = btrfs_read_fs_root_no_name(fs_info, &key);
2233         if (IS_ERR(root)) {
2234                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2235                 if (PTR_ERR(root) == -ENOENT)
2236                         return 0;
2237                 return PTR_ERR(root);
2238         }
2239         if (btrfs_root_refs(&root->root_item) == 0) {
2240                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2241                 /* parse ENOENT to 0 */
2242                 return 0;
2243         }
2244
2245         /* step 2: get inode */
2246         key.objectid = backref->inum;
2247         key.type = BTRFS_INODE_ITEM_KEY;
2248         key.offset = 0;
2249
2250         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2251         if (IS_ERR(inode)) {
2252                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2253                 return 0;
2254         }
2255
2256         srcu_read_unlock(&fs_info->subvol_srcu, index);
2257
2258         /* step 3: relink backref */
2259         lock_start = backref->file_pos;
2260         lock_end = backref->file_pos + backref->num_bytes - 1;
2261         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2262                          0, &cached);
2263
2264         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2265         if (ordered) {
2266                 btrfs_put_ordered_extent(ordered);
2267                 goto out_unlock;
2268         }
2269
2270         trans = btrfs_join_transaction(root);
2271         if (IS_ERR(trans)) {
2272                 ret = PTR_ERR(trans);
2273                 goto out_unlock;
2274         }
2275
2276         key.objectid = backref->inum;
2277         key.type = BTRFS_EXTENT_DATA_KEY;
2278         key.offset = backref->file_pos;
2279
2280         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2281         if (ret < 0) {
2282                 goto out_free_path;
2283         } else if (ret > 0) {
2284                 ret = 0;
2285                 goto out_free_path;
2286         }
2287
2288         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2289                                 struct btrfs_file_extent_item);
2290
2291         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2292             backref->generation)
2293                 goto out_free_path;
2294
2295         btrfs_release_path(path);
2296
2297         start = backref->file_pos;
2298         if (backref->extent_offset < old->extent_offset + old->offset)
2299                 start += old->extent_offset + old->offset -
2300                          backref->extent_offset;
2301
2302         len = min(backref->extent_offset + backref->num_bytes,
2303                   old->extent_offset + old->offset + old->len);
2304         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2305
2306         ret = btrfs_drop_extents(trans, root, inode, start,
2307                                  start + len, 1);
2308         if (ret)
2309                 goto out_free_path;
2310 again:
2311         key.objectid = btrfs_ino(inode);
2312         key.type = BTRFS_EXTENT_DATA_KEY;
2313         key.offset = start;
2314
2315         if (merge) {
2316                 struct btrfs_file_extent_item *fi;
2317                 u64 extent_len;
2318                 struct btrfs_key found_key;
2319
2320                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2321                 if (ret < 0)
2322                         goto out_free_path;
2323
2324                 path->slots[0]--;
2325                 leaf = path->nodes[0];
2326                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2327
2328                 fi = btrfs_item_ptr(leaf, path->slots[0],
2329                                     struct btrfs_file_extent_item);
2330                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2331
2332                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2333                     extent_len + found_key.offset == start) {
2334                         btrfs_set_file_extent_num_bytes(leaf, fi,
2335                                                         extent_len + len);
2336                         btrfs_mark_buffer_dirty(leaf);
2337                         inode_add_bytes(inode, len);
2338
2339                         ret = 1;
2340                         goto out_free_path;
2341                 } else {
2342                         merge = false;
2343                         btrfs_release_path(path);
2344                         goto again;
2345                 }
2346         }
2347
2348         ret = btrfs_insert_empty_item(trans, root, path, &key,
2349                                         sizeof(*extent));
2350         if (ret) {
2351                 btrfs_abort_transaction(trans, root, ret);
2352                 goto out_free_path;
2353         }
2354
2355         leaf = path->nodes[0];
2356         item = btrfs_item_ptr(leaf, path->slots[0],
2357                                 struct btrfs_file_extent_item);
2358         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2359         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2360         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2361         btrfs_set_file_extent_num_bytes(leaf, item, len);
2362         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2363         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2364         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2365         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2366         btrfs_set_file_extent_encryption(leaf, item, 0);
2367         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2368
2369         btrfs_mark_buffer_dirty(leaf);
2370         inode_add_bytes(inode, len);
2371
2372         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2373                         new->disk_len, 0,
2374                         backref->root_id, backref->inum,
2375                         new->file_pos, 0);      /* start - extent_offset */
2376         if (ret) {
2377                 btrfs_abort_transaction(trans, root, ret);
2378                 goto out_free_path;
2379         }
2380
2381         ret = 1;
2382 out_free_path:
2383         btrfs_release_path(path);
2384         btrfs_end_transaction(trans, root);
2385 out_unlock:
2386         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2387                              &cached, GFP_NOFS);
2388         iput(inode);
2389         return ret;
2390 }
2391
2392 static void relink_file_extents(struct new_sa_defrag_extent *new)
2393 {
2394         struct btrfs_path *path;
2395         struct old_sa_defrag_extent *old, *tmp;
2396         struct sa_defrag_extent_backref *backref;
2397         struct sa_defrag_extent_backref *prev = NULL;
2398         struct inode *inode;
2399         struct btrfs_root *root;
2400         struct rb_node *node;
2401         int ret;
2402
2403         inode = new->inode;
2404         root = BTRFS_I(inode)->root;
2405
2406         path = btrfs_alloc_path();
2407         if (!path)
2408                 return;
2409
2410         if (!record_extent_backrefs(path, new)) {
2411                 btrfs_free_path(path);
2412                 goto out;
2413         }
2414         btrfs_release_path(path);
2415
2416         while (1) {
2417                 node = rb_first(&new->root);
2418                 if (!node)
2419                         break;
2420                 rb_erase(node, &new->root);
2421
2422                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2423
2424                 ret = relink_extent_backref(path, prev, backref);
2425                 WARN_ON(ret < 0);
2426
2427                 kfree(prev);
2428
2429                 if (ret == 1)
2430                         prev = backref;
2431                 else
2432                         prev = NULL;
2433                 cond_resched();
2434         }
2435         kfree(prev);
2436
2437         btrfs_free_path(path);
2438
2439         list_for_each_entry_safe(old, tmp, &new->head, list) {
2440                 list_del(&old->list);
2441                 kfree(old);
2442         }
2443 out:
2444         atomic_dec(&root->fs_info->defrag_running);
2445         wake_up(&root->fs_info->transaction_wait);
2446
2447         kfree(new);
2448 }
2449
2450 static struct new_sa_defrag_extent *
2451 record_old_file_extents(struct inode *inode,
2452                         struct btrfs_ordered_extent *ordered)
2453 {
2454         struct btrfs_root *root = BTRFS_I(inode)->root;
2455         struct btrfs_path *path;
2456         struct btrfs_key key;
2457         struct old_sa_defrag_extent *old, *tmp;
2458         struct new_sa_defrag_extent *new;
2459         int ret;
2460
2461         new = kmalloc(sizeof(*new), GFP_NOFS);
2462         if (!new)
2463                 return NULL;
2464
2465         new->inode = inode;
2466         new->file_pos = ordered->file_offset;
2467         new->len = ordered->len;
2468         new->bytenr = ordered->start;
2469         new->disk_len = ordered->disk_len;
2470         new->compress_type = ordered->compress_type;
2471         new->root = RB_ROOT;
2472         INIT_LIST_HEAD(&new->head);
2473
2474         path = btrfs_alloc_path();
2475         if (!path)
2476                 goto out_kfree;
2477
2478         key.objectid = btrfs_ino(inode);
2479         key.type = BTRFS_EXTENT_DATA_KEY;
2480         key.offset = new->file_pos;
2481
2482         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2483         if (ret < 0)
2484                 goto out_free_path;
2485         if (ret > 0 && path->slots[0] > 0)
2486                 path->slots[0]--;
2487
2488         /* find out all the old extents for the file range */
2489         while (1) {
2490                 struct btrfs_file_extent_item *extent;
2491                 struct extent_buffer *l;
2492                 int slot;
2493                 u64 num_bytes;
2494                 u64 offset;
2495                 u64 end;
2496                 u64 disk_bytenr;
2497                 u64 extent_offset;
2498
2499                 l = path->nodes[0];
2500                 slot = path->slots[0];
2501
2502                 if (slot >= btrfs_header_nritems(l)) {
2503                         ret = btrfs_next_leaf(root, path);
2504                         if (ret < 0)
2505                                 goto out_free_list;
2506                         else if (ret > 0)
2507                                 break;
2508                         continue;
2509                 }
2510
2511                 btrfs_item_key_to_cpu(l, &key, slot);
2512
2513                 if (key.objectid != btrfs_ino(inode))
2514                         break;
2515                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2516                         break;
2517                 if (key.offset >= new->file_pos + new->len)
2518                         break;
2519
2520                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2521
2522                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2523                 if (key.offset + num_bytes < new->file_pos)
2524                         goto next;
2525
2526                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2527                 if (!disk_bytenr)
2528                         goto next;
2529
2530                 extent_offset = btrfs_file_extent_offset(l, extent);
2531
2532                 old = kmalloc(sizeof(*old), GFP_NOFS);
2533                 if (!old)
2534                         goto out_free_list;
2535
2536                 offset = max(new->file_pos, key.offset);
2537                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2538
2539                 old->bytenr = disk_bytenr;
2540                 old->extent_offset = extent_offset;
2541                 old->offset = offset - key.offset;
2542                 old->len = end - offset;
2543                 old->new = new;
2544                 old->count = 0;
2545                 list_add_tail(&old->list, &new->head);
2546 next:
2547                 path->slots[0]++;
2548                 cond_resched();
2549         }
2550
2551         btrfs_free_path(path);
2552         atomic_inc(&root->fs_info->defrag_running);
2553
2554         return new;
2555
2556 out_free_list:
2557         list_for_each_entry_safe(old, tmp, &new->head, list) {
2558                 list_del(&old->list);
2559                 kfree(old);
2560         }
2561 out_free_path:
2562         btrfs_free_path(path);
2563 out_kfree:
2564         kfree(new);
2565         return NULL;
2566 }
2567
2568 /*
2569  * helper function for btrfs_finish_ordered_io, this
2570  * just reads in some of the csum leaves to prime them into ram
2571  * before we start the transaction.  It limits the amount of btree
2572  * reads required while inside the transaction.
2573  */
2574 /* as ordered data IO finishes, this gets called so we can finish
2575  * an ordered extent if the range of bytes in the file it covers are
2576  * fully written.
2577  */
2578 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2579 {
2580         struct inode *inode = ordered_extent->inode;
2581         struct btrfs_root *root = BTRFS_I(inode)->root;
2582         struct btrfs_trans_handle *trans = NULL;
2583         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2584         struct extent_state *cached_state = NULL;
2585         struct new_sa_defrag_extent *new = NULL;
2586         int compress_type = 0;
2587         int ret;
2588         bool nolock;
2589
2590         nolock = btrfs_is_free_space_inode(inode);
2591
2592         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2593                 ret = -EIO;
2594                 goto out;
2595         }
2596
2597         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2598                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2599                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2600                 if (nolock)
2601                         trans = btrfs_join_transaction_nolock(root);
2602                 else
2603                         trans = btrfs_join_transaction(root);
2604                 if (IS_ERR(trans)) {
2605                         ret = PTR_ERR(trans);
2606                         trans = NULL;
2607                         goto out;
2608                 }
2609                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2610                 ret = btrfs_update_inode_fallback(trans, root, inode);
2611                 if (ret) /* -ENOMEM or corruption */
2612                         btrfs_abort_transaction(trans, root, ret);
2613                 goto out;
2614         }
2615
2616         lock_extent_bits(io_tree, ordered_extent->file_offset,
2617                          ordered_extent->file_offset + ordered_extent->len - 1,
2618                          0, &cached_state);
2619
2620         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2621                         ordered_extent->file_offset + ordered_extent->len - 1,
2622                         EXTENT_DEFRAG, 1, cached_state);
2623         if (ret) {
2624                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2625                 if (last_snapshot >= BTRFS_I(inode)->generation)
2626                         /* the inode is shared */
2627                         new = record_old_file_extents(inode, ordered_extent);
2628
2629                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2630                         ordered_extent->file_offset + ordered_extent->len - 1,
2631                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2632         }
2633
2634         if (nolock)
2635                 trans = btrfs_join_transaction_nolock(root);
2636         else
2637                 trans = btrfs_join_transaction(root);
2638         if (IS_ERR(trans)) {
2639                 ret = PTR_ERR(trans);
2640                 trans = NULL;
2641                 goto out_unlock;
2642         }
2643         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2644
2645         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2646                 compress_type = ordered_extent->compress_type;
2647         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2648                 BUG_ON(compress_type);
2649                 ret = btrfs_mark_extent_written(trans, inode,
2650                                                 ordered_extent->file_offset,
2651                                                 ordered_extent->file_offset +
2652                                                 ordered_extent->len);
2653         } else {
2654                 BUG_ON(root == root->fs_info->tree_root);
2655                 ret = insert_reserved_file_extent(trans, inode,
2656                                                 ordered_extent->file_offset,
2657                                                 ordered_extent->start,
2658                                                 ordered_extent->disk_len,
2659                                                 ordered_extent->len,
2660                                                 ordered_extent->len,
2661                                                 compress_type, 0, 0,
2662                                                 BTRFS_FILE_EXTENT_REG);
2663         }
2664         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2665                            ordered_extent->file_offset, ordered_extent->len,
2666                            trans->transid);
2667         if (ret < 0) {
2668                 btrfs_abort_transaction(trans, root, ret);
2669                 goto out_unlock;
2670         }
2671
2672         add_pending_csums(trans, inode, ordered_extent->file_offset,
2673                           &ordered_extent->list);
2674
2675         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2676         ret = btrfs_update_inode_fallback(trans, root, inode);
2677         if (ret) { /* -ENOMEM or corruption */
2678                 btrfs_abort_transaction(trans, root, ret);
2679                 goto out_unlock;
2680         }
2681         ret = 0;
2682 out_unlock:
2683         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2684                              ordered_extent->file_offset +
2685                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2686 out:
2687         if (root != root->fs_info->tree_root)
2688                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2689         if (trans)
2690                 btrfs_end_transaction(trans, root);
2691
2692         if (ret) {
2693                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2694                                       ordered_extent->file_offset +
2695                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2696
2697                 /*
2698                  * If the ordered extent had an IOERR or something else went
2699                  * wrong we need to return the space for this ordered extent
2700                  * back to the allocator.
2701                  */
2702                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2703                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2704                         btrfs_free_reserved_extent(root, ordered_extent->start,
2705                                                    ordered_extent->disk_len);
2706         }
2707
2708
2709         /*
2710          * This needs to be done to make sure anybody waiting knows we are done
2711          * updating everything for this ordered extent.
2712          */
2713         btrfs_remove_ordered_extent(inode, ordered_extent);
2714
2715         /* for snapshot-aware defrag */
2716         if (new)
2717                 relink_file_extents(new);
2718
2719         /* once for us */
2720         btrfs_put_ordered_extent(ordered_extent);
2721         /* once for the tree */
2722         btrfs_put_ordered_extent(ordered_extent);
2723
2724         return ret;
2725 }
2726
2727 static void finish_ordered_fn(struct btrfs_work *work)
2728 {
2729         struct btrfs_ordered_extent *ordered_extent;
2730         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2731         btrfs_finish_ordered_io(ordered_extent);
2732 }
2733
2734 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2735                                 struct extent_state *state, int uptodate)
2736 {
2737         struct inode *inode = page->mapping->host;
2738         struct btrfs_root *root = BTRFS_I(inode)->root;
2739         struct btrfs_ordered_extent *ordered_extent = NULL;
2740         struct btrfs_workers *workers;
2741
2742         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2743
2744         ClearPagePrivate2(page);
2745         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2746                                             end - start + 1, uptodate))
2747                 return 0;
2748
2749         ordered_extent->work.func = finish_ordered_fn;
2750         ordered_extent->work.flags = 0;
2751
2752         if (btrfs_is_free_space_inode(inode))
2753                 workers = &root->fs_info->endio_freespace_worker;
2754         else
2755                 workers = &root->fs_info->endio_write_workers;
2756         btrfs_queue_worker(workers, &ordered_extent->work);
2757
2758         return 0;
2759 }
2760
2761 /*
2762  * when reads are done, we need to check csums to verify the data is correct
2763  * if there's a match, we allow the bio to finish.  If not, the code in
2764  * extent_io.c will try to find good copies for us.
2765  */
2766 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2767                                struct extent_state *state, int mirror)
2768 {
2769         size_t offset = start - page_offset(page);
2770         struct inode *inode = page->mapping->host;
2771         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2772         char *kaddr;
2773         u64 private = ~(u32)0;
2774         int ret;
2775         struct btrfs_root *root = BTRFS_I(inode)->root;
2776         u32 csum = ~(u32)0;
2777
2778         if (PageChecked(page)) {
2779                 ClearPageChecked(page);
2780                 goto good;
2781         }
2782
2783         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2784                 goto good;
2785
2786         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2787             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2788                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2789                                   GFP_NOFS);
2790                 return 0;
2791         }
2792
2793         if (state && state->start == start) {
2794                 private = state->private;
2795                 ret = 0;
2796         } else {
2797                 ret = get_state_private(io_tree, start, &private);
2798         }
2799         kaddr = kmap_atomic(page);
2800         if (ret)
2801                 goto zeroit;
2802
2803         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2804         btrfs_csum_final(csum, (char *)&csum);
2805         if (csum != private)
2806                 goto zeroit;
2807
2808         kunmap_atomic(kaddr);
2809 good:
2810         return 0;
2811
2812 zeroit:
2813         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2814                        "private %llu\n",
2815                        (unsigned long long)btrfs_ino(page->mapping->host),
2816                        (unsigned long long)start, csum,
2817                        (unsigned long long)private);
2818         memset(kaddr + offset, 1, end - start + 1);
2819         flush_dcache_page(page);
2820         kunmap_atomic(kaddr);
2821         if (private == 0)
2822                 return 0;
2823         return -EIO;
2824 }
2825
2826 struct delayed_iput {
2827         struct list_head list;
2828         struct inode *inode;
2829 };
2830
2831 /* JDM: If this is fs-wide, why can't we add a pointer to
2832  * btrfs_inode instead and avoid the allocation? */
2833 void btrfs_add_delayed_iput(struct inode *inode)
2834 {
2835         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2836         struct delayed_iput *delayed;
2837
2838         if (atomic_add_unless(&inode->i_count, -1, 1))
2839                 return;
2840
2841         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2842         delayed->inode = inode;
2843
2844         spin_lock(&fs_info->delayed_iput_lock);
2845         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2846         spin_unlock(&fs_info->delayed_iput_lock);
2847 }
2848
2849 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2850 {
2851         LIST_HEAD(list);
2852         struct btrfs_fs_info *fs_info = root->fs_info;
2853         struct delayed_iput *delayed;
2854         int empty;
2855
2856         spin_lock(&fs_info->delayed_iput_lock);
2857         empty = list_empty(&fs_info->delayed_iputs);
2858         spin_unlock(&fs_info->delayed_iput_lock);
2859         if (empty)
2860                 return;
2861
2862         spin_lock(&fs_info->delayed_iput_lock);
2863         list_splice_init(&fs_info->delayed_iputs, &list);
2864         spin_unlock(&fs_info->delayed_iput_lock);
2865
2866         while (!list_empty(&list)) {
2867                 delayed = list_entry(list.next, struct delayed_iput, list);
2868                 list_del(&delayed->list);
2869                 iput(delayed->inode);
2870                 kfree(delayed);
2871         }
2872 }
2873
2874 /*
2875  * This is called in transaction commit time. If there are no orphan
2876  * files in the subvolume, it removes orphan item and frees block_rsv
2877  * structure.
2878  */
2879 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2880                               struct btrfs_root *root)
2881 {
2882         struct btrfs_block_rsv *block_rsv;
2883         int ret;
2884
2885         if (atomic_read(&root->orphan_inodes) ||
2886             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2887                 return;
2888
2889         spin_lock(&root->orphan_lock);
2890         if (atomic_read(&root->orphan_inodes)) {
2891                 spin_unlock(&root->orphan_lock);
2892                 return;
2893         }
2894
2895         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2896                 spin_unlock(&root->orphan_lock);
2897                 return;
2898         }
2899
2900         block_rsv = root->orphan_block_rsv;
2901         root->orphan_block_rsv = NULL;
2902         spin_unlock(&root->orphan_lock);
2903
2904         if (root->orphan_item_inserted &&
2905             btrfs_root_refs(&root->root_item) > 0) {
2906                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2907                                             root->root_key.objectid);
2908                 BUG_ON(ret);
2909                 root->orphan_item_inserted = 0;
2910         }
2911
2912         if (block_rsv) {
2913                 WARN_ON(block_rsv->size > 0);
2914                 btrfs_free_block_rsv(root, block_rsv);
2915         }
2916 }
2917
2918 /*
2919  * This creates an orphan entry for the given inode in case something goes
2920  * wrong in the middle of an unlink/truncate.
2921  *
2922  * NOTE: caller of this function should reserve 5 units of metadata for
2923  *       this function.
2924  */
2925 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2926 {
2927         struct btrfs_root *root = BTRFS_I(inode)->root;
2928         struct btrfs_block_rsv *block_rsv = NULL;
2929         int reserve = 0;
2930         int insert = 0;
2931         int ret;
2932
2933         if (!root->orphan_block_rsv) {
2934                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2935                 if (!block_rsv)
2936                         return -ENOMEM;
2937         }
2938
2939         spin_lock(&root->orphan_lock);
2940         if (!root->orphan_block_rsv) {
2941                 root->orphan_block_rsv = block_rsv;
2942         } else if (block_rsv) {
2943                 btrfs_free_block_rsv(root, block_rsv);
2944                 block_rsv = NULL;
2945         }
2946
2947         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2948                               &BTRFS_I(inode)->runtime_flags)) {
2949 #if 0
2950                 /*
2951                  * For proper ENOSPC handling, we should do orphan
2952                  * cleanup when mounting. But this introduces backward
2953                  * compatibility issue.
2954                  */
2955                 if (!xchg(&root->orphan_item_inserted, 1))
2956                         insert = 2;
2957                 else
2958                         insert = 1;
2959 #endif
2960                 insert = 1;
2961                 atomic_inc(&root->orphan_inodes);
2962         }
2963
2964         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2965                               &BTRFS_I(inode)->runtime_flags))
2966                 reserve = 1;
2967         spin_unlock(&root->orphan_lock);
2968
2969         /* grab metadata reservation from transaction handle */
2970         if (reserve) {
2971                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2972                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2973         }
2974
2975         /* insert an orphan item to track this unlinked/truncated file */
2976         if (insert >= 1) {
2977                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2978                 if (ret && ret != -EEXIST) {
2979                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980                                   &BTRFS_I(inode)->runtime_flags);
2981                         btrfs_abort_transaction(trans, root, ret);
2982                         return ret;
2983                 }
2984                 ret = 0;
2985         }
2986
2987         /* insert an orphan item to track subvolume contains orphan files */
2988         if (insert >= 2) {
2989                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2990                                                root->root_key.objectid);
2991                 if (ret && ret != -EEXIST) {
2992                         btrfs_abort_transaction(trans, root, ret);
2993                         return ret;
2994                 }
2995         }
2996         return 0;
2997 }
2998
2999 /*
3000  * We have done the truncate/delete so we can go ahead and remove the orphan
3001  * item for this particular inode.
3002  */
3003 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
3004 {
3005         struct btrfs_root *root = BTRFS_I(inode)->root;
3006         int delete_item = 0;
3007         int release_rsv = 0;
3008         int ret = 0;
3009
3010         spin_lock(&root->orphan_lock);
3011         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3012                                &BTRFS_I(inode)->runtime_flags))
3013                 delete_item = 1;
3014
3015         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3016                                &BTRFS_I(inode)->runtime_flags))
3017                 release_rsv = 1;
3018         spin_unlock(&root->orphan_lock);
3019
3020         if (trans && delete_item) {
3021                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3022                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3023         }
3024
3025         if (release_rsv) {
3026                 btrfs_orphan_release_metadata(inode);
3027                 atomic_dec(&root->orphan_inodes);
3028         }
3029
3030         return 0;
3031 }
3032
3033 /*
3034  * this cleans up any orphans that may be left on the list from the last use
3035  * of this root.
3036  */
3037 int btrfs_orphan_cleanup(struct btrfs_root *root)
3038 {
3039         struct btrfs_path *path;
3040         struct extent_buffer *leaf;
3041         struct btrfs_key key, found_key;
3042         struct btrfs_trans_handle *trans;
3043         struct inode *inode;
3044         u64 last_objectid = 0;
3045         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3046
3047         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3048                 return 0;
3049
3050         path = btrfs_alloc_path();
3051         if (!path) {
3052                 ret = -ENOMEM;
3053                 goto out;
3054         }
3055         path->reada = -1;
3056
3057         key.objectid = BTRFS_ORPHAN_OBJECTID;
3058         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3059         key.offset = (u64)-1;
3060
3061         while (1) {
3062                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3063                 if (ret < 0)
3064                         goto out;
3065
3066                 /*
3067                  * if ret == 0 means we found what we were searching for, which
3068                  * is weird, but possible, so only screw with path if we didn't
3069                  * find the key and see if we have stuff that matches
3070                  */
3071                 if (ret > 0) {
3072                         ret = 0;
3073                         if (path->slots[0] == 0)
3074                                 break;
3075                         path->slots[0]--;
3076                 }
3077
3078                 /* pull out the item */
3079                 leaf = path->nodes[0];
3080                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3081
3082                 /* make sure the item matches what we want */
3083                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3084                         break;
3085                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3086                         break;
3087
3088                 /* release the path since we're done with it */
3089                 btrfs_release_path(path);
3090
3091                 /*
3092                  * this is where we are basically btrfs_lookup, without the
3093                  * crossing root thing.  we store the inode number in the
3094                  * offset of the orphan item.
3095                  */
3096
3097                 if (found_key.offset == last_objectid) {
3098                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
3099                                "stopping orphan cleanup\n");
3100                         ret = -EINVAL;
3101                         goto out;
3102                 }
3103
3104                 last_objectid = found_key.offset;
3105
3106                 found_key.objectid = found_key.offset;
3107                 found_key.type = BTRFS_INODE_ITEM_KEY;
3108                 found_key.offset = 0;
3109                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3110                 ret = PTR_RET(inode);
3111                 if (ret && ret != -ESTALE)
3112                         goto out;
3113
3114                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3115                         struct btrfs_root *dead_root;
3116                         struct btrfs_fs_info *fs_info = root->fs_info;
3117                         int is_dead_root = 0;
3118
3119                         /*
3120                          * this is an orphan in the tree root. Currently these
3121                          * could come from 2 sources:
3122                          *  a) a snapshot deletion in progress
3123                          *  b) a free space cache inode
3124                          * We need to distinguish those two, as the snapshot
3125                          * orphan must not get deleted.
3126                          * find_dead_roots already ran before us, so if this
3127                          * is a snapshot deletion, we should find the root
3128                          * in the dead_roots list
3129                          */
3130                         spin_lock(&fs_info->trans_lock);
3131                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3132                                             root_list) {
3133                                 if (dead_root->root_key.objectid ==
3134                                     found_key.objectid) {
3135                                         is_dead_root = 1;
3136                                         break;
3137                                 }
3138                         }
3139                         spin_unlock(&fs_info->trans_lock);
3140                         if (is_dead_root) {
3141                                 /* prevent this orphan from being found again */
3142                                 key.offset = found_key.objectid - 1;
3143                                 continue;
3144                         }
3145                 }
3146                 /*
3147                  * Inode is already gone but the orphan item is still there,
3148                  * kill the orphan item.
3149                  */
3150                 if (ret == -ESTALE) {
3151                         trans = btrfs_start_transaction(root, 1);
3152                         if (IS_ERR(trans)) {
3153                                 ret = PTR_ERR(trans);
3154                                 goto out;
3155                         }
3156                         printk(KERN_ERR "auto deleting %Lu\n",
3157                                found_key.objectid);
3158                         ret = btrfs_del_orphan_item(trans, root,
3159                                                     found_key.objectid);
3160                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3161                         btrfs_end_transaction(trans, root);
3162                         continue;
3163                 }
3164
3165                 /*
3166                  * add this inode to the orphan list so btrfs_orphan_del does
3167                  * the proper thing when we hit it
3168                  */
3169                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3170                         &BTRFS_I(inode)->runtime_flags);
3171                 atomic_inc(&root->orphan_inodes);
3172
3173                 /* if we have links, this was a truncate, lets do that */
3174                 if (inode->i_nlink) {
3175                         if (!S_ISREG(inode->i_mode)) {
3176                                 WARN_ON(1);
3177                                 iput(inode);
3178                                 continue;
3179                         }
3180                         nr_truncate++;
3181
3182                         /* 1 for the orphan item deletion. */
3183                         trans = btrfs_start_transaction(root, 1);
3184                         if (IS_ERR(trans)) {
3185                                 ret = PTR_ERR(trans);
3186                                 goto out;
3187                         }
3188                         ret = btrfs_orphan_add(trans, inode);
3189                         btrfs_end_transaction(trans, root);
3190                         if (ret)
3191                                 goto out;
3192
3193                         ret = btrfs_truncate(inode);
3194                         if (ret)
3195                                 btrfs_orphan_del(NULL, inode);
3196                 } else {
3197                         nr_unlink++;
3198                 }
3199
3200                 /* this will do delete_inode and everything for us */
3201                 iput(inode);
3202                 if (ret)
3203                         goto out;
3204         }
3205         /* release the path since we're done with it */
3206         btrfs_release_path(path);
3207
3208         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3209
3210         if (root->orphan_block_rsv)
3211                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3212                                         (u64)-1);
3213
3214         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3215                 trans = btrfs_join_transaction(root);
3216                 if (!IS_ERR(trans))
3217                         btrfs_end_transaction(trans, root);
3218         }
3219
3220         if (nr_unlink)
3221                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
3222         if (nr_truncate)
3223                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
3224
3225 out:
3226         if (ret)
3227                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
3228         btrfs_free_path(path);
3229         return ret;
3230 }
3231
3232 /*
3233  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3234  * don't find any xattrs, we know there can't be any acls.
3235  *
3236  * slot is the slot the inode is in, objectid is the objectid of the inode
3237  */
3238 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3239                                           int slot, u64 objectid)
3240 {
3241         u32 nritems = btrfs_header_nritems(leaf);
3242         struct btrfs_key found_key;
3243         int scanned = 0;
3244
3245         slot++;
3246         while (slot < nritems) {
3247                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3248
3249                 /* we found a different objectid, there must not be acls */
3250                 if (found_key.objectid != objectid)
3251                         return 0;
3252
3253                 /* we found an xattr, assume we've got an acl */
3254                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3255                         return 1;
3256
3257                 /*
3258                  * we found a key greater than an xattr key, there can't
3259                  * be any acls later on
3260                  */
3261                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3262                         return 0;
3263
3264                 slot++;
3265                 scanned++;
3266
3267                 /*
3268                  * it goes inode, inode backrefs, xattrs, extents,
3269                  * so if there are a ton of hard links to an inode there can
3270                  * be a lot of backrefs.  Don't waste time searching too hard,
3271                  * this is just an optimization
3272                  */
3273                 if (scanned >= 8)
3274                         break;
3275         }
3276         /* we hit the end of the leaf before we found an xattr or
3277          * something larger than an xattr.  We have to assume the inode
3278          * has acls
3279          */
3280         return 1;
3281 }
3282
3283 /*
3284  * read an inode from the btree into the in-memory inode
3285  */
3286 static void btrfs_read_locked_inode(struct inode *inode)
3287 {
3288         struct btrfs_path *path;
3289         struct extent_buffer *leaf;
3290         struct btrfs_inode_item *inode_item;
3291         struct btrfs_timespec *tspec;
3292         struct btrfs_root *root = BTRFS_I(inode)->root;
3293         struct btrfs_key location;
3294         int maybe_acls;
3295         u32 rdev;
3296         int ret;
3297         bool filled = false;
3298
3299         ret = btrfs_fill_inode(inode, &rdev);
3300         if (!ret)
3301                 filled = true;
3302
3303         path = btrfs_alloc_path();
3304         if (!path)
3305                 goto make_bad;
3306
3307         path->leave_spinning = 1;
3308         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3309
3310         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3311         if (ret)
3312                 goto make_bad;
3313
3314         leaf = path->nodes[0];
3315
3316         if (filled)
3317                 goto cache_acl;
3318
3319         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3320                                     struct btrfs_inode_item);
3321         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3322         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3323         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3324         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3325         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3326
3327         tspec = btrfs_inode_atime(inode_item);
3328         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3329         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3330
3331         tspec = btrfs_inode_mtime(inode_item);
3332         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3333         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3334
3335         tspec = btrfs_inode_ctime(inode_item);
3336         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3337         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3338
3339         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3340         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3341         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3342
3343         /*
3344          * If we were modified in the current generation and evicted from memory
3345          * and then re-read we need to do a full sync since we don't have any
3346          * idea about which extents were modified before we were evicted from
3347          * cache.
3348          */
3349         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3350                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3351                         &BTRFS_I(inode)->runtime_flags);
3352
3353         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3354         inode->i_generation = BTRFS_I(inode)->generation;
3355         inode->i_rdev = 0;
3356         rdev = btrfs_inode_rdev(leaf, inode_item);
3357
3358         BTRFS_I(inode)->index_cnt = (u64)-1;
3359         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3360 cache_acl:
3361         /*
3362          * try to precache a NULL acl entry for files that don't have
3363          * any xattrs or acls
3364          */
3365         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3366                                            btrfs_ino(inode));
3367         if (!maybe_acls)
3368                 cache_no_acl(inode);
3369
3370         btrfs_free_path(path);
3371
3372         switch (inode->i_mode & S_IFMT) {
3373         case S_IFREG:
3374                 inode->i_mapping->a_ops = &btrfs_aops;
3375                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3376                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3377                 inode->i_fop = &btrfs_file_operations;
3378                 inode->i_op = &btrfs_file_inode_operations;
3379                 break;
3380         case S_IFDIR:
3381                 inode->i_fop = &btrfs_dir_file_operations;
3382                 if (root == root->fs_info->tree_root)
3383                         inode->i_op = &btrfs_dir_ro_inode_operations;
3384                 else
3385                         inode->i_op = &btrfs_dir_inode_operations;
3386                 break;
3387         case S_IFLNK:
3388                 inode->i_op = &btrfs_symlink_inode_operations;
3389                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3390                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3391                 break;
3392         default:
3393                 inode->i_op = &btrfs_special_inode_operations;
3394                 init_special_inode(inode, inode->i_mode, rdev);
3395                 break;
3396         }
3397
3398         btrfs_update_iflags(inode);
3399         return;
3400
3401 make_bad:
3402         btrfs_free_path(path);
3403         make_bad_inode(inode);
3404 }
3405
3406 /*
3407  * given a leaf and an inode, copy the inode fields into the leaf
3408  */
3409 static void fill_inode_item(struct btrfs_trans_handle *trans,
3410                             struct extent_buffer *leaf,
3411                             struct btrfs_inode_item *item,
3412                             struct inode *inode)
3413 {
3414         struct btrfs_map_token token;
3415
3416         btrfs_init_map_token(&token);
3417
3418         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3419         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3420         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3421                                    &token);
3422         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3423         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3424
3425         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3426                                      inode->i_atime.tv_sec, &token);
3427         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3428                                       inode->i_atime.tv_nsec, &token);
3429
3430         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3431                                      inode->i_mtime.tv_sec, &token);
3432         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3433                                       inode->i_mtime.tv_nsec, &token);
3434
3435         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3436                                      inode->i_ctime.tv_sec, &token);
3437         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3438                                       inode->i_ctime.tv_nsec, &token);
3439
3440         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3441                                      &token);
3442         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3443                                          &token);
3444         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3445         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3446         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3447         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3448         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3449 }
3450
3451 /*
3452  * copy everything in the in-memory inode into the btree.
3453  */
3454 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3455                                 struct btrfs_root *root, struct inode *inode)
3456 {
3457         struct btrfs_inode_item *inode_item;
3458         struct btrfs_path *path;
3459         struct extent_buffer *leaf;
3460         int ret;
3461
3462         path = btrfs_alloc_path();
3463         if (!path)
3464                 return -ENOMEM;
3465
3466         path->leave_spinning = 1;
3467         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3468                                  1);
3469         if (ret) {
3470                 if (ret > 0)
3471                         ret = -ENOENT;
3472                 goto failed;
3473         }
3474
3475         btrfs_unlock_up_safe(path, 1);
3476         leaf = path->nodes[0];
3477         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3478                                     struct btrfs_inode_item);
3479
3480         fill_inode_item(trans, leaf, inode_item, inode);
3481         btrfs_mark_buffer_dirty(leaf);
3482         btrfs_set_inode_last_trans(trans, inode);
3483         ret = 0;
3484 failed:
3485         btrfs_free_path(path);
3486         return ret;
3487 }
3488
3489 /*
3490  * copy everything in the in-memory inode into the btree.
3491  */
3492 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3493                                 struct btrfs_root *root, struct inode *inode)
3494 {
3495         int ret;
3496
3497         /*
3498          * If the inode is a free space inode, we can deadlock during commit
3499          * if we put it into the delayed code.
3500          *
3501          * The data relocation inode should also be directly updated
3502          * without delay
3503          */
3504         if (!btrfs_is_free_space_inode(inode)
3505             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3506                 btrfs_update_root_times(trans, root);
3507
3508                 ret = btrfs_delayed_update_inode(trans, root, inode);
3509                 if (!ret)
3510                         btrfs_set_inode_last_trans(trans, inode);
3511                 return ret;
3512         }
3513
3514         return btrfs_update_inode_item(trans, root, inode);
3515 }
3516
3517 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3518                                          struct btrfs_root *root,
3519                                          struct inode *inode)
3520 {
3521         int ret;
3522
3523         ret = btrfs_update_inode(trans, root, inode);
3524         if (ret == -ENOSPC)
3525                 return btrfs_update_inode_item(trans, root, inode);
3526         return ret;
3527 }
3528
3529 /*
3530  * unlink helper that gets used here in inode.c and in the tree logging
3531  * recovery code.  It remove a link in a directory with a given name, and
3532  * also drops the back refs in the inode to the directory
3533  */
3534 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3535                                 struct btrfs_root *root,
3536                                 struct inode *dir, struct inode *inode,
3537                                 const char *name, int name_len)
3538 {
3539         struct btrfs_path *path;
3540         int ret = 0;
3541         struct extent_buffer *leaf;
3542         struct btrfs_dir_item *di;
3543         struct btrfs_key key;
3544         u64 index;
3545         u64 ino = btrfs_ino(inode);
3546         u64 dir_ino = btrfs_ino(dir);
3547
3548         path = btrfs_alloc_path();
3549         if (!path) {
3550                 ret = -ENOMEM;
3551                 goto out;
3552         }
3553
3554         path->leave_spinning = 1;
3555         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3556                                     name, name_len, -1);
3557         if (IS_ERR(di)) {
3558                 ret = PTR_ERR(di);
3559                 goto err;
3560         }
3561         if (!di) {
3562                 ret = -ENOENT;
3563                 goto err;
3564         }
3565         leaf = path->nodes[0];
3566         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3567         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3568         if (ret)
3569                 goto err;
3570         btrfs_release_path(path);
3571
3572         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3573                                   dir_ino, &index);
3574         if (ret) {
3575                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
3576                        "inode %llu parent %llu\n", name_len, name,
3577                        (unsigned long long)ino, (unsigned long long)dir_ino);
3578                 btrfs_abort_transaction(trans, root, ret);
3579                 goto err;
3580         }
3581
3582         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3583         if (ret) {
3584                 btrfs_abort_transaction(trans, root, ret);
3585                 goto err;
3586         }
3587
3588         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3589                                          inode, dir_ino);
3590         if (ret != 0 && ret != -ENOENT) {
3591                 btrfs_abort_transaction(trans, root, ret);
3592                 goto err;
3593         }
3594
3595         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3596                                            dir, index);
3597         if (ret == -ENOENT)
3598                 ret = 0;
3599 err:
3600         btrfs_free_path(path);
3601         if (ret)
3602                 goto out;
3603
3604         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3605         inode_inc_iversion(inode);
3606         inode_inc_iversion(dir);
3607         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3608         ret = btrfs_update_inode(trans, root, dir);
3609 out:
3610         return ret;
3611 }
3612
3613 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3614                        struct btrfs_root *root,
3615                        struct inode *dir, struct inode *inode,
3616                        const char *name, int name_len)
3617 {
3618         int ret;
3619         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3620         if (!ret) {
3621                 btrfs_drop_nlink(inode);
3622                 ret = btrfs_update_inode(trans, root, inode);
3623         }
3624         return ret;
3625 }
3626
3627
3628 /* helper to check if there is any shared block in the path */
3629 static int check_path_shared(struct btrfs_root *root,
3630                              struct btrfs_path *path)
3631 {
3632         struct extent_buffer *eb;
3633         int level;
3634         u64 refs = 1;
3635
3636         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3637                 int ret;
3638
3639                 if (!path->nodes[level])
3640                         break;
3641                 eb = path->nodes[level];
3642                 if (!btrfs_block_can_be_shared(root, eb))
3643                         continue;
3644                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
3645                                                &refs, NULL);
3646                 if (refs > 1)
3647                         return 1;
3648         }
3649         return 0;
3650 }
3651
3652 /*
3653  * helper to start transaction for unlink and rmdir.
3654  *
3655  * unlink and rmdir are special in btrfs, they do not always free space.
3656  * so in enospc case, we should make sure they will free space before
3657  * allowing them to use the global metadata reservation.
3658  */
3659 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3660                                                        struct dentry *dentry)
3661 {
3662         struct btrfs_trans_handle *trans;
3663         struct btrfs_root *root = BTRFS_I(dir)->root;
3664         struct btrfs_path *path;
3665         struct btrfs_dir_item *di;
3666         struct inode *inode = dentry->d_inode;
3667         u64 index;
3668         int check_link = 1;
3669         int err = -ENOSPC;
3670         int ret;
3671         u64 ino = btrfs_ino(inode);
3672         u64 dir_ino = btrfs_ino(dir);
3673
3674         /*
3675          * 1 for the possible orphan item
3676          * 1 for the dir item
3677          * 1 for the dir index
3678          * 1 for the inode ref
3679          * 1 for the inode ref in the tree log
3680          * 2 for the dir entries in the log
3681          * 1 for the inode
3682          */
3683         trans = btrfs_start_transaction(root, 8);
3684         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3685                 return trans;
3686
3687         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3688                 return ERR_PTR(-ENOSPC);
3689
3690         /* check if there is someone else holds reference */
3691         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3692                 return ERR_PTR(-ENOSPC);
3693
3694         if (atomic_read(&inode->i_count) > 2)
3695                 return ERR_PTR(-ENOSPC);
3696
3697         if (xchg(&root->fs_info->enospc_unlink, 1))
3698                 return ERR_PTR(-ENOSPC);
3699
3700         path = btrfs_alloc_path();
3701         if (!path) {
3702                 root->fs_info->enospc_unlink = 0;
3703                 return ERR_PTR(-ENOMEM);
3704         }
3705
3706         /* 1 for the orphan item */
3707         trans = btrfs_start_transaction(root, 1);
3708         if (IS_ERR(trans)) {
3709                 btrfs_free_path(path);
3710                 root->fs_info->enospc_unlink = 0;
3711                 return trans;
3712         }
3713
3714         path->skip_locking = 1;
3715         path->search_commit_root = 1;
3716
3717         ret = btrfs_lookup_inode(trans, root, path,
3718                                 &BTRFS_I(dir)->location, 0);
3719         if (ret < 0) {
3720                 err = ret;
3721                 goto out;
3722         }
3723         if (ret == 0) {
3724                 if (check_path_shared(root, path))
3725                         goto out;
3726         } else {
3727                 check_link = 0;
3728         }
3729         btrfs_release_path(path);
3730
3731         ret = btrfs_lookup_inode(trans, root, path,
3732                                 &BTRFS_I(inode)->location, 0);
3733         if (ret < 0) {
3734                 err = ret;
3735                 goto out;
3736         }
3737         if (ret == 0) {
3738                 if (check_path_shared(root, path))
3739                         goto out;
3740         } else {
3741                 check_link = 0;
3742         }
3743         btrfs_release_path(path);
3744
3745         if (ret == 0 && S_ISREG(inode->i_mode)) {
3746                 ret = btrfs_lookup_file_extent(trans, root, path,
3747                                                ino, (u64)-1, 0);
3748                 if (ret < 0) {
3749                         err = ret;
3750                         goto out;
3751                 }
3752                 BUG_ON(ret == 0); /* Corruption */
3753                 if (check_path_shared(root, path))
3754                         goto out;
3755                 btrfs_release_path(path);
3756         }
3757
3758         if (!check_link) {
3759                 err = 0;
3760                 goto out;
3761         }
3762
3763         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3764                                 dentry->d_name.name, dentry->d_name.len, 0);
3765         if (IS_ERR(di)) {
3766                 err = PTR_ERR(di);
3767                 goto out;
3768         }
3769         if (di) {
3770                 if (check_path_shared(root, path))
3771                         goto out;
3772         } else {
3773                 err = 0;
3774                 goto out;
3775         }
3776         btrfs_release_path(path);
3777
3778         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3779                                         dentry->d_name.len, ino, dir_ino, 0,
3780                                         &index);
3781         if (ret) {
3782                 err = ret;
3783                 goto out;
3784         }
3785
3786         if (check_path_shared(root, path))
3787                 goto out;
3788
3789         btrfs_release_path(path);
3790
3791         /*
3792          * This is a commit root search, if we can lookup inode item and other
3793          * relative items in the commit root, it means the transaction of
3794          * dir/file creation has been committed, and the dir index item that we
3795          * delay to insert has also been inserted into the commit root. So
3796          * we needn't worry about the delayed insertion of the dir index item
3797          * here.
3798          */
3799         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3800                                 dentry->d_name.name, dentry->d_name.len, 0);
3801         if (IS_ERR(di)) {
3802                 err = PTR_ERR(di);
3803                 goto out;
3804         }
3805         BUG_ON(ret == -ENOENT);
3806         if (check_path_shared(root, path))
3807                 goto out;
3808
3809         err = 0;
3810 out:
3811         btrfs_free_path(path);
3812         /* Migrate the orphan reservation over */
3813         if (!err)
3814                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3815                                 &root->fs_info->global_block_rsv,
3816                                 trans->bytes_reserved);
3817
3818         if (err) {
3819                 btrfs_end_transaction(trans, root);
3820                 root->fs_info->enospc_unlink = 0;
3821                 return ERR_PTR(err);
3822         }
3823
3824         trans->block_rsv = &root->fs_info->global_block_rsv;
3825         return trans;
3826 }
3827
3828 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3829                                struct btrfs_root *root)
3830 {
3831         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3832                 btrfs_block_rsv_release(root, trans->block_rsv,
3833                                         trans->bytes_reserved);
3834                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3835                 BUG_ON(!root->fs_info->enospc_unlink);
3836                 root->fs_info->enospc_unlink = 0;
3837         }
3838         btrfs_end_transaction(trans, root);
3839 }
3840
3841 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3842 {
3843         struct btrfs_root *root = BTRFS_I(dir)->root;
3844         struct btrfs_trans_handle *trans;
3845         struct inode *inode = dentry->d_inode;
3846         int ret;
3847
3848         trans = __unlink_start_trans(dir, dentry);
3849         if (IS_ERR(trans))
3850                 return PTR_ERR(trans);
3851
3852         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3853
3854         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3855                                  dentry->d_name.name, dentry->d_name.len);
3856         if (ret)
3857                 goto out;
3858
3859         if (inode->i_nlink == 0) {
3860                 ret = btrfs_orphan_add(trans, inode);
3861                 if (ret)
3862                         goto out;
3863         }
3864
3865 out:
3866         __unlink_end_trans(trans, root);
3867         btrfs_btree_balance_dirty(root);
3868         return ret;
3869 }
3870
3871 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3872                         struct btrfs_root *root,
3873                         struct inode *dir, u64 objectid,
3874                         const char *name, int name_len)
3875 {
3876         struct btrfs_path *path;
3877         struct extent_buffer *leaf;
3878         struct btrfs_dir_item *di;
3879         struct btrfs_key key;
3880         u64 index;
3881         int ret;
3882         u64 dir_ino = btrfs_ino(dir);
3883
3884         path = btrfs_alloc_path();
3885         if (!path)
3886                 return -ENOMEM;
3887
3888         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3889                                    name, name_len, -1);
3890         if (IS_ERR_OR_NULL(di)) {
3891                 if (!di)
3892                         ret = -ENOENT;
3893                 else
3894                         ret = PTR_ERR(di);
3895                 goto out;
3896         }
3897
3898         leaf = path->nodes[0];
3899         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3900         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3901         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3902         if (ret) {
3903                 btrfs_abort_transaction(trans, root, ret);
3904                 goto out;
3905         }
3906         btrfs_release_path(path);
3907
3908         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3909                                  objectid, root->root_key.objectid,
3910                                  dir_ino, &index, name, name_len);
3911         if (ret < 0) {
3912                 if (ret != -ENOENT) {
3913                         btrfs_abort_transaction(trans, root, ret);
3914                         goto out;
3915                 }
3916                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3917                                                  name, name_len);
3918                 if (IS_ERR_OR_NULL(di)) {
3919                         if (!di)
3920                                 ret = -ENOENT;
3921                         else
3922                                 ret = PTR_ERR(di);
3923                         btrfs_abort_transaction(trans, root, ret);
3924                         goto out;
3925                 }
3926
3927                 leaf = path->nodes[0];
3928                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3929                 btrfs_release_path(path);
3930                 index = key.offset;
3931         }
3932         btrfs_release_path(path);
3933
3934         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3935         if (ret) {
3936                 btrfs_abort_transaction(trans, root, ret);
3937                 goto out;
3938         }
3939
3940         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3941         inode_inc_iversion(dir);
3942         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3943         ret = btrfs_update_inode_fallback(trans, root, dir);
3944         if (ret)
3945                 btrfs_abort_transaction(trans, root, ret);
3946 out:
3947         btrfs_free_path(path);
3948         return ret;
3949 }
3950
3951 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3952 {
3953         struct inode *inode = dentry->d_inode;
3954         int err = 0;
3955         struct btrfs_root *root = BTRFS_I(dir)->root;
3956         struct btrfs_trans_handle *trans;
3957
3958         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3959                 return -ENOTEMPTY;
3960         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3961                 return -EPERM;
3962
3963         trans = __unlink_start_trans(dir, dentry);
3964         if (IS_ERR(trans))
3965                 return PTR_ERR(trans);
3966
3967         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3968                 err = btrfs_unlink_subvol(trans, root, dir,
3969                                           BTRFS_I(inode)->location.objectid,
3970                                           dentry->d_name.name,
3971                                           dentry->d_name.len);
3972                 goto out;
3973         }
3974
3975         err = btrfs_orphan_add(trans, inode);
3976         if (err)
3977                 goto out;
3978
3979         /* now the directory is empty */
3980         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3981                                  dentry->d_name.name, dentry->d_name.len);
3982         if (!err)
3983                 btrfs_i_size_write(inode, 0);
3984 out:
3985         __unlink_end_trans(trans, root);
3986         btrfs_btree_balance_dirty(root);
3987
3988         return err;
3989 }
3990
3991 /*
3992  * this can truncate away extent items, csum items and directory items.
3993  * It starts at a high offset and removes keys until it can't find
3994  * any higher than new_size
3995  *
3996  * csum items that cross the new i_size are truncated to the new size
3997  * as well.
3998  *
3999  * min_type is the minimum key type to truncate down to.  If set to 0, this
4000  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4001  */
4002 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4003                                struct btrfs_root *root,
4004                                struct inode *inode,
4005                                u64 new_size, u32 min_type)
4006 {
4007         struct btrfs_path *path;
4008         struct extent_buffer *leaf;
4009         struct btrfs_file_extent_item *fi;
4010         struct btrfs_key key;
4011         struct btrfs_key found_key;
4012         u64 extent_start = 0;
4013         u64 extent_num_bytes = 0;
4014         u64 extent_offset = 0;
4015         u64 item_end = 0;
4016         u32 found_type = (u8)-1;
4017         int found_extent;
4018         int del_item;
4019         int pending_del_nr = 0;
4020         int pending_del_slot = 0;
4021         int extent_type = -1;
4022         int ret;
4023         int err = 0;
4024         u64 ino = btrfs_ino(inode);
4025
4026         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4027
4028         path = btrfs_alloc_path();
4029         if (!path)
4030                 return -ENOMEM;
4031         path->reada = -1;
4032
4033         /*
4034          * We want to drop from the next block forward in case this new size is
4035          * not block aligned since we will be keeping the last block of the
4036          * extent just the way it is.
4037          */
4038         if (root->ref_cows || root == root->fs_info->tree_root)
4039                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4040                                         root->sectorsize), (u64)-1, 0);
4041
4042         /*
4043          * This function is also used to drop the items in the log tree before
4044          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4045          * it is used to drop the loged items. So we shouldn't kill the delayed
4046          * items.
4047          */
4048         if (min_type == 0 && root == BTRFS_I(inode)->root)
4049                 btrfs_kill_delayed_inode_items(inode);
4050
4051         key.objectid = ino;
4052         key.offset = (u64)-1;
4053         key.type = (u8)-1;
4054
4055 search_again:
4056         path->leave_spinning = 1;
4057         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4058         if (ret < 0) {
4059                 err = ret;
4060                 goto out;
4061         }
4062
4063         if (ret > 0) {
4064                 /* there are no items in the tree for us to truncate, we're
4065                  * done
4066                  */
4067                 if (path->slots[0] == 0)
4068                         goto out;
4069                 path->slots[0]--;
4070         }
4071
4072         while (1) {
4073                 fi = NULL;
4074                 leaf = path->nodes[0];
4075                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4076                 found_type = btrfs_key_type(&found_key);
4077
4078                 if (found_key.objectid != ino)
4079                         break;
4080
4081                 if (found_type < min_type)
4082                         break;
4083
4084                 item_end = found_key.offset;
4085                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4086                         fi = btrfs_item_ptr(leaf, path->slots[0],
4087                                             struct btrfs_file_extent_item);
4088                         extent_type = btrfs_file_extent_type(leaf, fi);
4089                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4090                                 item_end +=
4091                                     btrfs_file_extent_num_bytes(leaf, fi);
4092                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4093                                 item_end += btrfs_file_extent_inline_len(leaf,
4094                                                                          fi);
4095                         }
4096                         item_end--;
4097                 }
4098                 if (found_type > min_type) {
4099                         del_item = 1;
4100                 } else {
4101                         if (item_end < new_size)
4102                                 break;
4103                         if (found_key.offset >= new_size)
4104                                 del_item = 1;
4105                         else
4106                                 del_item = 0;
4107                 }
4108                 found_extent = 0;
4109                 /* FIXME, shrink the extent if the ref count is only 1 */
4110                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4111                         goto delete;
4112
4113                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4114                         u64 num_dec;
4115                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4116                         if (!del_item) {
4117                                 u64 orig_num_bytes =
4118                                         btrfs_file_extent_num_bytes(leaf, fi);
4119                                 extent_num_bytes = ALIGN(new_size -
4120                                                 found_key.offset,
4121                                                 root->sectorsize);
4122                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4123                                                          extent_num_bytes);
4124                                 num_dec = (orig_num_bytes -
4125                                            extent_num_bytes);
4126                                 if (root->ref_cows && extent_start != 0)
4127                                         inode_sub_bytes(inode, num_dec);
4128                                 btrfs_mark_buffer_dirty(leaf);
4129                         } else {
4130                                 extent_num_bytes =
4131                                         btrfs_file_extent_disk_num_bytes(leaf,
4132                                                                          fi);
4133                                 extent_offset = found_key.offset -
4134                                         btrfs_file_extent_offset(leaf, fi);
4135
4136                                 /* FIXME blocksize != 4096 */
4137                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4138                                 if (extent_start != 0) {
4139                                         found_extent = 1;
4140                                         if (root->ref_cows)
4141                                                 inode_sub_bytes(inode, num_dec);
4142                                 }
4143                         }
4144                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4145                         /*
4146                          * we can't truncate inline items that have had
4147                          * special encodings
4148                          */
4149                         if (!del_item &&
4150                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4151                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4152                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4153                                 u32 size = new_size - found_key.offset;
4154
4155                                 if (root->ref_cows) {
4156                                         inode_sub_bytes(inode, item_end + 1 -
4157                                                         new_size);
4158                                 }
4159                                 size =
4160                                     btrfs_file_extent_calc_inline_size(size);
4161                                 btrfs_truncate_item(trans, root, path,
4162                                                     size, 1);
4163                         } else if (root->ref_cows) {
4164                                 inode_sub_bytes(inode, item_end + 1 -
4165                                                 found_key.offset);
4166                         }
4167                 }
4168 delete:
4169                 if (del_item) {
4170                         if (!pending_del_nr) {
4171                                 /* no pending yet, add ourselves */
4172                                 pending_del_slot = path->slots[0];
4173                                 pending_del_nr = 1;
4174                         } else if (pending_del_nr &&
4175                                    path->slots[0] + 1 == pending_del_slot) {
4176                                 /* hop on the pending chunk */
4177                                 pending_del_nr++;
4178                                 pending_del_slot = path->slots[0];
4179                         } else {
4180                                 BUG();
4181                         }
4182                 } else {
4183                         break;
4184                 }
4185                 if (found_extent && (root->ref_cows ||
4186                                      root == root->fs_info->tree_root)) {
4187                         btrfs_set_path_blocking(path);
4188                         ret = btrfs_free_extent(trans, root, extent_start,
4189                                                 extent_num_bytes, 0,
4190                                                 btrfs_header_owner(leaf),
4191                                                 ino, extent_offset, 0);
4192                         BUG_ON(ret);
4193                 }
4194
4195                 if (found_type == BTRFS_INODE_ITEM_KEY)
4196                         break;
4197
4198                 if (path->slots[0] == 0 ||
4199                     path->slots[0] != pending_del_slot) {
4200                         if (pending_del_nr) {
4201                                 ret = btrfs_del_items(trans, root, path,
4202                                                 pending_del_slot,
4203                                                 pending_del_nr);
4204                                 if (ret) {
4205                                         btrfs_abort_transaction(trans,
4206                                                                 root, ret);
4207                                         goto error;
4208                                 }
4209                                 pending_del_nr = 0;
4210                         }
4211                         btrfs_release_path(path);
4212                         goto search_again;
4213                 } else {
4214                         path->slots[0]--;
4215                 }
4216         }
4217 out:
4218         if (pending_del_nr) {
4219                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4220                                       pending_del_nr);
4221                 if (ret)
4222                         btrfs_abort_transaction(trans, root, ret);
4223         }
4224 error:
4225         btrfs_free_path(path);
4226         return err;
4227 }
4228
4229 /*
4230  * btrfs_truncate_page - read, zero a chunk and write a page
4231  * @inode - inode that we're zeroing
4232  * @from - the offset to start zeroing
4233  * @len - the length to zero, 0 to zero the entire range respective to the
4234  *      offset
4235  * @front - zero up to the offset instead of from the offset on
4236  *
4237  * This will find the page for the "from" offset and cow the page and zero the
4238  * part we want to zero.  This is used with truncate and hole punching.
4239  */
4240 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4241                         int front)
4242 {
4243         struct address_space *mapping = inode->i_mapping;
4244         struct btrfs_root *root = BTRFS_I(inode)->root;
4245         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4246         struct btrfs_ordered_extent *ordered;
4247         struct extent_state *cached_state = NULL;
4248         char *kaddr;
4249         u32 blocksize = root->sectorsize;
4250         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4251         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4252         struct page *page;
4253         gfp_t mask = btrfs_alloc_write_mask(mapping);
4254         int ret = 0;
4255         u64 page_start;
4256         u64 page_end;
4257
4258         if ((offset & (blocksize - 1)) == 0 &&
4259             (!len || ((len & (blocksize - 1)) == 0)))
4260                 goto out;
4261         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4262         if (ret)
4263                 goto out;
4264
4265 again:
4266         page = find_or_create_page(mapping, index, mask);
4267         if (!page) {
4268                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4269                 ret = -ENOMEM;
4270                 goto out;
4271         }
4272
4273         page_start = page_offset(page);
4274         page_end = page_start + PAGE_CACHE_SIZE - 1;
4275
4276         if (!PageUptodate(page)) {
4277                 ret = btrfs_readpage(NULL, page);
4278                 lock_page(page);
4279                 if (page->mapping != mapping) {
4280                         unlock_page(page);
4281                         page_cache_release(page);
4282                         goto again;
4283                 }
4284                 if (!PageUptodate(page)) {
4285                         ret = -EIO;
4286                         goto out_unlock;
4287                 }
4288         }
4289         wait_on_page_writeback(page);
4290
4291         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4292         set_page_extent_mapped(page);
4293
4294         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4295         if (ordered) {
4296                 unlock_extent_cached(io_tree, page_start, page_end,
4297                                      &cached_state, GFP_NOFS);
4298                 unlock_page(page);
4299                 page_cache_release(page);
4300                 btrfs_start_ordered_extent(inode, ordered, 1);
4301                 btrfs_put_ordered_extent(ordered);
4302                 goto again;
4303         }
4304
4305         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4306                           EXTENT_DIRTY | EXTENT_DELALLOC |
4307                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4308                           0, 0, &cached_state, GFP_NOFS);
4309
4310         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4311                                         &cached_state);
4312         if (ret) {
4313                 unlock_extent_cached(io_tree, page_start, page_end,
4314                                      &cached_state, GFP_NOFS);
4315                 goto out_unlock;
4316         }
4317
4318         if (offset != PAGE_CACHE_SIZE) {
4319                 if (!len)
4320                         len = PAGE_CACHE_SIZE - offset;
4321                 kaddr = kmap(page);
4322                 if (front)
4323                         memset(kaddr, 0, offset);
4324                 else
4325                         memset(kaddr + offset, 0, len);
4326                 flush_dcache_page(page);
4327                 kunmap(page);
4328         }
4329         ClearPageChecked(page);
4330         set_page_dirty(page);
4331         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4332                              GFP_NOFS);
4333
4334 out_unlock:
4335         if (ret)
4336                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4337         unlock_page(page);
4338         page_cache_release(page);
4339 out:
4340         return ret;
4341 }
4342
4343 /*
4344  * This function puts in dummy file extents for the area we're creating a hole
4345  * for.  So if we are truncating this file to a larger size we need to insert
4346  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4347  * the range between oldsize and size
4348  */
4349 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4350 {
4351         struct btrfs_trans_handle *trans;
4352         struct btrfs_root *root = BTRFS_I(inode)->root;
4353         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4354         struct extent_map *em = NULL;
4355         struct extent_state *cached_state = NULL;
4356         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4357         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4358         u64 block_end = ALIGN(size, root->sectorsize);
4359         u64 last_byte;
4360         u64 cur_offset;
4361         u64 hole_size;
4362         int err = 0;
4363
4364         if (size <= hole_start)
4365                 return 0;
4366
4367         while (1) {
4368                 struct btrfs_ordered_extent *ordered;
4369                 btrfs_wait_ordered_range(inode, hole_start,
4370                                          block_end - hole_start);
4371                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4372                                  &cached_state);
4373                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4374                 if (!ordered)
4375                         break;
4376                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4377                                      &cached_state, GFP_NOFS);
4378                 btrfs_put_ordered_extent(ordered);
4379         }
4380
4381         cur_offset = hole_start;
4382         while (1) {
4383                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4384                                 block_end - cur_offset, 0);
4385                 if (IS_ERR(em)) {
4386                         err = PTR_ERR(em);
4387                         em = NULL;
4388                         break;
4389                 }
4390                 last_byte = min(extent_map_end(em), block_end);
4391                 last_byte = ALIGN(last_byte , root->sectorsize);
4392                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4393                         struct extent_map *hole_em;
4394                         hole_size = last_byte - cur_offset;
4395
4396                         trans = btrfs_start_transaction(root, 3);
4397                         if (IS_ERR(trans)) {
4398                                 err = PTR_ERR(trans);
4399                                 break;
4400                         }
4401
4402                         err = btrfs_drop_extents(trans, root, inode,
4403                                                  cur_offset,
4404                                                  cur_offset + hole_size, 1);
4405                         if (err) {
4406                                 btrfs_abort_transaction(trans, root, err);
4407                                 btrfs_end_transaction(trans, root);
4408                                 break;
4409                         }
4410
4411                         err = btrfs_insert_file_extent(trans, root,
4412                                         btrfs_ino(inode), cur_offset, 0,
4413                                         0, hole_size, 0, hole_size,
4414                                         0, 0, 0);
4415                         if (err) {
4416                                 btrfs_abort_transaction(trans, root, err);
4417                                 btrfs_end_transaction(trans, root);
4418                                 break;
4419                         }
4420
4421                         btrfs_drop_extent_cache(inode, cur_offset,
4422                                                 cur_offset + hole_size - 1, 0);
4423                         hole_em = alloc_extent_map();
4424                         if (!hole_em) {
4425                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4426                                         &BTRFS_I(inode)->runtime_flags);
4427                                 goto next;
4428                         }
4429                         hole_em->start = cur_offset;
4430                         hole_em->len = hole_size;
4431                         hole_em->orig_start = cur_offset;
4432
4433                         hole_em->block_start = EXTENT_MAP_HOLE;
4434                         hole_em->block_len = 0;
4435                         hole_em->orig_block_len = 0;
4436                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4437                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4438                         hole_em->generation = trans->transid;
4439
4440                         while (1) {
4441                                 write_lock(&em_tree->lock);
4442                                 err = add_extent_mapping(em_tree, hole_em);
4443                                 if (!err)
4444                                         list_move(&hole_em->list,
4445                                                   &em_tree->modified_extents);
4446                                 write_unlock(&em_tree->lock);
4447                                 if (err != -EEXIST)
4448                                         break;
4449                                 btrfs_drop_extent_cache(inode, cur_offset,
4450                                                         cur_offset +
4451                                                         hole_size - 1, 0);
4452                         }
4453                         free_extent_map(hole_em);
4454 next:
4455                         btrfs_update_inode(trans, root, inode);
4456                         btrfs_end_transaction(trans, root);
4457                 }
4458                 free_extent_map(em);
4459                 em = NULL;
4460                 cur_offset = last_byte;
4461                 if (cur_offset >= block_end)
4462                         break;
4463         }
4464
4465         free_extent_map(em);
4466         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4467                              GFP_NOFS);
4468         return err;
4469 }
4470
4471 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4472 {
4473         struct btrfs_root *root = BTRFS_I(inode)->root;
4474         struct btrfs_trans_handle *trans;
4475         loff_t oldsize = i_size_read(inode);
4476         loff_t newsize = attr->ia_size;
4477         int mask = attr->ia_valid;
4478         int ret;
4479
4480         if (newsize == oldsize)
4481                 return 0;
4482
4483         /*
4484          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4485          * special case where we need to update the times despite not having
4486          * these flags set.  For all other operations the VFS set these flags
4487          * explicitly if it wants a timestamp update.
4488          */
4489         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4490                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4491
4492         if (newsize > oldsize) {
4493                 truncate_pagecache(inode, oldsize, newsize);
4494                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4495                 if (ret)
4496                         return ret;
4497
4498                 trans = btrfs_start_transaction(root, 1);
4499                 if (IS_ERR(trans))
4500                         return PTR_ERR(trans);
4501
4502                 i_size_write(inode, newsize);
4503                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4504                 ret = btrfs_update_inode(trans, root, inode);
4505                 btrfs_end_transaction(trans, root);
4506         } else {
4507
4508                 /*
4509                  * We're truncating a file that used to have good data down to
4510                  * zero. Make sure it gets into the ordered flush list so that
4511                  * any new writes get down to disk quickly.
4512                  */
4513                 if (newsize == 0)
4514                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4515                                 &BTRFS_I(inode)->runtime_flags);
4516
4517                 /*
4518                  * 1 for the orphan item we're going to add
4519                  * 1 for the orphan item deletion.
4520                  */
4521                 trans = btrfs_start_transaction(root, 2);
4522                 if (IS_ERR(trans))
4523                         return PTR_ERR(trans);
4524
4525                 /*
4526                  * We need to do this in case we fail at _any_ point during the
4527                  * actual truncate.  Once we do the truncate_setsize we could
4528                  * invalidate pages which forces any outstanding ordered io to
4529                  * be instantly completed which will give us extents that need
4530                  * to be truncated.  If we fail to get an orphan inode down we
4531                  * could have left over extents that were never meant to live,
4532                  * so we need to garuntee from this point on that everything
4533                  * will be consistent.
4534                  */
4535                 ret = btrfs_orphan_add(trans, inode);
4536                 btrfs_end_transaction(trans, root);
4537                 if (ret)
4538                         return ret;
4539
4540                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4541                 truncate_setsize(inode, newsize);
4542
4543                 /* Disable nonlocked read DIO to avoid the end less truncate */
4544                 btrfs_inode_block_unlocked_dio(inode);
4545                 inode_dio_wait(inode);
4546                 btrfs_inode_resume_unlocked_dio(inode);
4547
4548                 ret = btrfs_truncate(inode);
4549                 if (ret && inode->i_nlink)
4550                         btrfs_orphan_del(NULL, inode);
4551         }
4552
4553         return ret;
4554 }
4555
4556 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4557 {
4558         struct inode *inode = dentry->d_inode;
4559         struct btrfs_root *root = BTRFS_I(inode)->root;
4560         int err;
4561
4562         if (btrfs_root_readonly(root))
4563                 return -EROFS;
4564
4565         err = inode_change_ok(inode, attr);
4566         if (err)
4567                 return err;
4568
4569         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4570                 err = btrfs_setsize(inode, attr);
4571                 if (err)
4572                         return err;
4573         }
4574
4575         if (attr->ia_valid) {
4576                 setattr_copy(inode, attr);
4577                 inode_inc_iversion(inode);
4578                 err = btrfs_dirty_inode(inode);
4579
4580                 if (!err && attr->ia_valid & ATTR_MODE)
4581                         err = btrfs_acl_chmod(inode);
4582         }
4583
4584         return err;
4585 }
4586
4587 void btrfs_evict_inode(struct inode *inode)
4588 {
4589         struct btrfs_trans_handle *trans;
4590         struct btrfs_root *root = BTRFS_I(inode)->root;
4591         struct btrfs_block_rsv *rsv, *global_rsv;
4592         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4593         int ret;
4594
4595         trace_btrfs_inode_evict(inode);
4596
4597         truncate_inode_pages(&inode->i_data, 0);
4598         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4599                                btrfs_is_free_space_inode(inode)))
4600                 goto no_delete;
4601
4602         if (is_bad_inode(inode)) {
4603                 btrfs_orphan_del(NULL, inode);
4604                 goto no_delete;
4605         }
4606         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4607         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4608
4609         if (root->fs_info->log_root_recovering) {
4610                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4611                                  &BTRFS_I(inode)->runtime_flags));
4612                 goto no_delete;
4613         }
4614
4615         if (inode->i_nlink > 0) {
4616                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4617                 goto no_delete;
4618         }
4619
4620         ret = btrfs_commit_inode_delayed_inode(inode);
4621         if (ret) {
4622                 btrfs_orphan_del(NULL, inode);
4623                 goto no_delete;
4624         }
4625
4626         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4627         if (!rsv) {
4628                 btrfs_orphan_del(NULL, inode);
4629                 goto no_delete;
4630         }
4631         rsv->size = min_size;
4632         rsv->failfast = 1;
4633         global_rsv = &root->fs_info->global_block_rsv;
4634
4635         btrfs_i_size_write(inode, 0);
4636
4637         /*
4638          * This is a bit simpler than btrfs_truncate since we've already
4639          * reserved our space for our orphan item in the unlink, so we just
4640          * need to reserve some slack space in case we add bytes and update
4641          * inode item when doing the truncate.
4642          */
4643         while (1) {
4644                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4645                                              BTRFS_RESERVE_FLUSH_LIMIT);
4646
4647                 /*
4648                  * Try and steal from the global reserve since we will
4649                  * likely not use this space anyway, we want to try as
4650                  * hard as possible to get this to work.
4651                  */
4652                 if (ret)
4653                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4654
4655                 if (ret) {
4656                         printk(KERN_WARNING "Could not get space for a "
4657                                "delete, will truncate on mount %d\n", ret);
4658                         btrfs_orphan_del(NULL, inode);
4659                         btrfs_free_block_rsv(root, rsv);
4660                         goto no_delete;
4661                 }
4662
4663                 trans = btrfs_join_transaction(root);
4664                 if (IS_ERR(trans)) {
4665                         btrfs_orphan_del(NULL, inode);
4666                         btrfs_free_block_rsv(root, rsv);
4667                         goto no_delete;
4668                 }
4669
4670                 trans->block_rsv = rsv;
4671
4672                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4673                 if (ret != -ENOSPC)
4674                         break;
4675
4676                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4677                 btrfs_end_transaction(trans, root);
4678                 trans = NULL;
4679                 btrfs_btree_balance_dirty(root);
4680         }
4681
4682         btrfs_free_block_rsv(root, rsv);
4683
4684         if (ret == 0) {
4685                 trans->block_rsv = root->orphan_block_rsv;
4686                 ret = btrfs_orphan_del(trans, inode);
4687                 BUG_ON(ret);
4688         }
4689
4690         trans->block_rsv = &root->fs_info->trans_block_rsv;
4691         if (!(root == root->fs_info->tree_root ||
4692               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4693                 btrfs_return_ino(root, btrfs_ino(inode));
4694
4695         btrfs_end_transaction(trans, root);
4696         btrfs_btree_balance_dirty(root);
4697 no_delete:
4698         clear_inode(inode);
4699         return;
4700 }
4701
4702 /*
4703  * this returns the key found in the dir entry in the location pointer.
4704  * If no dir entries were found, location->objectid is 0.
4705  */
4706 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4707                                struct btrfs_key *location)
4708 {
4709         const char *name = dentry->d_name.name;
4710         int namelen = dentry->d_name.len;
4711         struct btrfs_dir_item *di;
4712         struct btrfs_path *path;
4713         struct btrfs_root *root = BTRFS_I(dir)->root;
4714         int ret = 0;
4715
4716         path = btrfs_alloc_path();
4717         if (!path)
4718                 return -ENOMEM;
4719
4720         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4721                                     namelen, 0);
4722         if (IS_ERR(di))
4723                 ret = PTR_ERR(di);
4724
4725         if (IS_ERR_OR_NULL(di))
4726                 goto out_err;
4727
4728         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4729 out:
4730         btrfs_free_path(path);
4731         return ret;
4732 out_err:
4733         location->objectid = 0;
4734         goto out;
4735 }
4736
4737 /*
4738  * when we hit a tree root in a directory, the btrfs part of the inode
4739  * needs to be changed to reflect the root directory of the tree root.  This
4740  * is kind of like crossing a mount point.
4741  */
4742 static int fixup_tree_root_location(struct btrfs_root *root,
4743                                     struct inode *dir,
4744                                     struct dentry *dentry,
4745                                     struct btrfs_key *location,
4746                                     struct btrfs_root **sub_root)
4747 {
4748         struct btrfs_path *path;
4749         struct btrfs_root *new_root;
4750         struct btrfs_root_ref *ref;
4751         struct extent_buffer *leaf;
4752         int ret;
4753         int err = 0;
4754
4755         path = btrfs_alloc_path();
4756         if (!path) {
4757                 err = -ENOMEM;
4758                 goto out;
4759         }
4760
4761         err = -ENOENT;
4762         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4763                                   BTRFS_I(dir)->root->root_key.objectid,
4764                                   location->objectid);
4765         if (ret) {
4766                 if (ret < 0)
4767                         err = ret;
4768                 goto out;
4769         }
4770
4771         leaf = path->nodes[0];
4772         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4773         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4774             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4775                 goto out;
4776
4777         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4778                                    (unsigned long)(ref + 1),
4779                                    dentry->d_name.len);
4780         if (ret)
4781                 goto out;
4782
4783         btrfs_release_path(path);
4784
4785         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4786         if (IS_ERR(new_root)) {
4787                 err = PTR_ERR(new_root);
4788                 goto out;
4789         }
4790
4791         if (btrfs_root_refs(&new_root->root_item) == 0) {
4792                 err = -ENOENT;
4793                 goto out;
4794         }
4795
4796         *sub_root = new_root;
4797         location->objectid = btrfs_root_dirid(&new_root->root_item);
4798         location->type = BTRFS_INODE_ITEM_KEY;
4799         location->offset = 0;
4800         err = 0;
4801 out:
4802         btrfs_free_path(path);
4803         return err;
4804 }
4805
4806 static void inode_tree_add(struct inode *inode)
4807 {
4808         struct btrfs_root *root = BTRFS_I(inode)->root;
4809         struct btrfs_inode *entry;
4810         struct rb_node **p;
4811         struct rb_node *parent;
4812         u64 ino = btrfs_ino(inode);
4813 again:
4814         p = &root->inode_tree.rb_node;
4815         parent = NULL;
4816
4817         if (inode_unhashed(inode))
4818                 return;
4819
4820         spin_lock(&root->inode_lock);
4821         while (*p) {
4822                 parent = *p;
4823                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4824
4825                 if (ino < btrfs_ino(&entry->vfs_inode))
4826                         p = &parent->rb_left;
4827                 else if (ino > btrfs_ino(&entry->vfs_inode))
4828                         p = &parent->rb_right;
4829                 else {
4830                         WARN_ON(!(entry->vfs_inode.i_state &
4831                                   (I_WILL_FREE | I_FREEING)));
4832                         rb_erase(parent, &root->inode_tree);
4833                         RB_CLEAR_NODE(parent);
4834                         spin_unlock(&root->inode_lock);
4835                         goto again;
4836                 }
4837         }
4838         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4839         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4840         spin_unlock(&root->inode_lock);
4841 }
4842
4843 static void inode_tree_del(struct inode *inode)
4844 {
4845         struct btrfs_root *root = BTRFS_I(inode)->root;
4846         int empty = 0;
4847
4848         spin_lock(&root->inode_lock);
4849         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4850                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4851                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4852                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4853         }
4854         spin_unlock(&root->inode_lock);
4855
4856         /*
4857          * Free space cache has inodes in the tree root, but the tree root has a
4858          * root_refs of 0, so this could end up dropping the tree root as a
4859          * snapshot, so we need the extra !root->fs_info->tree_root check to
4860          * make sure we don't drop it.
4861          */
4862         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4863             root != root->fs_info->tree_root) {
4864                 synchronize_srcu(&root->fs_info->subvol_srcu);
4865                 spin_lock(&root->inode_lock);
4866                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4867                 spin_unlock(&root->inode_lock);
4868                 if (empty)
4869                         btrfs_add_dead_root(root);
4870         }
4871 }
4872
4873 void btrfs_invalidate_inodes(struct btrfs_root *root)
4874 {
4875         struct rb_node *node;
4876         struct rb_node *prev;
4877         struct btrfs_inode *entry;
4878         struct inode *inode;
4879         u64 objectid = 0;
4880
4881         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4882
4883         spin_lock(&root->inode_lock);
4884 again:
4885         node = root->inode_tree.rb_node;
4886         prev = NULL;
4887         while (node) {
4888                 prev = node;
4889                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4890
4891                 if (objectid < btrfs_ino(&entry->vfs_inode))
4892                         node = node->rb_left;
4893                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4894                         node = node->rb_right;
4895                 else
4896                         break;
4897         }
4898         if (!node) {
4899                 while (prev) {
4900                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4901                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4902                                 node = prev;
4903                                 break;
4904                         }
4905                         prev = rb_next(prev);
4906                 }
4907         }
4908         while (node) {
4909                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4910                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4911                 inode = igrab(&entry->vfs_inode);
4912                 if (inode) {
4913                         spin_unlock(&root->inode_lock);
4914                         if (atomic_read(&inode->i_count) > 1)
4915                                 d_prune_aliases(inode);
4916                         /*
4917                          * btrfs_drop_inode will have it removed from
4918                          * the inode cache when its usage count
4919                          * hits zero.
4920                          */
4921                         iput(inode);
4922                         cond_resched();
4923                         spin_lock(&root->inode_lock);
4924                         goto again;
4925                 }
4926
4927                 if (cond_resched_lock(&root->inode_lock))
4928                         goto again;
4929
4930                 node = rb_next(node);
4931         }
4932         spin_unlock(&root->inode_lock);
4933 }
4934
4935 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4936 {
4937         struct btrfs_iget_args *args = p;
4938         inode->i_ino = args->ino;
4939         BTRFS_I(inode)->root = args->root;
4940         return 0;
4941 }
4942
4943 static int btrfs_find_actor(struct inode *inode, void *opaque)
4944 {
4945         struct btrfs_iget_args *args = opaque;
4946         return args->ino == btrfs_ino(inode) &&
4947                 args->root == BTRFS_I(inode)->root;
4948 }
4949
4950 static struct inode *btrfs_iget_locked(struct super_block *s,
4951                                        u64 objectid,
4952                                        struct btrfs_root *root)
4953 {
4954         struct inode *inode;
4955         struct btrfs_iget_args args;
4956         args.ino = objectid;
4957         args.root = root;
4958
4959         inode = iget5_locked(s, objectid, btrfs_find_actor,
4960                              btrfs_init_locked_inode,
4961                              (void *)&args);
4962         return inode;
4963 }
4964
4965 /* Get an inode object given its location and corresponding root.
4966  * Returns in *is_new if the inode was read from disk
4967  */
4968 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4969                          struct btrfs_root *root, int *new)
4970 {
4971         struct inode *inode;
4972
4973         inode = btrfs_iget_locked(s, location->objectid, root);
4974         if (!inode)
4975                 return ERR_PTR(-ENOMEM);
4976
4977         if (inode->i_state & I_NEW) {
4978                 BTRFS_I(inode)->root = root;
4979                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4980                 btrfs_read_locked_inode(inode);
4981                 if (!is_bad_inode(inode)) {
4982                         inode_tree_add(inode);
4983                         unlock_new_inode(inode);
4984                         if (new)
4985                                 *new = 1;
4986                 } else {
4987                         unlock_new_inode(inode);
4988                         iput(inode);
4989                         inode = ERR_PTR(-ESTALE);
4990                 }
4991         }
4992
4993         return inode;
4994 }
4995
4996 static struct inode *new_simple_dir(struct super_block *s,
4997                                     struct btrfs_key *key,
4998                                     struct btrfs_root *root)
4999 {
5000         struct inode *inode = new_inode(s);
5001
5002         if (!inode)
5003                 return ERR_PTR(-ENOMEM);
5004
5005         BTRFS_I(inode)->root = root;
5006         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5007         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5008
5009         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5010         inode->i_op = &btrfs_dir_ro_inode_operations;
5011         inode->i_fop = &simple_dir_operations;
5012         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5013         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5014
5015         return inode;
5016 }
5017
5018 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5019 {
5020         struct inode *inode;
5021         struct btrfs_root *root = BTRFS_I(dir)->root;
5022         struct btrfs_root *sub_root = root;
5023         struct btrfs_key location;
5024         int index;
5025         int ret = 0;
5026
5027         if (dentry->d_name.len > BTRFS_NAME_LEN)
5028                 return ERR_PTR(-ENAMETOOLONG);
5029
5030         ret = btrfs_inode_by_name(dir, dentry, &location);
5031         if (ret < 0)
5032                 return ERR_PTR(ret);
5033
5034         if (location.objectid == 0)
5035                 return NULL;
5036
5037         if (location.type == BTRFS_INODE_ITEM_KEY) {
5038                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5039                 return inode;
5040         }
5041
5042         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5043
5044         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5045         ret = fixup_tree_root_location(root, dir, dentry,
5046                                        &location, &sub_root);
5047         if (ret < 0) {
5048                 if (ret != -ENOENT)
5049                         inode = ERR_PTR(ret);
5050                 else
5051                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5052         } else {
5053                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5054         }
5055         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5056
5057         if (!IS_ERR(inode) && root != sub_root) {
5058                 down_read(&root->fs_info->cleanup_work_sem);
5059                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5060                         ret = btrfs_orphan_cleanup(sub_root);
5061                 up_read(&root->fs_info->cleanup_work_sem);
5062                 if (ret)
5063                         inode = ERR_PTR(ret);
5064         }
5065
5066         return inode;
5067 }
5068
5069 static int btrfs_dentry_delete(const struct dentry *dentry)
5070 {
5071         struct btrfs_root *root;
5072         struct inode *inode = dentry->d_inode;
5073
5074         if (!inode && !IS_ROOT(dentry))
5075                 inode = dentry->d_parent->d_inode;
5076
5077         if (inode) {
5078                 root = BTRFS_I(inode)->root;
5079                 if (btrfs_root_refs(&root->root_item) == 0)
5080                         return 1;
5081
5082                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5083                         return 1;
5084         }
5085         return 0;
5086 }
5087
5088 static void btrfs_dentry_release(struct dentry *dentry)
5089 {
5090         if (dentry->d_fsdata)
5091                 kfree(dentry->d_fsdata);
5092 }
5093
5094 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5095                                    unsigned int flags)
5096 {
5097         struct dentry *ret;
5098
5099         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5100         return ret;
5101 }
5102
5103 unsigned char btrfs_filetype_table[] = {
5104         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5105 };
5106
5107 static int btrfs_real_readdir(struct file *filp, void *dirent,
5108                               filldir_t filldir)
5109 {
5110         struct inode *inode = file_inode(filp);
5111         struct btrfs_root *root = BTRFS_I(inode)->root;
5112         struct btrfs_item *item;
5113         struct btrfs_dir_item *di;
5114         struct btrfs_key key;
5115         struct btrfs_key found_key;
5116         struct btrfs_path *path;
5117         struct list_head ins_list;
5118         struct list_head del_list;
5119         int ret;
5120         struct extent_buffer *leaf;
5121         int slot;
5122         unsigned char d_type;
5123         int over = 0;
5124         u32 di_cur;
5125         u32 di_total;
5126         u32 di_len;
5127         int key_type = BTRFS_DIR_INDEX_KEY;
5128         char tmp_name[32];
5129         char *name_ptr;
5130         int name_len;
5131         int is_curr = 0;        /* filp->f_pos points to the current index? */
5132
5133         /* FIXME, use a real flag for deciding about the key type */
5134         if (root->fs_info->tree_root == root)
5135                 key_type = BTRFS_DIR_ITEM_KEY;
5136
5137         /* special case for "." */
5138         if (filp->f_pos == 0) {
5139                 over = filldir(dirent, ".", 1,
5140                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5141                 if (over)
5142                         return 0;
5143                 filp->f_pos = 1;
5144         }
5145         /* special case for .., just use the back ref */
5146         if (filp->f_pos == 1) {
5147                 u64 pino = parent_ino(filp->f_path.dentry);
5148                 over = filldir(dirent, "..", 2,
5149                                filp->f_pos, pino, DT_DIR);
5150                 if (over)
5151                         return 0;
5152                 filp->f_pos = 2;
5153         }
5154         path = btrfs_alloc_path();
5155         if (!path)
5156                 return -ENOMEM;
5157
5158         path->reada = 1;
5159
5160         if (key_type == BTRFS_DIR_INDEX_KEY) {
5161                 INIT_LIST_HEAD(&ins_list);
5162                 INIT_LIST_HEAD(&del_list);
5163                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5164         }
5165
5166         btrfs_set_key_type(&key, key_type);
5167         key.offset = filp->f_pos;
5168         key.objectid = btrfs_ino(inode);
5169
5170         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5171         if (ret < 0)
5172                 goto err;
5173
5174         while (1) {
5175                 leaf = path->nodes[0];
5176                 slot = path->slots[0];
5177                 if (slot >= btrfs_header_nritems(leaf)) {
5178                         ret = btrfs_next_leaf(root, path);
5179                         if (ret < 0)
5180                                 goto err;
5181                         else if (ret > 0)
5182                                 break;
5183                         continue;
5184                 }
5185
5186                 item = btrfs_item_nr(leaf, slot);
5187                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5188
5189                 if (found_key.objectid != key.objectid)
5190                         break;
5191                 if (btrfs_key_type(&found_key) != key_type)
5192                         break;
5193                 if (found_key.offset < filp->f_pos)
5194                         goto next;
5195                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5196                     btrfs_should_delete_dir_index(&del_list,
5197                                                   found_key.offset))
5198                         goto next;
5199
5200                 filp->f_pos = found_key.offset;
5201                 is_curr = 1;
5202
5203                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5204                 di_cur = 0;
5205                 di_total = btrfs_item_size(leaf, item);
5206
5207                 while (di_cur < di_total) {
5208                         struct btrfs_key location;
5209
5210                         if (verify_dir_item(root, leaf, di))
5211                                 break;
5212
5213                         name_len = btrfs_dir_name_len(leaf, di);
5214                         if (name_len <= sizeof(tmp_name)) {
5215                                 name_ptr = tmp_name;
5216                         } else {
5217                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5218                                 if (!name_ptr) {
5219                                         ret = -ENOMEM;
5220                                         goto err;
5221                                 }
5222                         }
5223                         read_extent_buffer(leaf, name_ptr,
5224                                            (unsigned long)(di + 1), name_len);
5225
5226                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5227                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5228
5229
5230                         /* is this a reference to our own snapshot? If so
5231                          * skip it.
5232                          *
5233                          * In contrast to old kernels, we insert the snapshot's
5234                          * dir item and dir index after it has been created, so
5235                          * we won't find a reference to our own snapshot. We
5236                          * still keep the following code for backward
5237                          * compatibility.
5238                          */
5239                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5240                             location.objectid == root->root_key.objectid) {
5241                                 over = 0;
5242                                 goto skip;
5243                         }
5244                         over = filldir(dirent, name_ptr, name_len,
5245                                        found_key.offset, location.objectid,
5246                                        d_type);
5247
5248 skip:
5249                         if (name_ptr != tmp_name)
5250                                 kfree(name_ptr);
5251
5252                         if (over)
5253                                 goto nopos;
5254                         di_len = btrfs_dir_name_len(leaf, di) +
5255                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5256                         di_cur += di_len;
5257                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5258                 }
5259 next:
5260                 path->slots[0]++;
5261         }
5262
5263         if (key_type == BTRFS_DIR_INDEX_KEY) {
5264                 if (is_curr)
5265                         filp->f_pos++;
5266                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5267                                                       &ins_list);
5268                 if (ret)
5269                         goto nopos;
5270         }
5271
5272         /* Reached end of directory/root. Bump pos past the last item. */
5273         if (key_type == BTRFS_DIR_INDEX_KEY)
5274                 /*
5275                  * 32-bit glibc will use getdents64, but then strtol -
5276                  * so the last number we can serve is this.
5277                  */
5278                 filp->f_pos = 0x7fffffff;
5279         else
5280                 filp->f_pos++;
5281 nopos:
5282         ret = 0;
5283 err:
5284         if (key_type == BTRFS_DIR_INDEX_KEY)
5285                 btrfs_put_delayed_items(&ins_list, &del_list);
5286         btrfs_free_path(path);
5287         return ret;
5288 }
5289
5290 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5291 {
5292         struct btrfs_root *root = BTRFS_I(inode)->root;
5293         struct btrfs_trans_handle *trans;
5294         int ret = 0;
5295         bool nolock = false;
5296
5297         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5298                 return 0;
5299
5300         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5301                 nolock = true;
5302
5303         if (wbc->sync_mode == WB_SYNC_ALL) {
5304                 if (nolock)
5305                         trans = btrfs_join_transaction_nolock(root);
5306                 else
5307                         trans = btrfs_join_transaction(root);
5308                 if (IS_ERR(trans))
5309                         return PTR_ERR(trans);
5310                 ret = btrfs_commit_transaction(trans, root);
5311         }
5312         return ret;
5313 }
5314
5315 /*
5316  * This is somewhat expensive, updating the tree every time the
5317  * inode changes.  But, it is most likely to find the inode in cache.
5318  * FIXME, needs more benchmarking...there are no reasons other than performance
5319  * to keep or drop this code.
5320  */
5321 int btrfs_dirty_inode(struct inode *inode)
5322 {
5323         struct btrfs_root *root = BTRFS_I(inode)->root;
5324         struct btrfs_trans_handle *trans;
5325         int ret;
5326
5327         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5328                 return 0;
5329
5330         trans = btrfs_join_transaction(root);
5331         if (IS_ERR(trans))
5332                 return PTR_ERR(trans);
5333
5334         ret = btrfs_update_inode(trans, root, inode);
5335         if (ret && ret == -ENOSPC) {
5336                 /* whoops, lets try again with the full transaction */
5337                 btrfs_end_transaction(trans, root);
5338                 trans = btrfs_start_transaction(root, 1);
5339                 if (IS_ERR(trans))
5340                         return PTR_ERR(trans);
5341
5342                 ret = btrfs_update_inode(trans, root, inode);
5343         }
5344         btrfs_end_transaction(trans, root);
5345         if (BTRFS_I(inode)->delayed_node)
5346                 btrfs_balance_delayed_items(root);
5347
5348         return ret;
5349 }
5350
5351 /*
5352  * This is a copy of file_update_time.  We need this so we can return error on
5353  * ENOSPC for updating the inode in the case of file write and mmap writes.
5354  */
5355 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5356                              int flags)
5357 {
5358         struct btrfs_root *root = BTRFS_I(inode)->root;
5359
5360         if (btrfs_root_readonly(root))
5361                 return -EROFS;
5362
5363         if (flags & S_VERSION)
5364                 inode_inc_iversion(inode);
5365         if (flags & S_CTIME)
5366                 inode->i_ctime = *now;
5367         if (flags & S_MTIME)
5368                 inode->i_mtime = *now;
5369         if (flags & S_ATIME)
5370                 inode->i_atime = *now;
5371         return btrfs_dirty_inode(inode);
5372 }
5373
5374 /*
5375  * find the highest existing sequence number in a directory
5376  * and then set the in-memory index_cnt variable to reflect
5377  * free sequence numbers
5378  */
5379 static int btrfs_set_inode_index_count(struct inode *inode)
5380 {
5381         struct btrfs_root *root = BTRFS_I(inode)->root;
5382         struct btrfs_key key, found_key;
5383         struct btrfs_path *path;
5384         struct extent_buffer *leaf;
5385         int ret;
5386
5387         key.objectid = btrfs_ino(inode);
5388         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5389         key.offset = (u64)-1;
5390
5391         path = btrfs_alloc_path();
5392         if (!path)
5393                 return -ENOMEM;
5394
5395         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5396         if (ret < 0)
5397                 goto out;
5398         /* FIXME: we should be able to handle this */
5399         if (ret == 0)
5400                 goto out;
5401         ret = 0;
5402
5403         /*
5404          * MAGIC NUMBER EXPLANATION:
5405          * since we search a directory based on f_pos we have to start at 2
5406          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5407          * else has to start at 2
5408          */
5409         if (path->slots[0] == 0) {
5410                 BTRFS_I(inode)->index_cnt = 2;
5411                 goto out;
5412         }
5413
5414         path->slots[0]--;
5415
5416         leaf = path->nodes[0];
5417         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5418
5419         if (found_key.objectid != btrfs_ino(inode) ||
5420             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5421                 BTRFS_I(inode)->index_cnt = 2;
5422                 goto out;
5423         }
5424
5425         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5426 out:
5427         btrfs_free_path(path);
5428         return ret;
5429 }
5430
5431 /*
5432  * helper to find a free sequence number in a given directory.  This current
5433  * code is very simple, later versions will do smarter things in the btree
5434  */
5435 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5436 {
5437         int ret = 0;
5438
5439         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5440                 ret = btrfs_inode_delayed_dir_index_count(dir);
5441                 if (ret) {
5442                         ret = btrfs_set_inode_index_count(dir);
5443                         if (ret)
5444                                 return ret;
5445                 }
5446         }
5447
5448         *index = BTRFS_I(dir)->index_cnt;
5449         BTRFS_I(dir)->index_cnt++;
5450
5451         return ret;
5452 }
5453
5454 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5455                                      struct btrfs_root *root,
5456                                      struct inode *dir,
5457                                      const char *name, int name_len,
5458                                      u64 ref_objectid, u64 objectid,
5459                                      umode_t mode, u64 *index)
5460 {
5461         struct inode *inode;
5462         struct btrfs_inode_item *inode_item;
5463         struct btrfs_key *location;
5464         struct btrfs_path *path;
5465         struct btrfs_inode_ref *ref;
5466         struct btrfs_key key[2];
5467         u32 sizes[2];
5468         unsigned long ptr;
5469         int ret;
5470         int owner;
5471
5472         path = btrfs_alloc_path();
5473         if (!path)
5474                 return ERR_PTR(-ENOMEM);
5475
5476         inode = new_inode(root->fs_info->sb);
5477         if (!inode) {
5478                 btrfs_free_path(path);
5479                 return ERR_PTR(-ENOMEM);
5480         }
5481
5482         /*
5483          * we have to initialize this early, so we can reclaim the inode
5484          * number if we fail afterwards in this function.
5485          */
5486         inode->i_ino = objectid;
5487
5488         if (dir) {
5489                 trace_btrfs_inode_request(dir);
5490
5491                 ret = btrfs_set_inode_index(dir, index);
5492                 if (ret) {
5493                         btrfs_free_path(path);
5494                         iput(inode);
5495                         return ERR_PTR(ret);
5496                 }
5497         }
5498         /*
5499          * index_cnt is ignored for everything but a dir,
5500          * btrfs_get_inode_index_count has an explanation for the magic
5501          * number
5502          */
5503         BTRFS_I(inode)->index_cnt = 2;
5504         BTRFS_I(inode)->root = root;
5505         BTRFS_I(inode)->generation = trans->transid;
5506         inode->i_generation = BTRFS_I(inode)->generation;
5507
5508         /*
5509          * We could have gotten an inode number from somebody who was fsynced
5510          * and then removed in this same transaction, so let's just set full
5511          * sync since it will be a full sync anyway and this will blow away the
5512          * old info in the log.
5513          */
5514         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5515
5516         if (S_ISDIR(mode))
5517                 owner = 0;
5518         else
5519                 owner = 1;
5520
5521         key[0].objectid = objectid;
5522         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5523         key[0].offset = 0;
5524
5525         /*
5526          * Start new inodes with an inode_ref. This is slightly more
5527          * efficient for small numbers of hard links since they will
5528          * be packed into one item. Extended refs will kick in if we
5529          * add more hard links than can fit in the ref item.
5530          */
5531         key[1].objectid = objectid;
5532         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5533         key[1].offset = ref_objectid;
5534
5535         sizes[0] = sizeof(struct btrfs_inode_item);
5536         sizes[1] = name_len + sizeof(*ref);
5537
5538         path->leave_spinning = 1;
5539         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5540         if (ret != 0)
5541                 goto fail;
5542
5543         inode_init_owner(inode, dir, mode);
5544         inode_set_bytes(inode, 0);
5545         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5546         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5547                                   struct btrfs_inode_item);
5548         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5549                              sizeof(*inode_item));
5550         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5551
5552         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5553                              struct btrfs_inode_ref);
5554         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5555         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5556         ptr = (unsigned long)(ref + 1);
5557         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5558
5559         btrfs_mark_buffer_dirty(path->nodes[0]);
5560         btrfs_free_path(path);
5561
5562         location = &BTRFS_I(inode)->location;
5563         location->objectid = objectid;
5564         location->offset = 0;
5565         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5566
5567         btrfs_inherit_iflags(inode, dir);
5568
5569         if (S_ISREG(mode)) {
5570                 if (btrfs_test_opt(root, NODATASUM))
5571                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5572                 if (btrfs_test_opt(root, NODATACOW))
5573                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5574                                 BTRFS_INODE_NODATASUM;
5575         }
5576
5577         insert_inode_hash(inode);
5578         inode_tree_add(inode);
5579
5580         trace_btrfs_inode_new(inode);
5581         btrfs_set_inode_last_trans(trans, inode);
5582
5583         btrfs_update_root_times(trans, root);
5584
5585         return inode;
5586 fail:
5587         if (dir)
5588                 BTRFS_I(dir)->index_cnt--;
5589         btrfs_free_path(path);
5590         iput(inode);
5591         return ERR_PTR(ret);
5592 }
5593
5594 static inline u8 btrfs_inode_type(struct inode *inode)
5595 {
5596         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5597 }
5598
5599 /*
5600  * utility function to add 'inode' into 'parent_inode' with
5601  * a give name and a given sequence number.
5602  * if 'add_backref' is true, also insert a backref from the
5603  * inode to the parent directory.
5604  */
5605 int btrfs_add_link(struct btrfs_trans_handle *trans,
5606                    struct inode *parent_inode, struct inode *inode,
5607                    const char *name, int name_len, int add_backref, u64 index)
5608 {
5609         int ret = 0;
5610         struct btrfs_key key;
5611         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5612         u64 ino = btrfs_ino(inode);
5613         u64 parent_ino = btrfs_ino(parent_inode);
5614
5615         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5616                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5617         } else {
5618                 key.objectid = ino;
5619                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5620                 key.offset = 0;
5621         }
5622
5623         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5624                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5625                                          key.objectid, root->root_key.objectid,
5626                                          parent_ino, index, name, name_len);
5627         } else if (add_backref) {
5628                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5629                                              parent_ino, index);
5630         }
5631
5632         /* Nothing to clean up yet */
5633         if (ret)
5634                 return ret;
5635
5636         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5637                                     parent_inode, &key,
5638                                     btrfs_inode_type(inode), index);
5639         if (ret == -EEXIST || ret == -EOVERFLOW)
5640                 goto fail_dir_item;
5641         else if (ret) {
5642                 btrfs_abort_transaction(trans, root, ret);
5643                 return ret;
5644         }
5645
5646         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5647                            name_len * 2);
5648         inode_inc_iversion(parent_inode);
5649         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5650         ret = btrfs_update_inode(trans, root, parent_inode);
5651         if (ret)
5652                 btrfs_abort_transaction(trans, root, ret);
5653         return ret;
5654
5655 fail_dir_item:
5656         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5657                 u64 local_index;
5658                 int err;
5659                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5660                                  key.objectid, root->root_key.objectid,
5661                                  parent_ino, &local_index, name, name_len);
5662
5663         } else if (add_backref) {
5664                 u64 local_index;
5665                 int err;
5666
5667                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5668                                           ino, parent_ino, &local_index);
5669         }
5670         return ret;
5671 }
5672
5673 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5674                             struct inode *dir, struct dentry *dentry,
5675                             struct inode *inode, int backref, u64 index)
5676 {
5677         int err = btrfs_add_link(trans, dir, inode,
5678                                  dentry->d_name.name, dentry->d_name.len,
5679                                  backref, index);
5680         if (err > 0)
5681                 err = -EEXIST;
5682         return err;
5683 }
5684
5685 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5686                         umode_t mode, dev_t rdev)
5687 {
5688         struct btrfs_trans_handle *trans;
5689         struct btrfs_root *root = BTRFS_I(dir)->root;
5690         struct inode *inode = NULL;
5691         int err;
5692         int drop_inode = 0;
5693         u64 objectid;
5694         u64 index = 0;
5695
5696         if (!new_valid_dev(rdev))
5697                 return -EINVAL;
5698
5699         /*
5700          * 2 for inode item and ref
5701          * 2 for dir items
5702          * 1 for xattr if selinux is on
5703          */
5704         trans = btrfs_start_transaction(root, 5);
5705         if (IS_ERR(trans))
5706                 return PTR_ERR(trans);
5707
5708         err = btrfs_find_free_ino(root, &objectid);
5709         if (err)
5710                 goto out_unlock;
5711
5712         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5713                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5714                                 mode, &index);
5715         if (IS_ERR(inode)) {
5716                 err = PTR_ERR(inode);
5717                 goto out_unlock;
5718         }
5719
5720         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5721         if (err) {
5722                 drop_inode = 1;
5723                 goto out_unlock;
5724         }
5725
5726         /*
5727         * If the active LSM wants to access the inode during
5728         * d_instantiate it needs these. Smack checks to see
5729         * if the filesystem supports xattrs by looking at the
5730         * ops vector.
5731         */
5732
5733         inode->i_op = &btrfs_special_inode_operations;
5734         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5735         if (err)
5736                 drop_inode = 1;
5737         else {
5738                 init_special_inode(inode, inode->i_mode, rdev);
5739                 btrfs_update_inode(trans, root, inode);
5740                 d_instantiate(dentry, inode);
5741         }
5742 out_unlock:
5743         btrfs_end_transaction(trans, root);
5744         btrfs_btree_balance_dirty(root);
5745         if (drop_inode) {
5746                 inode_dec_link_count(inode);
5747                 iput(inode);
5748         }
5749         return err;
5750 }
5751
5752 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5753                         umode_t mode, bool excl)
5754 {
5755         struct btrfs_trans_handle *trans;
5756         struct btrfs_root *root = BTRFS_I(dir)->root;
5757         struct inode *inode = NULL;
5758         int drop_inode_on_err = 0;
5759         int err;
5760         u64 objectid;
5761         u64 index = 0;
5762
5763         /*
5764          * 2 for inode item and ref
5765          * 2 for dir items
5766          * 1 for xattr if selinux is on
5767          */
5768         trans = btrfs_start_transaction(root, 5);
5769         if (IS_ERR(trans))
5770                 return PTR_ERR(trans);
5771
5772         err = btrfs_find_free_ino(root, &objectid);
5773         if (err)
5774                 goto out_unlock;
5775
5776         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5777                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5778                                 mode, &index);
5779         if (IS_ERR(inode)) {
5780                 err = PTR_ERR(inode);
5781                 goto out_unlock;
5782         }
5783         drop_inode_on_err = 1;
5784
5785         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5786         if (err)
5787                 goto out_unlock;
5788
5789         err = btrfs_update_inode(trans, root, inode);
5790         if (err)
5791                 goto out_unlock;
5792
5793         /*
5794         * If the active LSM wants to access the inode during
5795         * d_instantiate it needs these. Smack checks to see
5796         * if the filesystem supports xattrs by looking at the
5797         * ops vector.
5798         */
5799         inode->i_fop = &btrfs_file_operations;
5800         inode->i_op = &btrfs_file_inode_operations;
5801
5802         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5803         if (err)
5804                 goto out_unlock;
5805
5806         inode->i_mapping->a_ops = &btrfs_aops;
5807         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5808         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5809         d_instantiate(dentry, inode);
5810
5811 out_unlock:
5812         btrfs_end_transaction(trans, root);
5813         if (err && drop_inode_on_err) {
5814                 inode_dec_link_count(inode);
5815                 iput(inode);
5816         }
5817         btrfs_btree_balance_dirty(root);
5818         return err;
5819 }
5820
5821 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5822                       struct dentry *dentry)
5823 {
5824         struct btrfs_trans_handle *trans;
5825         struct btrfs_root *root = BTRFS_I(dir)->root;
5826         struct inode *inode = old_dentry->d_inode;
5827         u64 index;
5828         int err;
5829         int drop_inode = 0;
5830
5831         /* do not allow sys_link's with other subvols of the same device */
5832         if (root->objectid != BTRFS_I(inode)->root->objectid)
5833                 return -EXDEV;
5834
5835         if (inode->i_nlink >= BTRFS_LINK_MAX)
5836                 return -EMLINK;
5837
5838         err = btrfs_set_inode_index(dir, &index);
5839         if (err)
5840                 goto fail;
5841
5842         /*
5843          * 2 items for inode and inode ref
5844          * 2 items for dir items
5845          * 1 item for parent inode
5846          */
5847         trans = btrfs_start_transaction(root, 5);
5848         if (IS_ERR(trans)) {
5849                 err = PTR_ERR(trans);
5850                 goto fail;
5851         }
5852
5853         btrfs_inc_nlink(inode);
5854         inode_inc_iversion(inode);
5855         inode->i_ctime = CURRENT_TIME;
5856         ihold(inode);
5857         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5858
5859         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5860
5861         if (err) {
5862                 drop_inode = 1;
5863         } else {
5864                 struct dentry *parent = dentry->d_parent;
5865                 err = btrfs_update_inode(trans, root, inode);
5866                 if (err)
5867                         goto fail;
5868                 d_instantiate(dentry, inode);
5869                 btrfs_log_new_name(trans, inode, NULL, parent);
5870         }
5871
5872         btrfs_end_transaction(trans, root);
5873 fail:
5874         if (drop_inode) {
5875                 inode_dec_link_count(inode);
5876                 iput(inode);
5877         }
5878         btrfs_btree_balance_dirty(root);
5879         return err;
5880 }
5881
5882 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5883 {
5884         struct inode *inode = NULL;
5885         struct btrfs_trans_handle *trans;
5886         struct btrfs_root *root = BTRFS_I(dir)->root;
5887         int err = 0;
5888         int drop_on_err = 0;
5889         u64 objectid = 0;
5890         u64 index = 0;
5891
5892         /*
5893          * 2 items for inode and ref
5894          * 2 items for dir items
5895          * 1 for xattr if selinux is on
5896          */
5897         trans = btrfs_start_transaction(root, 5);
5898         if (IS_ERR(trans))
5899                 return PTR_ERR(trans);
5900
5901         err = btrfs_find_free_ino(root, &objectid);
5902         if (err)
5903                 goto out_fail;
5904
5905         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5906                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5907                                 S_IFDIR | mode, &index);
5908         if (IS_ERR(inode)) {
5909                 err = PTR_ERR(inode);
5910                 goto out_fail;
5911         }
5912
5913         drop_on_err = 1;
5914
5915         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5916         if (err)
5917                 goto out_fail;
5918
5919         inode->i_op = &btrfs_dir_inode_operations;
5920         inode->i_fop = &btrfs_dir_file_operations;
5921
5922         btrfs_i_size_write(inode, 0);
5923         err = btrfs_update_inode(trans, root, inode);
5924         if (err)
5925                 goto out_fail;
5926
5927         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5928                              dentry->d_name.len, 0, index);
5929         if (err)
5930                 goto out_fail;
5931
5932         d_instantiate(dentry, inode);
5933         drop_on_err = 0;
5934
5935 out_fail:
5936         btrfs_end_transaction(trans, root);
5937         if (drop_on_err)
5938                 iput(inode);
5939         btrfs_btree_balance_dirty(root);
5940         return err;
5941 }
5942
5943 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5944  * and an extent that you want to insert, deal with overlap and insert
5945  * the new extent into the tree.
5946  */
5947 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5948                                 struct extent_map *existing,
5949                                 struct extent_map *em,
5950                                 u64 map_start, u64 map_len)
5951 {
5952         u64 start_diff;
5953
5954         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5955         start_diff = map_start - em->start;
5956         em->start = map_start;
5957         em->len = map_len;
5958         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5959             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5960                 em->block_start += start_diff;
5961                 em->block_len -= start_diff;
5962         }
5963         return add_extent_mapping(em_tree, em);
5964 }
5965
5966 static noinline int uncompress_inline(struct btrfs_path *path,
5967                                       struct inode *inode, struct page *page,
5968                                       size_t pg_offset, u64 extent_offset,
5969                                       struct btrfs_file_extent_item *item)
5970 {
5971         int ret;
5972         struct extent_buffer *leaf = path->nodes[0];
5973         char *tmp;
5974         size_t max_size;
5975         unsigned long inline_size;
5976         unsigned long ptr;
5977         int compress_type;
5978
5979         WARN_ON(pg_offset != 0);
5980         compress_type = btrfs_file_extent_compression(leaf, item);
5981         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5982         inline_size = btrfs_file_extent_inline_item_len(leaf,
5983                                         btrfs_item_nr(leaf, path->slots[0]));
5984         tmp = kmalloc(inline_size, GFP_NOFS);
5985         if (!tmp)
5986                 return -ENOMEM;
5987         ptr = btrfs_file_extent_inline_start(item);
5988
5989         read_extent_buffer(leaf, tmp, ptr, inline_size);
5990
5991         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5992         ret = btrfs_decompress(compress_type, tmp, page,
5993                                extent_offset, inline_size, max_size);
5994         if (ret) {
5995                 char *kaddr = kmap_atomic(page);
5996                 unsigned long copy_size = min_t(u64,
5997                                   PAGE_CACHE_SIZE - pg_offset,
5998                                   max_size - extent_offset);
5999                 memset(kaddr + pg_offset, 0, copy_size);
6000                 kunmap_atomic(kaddr);
6001         }
6002         kfree(tmp);
6003         return 0;
6004 }
6005
6006 /*
6007  * a bit scary, this does extent mapping from logical file offset to the disk.
6008  * the ugly parts come from merging extents from the disk with the in-ram
6009  * representation.  This gets more complex because of the data=ordered code,
6010  * where the in-ram extents might be locked pending data=ordered completion.
6011  *
6012  * This also copies inline extents directly into the page.
6013  */
6014
6015 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6016                                     size_t pg_offset, u64 start, u64 len,
6017                                     int create)
6018 {
6019         int ret;
6020         int err = 0;
6021         u64 bytenr;
6022         u64 extent_start = 0;
6023         u64 extent_end = 0;
6024         u64 objectid = btrfs_ino(inode);
6025         u32 found_type;
6026         struct btrfs_path *path = NULL;
6027         struct btrfs_root *root = BTRFS_I(inode)->root;
6028         struct btrfs_file_extent_item *item;
6029         struct extent_buffer *leaf;
6030         struct btrfs_key found_key;
6031         struct extent_map *em = NULL;
6032         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6033         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6034         struct btrfs_trans_handle *trans = NULL;
6035         int compress_type;
6036
6037 again:
6038         read_lock(&em_tree->lock);
6039         em = lookup_extent_mapping(em_tree, start, len);
6040         if (em)
6041                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6042         read_unlock(&em_tree->lock);
6043
6044         if (em) {
6045                 if (em->start > start || em->start + em->len <= start)
6046                         free_extent_map(em);
6047                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6048                         free_extent_map(em);
6049                 else
6050                         goto out;
6051         }
6052         em = alloc_extent_map();
6053         if (!em) {
6054                 err = -ENOMEM;
6055                 goto out;
6056         }
6057         em->bdev = root->fs_info->fs_devices->latest_bdev;
6058         em->start = EXTENT_MAP_HOLE;
6059         em->orig_start = EXTENT_MAP_HOLE;
6060         em->len = (u64)-1;
6061         em->block_len = (u64)-1;
6062
6063         if (!path) {
6064                 path = btrfs_alloc_path();
6065                 if (!path) {
6066                         err = -ENOMEM;
6067                         goto out;
6068                 }
6069                 /*
6070                  * Chances are we'll be called again, so go ahead and do
6071                  * readahead
6072                  */
6073                 path->reada = 1;
6074         }
6075
6076         ret = btrfs_lookup_file_extent(trans, root, path,
6077                                        objectid, start, trans != NULL);
6078         if (ret < 0) {
6079                 err = ret;
6080                 goto out;
6081         }
6082
6083         if (ret != 0) {
6084                 if (path->slots[0] == 0)
6085                         goto not_found;
6086                 path->slots[0]--;
6087         }
6088
6089         leaf = path->nodes[0];
6090         item = btrfs_item_ptr(leaf, path->slots[0],
6091                               struct btrfs_file_extent_item);
6092         /* are we inside the extent that was found? */
6093         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6094         found_type = btrfs_key_type(&found_key);
6095         if (found_key.objectid != objectid ||
6096             found_type != BTRFS_EXTENT_DATA_KEY) {
6097                 goto not_found;
6098         }
6099
6100         found_type = btrfs_file_extent_type(leaf, item);
6101         extent_start = found_key.offset;
6102         compress_type = btrfs_file_extent_compression(leaf, item);
6103         if (found_type == BTRFS_FILE_EXTENT_REG ||
6104             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6105                 extent_end = extent_start +
6106                        btrfs_file_extent_num_bytes(leaf, item);
6107         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6108                 size_t size;
6109                 size = btrfs_file_extent_inline_len(leaf, item);
6110                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6111         }
6112
6113         if (start >= extent_end) {
6114                 path->slots[0]++;
6115                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6116                         ret = btrfs_next_leaf(root, path);
6117                         if (ret < 0) {
6118                                 err = ret;
6119                                 goto out;
6120                         }
6121                         if (ret > 0)
6122                                 goto not_found;
6123                         leaf = path->nodes[0];
6124                 }
6125                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6126                 if (found_key.objectid != objectid ||
6127                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6128                         goto not_found;
6129                 if (start + len <= found_key.offset)
6130                         goto not_found;
6131                 em->start = start;
6132                 em->orig_start = start;
6133                 em->len = found_key.offset - start;
6134                 goto not_found_em;
6135         }
6136
6137         if (found_type == BTRFS_FILE_EXTENT_REG ||
6138             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6139                 em->start = extent_start;
6140                 em->len = extent_end - extent_start;
6141                 em->orig_start = extent_start -
6142                                  btrfs_file_extent_offset(leaf, item);
6143                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6144                                                                       item);
6145                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6146                 if (bytenr == 0) {
6147                         em->block_start = EXTENT_MAP_HOLE;
6148                         goto insert;
6149                 }
6150                 if (compress_type != BTRFS_COMPRESS_NONE) {
6151                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6152                         em->compress_type = compress_type;
6153                         em->block_start = bytenr;
6154                         em->block_len = em->orig_block_len;
6155                 } else {
6156                         bytenr += btrfs_file_extent_offset(leaf, item);
6157                         em->block_start = bytenr;
6158                         em->block_len = em->len;
6159                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6160                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6161                 }
6162                 goto insert;
6163         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6164                 unsigned long ptr;
6165                 char *map;
6166                 size_t size;
6167                 size_t extent_offset;
6168                 size_t copy_size;
6169
6170                 em->block_start = EXTENT_MAP_INLINE;
6171                 if (!page || create) {
6172                         em->start = extent_start;
6173                         em->len = extent_end - extent_start;
6174                         goto out;
6175                 }
6176
6177                 size = btrfs_file_extent_inline_len(leaf, item);
6178                 extent_offset = page_offset(page) + pg_offset - extent_start;
6179                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6180                                 size - extent_offset);
6181                 em->start = extent_start + extent_offset;
6182                 em->len = ALIGN(copy_size, root->sectorsize);
6183                 em->orig_block_len = em->len;
6184                 em->orig_start = em->start;
6185                 if (compress_type) {
6186                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6187                         em->compress_type = compress_type;
6188                 }
6189                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6190                 if (create == 0 && !PageUptodate(page)) {
6191                         if (btrfs_file_extent_compression(leaf, item) !=
6192                             BTRFS_COMPRESS_NONE) {
6193                                 ret = uncompress_inline(path, inode, page,
6194                                                         pg_offset,
6195                                                         extent_offset, item);
6196                                 BUG_ON(ret); /* -ENOMEM */
6197                         } else {
6198                                 map = kmap(page);
6199                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6200                                                    copy_size);
6201                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6202                                         memset(map + pg_offset + copy_size, 0,
6203                                                PAGE_CACHE_SIZE - pg_offset -
6204                                                copy_size);
6205                                 }
6206                                 kunmap(page);
6207                         }
6208                         flush_dcache_page(page);
6209                 } else if (create && PageUptodate(page)) {
6210                         BUG();
6211                         if (!trans) {
6212                                 kunmap(page);
6213                                 free_extent_map(em);
6214                                 em = NULL;
6215
6216                                 btrfs_release_path(path);
6217                                 trans = btrfs_join_transaction(root);
6218
6219                                 if (IS_ERR(trans))
6220                                         return ERR_CAST(trans);
6221                                 goto again;
6222                         }
6223                         map = kmap(page);
6224                         write_extent_buffer(leaf, map + pg_offset, ptr,
6225                                             copy_size);
6226                         kunmap(page);
6227                         btrfs_mark_buffer_dirty(leaf);
6228                 }
6229                 set_extent_uptodate(io_tree, em->start,
6230                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6231                 goto insert;
6232         } else {
6233                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6234         }
6235 not_found:
6236         em->start = start;
6237         em->orig_start = start;
6238         em->len = len;
6239 not_found_em:
6240         em->block_start = EXTENT_MAP_HOLE;
6241         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6242 insert:
6243         btrfs_release_path(path);
6244         if (em->start > start || extent_map_end(em) <= start) {
6245                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
6246                        "[%llu %llu]\n", (unsigned long long)em->start,
6247                        (unsigned long long)em->len,
6248                        (unsigned long long)start,
6249                        (unsigned long long)len);
6250                 err = -EIO;
6251                 goto out;
6252         }
6253
6254         err = 0;
6255         write_lock(&em_tree->lock);
6256         ret = add_extent_mapping(em_tree, em);
6257         /* it is possible that someone inserted the extent into the tree
6258          * while we had the lock dropped.  It is also possible that
6259          * an overlapping map exists in the tree
6260          */
6261         if (ret == -EEXIST) {
6262                 struct extent_map *existing;
6263
6264                 ret = 0;
6265
6266                 existing = lookup_extent_mapping(em_tree, start, len);
6267                 if (existing && (existing->start > start ||
6268                     existing->start + existing->len <= start)) {
6269                         free_extent_map(existing);
6270                         existing = NULL;
6271                 }
6272                 if (!existing) {
6273                         existing = lookup_extent_mapping(em_tree, em->start,
6274                                                          em->len);
6275                         if (existing) {
6276                                 err = merge_extent_mapping(em_tree, existing,
6277                                                            em, start,
6278                                                            root->sectorsize);
6279                                 free_extent_map(existing);
6280                                 if (err) {
6281                                         free_extent_map(em);
6282                                         em = NULL;
6283                                 }
6284                         } else {
6285                                 err = -EIO;
6286                                 free_extent_map(em);
6287                                 em = NULL;
6288                         }
6289                 } else {
6290                         free_extent_map(em);
6291                         em = existing;
6292                         err = 0;
6293                 }
6294         }
6295         write_unlock(&em_tree->lock);
6296 out:
6297
6298         if (em)
6299                 trace_btrfs_get_extent(root, em);
6300
6301         if (path)
6302                 btrfs_free_path(path);
6303         if (trans) {
6304                 ret = btrfs_end_transaction(trans, root);
6305                 if (!err)
6306                         err = ret;
6307         }
6308         if (err) {
6309                 free_extent_map(em);
6310                 return ERR_PTR(err);
6311         }
6312         BUG_ON(!em); /* Error is always set */
6313         return em;
6314 }
6315
6316 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6317                                            size_t pg_offset, u64 start, u64 len,
6318                                            int create)
6319 {
6320         struct extent_map *em;
6321         struct extent_map *hole_em = NULL;
6322         u64 range_start = start;
6323         u64 end;
6324         u64 found;
6325         u64 found_end;
6326         int err = 0;
6327
6328         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6329         if (IS_ERR(em))
6330                 return em;
6331         if (em) {
6332                 /*
6333                  * if our em maps to
6334                  * -  a hole or
6335                  * -  a pre-alloc extent,
6336                  * there might actually be delalloc bytes behind it.
6337                  */
6338                 if (em->block_start != EXTENT_MAP_HOLE &&
6339                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6340                         return em;
6341                 else
6342                         hole_em = em;
6343         }
6344
6345         /* check to see if we've wrapped (len == -1 or similar) */
6346         end = start + len;
6347         if (end < start)
6348                 end = (u64)-1;
6349         else
6350                 end -= 1;
6351
6352         em = NULL;
6353
6354         /* ok, we didn't find anything, lets look for delalloc */
6355         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6356                                  end, len, EXTENT_DELALLOC, 1);
6357         found_end = range_start + found;
6358         if (found_end < range_start)
6359                 found_end = (u64)-1;
6360
6361         /*
6362          * we didn't find anything useful, return
6363          * the original results from get_extent()
6364          */
6365         if (range_start > end || found_end <= start) {
6366                 em = hole_em;
6367                 hole_em = NULL;
6368                 goto out;
6369         }
6370
6371         /* adjust the range_start to make sure it doesn't
6372          * go backwards from the start they passed in
6373          */
6374         range_start = max(start,range_start);
6375         found = found_end - range_start;
6376
6377         if (found > 0) {
6378                 u64 hole_start = start;
6379                 u64 hole_len = len;
6380
6381                 em = alloc_extent_map();
6382                 if (!em) {
6383                         err = -ENOMEM;
6384                         goto out;
6385                 }
6386                 /*
6387                  * when btrfs_get_extent can't find anything it
6388                  * returns one huge hole
6389                  *
6390                  * make sure what it found really fits our range, and
6391                  * adjust to make sure it is based on the start from
6392                  * the caller
6393                  */
6394                 if (hole_em) {
6395                         u64 calc_end = extent_map_end(hole_em);
6396
6397                         if (calc_end <= start || (hole_em->start > end)) {
6398                                 free_extent_map(hole_em);
6399                                 hole_em = NULL;
6400                         } else {
6401                                 hole_start = max(hole_em->start, start);
6402                                 hole_len = calc_end - hole_start;
6403                         }
6404                 }
6405                 em->bdev = NULL;
6406                 if (hole_em && range_start > hole_start) {
6407                         /* our hole starts before our delalloc, so we
6408                          * have to return just the parts of the hole
6409                          * that go until  the delalloc starts
6410                          */
6411                         em->len = min(hole_len,
6412                                       range_start - hole_start);
6413                         em->start = hole_start;
6414                         em->orig_start = hole_start;
6415                         /*
6416                          * don't adjust block start at all,
6417                          * it is fixed at EXTENT_MAP_HOLE
6418                          */
6419                         em->block_start = hole_em->block_start;
6420                         em->block_len = hole_len;
6421                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6422                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6423                 } else {
6424                         em->start = range_start;
6425                         em->len = found;
6426                         em->orig_start = range_start;
6427                         em->block_start = EXTENT_MAP_DELALLOC;
6428                         em->block_len = found;
6429                 }
6430         } else if (hole_em) {
6431                 return hole_em;
6432         }
6433 out:
6434
6435         free_extent_map(hole_em);
6436         if (err) {
6437                 free_extent_map(em);
6438                 return ERR_PTR(err);
6439         }
6440         return em;
6441 }
6442
6443 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6444                                                   u64 start, u64 len)
6445 {
6446         struct btrfs_root *root = BTRFS_I(inode)->root;
6447         struct btrfs_trans_handle *trans;
6448         struct extent_map *em;
6449         struct btrfs_key ins;
6450         u64 alloc_hint;
6451         int ret;
6452
6453         trans = btrfs_join_transaction(root);
6454         if (IS_ERR(trans))
6455                 return ERR_CAST(trans);
6456
6457         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6458
6459         alloc_hint = get_extent_allocation_hint(inode, start, len);
6460         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6461                                    alloc_hint, &ins, 1);
6462         if (ret) {
6463                 em = ERR_PTR(ret);
6464                 goto out;
6465         }
6466
6467         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6468                               ins.offset, ins.offset, 0);
6469         if (IS_ERR(em))
6470                 goto out;
6471
6472         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6473                                            ins.offset, ins.offset, 0);
6474         if (ret) {
6475                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6476                 em = ERR_PTR(ret);
6477         }
6478 out:
6479         btrfs_end_transaction(trans, root);
6480         return em;
6481 }
6482
6483 /*
6484  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6485  * block must be cow'd
6486  */
6487 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6488                                       struct inode *inode, u64 offset, u64 len)
6489 {
6490         struct btrfs_path *path;
6491         int ret;
6492         struct extent_buffer *leaf;
6493         struct btrfs_root *root = BTRFS_I(inode)->root;
6494         struct btrfs_file_extent_item *fi;
6495         struct btrfs_key key;
6496         u64 disk_bytenr;
6497         u64 backref_offset;
6498         u64 extent_end;
6499         u64 num_bytes;
6500         int slot;
6501         int found_type;
6502
6503         path = btrfs_alloc_path();
6504         if (!path)
6505                 return -ENOMEM;
6506
6507         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6508                                        offset, 0);
6509         if (ret < 0)
6510                 goto out;
6511
6512         slot = path->slots[0];
6513         if (ret == 1) {
6514                 if (slot == 0) {
6515                         /* can't find the item, must cow */
6516                         ret = 0;
6517                         goto out;
6518                 }
6519                 slot--;
6520         }
6521         ret = 0;
6522         leaf = path->nodes[0];
6523         btrfs_item_key_to_cpu(leaf, &key, slot);
6524         if (key.objectid != btrfs_ino(inode) ||
6525             key.type != BTRFS_EXTENT_DATA_KEY) {
6526                 /* not our file or wrong item type, must cow */
6527                 goto out;
6528         }
6529
6530         if (key.offset > offset) {
6531                 /* Wrong offset, must cow */
6532                 goto out;
6533         }
6534
6535         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6536         found_type = btrfs_file_extent_type(leaf, fi);
6537         if (found_type != BTRFS_FILE_EXTENT_REG &&
6538             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6539                 /* not a regular extent, must cow */
6540                 goto out;
6541         }
6542         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6543         backref_offset = btrfs_file_extent_offset(leaf, fi);
6544
6545         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6546         if (extent_end < offset + len) {
6547                 /* extent doesn't include our full range, must cow */
6548                 goto out;
6549         }
6550
6551         if (btrfs_extent_readonly(root, disk_bytenr))
6552                 goto out;
6553
6554         /*
6555          * look for other files referencing this extent, if we
6556          * find any we must cow
6557          */
6558         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6559                                   key.offset - backref_offset, disk_bytenr))
6560                 goto out;
6561
6562         /*
6563          * adjust disk_bytenr and num_bytes to cover just the bytes
6564          * in this extent we are about to write.  If there
6565          * are any csums in that range we have to cow in order
6566          * to keep the csums correct
6567          */
6568         disk_bytenr += backref_offset;
6569         disk_bytenr += offset - key.offset;
6570         num_bytes = min(offset + len, extent_end) - offset;
6571         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6572                                 goto out;
6573         /*
6574          * all of the above have passed, it is safe to overwrite this extent
6575          * without cow
6576          */
6577         ret = 1;
6578 out:
6579         btrfs_free_path(path);
6580         return ret;
6581 }
6582
6583 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6584                               struct extent_state **cached_state, int writing)
6585 {
6586         struct btrfs_ordered_extent *ordered;
6587         int ret = 0;
6588
6589         while (1) {
6590                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6591                                  0, cached_state);
6592                 /*
6593                  * We're concerned with the entire range that we're going to be
6594                  * doing DIO to, so we need to make sure theres no ordered
6595                  * extents in this range.
6596                  */
6597                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6598                                                      lockend - lockstart + 1);
6599
6600                 /*
6601                  * We need to make sure there are no buffered pages in this
6602                  * range either, we could have raced between the invalidate in
6603                  * generic_file_direct_write and locking the extent.  The
6604                  * invalidate needs to happen so that reads after a write do not
6605                  * get stale data.
6606                  */
6607                 if (!ordered && (!writing ||
6608                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6609                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6610                                     *cached_state)))
6611                         break;
6612
6613                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6614                                      cached_state, GFP_NOFS);
6615
6616                 if (ordered) {
6617                         btrfs_start_ordered_extent(inode, ordered, 1);
6618                         btrfs_put_ordered_extent(ordered);
6619                 } else {
6620                         /* Screw you mmap */
6621                         ret = filemap_write_and_wait_range(inode->i_mapping,
6622                                                            lockstart,
6623                                                            lockend);
6624                         if (ret)
6625                                 break;
6626
6627                         /*
6628                          * If we found a page that couldn't be invalidated just
6629                          * fall back to buffered.
6630                          */
6631                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6632                                         lockstart >> PAGE_CACHE_SHIFT,
6633                                         lockend >> PAGE_CACHE_SHIFT);
6634                         if (ret)
6635                                 break;
6636                 }
6637
6638                 cond_resched();
6639         }
6640
6641         return ret;
6642 }
6643
6644 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6645                                            u64 len, u64 orig_start,
6646                                            u64 block_start, u64 block_len,
6647                                            u64 orig_block_len, int type)
6648 {
6649         struct extent_map_tree *em_tree;
6650         struct extent_map *em;
6651         struct btrfs_root *root = BTRFS_I(inode)->root;
6652         int ret;
6653
6654         em_tree = &BTRFS_I(inode)->extent_tree;
6655         em = alloc_extent_map();
6656         if (!em)
6657                 return ERR_PTR(-ENOMEM);
6658
6659         em->start = start;
6660         em->orig_start = orig_start;
6661         em->mod_start = start;
6662         em->mod_len = len;
6663         em->len = len;
6664         em->block_len = block_len;
6665         em->block_start = block_start;
6666         em->bdev = root->fs_info->fs_devices->latest_bdev;
6667         em->orig_block_len = orig_block_len;
6668         em->generation = -1;
6669         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6670         if (type == BTRFS_ORDERED_PREALLOC)
6671                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6672
6673         do {
6674                 btrfs_drop_extent_cache(inode, em->start,
6675                                 em->start + em->len - 1, 0);
6676                 write_lock(&em_tree->lock);
6677                 ret = add_extent_mapping(em_tree, em);
6678                 if (!ret)
6679                         list_move(&em->list,
6680                                   &em_tree->modified_extents);
6681                 write_unlock(&em_tree->lock);
6682         } while (ret == -EEXIST);
6683
6684         if (ret) {
6685                 free_extent_map(em);
6686                 return ERR_PTR(ret);
6687         }
6688
6689         return em;
6690 }
6691
6692
6693 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6694                                    struct buffer_head *bh_result, int create)
6695 {
6696         struct extent_map *em;
6697         struct btrfs_root *root = BTRFS_I(inode)->root;
6698         struct extent_state *cached_state = NULL;
6699         u64 start = iblock << inode->i_blkbits;
6700         u64 lockstart, lockend;
6701         u64 len = bh_result->b_size;
6702         struct btrfs_trans_handle *trans;
6703         int unlock_bits = EXTENT_LOCKED;
6704         int ret = 0;
6705
6706         if (create)
6707                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6708         else
6709                 len = min_t(u64, len, root->sectorsize);
6710
6711         lockstart = start;
6712         lockend = start + len - 1;
6713
6714         /*
6715          * If this errors out it's because we couldn't invalidate pagecache for
6716          * this range and we need to fallback to buffered.
6717          */
6718         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6719                 return -ENOTBLK;
6720
6721         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6722         if (IS_ERR(em)) {
6723                 ret = PTR_ERR(em);
6724                 goto unlock_err;
6725         }
6726
6727         /*
6728          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6729          * io.  INLINE is special, and we could probably kludge it in here, but
6730          * it's still buffered so for safety lets just fall back to the generic
6731          * buffered path.
6732          *
6733          * For COMPRESSED we _have_ to read the entire extent in so we can
6734          * decompress it, so there will be buffering required no matter what we
6735          * do, so go ahead and fallback to buffered.
6736          *
6737          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6738          * to buffered IO.  Don't blame me, this is the price we pay for using
6739          * the generic code.
6740          */
6741         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6742             em->block_start == EXTENT_MAP_INLINE) {
6743                 free_extent_map(em);
6744                 ret = -ENOTBLK;
6745                 goto unlock_err;
6746         }
6747
6748         /* Just a good old fashioned hole, return */
6749         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6750                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6751                 free_extent_map(em);
6752                 goto unlock_err;
6753         }
6754
6755         /*
6756          * We don't allocate a new extent in the following cases
6757          *
6758          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6759          * existing extent.
6760          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6761          * just use the extent.
6762          *
6763          */
6764         if (!create) {
6765                 len = min(len, em->len - (start - em->start));
6766                 lockstart = start + len;
6767                 goto unlock;
6768         }
6769
6770         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6771             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6772              em->block_start != EXTENT_MAP_HOLE)) {
6773                 int type;
6774                 int ret;
6775                 u64 block_start;
6776
6777                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6778                         type = BTRFS_ORDERED_PREALLOC;
6779                 else
6780                         type = BTRFS_ORDERED_NOCOW;
6781                 len = min(len, em->len - (start - em->start));
6782                 block_start = em->block_start + (start - em->start);
6783
6784                 /*
6785                  * we're not going to log anything, but we do need
6786                  * to make sure the current transaction stays open
6787                  * while we look for nocow cross refs
6788                  */
6789                 trans = btrfs_join_transaction(root);
6790                 if (IS_ERR(trans))
6791                         goto must_cow;
6792
6793                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6794                         u64 orig_start = em->orig_start;
6795                         u64 orig_block_len = em->orig_block_len;
6796
6797                         if (type == BTRFS_ORDERED_PREALLOC) {
6798                                 free_extent_map(em);
6799                                 em = create_pinned_em(inode, start, len,
6800                                                        orig_start,
6801                                                        block_start, len,
6802                                                        orig_block_len, type);
6803                                 if (IS_ERR(em)) {
6804                                         btrfs_end_transaction(trans, root);
6805                                         goto unlock_err;
6806                                 }
6807                         }
6808
6809                         ret = btrfs_add_ordered_extent_dio(inode, start,
6810                                            block_start, len, len, type);
6811                         btrfs_end_transaction(trans, root);
6812                         if (ret) {
6813                                 free_extent_map(em);
6814                                 goto unlock_err;
6815                         }
6816                         goto unlock;
6817                 }
6818                 btrfs_end_transaction(trans, root);
6819         }
6820 must_cow:
6821         /*
6822          * this will cow the extent, reset the len in case we changed
6823          * it above
6824          */
6825         len = bh_result->b_size;
6826         free_extent_map(em);
6827         em = btrfs_new_extent_direct(inode, start, len);
6828         if (IS_ERR(em)) {
6829                 ret = PTR_ERR(em);
6830                 goto unlock_err;
6831         }
6832         len = min(len, em->len - (start - em->start));
6833 unlock:
6834         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6835                 inode->i_blkbits;
6836         bh_result->b_size = len;
6837         bh_result->b_bdev = em->bdev;
6838         set_buffer_mapped(bh_result);
6839         if (create) {
6840                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6841                         set_buffer_new(bh_result);
6842
6843                 /*
6844                  * Need to update the i_size under the extent lock so buffered
6845                  * readers will get the updated i_size when we unlock.
6846                  */
6847                 if (start + len > i_size_read(inode))
6848                         i_size_write(inode, start + len);
6849
6850                 spin_lock(&BTRFS_I(inode)->lock);
6851                 BTRFS_I(inode)->outstanding_extents++;
6852                 spin_unlock(&BTRFS_I(inode)->lock);
6853
6854                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6855                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6856                                      &cached_state, GFP_NOFS);
6857                 BUG_ON(ret);
6858         }
6859
6860         /*
6861          * In the case of write we need to clear and unlock the entire range,
6862          * in the case of read we need to unlock only the end area that we
6863          * aren't using if there is any left over space.
6864          */
6865         if (lockstart < lockend) {
6866                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6867                                  lockend, unlock_bits, 1, 0,
6868                                  &cached_state, GFP_NOFS);
6869         } else {
6870                 free_extent_state(cached_state);
6871         }
6872
6873         free_extent_map(em);
6874
6875         return 0;
6876
6877 unlock_err:
6878         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6879                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6880         return ret;
6881 }
6882
6883 struct btrfs_dio_private {
6884         struct inode *inode;
6885         u64 logical_offset;
6886         u64 disk_bytenr;
6887         u64 bytes;
6888         void *private;
6889
6890         /* number of bios pending for this dio */
6891         atomic_t pending_bios;
6892
6893         /* IO errors */
6894         int errors;
6895
6896         struct bio *orig_bio;
6897 };
6898
6899 static void btrfs_endio_direct_read(struct bio *bio, int err)
6900 {
6901         struct btrfs_dio_private *dip = bio->bi_private;
6902         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6903         struct bio_vec *bvec = bio->bi_io_vec;
6904         struct inode *inode = dip->inode;
6905         struct btrfs_root *root = BTRFS_I(inode)->root;
6906         u64 start;
6907
6908         start = dip->logical_offset;
6909         do {
6910                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6911                         struct page *page = bvec->bv_page;
6912                         char *kaddr;
6913                         u32 csum = ~(u32)0;
6914                         u64 private = ~(u32)0;
6915                         unsigned long flags;
6916
6917                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6918                                               start, &private))
6919                                 goto failed;
6920                         local_irq_save(flags);
6921                         kaddr = kmap_atomic(page);
6922                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6923                                                csum, bvec->bv_len);
6924                         btrfs_csum_final(csum, (char *)&csum);
6925                         kunmap_atomic(kaddr);
6926                         local_irq_restore(flags);
6927
6928                         flush_dcache_page(bvec->bv_page);
6929                         if (csum != private) {
6930 failed:
6931                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6932                                       " %llu csum %u private %u\n",
6933                                       (unsigned long long)btrfs_ino(inode),
6934                                       (unsigned long long)start,
6935                                       csum, (unsigned)private);
6936                                 err = -EIO;
6937                         }
6938                 }
6939
6940                 start += bvec->bv_len;
6941                 bvec++;
6942         } while (bvec <= bvec_end);
6943
6944         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6945                       dip->logical_offset + dip->bytes - 1);
6946         bio->bi_private = dip->private;
6947
6948         kfree(dip);
6949
6950         /* If we had a csum failure make sure to clear the uptodate flag */
6951         if (err)
6952                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6953         dio_end_io(bio, err);
6954 }
6955
6956 static void btrfs_endio_direct_write(struct bio *bio, int err)
6957 {
6958         struct btrfs_dio_private *dip = bio->bi_private;
6959         struct inode *inode = dip->inode;
6960         struct btrfs_root *root = BTRFS_I(inode)->root;
6961         struct btrfs_ordered_extent *ordered = NULL;
6962         u64 ordered_offset = dip->logical_offset;
6963         u64 ordered_bytes = dip->bytes;
6964         int ret;
6965
6966         if (err)
6967                 goto out_done;
6968 again:
6969         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6970                                                    &ordered_offset,
6971                                                    ordered_bytes, !err);
6972         if (!ret)
6973                 goto out_test;
6974
6975         ordered->work.func = finish_ordered_fn;
6976         ordered->work.flags = 0;
6977         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6978                            &ordered->work);
6979 out_test:
6980         /*
6981          * our bio might span multiple ordered extents.  If we haven't
6982          * completed the accounting for the whole dio, go back and try again
6983          */
6984         if (ordered_offset < dip->logical_offset + dip->bytes) {
6985                 ordered_bytes = dip->logical_offset + dip->bytes -
6986                         ordered_offset;
6987                 ordered = NULL;
6988                 goto again;
6989         }
6990 out_done:
6991         bio->bi_private = dip->private;
6992
6993         kfree(dip);
6994
6995         /* If we had an error make sure to clear the uptodate flag */
6996         if (err)
6997                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6998         dio_end_io(bio, err);
6999 }
7000
7001 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7002                                     struct bio *bio, int mirror_num,
7003                                     unsigned long bio_flags, u64 offset)
7004 {
7005         int ret;
7006         struct btrfs_root *root = BTRFS_I(inode)->root;
7007         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7008         BUG_ON(ret); /* -ENOMEM */
7009         return 0;
7010 }
7011
7012 static void btrfs_end_dio_bio(struct bio *bio, int err)
7013 {
7014         struct btrfs_dio_private *dip = bio->bi_private;
7015
7016         if (err) {
7017                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7018                       "sector %#Lx len %u err no %d\n",
7019                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7020                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7021                 dip->errors = 1;
7022
7023                 /*
7024                  * before atomic variable goto zero, we must make sure
7025                  * dip->errors is perceived to be set.
7026                  */
7027                 smp_mb__before_atomic_dec();
7028         }
7029
7030         /* if there are more bios still pending for this dio, just exit */
7031         if (!atomic_dec_and_test(&dip->pending_bios))
7032                 goto out;
7033
7034         if (dip->errors)
7035                 bio_io_error(dip->orig_bio);
7036         else {
7037                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
7038                 bio_endio(dip->orig_bio, 0);
7039         }
7040 out:
7041         bio_put(bio);
7042 }
7043
7044 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7045                                        u64 first_sector, gfp_t gfp_flags)
7046 {
7047         int nr_vecs = bio_get_nr_vecs(bdev);
7048         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7049 }
7050
7051 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7052                                          int rw, u64 file_offset, int skip_sum,
7053                                          int async_submit)
7054 {
7055         int write = rw & REQ_WRITE;
7056         struct btrfs_root *root = BTRFS_I(inode)->root;
7057         int ret;
7058
7059         if (async_submit)
7060                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7061
7062         bio_get(bio);
7063
7064         if (!write) {
7065                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7066                 if (ret)
7067                         goto err;
7068         }
7069
7070         if (skip_sum)
7071                 goto map;
7072
7073         if (write && async_submit) {
7074                 ret = btrfs_wq_submit_bio(root->fs_info,
7075                                    inode, rw, bio, 0, 0,
7076                                    file_offset,
7077                                    __btrfs_submit_bio_start_direct_io,
7078                                    __btrfs_submit_bio_done);
7079                 goto err;
7080         } else if (write) {
7081                 /*
7082                  * If we aren't doing async submit, calculate the csum of the
7083                  * bio now.
7084                  */
7085                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7086                 if (ret)
7087                         goto err;
7088         } else if (!skip_sum) {
7089                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7090                 if (ret)
7091                         goto err;
7092         }
7093
7094 map:
7095         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7096 err:
7097         bio_put(bio);
7098         return ret;
7099 }
7100
7101 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7102                                     int skip_sum)
7103 {
7104         struct inode *inode = dip->inode;
7105         struct btrfs_root *root = BTRFS_I(inode)->root;
7106         struct bio *bio;
7107         struct bio *orig_bio = dip->orig_bio;
7108         struct bio_vec *bvec = orig_bio->bi_io_vec;
7109         u64 start_sector = orig_bio->bi_sector;
7110         u64 file_offset = dip->logical_offset;
7111         u64 submit_len = 0;
7112         u64 map_length;
7113         int nr_pages = 0;
7114         int ret = 0;
7115         int async_submit = 0;
7116
7117         map_length = orig_bio->bi_size;
7118         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7119                               &map_length, NULL, 0);
7120         if (ret) {
7121                 bio_put(orig_bio);
7122                 return -EIO;
7123         }
7124         if (map_length >= orig_bio->bi_size) {
7125                 bio = orig_bio;
7126                 goto submit;
7127         }
7128
7129         /* async crcs make it difficult to collect full stripe writes. */
7130         if (btrfs_get_alloc_profile(root, 1) &
7131             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7132                 async_submit = 0;
7133         else
7134                 async_submit = 1;
7135
7136         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7137         if (!bio)
7138                 return -ENOMEM;
7139         bio->bi_private = dip;
7140         bio->bi_end_io = btrfs_end_dio_bio;
7141         atomic_inc(&dip->pending_bios);
7142
7143         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7144                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7145                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7146                                  bvec->bv_offset) < bvec->bv_len)) {
7147                         /*
7148                          * inc the count before we submit the bio so
7149                          * we know the end IO handler won't happen before
7150                          * we inc the count. Otherwise, the dip might get freed
7151                          * before we're done setting it up
7152                          */
7153                         atomic_inc(&dip->pending_bios);
7154                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7155                                                      file_offset, skip_sum,
7156                                                      async_submit);
7157                         if (ret) {
7158                                 bio_put(bio);
7159                                 atomic_dec(&dip->pending_bios);
7160                                 goto out_err;
7161                         }
7162
7163                         start_sector += submit_len >> 9;
7164                         file_offset += submit_len;
7165
7166                         submit_len = 0;
7167                         nr_pages = 0;
7168
7169                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7170                                                   start_sector, GFP_NOFS);
7171                         if (!bio)
7172                                 goto out_err;
7173                         bio->bi_private = dip;
7174                         bio->bi_end_io = btrfs_end_dio_bio;
7175
7176                         map_length = orig_bio->bi_size;
7177                         ret = btrfs_map_block(root->fs_info, rw,
7178                                               start_sector << 9,
7179                                               &map_length, NULL, 0);
7180                         if (ret) {
7181                                 bio_put(bio);
7182                                 goto out_err;
7183                         }
7184                 } else {
7185                         submit_len += bvec->bv_len;
7186                         nr_pages ++;
7187                         bvec++;
7188                 }
7189         }
7190
7191 submit:
7192         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7193                                      async_submit);
7194         if (!ret)
7195                 return 0;
7196
7197         bio_put(bio);
7198 out_err:
7199         dip->errors = 1;
7200         /*
7201          * before atomic variable goto zero, we must
7202          * make sure dip->errors is perceived to be set.
7203          */
7204         smp_mb__before_atomic_dec();
7205         if (atomic_dec_and_test(&dip->pending_bios))
7206                 bio_io_error(dip->orig_bio);
7207
7208         /* bio_end_io() will handle error, so we needn't return it */
7209         return 0;
7210 }
7211
7212 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
7213                                 loff_t file_offset)
7214 {
7215         struct btrfs_root *root = BTRFS_I(inode)->root;
7216         struct btrfs_dio_private *dip;
7217         struct bio_vec *bvec = bio->bi_io_vec;
7218         int skip_sum;
7219         int write = rw & REQ_WRITE;
7220         int ret = 0;
7221
7222         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7223
7224         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7225         if (!dip) {
7226                 ret = -ENOMEM;
7227                 goto free_ordered;
7228         }
7229
7230         dip->private = bio->bi_private;
7231         dip->inode = inode;
7232         dip->logical_offset = file_offset;
7233
7234         dip->bytes = 0;
7235         do {
7236                 dip->bytes += bvec->bv_len;
7237                 bvec++;
7238         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
7239
7240         dip->disk_bytenr = (u64)bio->bi_sector << 9;
7241         bio->bi_private = dip;
7242         dip->errors = 0;
7243         dip->orig_bio = bio;
7244         atomic_set(&dip->pending_bios, 0);
7245
7246         if (write)
7247                 bio->bi_end_io = btrfs_endio_direct_write;
7248         else
7249                 bio->bi_end_io = btrfs_endio_direct_read;
7250
7251         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7252         if (!ret)
7253                 return;
7254 free_ordered:
7255         /*
7256          * If this is a write, we need to clean up the reserved space and kill
7257          * the ordered extent.
7258          */
7259         if (write) {
7260                 struct btrfs_ordered_extent *ordered;
7261                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7262                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7263                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7264                         btrfs_free_reserved_extent(root, ordered->start,
7265                                                    ordered->disk_len);
7266                 btrfs_put_ordered_extent(ordered);
7267                 btrfs_put_ordered_extent(ordered);
7268         }
7269         bio_endio(bio, ret);
7270 }
7271
7272 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7273                         const struct iovec *iov, loff_t offset,
7274                         unsigned long nr_segs)
7275 {
7276         int seg;
7277         int i;
7278         size_t size;
7279         unsigned long addr;
7280         unsigned blocksize_mask = root->sectorsize - 1;
7281         ssize_t retval = -EINVAL;
7282         loff_t end = offset;
7283
7284         if (offset & blocksize_mask)
7285                 goto out;
7286
7287         /* Check the memory alignment.  Blocks cannot straddle pages */
7288         for (seg = 0; seg < nr_segs; seg++) {
7289                 addr = (unsigned long)iov[seg].iov_base;
7290                 size = iov[seg].iov_len;
7291                 end += size;
7292                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7293                         goto out;
7294
7295                 /* If this is a write we don't need to check anymore */
7296                 if (rw & WRITE)
7297                         continue;
7298
7299                 /*
7300                  * Check to make sure we don't have duplicate iov_base's in this
7301                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7302                  * when reading back.
7303                  */
7304                 for (i = seg + 1; i < nr_segs; i++) {
7305                         if (iov[seg].iov_base == iov[i].iov_base)
7306                                 goto out;
7307                 }
7308         }
7309         retval = 0;
7310 out:
7311         return retval;
7312 }
7313
7314 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7315                         const struct iovec *iov, loff_t offset,
7316                         unsigned long nr_segs)
7317 {
7318         struct file *file = iocb->ki_filp;
7319         struct inode *inode = file->f_mapping->host;
7320         size_t count = 0;
7321         int flags = 0;
7322         bool wakeup = true;
7323         bool relock = false;
7324         ssize_t ret;
7325
7326         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7327                             offset, nr_segs))
7328                 return 0;
7329
7330         atomic_inc(&inode->i_dio_count);
7331         smp_mb__after_atomic_inc();
7332
7333         if (rw & WRITE) {
7334                 count = iov_length(iov, nr_segs);
7335                 /*
7336                  * If the write DIO is beyond the EOF, we need update
7337                  * the isize, but it is protected by i_mutex. So we can
7338                  * not unlock the i_mutex at this case.
7339                  */
7340                 if (offset + count <= inode->i_size) {
7341                         mutex_unlock(&inode->i_mutex);
7342                         relock = true;
7343                 }
7344                 ret = btrfs_delalloc_reserve_space(inode, count);
7345                 if (ret)
7346                         goto out;
7347         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7348                                      &BTRFS_I(inode)->runtime_flags))) {
7349                 inode_dio_done(inode);
7350                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7351                 wakeup = false;
7352         }
7353
7354         ret = __blockdev_direct_IO(rw, iocb, inode,
7355                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7356                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7357                         btrfs_submit_direct, flags);
7358         if (rw & WRITE) {
7359                 if (ret < 0 && ret != -EIOCBQUEUED)
7360                         btrfs_delalloc_release_space(inode, count);
7361                 else if (ret >= 0 && (size_t)ret < count)
7362                         btrfs_delalloc_release_space(inode,
7363                                                      count - (size_t)ret);
7364                 else
7365                         btrfs_delalloc_release_metadata(inode, 0);
7366         }
7367 out:
7368         if (wakeup)
7369                 inode_dio_done(inode);
7370         if (relock)
7371                 mutex_lock(&inode->i_mutex);
7372
7373         return ret;
7374 }
7375
7376 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7377
7378 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7379                 __u64 start, __u64 len)
7380 {
7381         int     ret;
7382
7383         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7384         if (ret)
7385                 return ret;
7386
7387         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7388 }
7389
7390 int btrfs_readpage(struct file *file, struct page *page)
7391 {
7392         struct extent_io_tree *tree;
7393         tree = &BTRFS_I(page->mapping->host)->io_tree;
7394         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7395 }
7396
7397 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7398 {
7399         struct extent_io_tree *tree;
7400
7401
7402         if (current->flags & PF_MEMALLOC) {
7403                 redirty_page_for_writepage(wbc, page);
7404                 unlock_page(page);
7405                 return 0;
7406         }
7407         tree = &BTRFS_I(page->mapping->host)->io_tree;
7408         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7409 }
7410
7411 int btrfs_writepages(struct address_space *mapping,
7412                      struct writeback_control *wbc)
7413 {
7414         struct extent_io_tree *tree;
7415
7416         tree = &BTRFS_I(mapping->host)->io_tree;
7417         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7418 }
7419
7420 static int
7421 btrfs_readpages(struct file *file, struct address_space *mapping,
7422                 struct list_head *pages, unsigned nr_pages)
7423 {
7424         struct extent_io_tree *tree;
7425         tree = &BTRFS_I(mapping->host)->io_tree;
7426         return extent_readpages(tree, mapping, pages, nr_pages,
7427                                 btrfs_get_extent);
7428 }
7429 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7430 {
7431         struct extent_io_tree *tree;
7432         struct extent_map_tree *map;
7433         int ret;
7434
7435         tree = &BTRFS_I(page->mapping->host)->io_tree;
7436         map = &BTRFS_I(page->mapping->host)->extent_tree;
7437         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7438         if (ret == 1) {
7439                 ClearPagePrivate(page);
7440                 set_page_private(page, 0);
7441                 page_cache_release(page);
7442         }
7443         return ret;
7444 }
7445
7446 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7447 {
7448         if (PageWriteback(page) || PageDirty(page))
7449                 return 0;
7450         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7451 }
7452
7453 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7454 {
7455         struct inode *inode = page->mapping->host;
7456         struct extent_io_tree *tree;
7457         struct btrfs_ordered_extent *ordered;
7458         struct extent_state *cached_state = NULL;
7459         u64 page_start = page_offset(page);
7460         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7461
7462         /*
7463          * we have the page locked, so new writeback can't start,
7464          * and the dirty bit won't be cleared while we are here.
7465          *
7466          * Wait for IO on this page so that we can safely clear
7467          * the PagePrivate2 bit and do ordered accounting
7468          */
7469         wait_on_page_writeback(page);
7470
7471         tree = &BTRFS_I(inode)->io_tree;
7472         if (offset) {
7473                 btrfs_releasepage(page, GFP_NOFS);
7474                 return;
7475         }
7476         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7477         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7478         if (ordered) {
7479                 /*
7480                  * IO on this page will never be started, so we need
7481                  * to account for any ordered extents now
7482                  */
7483                 clear_extent_bit(tree, page_start, page_end,
7484                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7485                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7486                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7487                 /*
7488                  * whoever cleared the private bit is responsible
7489                  * for the finish_ordered_io
7490                  */
7491                 if (TestClearPagePrivate2(page) &&
7492                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7493                                                    PAGE_CACHE_SIZE, 1)) {
7494                         btrfs_finish_ordered_io(ordered);
7495                 }
7496                 btrfs_put_ordered_extent(ordered);
7497                 cached_state = NULL;
7498                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7499         }
7500         clear_extent_bit(tree, page_start, page_end,
7501                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7502                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7503                  &cached_state, GFP_NOFS);
7504         __btrfs_releasepage(page, GFP_NOFS);
7505
7506         ClearPageChecked(page);
7507         if (PagePrivate(page)) {
7508                 ClearPagePrivate(page);
7509                 set_page_private(page, 0);
7510                 page_cache_release(page);
7511         }
7512 }
7513
7514 /*
7515  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7516  * called from a page fault handler when a page is first dirtied. Hence we must
7517  * be careful to check for EOF conditions here. We set the page up correctly
7518  * for a written page which means we get ENOSPC checking when writing into
7519  * holes and correct delalloc and unwritten extent mapping on filesystems that
7520  * support these features.
7521  *
7522  * We are not allowed to take the i_mutex here so we have to play games to
7523  * protect against truncate races as the page could now be beyond EOF.  Because
7524  * vmtruncate() writes the inode size before removing pages, once we have the
7525  * page lock we can determine safely if the page is beyond EOF. If it is not
7526  * beyond EOF, then the page is guaranteed safe against truncation until we
7527  * unlock the page.
7528  */
7529 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7530 {
7531         struct page *page = vmf->page;
7532         struct inode *inode = file_inode(vma->vm_file);
7533         struct btrfs_root *root = BTRFS_I(inode)->root;
7534         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7535         struct btrfs_ordered_extent *ordered;
7536         struct extent_state *cached_state = NULL;
7537         char *kaddr;
7538         unsigned long zero_start;
7539         loff_t size;
7540         int ret;
7541         int reserved = 0;
7542         u64 page_start;
7543         u64 page_end;
7544
7545         sb_start_pagefault(inode->i_sb);
7546         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7547         if (!ret) {
7548                 ret = file_update_time(vma->vm_file);
7549                 reserved = 1;
7550         }
7551         if (ret) {
7552                 if (ret == -ENOMEM)
7553                         ret = VM_FAULT_OOM;
7554                 else /* -ENOSPC, -EIO, etc */
7555                         ret = VM_FAULT_SIGBUS;
7556                 if (reserved)
7557                         goto out;
7558                 goto out_noreserve;
7559         }
7560
7561         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7562 again:
7563         lock_page(page);
7564         size = i_size_read(inode);
7565         page_start = page_offset(page);
7566         page_end = page_start + PAGE_CACHE_SIZE - 1;
7567
7568         if ((page->mapping != inode->i_mapping) ||
7569             (page_start >= size)) {
7570                 /* page got truncated out from underneath us */
7571                 goto out_unlock;
7572         }
7573         wait_on_page_writeback(page);
7574
7575         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7576         set_page_extent_mapped(page);
7577
7578         /*
7579          * we can't set the delalloc bits if there are pending ordered
7580          * extents.  Drop our locks and wait for them to finish
7581          */
7582         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7583         if (ordered) {
7584                 unlock_extent_cached(io_tree, page_start, page_end,
7585                                      &cached_state, GFP_NOFS);
7586                 unlock_page(page);
7587                 btrfs_start_ordered_extent(inode, ordered, 1);
7588                 btrfs_put_ordered_extent(ordered);
7589                 goto again;
7590         }
7591
7592         /*
7593          * XXX - page_mkwrite gets called every time the page is dirtied, even
7594          * if it was already dirty, so for space accounting reasons we need to
7595          * clear any delalloc bits for the range we are fixing to save.  There
7596          * is probably a better way to do this, but for now keep consistent with
7597          * prepare_pages in the normal write path.
7598          */
7599         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7600                           EXTENT_DIRTY | EXTENT_DELALLOC |
7601                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7602                           0, 0, &cached_state, GFP_NOFS);
7603
7604         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7605                                         &cached_state);
7606         if (ret) {
7607                 unlock_extent_cached(io_tree, page_start, page_end,
7608                                      &cached_state, GFP_NOFS);
7609                 ret = VM_FAULT_SIGBUS;
7610                 goto out_unlock;
7611         }
7612         ret = 0;
7613
7614         /* page is wholly or partially inside EOF */
7615         if (page_start + PAGE_CACHE_SIZE > size)
7616                 zero_start = size & ~PAGE_CACHE_MASK;
7617         else
7618                 zero_start = PAGE_CACHE_SIZE;
7619
7620         if (zero_start != PAGE_CACHE_SIZE) {
7621                 kaddr = kmap(page);
7622                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7623                 flush_dcache_page(page);
7624                 kunmap(page);
7625         }
7626         ClearPageChecked(page);
7627         set_page_dirty(page);
7628         SetPageUptodate(page);
7629
7630         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7631         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7632         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7633
7634         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7635
7636 out_unlock:
7637         if (!ret) {
7638                 sb_end_pagefault(inode->i_sb);
7639                 return VM_FAULT_LOCKED;
7640         }
7641         unlock_page(page);
7642 out:
7643         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7644 out_noreserve:
7645         sb_end_pagefault(inode->i_sb);
7646         return ret;
7647 }
7648
7649 static int btrfs_truncate(struct inode *inode)
7650 {
7651         struct btrfs_root *root = BTRFS_I(inode)->root;
7652         struct btrfs_block_rsv *rsv;
7653         int ret;
7654         int err = 0;
7655         struct btrfs_trans_handle *trans;
7656         u64 mask = root->sectorsize - 1;
7657         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7658
7659         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7660         if (ret)
7661                 return ret;
7662
7663         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7664         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7665
7666         /*
7667          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7668          * 3 things going on here
7669          *
7670          * 1) We need to reserve space for our orphan item and the space to
7671          * delete our orphan item.  Lord knows we don't want to have a dangling
7672          * orphan item because we didn't reserve space to remove it.
7673          *
7674          * 2) We need to reserve space to update our inode.
7675          *
7676          * 3) We need to have something to cache all the space that is going to
7677          * be free'd up by the truncate operation, but also have some slack
7678          * space reserved in case it uses space during the truncate (thank you
7679          * very much snapshotting).
7680          *
7681          * And we need these to all be seperate.  The fact is we can use alot of
7682          * space doing the truncate, and we have no earthly idea how much space
7683          * we will use, so we need the truncate reservation to be seperate so it
7684          * doesn't end up using space reserved for updating the inode or
7685          * removing the orphan item.  We also need to be able to stop the
7686          * transaction and start a new one, which means we need to be able to
7687          * update the inode several times, and we have no idea of knowing how
7688          * many times that will be, so we can't just reserve 1 item for the
7689          * entirety of the opration, so that has to be done seperately as well.
7690          * Then there is the orphan item, which does indeed need to be held on
7691          * to for the whole operation, and we need nobody to touch this reserved
7692          * space except the orphan code.
7693          *
7694          * So that leaves us with
7695          *
7696          * 1) root->orphan_block_rsv - for the orphan deletion.
7697          * 2) rsv - for the truncate reservation, which we will steal from the
7698          * transaction reservation.
7699          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7700          * updating the inode.
7701          */
7702         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7703         if (!rsv)
7704                 return -ENOMEM;
7705         rsv->size = min_size;
7706         rsv->failfast = 1;
7707
7708         /*
7709          * 1 for the truncate slack space
7710          * 1 for updating the inode.
7711          */
7712         trans = btrfs_start_transaction(root, 2);
7713         if (IS_ERR(trans)) {
7714                 err = PTR_ERR(trans);
7715                 goto out;
7716         }
7717
7718         /* Migrate the slack space for the truncate to our reserve */
7719         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7720                                       min_size);
7721         BUG_ON(ret);
7722
7723         /*
7724          * setattr is responsible for setting the ordered_data_close flag,
7725          * but that is only tested during the last file release.  That
7726          * could happen well after the next commit, leaving a great big
7727          * window where new writes may get lost if someone chooses to write
7728          * to this file after truncating to zero
7729          *
7730          * The inode doesn't have any dirty data here, and so if we commit
7731          * this is a noop.  If someone immediately starts writing to the inode
7732          * it is very likely we'll catch some of their writes in this
7733          * transaction, and the commit will find this file on the ordered
7734          * data list with good things to send down.
7735          *
7736          * This is a best effort solution, there is still a window where
7737          * using truncate to replace the contents of the file will
7738          * end up with a zero length file after a crash.
7739          */
7740         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7741                                            &BTRFS_I(inode)->runtime_flags))
7742                 btrfs_add_ordered_operation(trans, root, inode);
7743
7744         /*
7745          * So if we truncate and then write and fsync we normally would just
7746          * write the extents that changed, which is a problem if we need to
7747          * first truncate that entire inode.  So set this flag so we write out
7748          * all of the extents in the inode to the sync log so we're completely
7749          * safe.
7750          */
7751         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7752         trans->block_rsv = rsv;
7753
7754         while (1) {
7755                 ret = btrfs_truncate_inode_items(trans, root, inode,
7756                                                  inode->i_size,
7757                                                  BTRFS_EXTENT_DATA_KEY);
7758                 if (ret != -ENOSPC) {
7759                         err = ret;
7760                         break;
7761                 }
7762
7763                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7764                 ret = btrfs_update_inode(trans, root, inode);
7765                 if (ret) {
7766                         err = ret;
7767                         break;
7768                 }
7769
7770                 btrfs_end_transaction(trans, root);
7771                 btrfs_btree_balance_dirty(root);
7772
7773                 trans = btrfs_start_transaction(root, 2);
7774                 if (IS_ERR(trans)) {
7775                         ret = err = PTR_ERR(trans);
7776                         trans = NULL;
7777                         break;
7778                 }
7779
7780                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7781                                               rsv, min_size);
7782                 BUG_ON(ret);    /* shouldn't happen */
7783                 trans->block_rsv = rsv;
7784         }
7785
7786         if (ret == 0 && inode->i_nlink > 0) {
7787                 trans->block_rsv = root->orphan_block_rsv;
7788                 ret = btrfs_orphan_del(trans, inode);
7789                 if (ret)
7790                         err = ret;
7791         }
7792
7793         if (trans) {
7794                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7795                 ret = btrfs_update_inode(trans, root, inode);
7796                 if (ret && !err)
7797                         err = ret;
7798
7799                 ret = btrfs_end_transaction(trans, root);
7800                 btrfs_btree_balance_dirty(root);
7801         }
7802
7803 out:
7804         btrfs_free_block_rsv(root, rsv);
7805
7806         if (ret && !err)
7807                 err = ret;
7808
7809         return err;
7810 }
7811
7812 /*
7813  * create a new subvolume directory/inode (helper for the ioctl).
7814  */
7815 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7816                              struct btrfs_root *new_root, u64 new_dirid)
7817 {
7818         struct inode *inode;
7819         int err;
7820         u64 index = 0;
7821
7822         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7823                                 new_dirid, new_dirid,
7824                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7825                                 &index);
7826         if (IS_ERR(inode))
7827                 return PTR_ERR(inode);
7828         inode->i_op = &btrfs_dir_inode_operations;
7829         inode->i_fop = &btrfs_dir_file_operations;
7830
7831         set_nlink(inode, 1);
7832         btrfs_i_size_write(inode, 0);
7833
7834         err = btrfs_update_inode(trans, new_root, inode);
7835
7836         iput(inode);
7837         return err;
7838 }
7839
7840 struct inode *btrfs_alloc_inode(struct super_block *sb)
7841 {
7842         struct btrfs_inode *ei;
7843         struct inode *inode;
7844
7845         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7846         if (!ei)
7847                 return NULL;
7848
7849         ei->root = NULL;
7850         ei->generation = 0;
7851         ei->last_trans = 0;
7852         ei->last_sub_trans = 0;
7853         ei->logged_trans = 0;
7854         ei->delalloc_bytes = 0;
7855         ei->disk_i_size = 0;
7856         ei->flags = 0;
7857         ei->csum_bytes = 0;
7858         ei->index_cnt = (u64)-1;
7859         ei->last_unlink_trans = 0;
7860         ei->last_log_commit = 0;
7861
7862         spin_lock_init(&ei->lock);
7863         ei->outstanding_extents = 0;
7864         ei->reserved_extents = 0;
7865
7866         ei->runtime_flags = 0;
7867         ei->force_compress = BTRFS_COMPRESS_NONE;
7868
7869         ei->delayed_node = NULL;
7870
7871         inode = &ei->vfs_inode;
7872         extent_map_tree_init(&ei->extent_tree);
7873         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7874         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7875         ei->io_tree.track_uptodate = 1;
7876         ei->io_failure_tree.track_uptodate = 1;
7877         atomic_set(&ei->sync_writers, 0);
7878         mutex_init(&ei->log_mutex);
7879         mutex_init(&ei->delalloc_mutex);
7880         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7881         INIT_LIST_HEAD(&ei->delalloc_inodes);
7882         INIT_LIST_HEAD(&ei->ordered_operations);
7883         RB_CLEAR_NODE(&ei->rb_node);
7884
7885         return inode;
7886 }
7887
7888 static void btrfs_i_callback(struct rcu_head *head)
7889 {
7890         struct inode *inode = container_of(head, struct inode, i_rcu);
7891         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7892 }
7893
7894 void btrfs_destroy_inode(struct inode *inode)
7895 {
7896         struct btrfs_ordered_extent *ordered;
7897         struct btrfs_root *root = BTRFS_I(inode)->root;
7898
7899         WARN_ON(!hlist_empty(&inode->i_dentry));
7900         WARN_ON(inode->i_data.nrpages);
7901         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7902         WARN_ON(BTRFS_I(inode)->reserved_extents);
7903         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7904         WARN_ON(BTRFS_I(inode)->csum_bytes);
7905
7906         /*
7907          * This can happen where we create an inode, but somebody else also
7908          * created the same inode and we need to destroy the one we already
7909          * created.
7910          */
7911         if (!root)
7912                 goto free;
7913
7914         /*
7915          * Make sure we're properly removed from the ordered operation
7916          * lists.
7917          */
7918         smp_mb();
7919         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7920                 spin_lock(&root->fs_info->ordered_extent_lock);
7921                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7922                 spin_unlock(&root->fs_info->ordered_extent_lock);
7923         }
7924
7925         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7926                      &BTRFS_I(inode)->runtime_flags)) {
7927                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7928                        (unsigned long long)btrfs_ino(inode));
7929                 atomic_dec(&root->orphan_inodes);
7930         }
7931
7932         while (1) {
7933                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7934                 if (!ordered)
7935                         break;
7936                 else {
7937                         printk(KERN_ERR "btrfs found ordered "
7938                                "extent %llu %llu on inode cleanup\n",
7939                                (unsigned long long)ordered->file_offset,
7940                                (unsigned long long)ordered->len);
7941                         btrfs_remove_ordered_extent(inode, ordered);
7942                         btrfs_put_ordered_extent(ordered);
7943                         btrfs_put_ordered_extent(ordered);
7944                 }
7945         }
7946         inode_tree_del(inode);
7947         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7948 free:
7949         btrfs_remove_delayed_node(inode);
7950         call_rcu(&inode->i_rcu, btrfs_i_callback);
7951 }
7952
7953 int btrfs_drop_inode(struct inode *inode)
7954 {
7955         struct btrfs_root *root = BTRFS_I(inode)->root;
7956
7957         /* the snap/subvol tree is on deleting */
7958         if (btrfs_root_refs(&root->root_item) == 0 &&
7959             root != root->fs_info->tree_root)
7960                 return 1;
7961         else
7962                 return generic_drop_inode(inode);
7963 }
7964
7965 static void init_once(void *foo)
7966 {
7967         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7968
7969         inode_init_once(&ei->vfs_inode);
7970 }
7971
7972 void btrfs_destroy_cachep(void)
7973 {
7974         /*
7975          * Make sure all delayed rcu free inodes are flushed before we
7976          * destroy cache.
7977          */
7978         rcu_barrier();
7979         if (btrfs_inode_cachep)
7980                 kmem_cache_destroy(btrfs_inode_cachep);
7981         if (btrfs_trans_handle_cachep)
7982                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7983         if (btrfs_transaction_cachep)
7984                 kmem_cache_destroy(btrfs_transaction_cachep);
7985         if (btrfs_path_cachep)
7986                 kmem_cache_destroy(btrfs_path_cachep);
7987         if (btrfs_free_space_cachep)
7988                 kmem_cache_destroy(btrfs_free_space_cachep);
7989         if (btrfs_delalloc_work_cachep)
7990                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7991 }
7992
7993 int btrfs_init_cachep(void)
7994 {
7995         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7996                         sizeof(struct btrfs_inode), 0,
7997                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7998         if (!btrfs_inode_cachep)
7999                 goto fail;
8000
8001         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8002                         sizeof(struct btrfs_trans_handle), 0,
8003                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8004         if (!btrfs_trans_handle_cachep)
8005                 goto fail;
8006
8007         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8008                         sizeof(struct btrfs_transaction), 0,
8009                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8010         if (!btrfs_transaction_cachep)
8011                 goto fail;
8012
8013         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8014                         sizeof(struct btrfs_path), 0,
8015                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8016         if (!btrfs_path_cachep)
8017                 goto fail;
8018
8019         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8020                         sizeof(struct btrfs_free_space), 0,
8021                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8022         if (!btrfs_free_space_cachep)
8023                 goto fail;
8024
8025         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8026                         sizeof(struct btrfs_delalloc_work), 0,
8027                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8028                         NULL);
8029         if (!btrfs_delalloc_work_cachep)
8030                 goto fail;
8031
8032         return 0;
8033 fail:
8034         btrfs_destroy_cachep();
8035         return -ENOMEM;
8036 }
8037
8038 static int btrfs_getattr(struct vfsmount *mnt,
8039                          struct dentry *dentry, struct kstat *stat)
8040 {
8041         u64 delalloc_bytes;
8042         struct inode *inode = dentry->d_inode;
8043         u32 blocksize = inode->i_sb->s_blocksize;
8044
8045         generic_fillattr(inode, stat);
8046         stat->dev = BTRFS_I(inode)->root->anon_dev;
8047         stat->blksize = PAGE_CACHE_SIZE;
8048
8049         spin_lock(&BTRFS_I(inode)->lock);
8050         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8051         spin_unlock(&BTRFS_I(inode)->lock);
8052         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8053                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8054         return 0;
8055 }
8056
8057 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8058                            struct inode *new_dir, struct dentry *new_dentry)
8059 {
8060         struct btrfs_trans_handle *trans;
8061         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8062         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8063         struct inode *new_inode = new_dentry->d_inode;
8064         struct inode *old_inode = old_dentry->d_inode;
8065         struct timespec ctime = CURRENT_TIME;
8066         u64 index = 0;
8067         u64 root_objectid;
8068         int ret;
8069         u64 old_ino = btrfs_ino(old_inode);
8070
8071         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8072                 return -EPERM;
8073
8074         /* we only allow rename subvolume link between subvolumes */
8075         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8076                 return -EXDEV;
8077
8078         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8079             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8080                 return -ENOTEMPTY;
8081
8082         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8083             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8084                 return -ENOTEMPTY;
8085
8086
8087         /* check for collisions, even if the  name isn't there */
8088         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8089                              new_dentry->d_name.name,
8090                              new_dentry->d_name.len);
8091
8092         if (ret) {
8093                 if (ret == -EEXIST) {
8094                         /* we shouldn't get
8095                          * eexist without a new_inode */
8096                         if (!new_inode) {
8097                                 WARN_ON(1);
8098                                 return ret;
8099                         }
8100                 } else {
8101                         /* maybe -EOVERFLOW */
8102                         return ret;
8103                 }
8104         }
8105         ret = 0;
8106
8107         /*
8108          * we're using rename to replace one file with another.
8109          * and the replacement file is large.  Start IO on it now so
8110          * we don't add too much work to the end of the transaction
8111          */
8112         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8113             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8114                 filemap_flush(old_inode->i_mapping);
8115
8116         /* close the racy window with snapshot create/destroy ioctl */
8117         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8118                 down_read(&root->fs_info->subvol_sem);
8119         /*
8120          * We want to reserve the absolute worst case amount of items.  So if
8121          * both inodes are subvols and we need to unlink them then that would
8122          * require 4 item modifications, but if they are both normal inodes it
8123          * would require 5 item modifications, so we'll assume their normal
8124          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8125          * should cover the worst case number of items we'll modify.
8126          */
8127         trans = btrfs_start_transaction(root, 20);
8128         if (IS_ERR(trans)) {
8129                 ret = PTR_ERR(trans);
8130                 goto out_notrans;
8131         }
8132
8133         if (dest != root)
8134                 btrfs_record_root_in_trans(trans, dest);
8135
8136         ret = btrfs_set_inode_index(new_dir, &index);
8137         if (ret)
8138                 goto out_fail;
8139
8140         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8141                 /* force full log commit if subvolume involved. */
8142                 root->fs_info->last_trans_log_full_commit = trans->transid;
8143         } else {
8144                 ret = btrfs_insert_inode_ref(trans, dest,
8145                                              new_dentry->d_name.name,
8146                                              new_dentry->d_name.len,
8147                                              old_ino,
8148                                              btrfs_ino(new_dir), index);
8149                 if (ret)
8150                         goto out_fail;
8151                 /*
8152                  * this is an ugly little race, but the rename is required
8153                  * to make sure that if we crash, the inode is either at the
8154                  * old name or the new one.  pinning the log transaction lets
8155                  * us make sure we don't allow a log commit to come in after
8156                  * we unlink the name but before we add the new name back in.
8157                  */
8158                 btrfs_pin_log_trans(root);
8159         }
8160         /*
8161          * make sure the inode gets flushed if it is replacing
8162          * something.
8163          */
8164         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8165                 btrfs_add_ordered_operation(trans, root, old_inode);
8166
8167         inode_inc_iversion(old_dir);
8168         inode_inc_iversion(new_dir);
8169         inode_inc_iversion(old_inode);
8170         old_dir->i_ctime = old_dir->i_mtime = ctime;
8171         new_dir->i_ctime = new_dir->i_mtime = ctime;
8172         old_inode->i_ctime = ctime;
8173
8174         if (old_dentry->d_parent != new_dentry->d_parent)
8175                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8176
8177         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8178                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8179                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8180                                         old_dentry->d_name.name,
8181                                         old_dentry->d_name.len);
8182         } else {
8183                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8184                                         old_dentry->d_inode,
8185                                         old_dentry->d_name.name,
8186                                         old_dentry->d_name.len);
8187                 if (!ret)
8188                         ret = btrfs_update_inode(trans, root, old_inode);
8189         }
8190         if (ret) {
8191                 btrfs_abort_transaction(trans, root, ret);
8192                 goto out_fail;
8193         }
8194
8195         if (new_inode) {
8196                 inode_inc_iversion(new_inode);
8197                 new_inode->i_ctime = CURRENT_TIME;
8198                 if (unlikely(btrfs_ino(new_inode) ==
8199                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8200                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8201                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8202                                                 root_objectid,
8203                                                 new_dentry->d_name.name,
8204                                                 new_dentry->d_name.len);
8205                         BUG_ON(new_inode->i_nlink == 0);
8206                 } else {
8207                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8208                                                  new_dentry->d_inode,
8209                                                  new_dentry->d_name.name,
8210                                                  new_dentry->d_name.len);
8211                 }
8212                 if (!ret && new_inode->i_nlink == 0) {
8213                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8214                         BUG_ON(ret);
8215                 }
8216                 if (ret) {
8217                         btrfs_abort_transaction(trans, root, ret);
8218                         goto out_fail;
8219                 }
8220         }
8221
8222         ret = btrfs_add_link(trans, new_dir, old_inode,
8223                              new_dentry->d_name.name,
8224                              new_dentry->d_name.len, 0, index);
8225         if (ret) {
8226                 btrfs_abort_transaction(trans, root, ret);
8227                 goto out_fail;
8228         }
8229
8230         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8231                 struct dentry *parent = new_dentry->d_parent;
8232                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8233                 btrfs_end_log_trans(root);
8234         }
8235 out_fail:
8236         btrfs_end_transaction(trans, root);
8237 out_notrans:
8238         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8239                 up_read(&root->fs_info->subvol_sem);
8240
8241         return ret;
8242 }
8243
8244 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8245 {
8246         struct btrfs_delalloc_work *delalloc_work;
8247
8248         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8249                                      work);
8250         if (delalloc_work->wait)
8251                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8252         else
8253                 filemap_flush(delalloc_work->inode->i_mapping);
8254
8255         if (delalloc_work->delay_iput)
8256                 btrfs_add_delayed_iput(delalloc_work->inode);
8257         else
8258                 iput(delalloc_work->inode);
8259         complete(&delalloc_work->completion);
8260 }
8261
8262 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8263                                                     int wait, int delay_iput)
8264 {
8265         struct btrfs_delalloc_work *work;
8266
8267         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8268         if (!work)
8269                 return NULL;
8270
8271         init_completion(&work->completion);
8272         INIT_LIST_HEAD(&work->list);
8273         work->inode = inode;
8274         work->wait = wait;
8275         work->delay_iput = delay_iput;
8276         work->work.func = btrfs_run_delalloc_work;
8277
8278         return work;
8279 }
8280
8281 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8282 {
8283         wait_for_completion(&work->completion);
8284         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8285 }
8286
8287 /*
8288  * some fairly slow code that needs optimization. This walks the list
8289  * of all the inodes with pending delalloc and forces them to disk.
8290  */
8291 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8292 {
8293         struct btrfs_inode *binode;
8294         struct inode *inode;
8295         struct btrfs_delalloc_work *work, *next;
8296         struct list_head works;
8297         struct list_head splice;
8298         int ret = 0;
8299
8300         if (root->fs_info->sb->s_flags & MS_RDONLY)
8301                 return -EROFS;
8302
8303         INIT_LIST_HEAD(&works);
8304         INIT_LIST_HEAD(&splice);
8305
8306         spin_lock(&root->fs_info->delalloc_lock);
8307         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8308         while (!list_empty(&splice)) {
8309                 binode = list_entry(splice.next, struct btrfs_inode,
8310                                     delalloc_inodes);
8311
8312                 list_del_init(&binode->delalloc_inodes);
8313
8314                 inode = igrab(&binode->vfs_inode);
8315                 if (!inode) {
8316                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8317                                   &binode->runtime_flags);
8318                         continue;
8319                 }
8320
8321                 list_add_tail(&binode->delalloc_inodes,
8322                               &root->fs_info->delalloc_inodes);
8323                 spin_unlock(&root->fs_info->delalloc_lock);
8324
8325                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8326                 if (unlikely(!work)) {
8327                         ret = -ENOMEM;
8328                         goto out;
8329                 }
8330                 list_add_tail(&work->list, &works);
8331                 btrfs_queue_worker(&root->fs_info->flush_workers,
8332                                    &work->work);
8333
8334                 cond_resched();
8335                 spin_lock(&root->fs_info->delalloc_lock);
8336         }
8337         spin_unlock(&root->fs_info->delalloc_lock);
8338
8339         list_for_each_entry_safe(work, next, &works, list) {
8340                 list_del_init(&work->list);
8341                 btrfs_wait_and_free_delalloc_work(work);
8342         }
8343
8344         /* the filemap_flush will queue IO into the worker threads, but
8345          * we have to make sure the IO is actually started and that
8346          * ordered extents get created before we return
8347          */
8348         atomic_inc(&root->fs_info->async_submit_draining);
8349         while (atomic_read(&root->fs_info->nr_async_submits) ||
8350               atomic_read(&root->fs_info->async_delalloc_pages)) {
8351                 wait_event(root->fs_info->async_submit_wait,
8352                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8353                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8354         }
8355         atomic_dec(&root->fs_info->async_submit_draining);
8356         return 0;
8357 out:
8358         list_for_each_entry_safe(work, next, &works, list) {
8359                 list_del_init(&work->list);
8360                 btrfs_wait_and_free_delalloc_work(work);
8361         }
8362
8363         if (!list_empty_careful(&splice)) {
8364                 spin_lock(&root->fs_info->delalloc_lock);
8365                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8366                 spin_unlock(&root->fs_info->delalloc_lock);
8367         }
8368         return ret;
8369 }
8370
8371 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8372                          const char *symname)
8373 {
8374         struct btrfs_trans_handle *trans;
8375         struct btrfs_root *root = BTRFS_I(dir)->root;
8376         struct btrfs_path *path;
8377         struct btrfs_key key;
8378         struct inode *inode = NULL;
8379         int err;
8380         int drop_inode = 0;
8381         u64 objectid;
8382         u64 index = 0 ;
8383         int name_len;
8384         int datasize;
8385         unsigned long ptr;
8386         struct btrfs_file_extent_item *ei;
8387         struct extent_buffer *leaf;
8388
8389         name_len = strlen(symname) + 1;
8390         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8391                 return -ENAMETOOLONG;
8392
8393         /*
8394          * 2 items for inode item and ref
8395          * 2 items for dir items
8396          * 1 item for xattr if selinux is on
8397          */
8398         trans = btrfs_start_transaction(root, 5);
8399         if (IS_ERR(trans))
8400                 return PTR_ERR(trans);
8401
8402         err = btrfs_find_free_ino(root, &objectid);
8403         if (err)
8404                 goto out_unlock;
8405
8406         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8407                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8408                                 S_IFLNK|S_IRWXUGO, &index);
8409         if (IS_ERR(inode)) {
8410                 err = PTR_ERR(inode);
8411                 goto out_unlock;
8412         }
8413
8414         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8415         if (err) {
8416                 drop_inode = 1;
8417                 goto out_unlock;
8418         }
8419
8420         /*
8421         * If the active LSM wants to access the inode during
8422         * d_instantiate it needs these. Smack checks to see
8423         * if the filesystem supports xattrs by looking at the
8424         * ops vector.
8425         */
8426         inode->i_fop = &btrfs_file_operations;
8427         inode->i_op = &btrfs_file_inode_operations;
8428
8429         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8430         if (err)
8431                 drop_inode = 1;
8432         else {
8433                 inode->i_mapping->a_ops = &btrfs_aops;
8434                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8435                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8436         }
8437         if (drop_inode)
8438                 goto out_unlock;
8439
8440         path = btrfs_alloc_path();
8441         if (!path) {
8442                 err = -ENOMEM;
8443                 drop_inode = 1;
8444                 goto out_unlock;
8445         }
8446         key.objectid = btrfs_ino(inode);
8447         key.offset = 0;
8448         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8449         datasize = btrfs_file_extent_calc_inline_size(name_len);
8450         err = btrfs_insert_empty_item(trans, root, path, &key,
8451                                       datasize);
8452         if (err) {
8453                 drop_inode = 1;
8454                 btrfs_free_path(path);
8455                 goto out_unlock;
8456         }
8457         leaf = path->nodes[0];
8458         ei = btrfs_item_ptr(leaf, path->slots[0],
8459                             struct btrfs_file_extent_item);
8460         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8461         btrfs_set_file_extent_type(leaf, ei,
8462                                    BTRFS_FILE_EXTENT_INLINE);
8463         btrfs_set_file_extent_encryption(leaf, ei, 0);
8464         btrfs_set_file_extent_compression(leaf, ei, 0);
8465         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8466         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8467
8468         ptr = btrfs_file_extent_inline_start(ei);
8469         write_extent_buffer(leaf, symname, ptr, name_len);
8470         btrfs_mark_buffer_dirty(leaf);
8471         btrfs_free_path(path);
8472
8473         inode->i_op = &btrfs_symlink_inode_operations;
8474         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8475         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8476         inode_set_bytes(inode, name_len);
8477         btrfs_i_size_write(inode, name_len - 1);
8478         err = btrfs_update_inode(trans, root, inode);
8479         if (err)
8480                 drop_inode = 1;
8481
8482 out_unlock:
8483         if (!err)
8484                 d_instantiate(dentry, inode);
8485         btrfs_end_transaction(trans, root);
8486         if (drop_inode) {
8487                 inode_dec_link_count(inode);
8488                 iput(inode);
8489         }
8490         btrfs_btree_balance_dirty(root);
8491         return err;
8492 }
8493
8494 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8495                                        u64 start, u64 num_bytes, u64 min_size,
8496                                        loff_t actual_len, u64 *alloc_hint,
8497                                        struct btrfs_trans_handle *trans)
8498 {
8499         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8500         struct extent_map *em;
8501         struct btrfs_root *root = BTRFS_I(inode)->root;
8502         struct btrfs_key ins;
8503         u64 cur_offset = start;
8504         u64 i_size;
8505         u64 cur_bytes;
8506         int ret = 0;
8507         bool own_trans = true;
8508
8509         if (trans)
8510                 own_trans = false;
8511         while (num_bytes > 0) {
8512                 if (own_trans) {
8513                         trans = btrfs_start_transaction(root, 3);
8514                         if (IS_ERR(trans)) {
8515                                 ret = PTR_ERR(trans);
8516                                 break;
8517                         }
8518                 }
8519
8520                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8521                 cur_bytes = max(cur_bytes, min_size);
8522                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8523                                            min_size, 0, *alloc_hint, &ins, 1);
8524                 if (ret) {
8525                         if (own_trans)
8526                                 btrfs_end_transaction(trans, root);
8527                         break;
8528                 }
8529
8530                 ret = insert_reserved_file_extent(trans, inode,
8531                                                   cur_offset, ins.objectid,
8532                                                   ins.offset, ins.offset,
8533                                                   ins.offset, 0, 0, 0,
8534                                                   BTRFS_FILE_EXTENT_PREALLOC);
8535                 if (ret) {
8536                         btrfs_abort_transaction(trans, root, ret);
8537                         if (own_trans)
8538                                 btrfs_end_transaction(trans, root);
8539                         break;
8540                 }
8541                 btrfs_drop_extent_cache(inode, cur_offset,
8542                                         cur_offset + ins.offset -1, 0);
8543
8544                 em = alloc_extent_map();
8545                 if (!em) {
8546                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8547                                 &BTRFS_I(inode)->runtime_flags);
8548                         goto next;
8549                 }
8550
8551                 em->start = cur_offset;
8552                 em->orig_start = cur_offset;
8553                 em->len = ins.offset;
8554                 em->block_start = ins.objectid;
8555                 em->block_len = ins.offset;
8556                 em->orig_block_len = ins.offset;
8557                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8558                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8559                 em->generation = trans->transid;
8560
8561                 while (1) {
8562                         write_lock(&em_tree->lock);
8563                         ret = add_extent_mapping(em_tree, em);
8564                         if (!ret)
8565                                 list_move(&em->list,
8566                                           &em_tree->modified_extents);
8567                         write_unlock(&em_tree->lock);
8568                         if (ret != -EEXIST)
8569                                 break;
8570                         btrfs_drop_extent_cache(inode, cur_offset,
8571                                                 cur_offset + ins.offset - 1,
8572                                                 0);
8573                 }
8574                 free_extent_map(em);
8575 next:
8576                 num_bytes -= ins.offset;
8577                 cur_offset += ins.offset;
8578                 *alloc_hint = ins.objectid + ins.offset;
8579
8580                 inode_inc_iversion(inode);
8581                 inode->i_ctime = CURRENT_TIME;
8582                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8583                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8584                     (actual_len > inode->i_size) &&
8585                     (cur_offset > inode->i_size)) {
8586                         if (cur_offset > actual_len)
8587                                 i_size = actual_len;
8588                         else
8589                                 i_size = cur_offset;
8590                         i_size_write(inode, i_size);
8591                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8592                 }
8593
8594                 ret = btrfs_update_inode(trans, root, inode);
8595
8596                 if (ret) {
8597                         btrfs_abort_transaction(trans, root, ret);
8598                         if (own_trans)
8599                                 btrfs_end_transaction(trans, root);
8600                         break;
8601                 }
8602
8603                 if (own_trans)
8604                         btrfs_end_transaction(trans, root);
8605         }
8606         return ret;
8607 }
8608
8609 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8610                               u64 start, u64 num_bytes, u64 min_size,
8611                               loff_t actual_len, u64 *alloc_hint)
8612 {
8613         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8614                                            min_size, actual_len, alloc_hint,
8615                                            NULL);
8616 }
8617
8618 int btrfs_prealloc_file_range_trans(struct inode *inode,
8619                                     struct btrfs_trans_handle *trans, int mode,
8620                                     u64 start, u64 num_bytes, u64 min_size,
8621                                     loff_t actual_len, u64 *alloc_hint)
8622 {
8623         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8624                                            min_size, actual_len, alloc_hint, trans);
8625 }
8626
8627 static int btrfs_set_page_dirty(struct page *page)
8628 {
8629         return __set_page_dirty_nobuffers(page);
8630 }
8631
8632 static int btrfs_permission(struct inode *inode, int mask)
8633 {
8634         struct btrfs_root *root = BTRFS_I(inode)->root;
8635         umode_t mode = inode->i_mode;
8636
8637         if (mask & MAY_WRITE &&
8638             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8639                 if (btrfs_root_readonly(root))
8640                         return -EROFS;
8641                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8642                         return -EACCES;
8643         }
8644         return generic_permission(inode, mask);
8645 }
8646
8647 static const struct inode_operations btrfs_dir_inode_operations = {
8648         .getattr        = btrfs_getattr,
8649         .lookup         = btrfs_lookup,
8650         .create         = btrfs_create,
8651         .unlink         = btrfs_unlink,
8652         .link           = btrfs_link,
8653         .mkdir          = btrfs_mkdir,
8654         .rmdir          = btrfs_rmdir,
8655         .rename         = btrfs_rename,
8656         .symlink        = btrfs_symlink,
8657         .setattr        = btrfs_setattr,
8658         .mknod          = btrfs_mknod,
8659         .setxattr       = btrfs_setxattr,
8660         .getxattr       = btrfs_getxattr,
8661         .listxattr      = btrfs_listxattr,
8662         .removexattr    = btrfs_removexattr,
8663         .permission     = btrfs_permission,
8664         .get_acl        = btrfs_get_acl,
8665 };
8666 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8667         .lookup         = btrfs_lookup,
8668         .permission     = btrfs_permission,
8669         .get_acl        = btrfs_get_acl,
8670 };
8671
8672 static const struct file_operations btrfs_dir_file_operations = {
8673         .llseek         = generic_file_llseek,
8674         .read           = generic_read_dir,
8675         .readdir        = btrfs_real_readdir,
8676         .unlocked_ioctl = btrfs_ioctl,
8677 #ifdef CONFIG_COMPAT
8678         .compat_ioctl   = btrfs_ioctl,
8679 #endif
8680         .release        = btrfs_release_file,
8681         .fsync          = btrfs_sync_file,
8682 };
8683
8684 static struct extent_io_ops btrfs_extent_io_ops = {
8685         .fill_delalloc = run_delalloc_range,
8686         .submit_bio_hook = btrfs_submit_bio_hook,
8687         .merge_bio_hook = btrfs_merge_bio_hook,
8688         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8689         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8690         .writepage_start_hook = btrfs_writepage_start_hook,
8691         .set_bit_hook = btrfs_set_bit_hook,
8692         .clear_bit_hook = btrfs_clear_bit_hook,
8693         .merge_extent_hook = btrfs_merge_extent_hook,
8694         .split_extent_hook = btrfs_split_extent_hook,
8695 };
8696
8697 /*
8698  * btrfs doesn't support the bmap operation because swapfiles
8699  * use bmap to make a mapping of extents in the file.  They assume
8700  * these extents won't change over the life of the file and they
8701  * use the bmap result to do IO directly to the drive.
8702  *
8703  * the btrfs bmap call would return logical addresses that aren't
8704  * suitable for IO and they also will change frequently as COW
8705  * operations happen.  So, swapfile + btrfs == corruption.
8706  *
8707  * For now we're avoiding this by dropping bmap.
8708  */
8709 static const struct address_space_operations btrfs_aops = {
8710         .readpage       = btrfs_readpage,
8711         .writepage      = btrfs_writepage,
8712         .writepages     = btrfs_writepages,
8713         .readpages      = btrfs_readpages,
8714         .direct_IO      = btrfs_direct_IO,
8715         .invalidatepage = btrfs_invalidatepage,
8716         .releasepage    = btrfs_releasepage,
8717         .set_page_dirty = btrfs_set_page_dirty,
8718         .error_remove_page = generic_error_remove_page,
8719 };
8720
8721 static const struct address_space_operations btrfs_symlink_aops = {
8722         .readpage       = btrfs_readpage,
8723         .writepage      = btrfs_writepage,
8724         .invalidatepage = btrfs_invalidatepage,
8725         .releasepage    = btrfs_releasepage,
8726 };
8727
8728 static const struct inode_operations btrfs_file_inode_operations = {
8729         .getattr        = btrfs_getattr,
8730         .setattr        = btrfs_setattr,
8731         .setxattr       = btrfs_setxattr,
8732         .getxattr       = btrfs_getxattr,
8733         .listxattr      = btrfs_listxattr,
8734         .removexattr    = btrfs_removexattr,
8735         .permission     = btrfs_permission,
8736         .fiemap         = btrfs_fiemap,
8737         .get_acl        = btrfs_get_acl,
8738         .update_time    = btrfs_update_time,
8739 };
8740 static const struct inode_operations btrfs_special_inode_operations = {
8741         .getattr        = btrfs_getattr,
8742         .setattr        = btrfs_setattr,
8743         .permission     = btrfs_permission,
8744         .setxattr       = btrfs_setxattr,
8745         .getxattr       = btrfs_getxattr,
8746         .listxattr      = btrfs_listxattr,
8747         .removexattr    = btrfs_removexattr,
8748         .get_acl        = btrfs_get_acl,
8749         .update_time    = btrfs_update_time,
8750 };
8751 static const struct inode_operations btrfs_symlink_inode_operations = {
8752         .readlink       = generic_readlink,
8753         .follow_link    = page_follow_link_light,
8754         .put_link       = page_put_link,
8755         .getattr        = btrfs_getattr,
8756         .setattr        = btrfs_setattr,
8757         .permission     = btrfs_permission,
8758         .setxattr       = btrfs_setxattr,
8759         .getxattr       = btrfs_getxattr,
8760         .listxattr      = btrfs_listxattr,
8761         .removexattr    = btrfs_removexattr,
8762         .get_acl        = btrfs_get_acl,
8763         .update_time    = btrfs_update_time,
8764 };
8765
8766 const struct dentry_operations btrfs_dentry_operations = {
8767         .d_delete       = btrfs_dentry_delete,
8768         .d_release      = btrfs_dentry_release,
8769 };