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