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/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62
  63 struct btrfs_iget_args {
  64         struct btrfs_key *location;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end >= PAGE_CACHE_SIZE ||
 253             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 /*
 352  * we create compressed extents in two phases.  The first
 353  * phase compresses a range of pages that have already been
 354  * locked (both pages and state bits are locked).
 355  *
 356  * This is done inside an ordered work queue, and the compression
 357  * is spread across many cpus.  The actual IO submission is step
 358  * two, and the ordered work queue takes care of making sure that
 359  * happens in the same order things were put onto the queue by
 360  * writepages and friends.
 361  *
 362  * If this code finds it can't get good compression, it puts an
 363  * entry onto the work queue to write the uncompressed bytes.  This
 364  * makes sure that both compressed inodes and uncompressed inodes
 365  * are written in the same order that the flusher thread sent them
 366  * down.
 367  */
 368 static noinline int compress_file_range(struct inode *inode,
 369                                         struct page *locked_page,
 370                                         u64 start, u64 end,
 371                                         struct async_cow *async_cow,
 372                                         int *num_added)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375         u64 num_bytes;
 376         u64 blocksize = root->sectorsize;
 377         u64 actual_end;
 378         u64 isize = i_size_read(inode);
 379         int ret = 0;
 380         struct page **pages = NULL;
 381         unsigned long nr_pages;
 382         unsigned long nr_pages_ret = 0;
 383         unsigned long total_compressed = 0;
 384         unsigned long total_in = 0;
 385         unsigned long max_compressed = 128 * 1024;
 386         unsigned long max_uncompressed = 128 * 1024;
 387         int i;
 388         int will_compress;
 389         int compress_type = root->fs_info->compress_type;
 390         int redirty = 0;
 391
 392         /* if this is a small write inside eof, kick off a defrag */
 393         if ((end - start + 1) < 16 * 1024 &&
 394             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 395                 btrfs_add_inode_defrag(NULL, inode);
 396
 397         actual_end = min_t(u64, isize, end + 1);
 398 again:
 399         will_compress = 0;
 400         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 401         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 402
 403         /*
 404          * we don't want to send crud past the end of i_size through
 405          * compression, that's just a waste of CPU time.  So, if the
 406          * end of the file is before the start of our current
 407          * requested range of bytes, we bail out to the uncompressed
 408          * cleanup code that can deal with all of this.
 409          *
 410          * It isn't really the fastest way to fix things, but this is a
 411          * very uncommon corner.
 412          */
 413         if (actual_end <= start)
 414                 goto cleanup_and_bail_uncompressed;
 415
 416         total_compressed = actual_end - start;
 417
 418         /* we want to make sure that amount of ram required to uncompress
 419          * an extent is reasonable, so we limit the total size in ram
 420          * of a compressed extent to 128k.  This is a crucial number
 421          * because it also controls how easily we can spread reads across
 422          * cpus for decompression.
 423          *
 424          * We also want to make sure the amount of IO required to do
 425          * a random read is reasonably small, so we limit the size of
 426          * a compressed extent to 128k.
 427          */
 428         total_compressed = min(total_compressed, max_uncompressed);
 429         num_bytes = ALIGN(end - start + 1, blocksize);
 430         num_bytes = max(blocksize,  num_bytes);
 431         total_in = 0;
 432         ret = 0;
 433
 434         /*
 435          * we do compression for mount -o compress and when the
 436          * inode has not been flagged as nocompress.  This flag can
 437          * change at any time if we discover bad compression ratios.
 438          */
 439         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 440             (btrfs_test_opt(root, COMPRESS) ||
 441              (BTRFS_I(inode)->force_compress) ||
 442              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 443                 WARN_ON(pages);
 444                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 445                 if (!pages) {
 446                         /* just bail out to the uncompressed code */
 447                         goto cont;
 448                 }
 449
 450                 if (BTRFS_I(inode)->force_compress)
 451                         compress_type = BTRFS_I(inode)->force_compress;
 452
 453                 /*
 454                  * we need to call clear_page_dirty_for_io on each
 455                  * page in the range.  Otherwise applications with the file
 456                  * mmap'd can wander in and change the page contents while
 457                  * we are compressing them.
 458                  *
 459                  * If the compression fails for any reason, we set the pages
 460                  * dirty again later on.
 461                  */
 462                 extent_range_clear_dirty_for_io(inode, start, end);
 463                 redirty = 1;
 464                 ret = btrfs_compress_pages(compress_type,
 465                                            inode->i_mapping, start,
 466                                            total_compressed, pages,
 467                                            nr_pages, &nr_pages_ret,
 468                                            &total_in,
 469                                            &total_compressed,
 470                                            max_compressed);
 471
 472                 if (!ret) {
 473                         unsigned long offset = total_compressed &
 474                                 (PAGE_CACHE_SIZE - 1);
 475                         struct page *page = pages[nr_pages_ret - 1];
 476                         char *kaddr;
 477
 478                         /* zero the tail end of the last page, we might be
 479                          * sending it down to disk
 480                          */
 481                         if (offset) {
 482                                 kaddr = kmap_atomic(page);
 483                                 memset(kaddr + offset, 0,
 484                                        PAGE_CACHE_SIZE - offset);
 485                                 kunmap_atomic(kaddr);
 486                         }
 487                         will_compress = 1;
 488                 }
 489         }
 490 cont:
 491         if (start == 0) {
 492                 /* lets try to make an inline extent */
 493                 if (ret || total_in < (actual_end - start)) {
 494                         /* we didn't compress the entire range, try
 495                          * to make an uncompressed inline extent.
 496                          */
 497                         ret = cow_file_range_inline(root, inode, start, end,
 498                                                     0, 0, NULL);
 499                 } else {
 500                         /* try making a compressed inline extent */
 501                         ret = cow_file_range_inline(root, inode, start, end,
 502                                                     total_compressed,
 503                                                     compress_type, pages);
 504                 }
 505                 if (ret <= 0) {
 506                         unsigned long clear_flags = EXTENT_DELALLOC |
 507                                 EXTENT_DEFRAG;
 508                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 509
 510                         /*
 511                          * inline extent creation worked or returned error,
 512                          * we don't need to create any more async work items.
 513                          * Unlock and free up our temp pages.
 514                          */
 515                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 516                                                      clear_flags, PAGE_UNLOCK |
 517                                                      PAGE_CLEAR_DIRTY |
 518                                                      PAGE_SET_WRITEBACK |
 519                                                      PAGE_END_WRITEBACK);
 520                         goto free_pages_out;
 521                 }
 522         }
 523
 524         if (will_compress) {
 525                 /*
 526                  * we aren't doing an inline extent round the compressed size
 527                  * up to a block size boundary so the allocator does sane
 528                  * things
 529                  */
 530                 total_compressed = ALIGN(total_compressed, blocksize);
 531
 532                 /*
 533                  * one last check to make sure the compression is really a
 534                  * win, compare the page count read with the blocks on disk
 535                  */
 536                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 537                 if (total_compressed >= total_in) {
 538                         will_compress = 0;
 539                 } else {
 540                         num_bytes = total_in;
 541                 }
 542         }
 543         if (!will_compress && pages) {
 544                 /*
 545                  * the compression code ran but failed to make things smaller,
 546                  * free any pages it allocated and our page pointer array
 547                  */
 548                 for (i = 0; i < nr_pages_ret; i++) {
 549                         WARN_ON(pages[i]->mapping);
 550                         page_cache_release(pages[i]);
 551                 }
 552                 kfree(pages);
 553                 pages = NULL;
 554                 total_compressed = 0;
 555                 nr_pages_ret = 0;
 556
 557                 /* flag the file so we don't compress in the future */
 558                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 559                     !(BTRFS_I(inode)->force_compress)) {
 560                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 561                 }
 562         }
 563         if (will_compress) {
 564                 *num_added += 1;
 565
 566                 /* the async work queues will take care of doing actual
 567                  * allocation on disk for these compressed pages,
 568                  * and will submit them to the elevator.
 569                  */
 570                 add_async_extent(async_cow, start, num_bytes,
 571                                  total_compressed, pages, nr_pages_ret,
 572                                  compress_type);
 573
 574                 if (start + num_bytes < end) {
 575                         start += num_bytes;
 576                         pages = NULL;
 577                         cond_resched();
 578                         goto again;
 579                 }
 580         } else {
 581 cleanup_and_bail_uncompressed:
 582                 /*
 583                  * No compression, but we still need to write the pages in
 584                  * the file we've been given so far.  redirty the locked
 585                  * page if it corresponds to our extent and set things up
 586                  * for the async work queue to run cow_file_range to do
 587                  * the normal delalloc dance
 588                  */
 589                 if (page_offset(locked_page) >= start &&
 590                     page_offset(locked_page) <= end) {
 591                         __set_page_dirty_nobuffers(locked_page);
 592                         /* unlocked later on in the async handlers */
 593                 }
 594                 if (redirty)
 595                         extent_range_redirty_for_io(inode, start, end);
 596                 add_async_extent(async_cow, start, end - start + 1,
 597                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 598                 *num_added += 1;
 599         }
 600
 601 out:
 602         return ret;
 603
 604 free_pages_out:
 605         for (i = 0; i < nr_pages_ret; i++) {
 606                 WARN_ON(pages[i]->mapping);
 607                 page_cache_release(pages[i]);
 608         }
 609         kfree(pages);
 610
 611         goto out;
 612 }
 613
 614 /*
 615  * phase two of compressed writeback.  This is the ordered portion
 616  * of the code, which only gets called in the order the work was
 617  * queued.  We walk all the async extents created by compress_file_range
 618  * and send them down to the disk.
 619  */
 620 static noinline int submit_compressed_extents(struct inode *inode,
 621                                               struct async_cow *async_cow)
 622 {
 623         struct async_extent *async_extent;
 624         u64 alloc_hint = 0;
 625         struct btrfs_key ins;
 626         struct extent_map *em;
 627         struct btrfs_root *root = BTRFS_I(inode)->root;
 628         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 629         struct extent_io_tree *io_tree;
 630         int ret = 0;
 631
 632         if (list_empty(&async_cow->extents))
 633                 return 0;
 634
 635 again:
 636         while (!list_empty(&async_cow->extents)) {
 637                 async_extent = list_entry(async_cow->extents.next,
 638                                           struct async_extent, list);
 639                 list_del(&async_extent->list);
 640
 641                 io_tree = &BTRFS_I(inode)->io_tree;
 642
 643 retry:
 644                 /* did the compression code fall back to uncompressed IO? */
 645                 if (!async_extent->pages) {
 646                         int page_started = 0;
 647                         unsigned long nr_written = 0;
 648
 649                         lock_extent(io_tree, async_extent->start,
 650                                          async_extent->start +
 651                                          async_extent->ram_size - 1);
 652
 653                         /* allocate blocks */
 654                         ret = cow_file_range(inode, async_cow->locked_page,
 655                                              async_extent->start,
 656                                              async_extent->start +
 657                                              async_extent->ram_size - 1,
 658                                              &page_started, &nr_written, 0);
 659
 660                         /* JDM XXX */
 661
 662                         /*
 663                          * if page_started, cow_file_range inserted an
 664                          * inline extent and took care of all the unlocking
 665                          * and IO for us.  Otherwise, we need to submit
 666                          * all those pages down to the drive.
 667                          */
 668                         if (!page_started && !ret)
 669                                 extent_write_locked_range(io_tree,
 670                                                   inode, async_extent->start,
 671                                                   async_extent->start +
 672                                                   async_extent->ram_size - 1,
 673                                                   btrfs_get_extent,
 674                                                   WB_SYNC_ALL);
 675                         else if (ret)
 676                                 unlock_page(async_cow->locked_page);
 677                         kfree(async_extent);
 678                         cond_resched();
 679                         continue;
 680                 }
 681
 682                 lock_extent(io_tree, async_extent->start,
 683                             async_extent->start + async_extent->ram_size - 1);
 684
 685                 ret = btrfs_reserve_extent(root,
 686                                            async_extent->compressed_size,
 687                                            async_extent->compressed_size,
 688                                            0, alloc_hint, &ins, 1);
 689                 if (ret) {
 690                         int i;
 691
 692                         for (i = 0; i < async_extent->nr_pages; i++) {
 693                                 WARN_ON(async_extent->pages[i]->mapping);
 694                                 page_cache_release(async_extent->pages[i]);
 695                         }
 696                         kfree(async_extent->pages);
 697                         async_extent->nr_pages = 0;
 698                         async_extent->pages = NULL;
 699
 700                         if (ret == -ENOSPC) {
 701                                 unlock_extent(io_tree, async_extent->start,
 702                                               async_extent->start +
 703                                               async_extent->ram_size - 1);
 704                                 goto retry;
 705                         }
 706                         goto out_free;
 707                 }
 708
 709                 /*
 710                  * here we're doing allocation and writeback of the
 711                  * compressed pages
 712                  */
 713                 btrfs_drop_extent_cache(inode, async_extent->start,
 714                                         async_extent->start +
 715                                         async_extent->ram_size - 1, 0);
 716
 717                 em = alloc_extent_map();
 718                 if (!em) {
 719                         ret = -ENOMEM;
 720                         goto out_free_reserve;
 721                 }
 722                 em->start = async_extent->start;
 723                 em->len = async_extent->ram_size;
 724                 em->orig_start = em->start;
 725                 em->mod_start = em->start;
 726                 em->mod_len = em->len;
 727
 728                 em->block_start = ins.objectid;
 729                 em->block_len = ins.offset;
 730                 em->orig_block_len = ins.offset;
 731                 em->ram_bytes = async_extent->ram_size;
 732                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 733                 em->compress_type = async_extent->compress_type;
 734                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 735                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 736                 em->generation = -1;
 737
 738                 while (1) {
 739                         write_lock(&em_tree->lock);
 740                         ret = add_extent_mapping(em_tree, em, 1);
 741                         write_unlock(&em_tree->lock);
 742                         if (ret != -EEXIST) {
 743                                 free_extent_map(em);
 744                                 break;
 745                         }
 746                         btrfs_drop_extent_cache(inode, async_extent->start,
 747                                                 async_extent->start +
 748                                                 async_extent->ram_size - 1, 0);
 749                 }
 750
 751                 if (ret)
 752                         goto out_free_reserve;
 753
 754                 ret = btrfs_add_ordered_extent_compress(inode,
 755                                                 async_extent->start,
 756                                                 ins.objectid,
 757                                                 async_extent->ram_size,
 758                                                 ins.offset,
 759                                                 BTRFS_ORDERED_COMPRESSED,
 760                                                 async_extent->compress_type);
 761                 if (ret)
 762                         goto out_free_reserve;
 763
 764                 /*
 765                  * clear dirty, set writeback and unlock the pages.
 766                  */
 767                 extent_clear_unlock_delalloc(inode, async_extent->start,
 768                                 async_extent->start +
 769                                 async_extent->ram_size - 1,
 770                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 771                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 772                                 PAGE_SET_WRITEBACK);
 773                 ret = btrfs_submit_compressed_write(inode,
 774                                     async_extent->start,
 775                                     async_extent->ram_size,
 776                                     ins.objectid,
 777                                     ins.offset, async_extent->pages,
 778                                     async_extent->nr_pages);
 779                 alloc_hint = ins.objectid + ins.offset;
 780                 kfree(async_extent);
 781                 if (ret)
 782                         goto out;
 783                 cond_resched();
 784         }
 785         ret = 0;
 786 out:
 787         return ret;
 788 out_free_reserve:
 789         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 790 out_free:
 791         extent_clear_unlock_delalloc(inode, async_extent->start,
 792                                      async_extent->start +
 793                                      async_extent->ram_size - 1,
 794                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 795                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 796                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 797                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 798         kfree(async_extent);
 799         goto again;
 800 }
 801
 802 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 803                                       u64 num_bytes)
 804 {
 805         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 806         struct extent_map *em;
 807         u64 alloc_hint = 0;
 808
 809         read_lock(&em_tree->lock);
 810         em = search_extent_mapping(em_tree, start, num_bytes);
 811         if (em) {
 812                 /*
 813                  * if block start isn't an actual block number then find the
 814                  * first block in this inode and use that as a hint.  If that
 815                  * block is also bogus then just don't worry about it.
 816                  */
 817                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 818                         free_extent_map(em);
 819                         em = search_extent_mapping(em_tree, 0, 0);
 820                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 821                                 alloc_hint = em->block_start;
 822                         if (em)
 823                                 free_extent_map(em);
 824                 } else {
 825                         alloc_hint = em->block_start;
 826                         free_extent_map(em);
 827                 }
 828         }
 829         read_unlock(&em_tree->lock);
 830
 831         return alloc_hint;
 832 }
 833
 834 /*
 835  * when extent_io.c finds a delayed allocation range in the file,
 836  * the call backs end up in this code.  The basic idea is to
 837  * allocate extents on disk for the range, and create ordered data structs
 838  * in ram to track those extents.
 839  *
 840  * locked_page is the page that writepage had locked already.  We use
 841  * it to make sure we don't do extra locks or unlocks.
 842  *
 843  * *page_started is set to one if we unlock locked_page and do everything
 844  * required to start IO on it.  It may be clean and already done with
 845  * IO when we return.
 846  */
 847 static noinline int cow_file_range(struct inode *inode,
 848                                    struct page *locked_page,
 849                                    u64 start, u64 end, int *page_started,
 850                                    unsigned long *nr_written,
 851                                    int unlock)
 852 {
 853         struct btrfs_root *root = BTRFS_I(inode)->root;
 854         u64 alloc_hint = 0;
 855         u64 num_bytes;
 856         unsigned long ram_size;
 857         u64 disk_num_bytes;
 858         u64 cur_alloc_size;
 859         u64 blocksize = root->sectorsize;
 860         struct btrfs_key ins;
 861         struct extent_map *em;
 862         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 863         int ret = 0;
 864
 865         if (btrfs_is_free_space_inode(inode)) {
 866                 WARN_ON_ONCE(1);
 867                 ret = -EINVAL;
 868                 goto out_unlock;
 869         }
 870
 871         num_bytes = ALIGN(end - start + 1, blocksize);
 872         num_bytes = max(blocksize,  num_bytes);
 873         disk_num_bytes = num_bytes;
 874
 875         /* if this is a small write inside eof, kick off defrag */
 876         if (num_bytes < 64 * 1024 &&
 877             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 878                 btrfs_add_inode_defrag(NULL, inode);
 879
 880         if (start == 0) {
 881                 /* lets try to make an inline extent */
 882                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 883                                             NULL);
 884                 if (ret == 0) {
 885                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 886                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 887                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 888                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 889                                      PAGE_END_WRITEBACK);
 890
 891                         *nr_written = *nr_written +
 892                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 893                         *page_started = 1;
 894                         goto out;
 895                 } else if (ret < 0) {
 896                         goto out_unlock;
 897                 }
 898         }
 899
 900         BUG_ON(disk_num_bytes >
 901                btrfs_super_total_bytes(root->fs_info->super_copy));
 902
 903         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 904         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 905
 906         while (disk_num_bytes > 0) {
 907                 unsigned long op;
 908
 909                 cur_alloc_size = disk_num_bytes;
 910                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 911                                            root->sectorsize, 0, alloc_hint,
 912                                            &ins, 1);
 913                 if (ret < 0)
 914                         goto out_unlock;
 915
 916                 em = alloc_extent_map();
 917                 if (!em) {
 918                         ret = -ENOMEM;
 919                         goto out_reserve;
 920                 }
 921                 em->start = start;
 922                 em->orig_start = em->start;
 923                 ram_size = ins.offset;
 924                 em->len = ins.offset;
 925                 em->mod_start = em->start;
 926                 em->mod_len = em->len;
 927
 928                 em->block_start = ins.objectid;
 929                 em->block_len = ins.offset;
 930                 em->orig_block_len = ins.offset;
 931                 em->ram_bytes = ram_size;
 932                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 933                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 934                 em->generation = -1;
 935
 936                 while (1) {
 937                         write_lock(&em_tree->lock);
 938                         ret = add_extent_mapping(em_tree, em, 1);
 939                         write_unlock(&em_tree->lock);
 940                         if (ret != -EEXIST) {
 941                                 free_extent_map(em);
 942                                 break;
 943                         }
 944                         btrfs_drop_extent_cache(inode, start,
 945                                                 start + ram_size - 1, 0);
 946                 }
 947                 if (ret)
 948                         goto out_reserve;
 949
 950                 cur_alloc_size = ins.offset;
 951                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 952                                                ram_size, cur_alloc_size, 0);
 953                 if (ret)
 954                         goto out_reserve;
 955
 956                 if (root->root_key.objectid ==
 957                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 958                         ret = btrfs_reloc_clone_csums(inode, start,
 959                                                       cur_alloc_size);
 960                         if (ret)
 961                                 goto out_reserve;
 962                 }
 963
 964                 if (disk_num_bytes < cur_alloc_size)
 965                         break;
 966
 967                 /* we're not doing compressed IO, don't unlock the first
 968                  * page (which the caller expects to stay locked), don't
 969                  * clear any dirty bits and don't set any writeback bits
 970                  *
 971                  * Do set the Private2 bit so we know this page was properly
 972                  * setup for writepage
 973                  */
 974                 op = unlock ? PAGE_UNLOCK : 0;
 975                 op |= PAGE_SET_PRIVATE2;
 976
 977                 extent_clear_unlock_delalloc(inode, start,
 978                                              start + ram_size - 1, locked_page,
 979                                              EXTENT_LOCKED | EXTENT_DELALLOC,
 980                                              op);
 981                 disk_num_bytes -= cur_alloc_size;
 982                 num_bytes -= cur_alloc_size;
 983                 alloc_hint = ins.objectid + ins.offset;
 984                 start += cur_alloc_size;
 985         }
 986 out:
 987         return ret;
 988
 989 out_reserve:
 990         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 991 out_unlock:
 992         extent_clear_unlock_delalloc(inode, start, end, locked_page,
 993                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
 994                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
 995                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 996                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 997         goto out;
 998 }
 999
1000 /*
1001  * work queue call back to started compression on a file and pages
1002  */
1003 static noinline void async_cow_start(struct btrfs_work *work)
1004 {
1005         struct async_cow *async_cow;
1006         int num_added = 0;
1007         async_cow = container_of(work, struct async_cow, work);
1008
1009         compress_file_range(async_cow->inode, async_cow->locked_page,
1010                             async_cow->start, async_cow->end, async_cow,
1011                             &num_added);
1012         if (num_added == 0) {
1013                 btrfs_add_delayed_iput(async_cow->inode);
1014                 async_cow->inode = NULL;
1015         }
1016 }
1017
1018 /*
1019  * work queue call back to submit previously compressed pages
1020  */
1021 static noinline void async_cow_submit(struct btrfs_work *work)
1022 {
1023         struct async_cow *async_cow;
1024         struct btrfs_root *root;
1025         unsigned long nr_pages;
1026
1027         async_cow = container_of(work, struct async_cow, work);
1028
1029         root = async_cow->root;
1030         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1031                 PAGE_CACHE_SHIFT;
1032
1033         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1034             5 * 1024 * 1024 &&
1035             waitqueue_active(&root->fs_info->async_submit_wait))
1036                 wake_up(&root->fs_info->async_submit_wait);
1037
1038         if (async_cow->inode)
1039                 submit_compressed_extents(async_cow->inode, async_cow);
1040 }
1041
1042 static noinline void async_cow_free(struct btrfs_work *work)
1043 {
1044         struct async_cow *async_cow;
1045         async_cow = container_of(work, struct async_cow, work);
1046         if (async_cow->inode)
1047                 btrfs_add_delayed_iput(async_cow->inode);
1048         kfree(async_cow);
1049 }
1050
1051 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1052                                 u64 start, u64 end, int *page_started,
1053                                 unsigned long *nr_written)
1054 {
1055         struct async_cow *async_cow;
1056         struct btrfs_root *root = BTRFS_I(inode)->root;
1057         unsigned long nr_pages;
1058         u64 cur_end;
1059         int limit = 10 * 1024 * 1024;
1060
1061         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1062                          1, 0, NULL, GFP_NOFS);
1063         while (start < end) {
1064                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1065                 BUG_ON(!async_cow); /* -ENOMEM */
1066                 async_cow->inode = igrab(inode);
1067                 async_cow->root = root;
1068                 async_cow->locked_page = locked_page;
1069                 async_cow->start = start;
1070
1071                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1072                         cur_end = end;
1073                 else
1074                         cur_end = min(end, start + 512 * 1024 - 1);
1075
1076                 async_cow->end = cur_end;
1077                 INIT_LIST_HEAD(&async_cow->extents);
1078
1079                 async_cow->work.func = async_cow_start;
1080                 async_cow->work.ordered_func = async_cow_submit;
1081                 async_cow->work.ordered_free = async_cow_free;
1082                 async_cow->work.flags = 0;
1083
1084                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1085                         PAGE_CACHE_SHIFT;
1086                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1087
1088                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1089                                    &async_cow->work);
1090
1091                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1092                         wait_event(root->fs_info->async_submit_wait,
1093                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1094                             limit));
1095                 }
1096
1097                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1098                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1099                         wait_event(root->fs_info->async_submit_wait,
1100                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1101                            0));
1102                 }
1103
1104                 *nr_written += nr_pages;
1105                 start = cur_end + 1;
1106         }
1107         *page_started = 1;
1108         return 0;
1109 }
1110
1111 static noinline int csum_exist_in_range(struct btrfs_root *root,
1112                                         u64 bytenr, u64 num_bytes)
1113 {
1114         int ret;
1115         struct btrfs_ordered_sum *sums;
1116         LIST_HEAD(list);
1117
1118         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1119                                        bytenr + num_bytes - 1, &list, 0);
1120         if (ret == 0 && list_empty(&list))
1121                 return 0;
1122
1123         while (!list_empty(&list)) {
1124                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1125                 list_del(&sums->list);
1126                 kfree(sums);
1127         }
1128         return 1;
1129 }
1130
1131 /*
1132  * when nowcow writeback call back.  This checks for snapshots or COW copies
1133  * of the extents that exist in the file, and COWs the file as required.
1134  *
1135  * If no cow copies or snapshots exist, we write directly to the existing
1136  * blocks on disk
1137  */
1138 static noinline int run_delalloc_nocow(struct inode *inode,
1139                                        struct page *locked_page,
1140                               u64 start, u64 end, int *page_started, int force,
1141                               unsigned long *nr_written)
1142 {
1143         struct btrfs_root *root = BTRFS_I(inode)->root;
1144         struct btrfs_trans_handle *trans;
1145         struct extent_buffer *leaf;
1146         struct btrfs_path *path;
1147         struct btrfs_file_extent_item *fi;
1148         struct btrfs_key found_key;
1149         u64 cow_start;
1150         u64 cur_offset;
1151         u64 extent_end;
1152         u64 extent_offset;
1153         u64 disk_bytenr;
1154         u64 num_bytes;
1155         u64 disk_num_bytes;
1156         u64 ram_bytes;
1157         int extent_type;
1158         int ret, err;
1159         int type;
1160         int nocow;
1161         int check_prev = 1;
1162         bool nolock;
1163         u64 ino = btrfs_ino(inode);
1164
1165         path = btrfs_alloc_path();
1166         if (!path) {
1167                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1168                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1169                                              EXTENT_DO_ACCOUNTING |
1170                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1171                                              PAGE_CLEAR_DIRTY |
1172                                              PAGE_SET_WRITEBACK |
1173                                              PAGE_END_WRITEBACK);
1174                 return -ENOMEM;
1175         }
1176
1177         nolock = btrfs_is_free_space_inode(inode);
1178
1179         if (nolock)
1180                 trans = btrfs_join_transaction_nolock(root);
1181         else
1182                 trans = btrfs_join_transaction(root);
1183
1184         if (IS_ERR(trans)) {
1185                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1186                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1187                                              EXTENT_DO_ACCOUNTING |
1188                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1189                                              PAGE_CLEAR_DIRTY |
1190                                              PAGE_SET_WRITEBACK |
1191                                              PAGE_END_WRITEBACK);
1192                 btrfs_free_path(path);
1193                 return PTR_ERR(trans);
1194         }
1195
1196         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1197
1198         cow_start = (u64)-1;
1199         cur_offset = start;
1200         while (1) {
1201                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1202                                                cur_offset, 0);
1203                 if (ret < 0)
1204                         goto error;
1205                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1206                         leaf = path->nodes[0];
1207                         btrfs_item_key_to_cpu(leaf, &found_key,
1208                                               path->slots[0] - 1);
1209                         if (found_key.objectid == ino &&
1210                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1211                                 path->slots[0]--;
1212                 }
1213                 check_prev = 0;
1214 next_slot:
1215                 leaf = path->nodes[0];
1216                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1217                         ret = btrfs_next_leaf(root, path);
1218                         if (ret < 0)
1219                                 goto error;
1220                         if (ret > 0)
1221                                 break;
1222                         leaf = path->nodes[0];
1223                 }
1224
1225                 nocow = 0;
1226                 disk_bytenr = 0;
1227                 num_bytes = 0;
1228                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1229
1230                 if (found_key.objectid > ino ||
1231                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1232                     found_key.offset > end)
1233                         break;
1234
1235                 if (found_key.offset > cur_offset) {
1236                         extent_end = found_key.offset;
1237                         extent_type = 0;
1238                         goto out_check;
1239                 }
1240
1241                 fi = btrfs_item_ptr(leaf, path->slots[0],
1242                                     struct btrfs_file_extent_item);
1243                 extent_type = btrfs_file_extent_type(leaf, fi);
1244
1245                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1246                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1247                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1248                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1249                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1250                         extent_end = found_key.offset +
1251                                 btrfs_file_extent_num_bytes(leaf, fi);
1252                         disk_num_bytes =
1253                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1254                         if (extent_end <= start) {
1255                                 path->slots[0]++;
1256                                 goto next_slot;
1257                         }
1258                         if (disk_bytenr == 0)
1259                                 goto out_check;
1260                         if (btrfs_file_extent_compression(leaf, fi) ||
1261                             btrfs_file_extent_encryption(leaf, fi) ||
1262                             btrfs_file_extent_other_encoding(leaf, fi))
1263                                 goto out_check;
1264                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1265                                 goto out_check;
1266                         if (btrfs_extent_readonly(root, disk_bytenr))
1267                                 goto out_check;
1268                         if (btrfs_cross_ref_exist(trans, root, ino,
1269                                                   found_key.offset -
1270                                                   extent_offset, disk_bytenr))
1271                                 goto out_check;
1272                         disk_bytenr += extent_offset;
1273                         disk_bytenr += cur_offset - found_key.offset;
1274                         num_bytes = min(end + 1, extent_end) - cur_offset;
1275                         /*
1276                          * force cow if csum exists in the range.
1277                          * this ensure that csum for a given extent are
1278                          * either valid or do not exist.
1279                          */
1280                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1281                                 goto out_check;
1282                         nocow = 1;
1283                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1284                         extent_end = found_key.offset +
1285                                 btrfs_file_extent_inline_len(leaf,
1286                                                      path->slots[0], fi);
1287                         extent_end = ALIGN(extent_end, root->sectorsize);
1288                 } else {
1289                         BUG_ON(1);
1290                 }
1291 out_check:
1292                 if (extent_end <= start) {
1293                         path->slots[0]++;
1294                         goto next_slot;
1295                 }
1296                 if (!nocow) {
1297                         if (cow_start == (u64)-1)
1298                                 cow_start = cur_offset;
1299                         cur_offset = extent_end;
1300                         if (cur_offset > end)
1301                                 break;
1302                         path->slots[0]++;
1303                         goto next_slot;
1304                 }
1305
1306                 btrfs_release_path(path);
1307                 if (cow_start != (u64)-1) {
1308                         ret = cow_file_range(inode, locked_page,
1309                                              cow_start, found_key.offset - 1,
1310                                              page_started, nr_written, 1);
1311                         if (ret)
1312                                 goto error;
1313                         cow_start = (u64)-1;
1314                 }
1315
1316                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1317                         struct extent_map *em;
1318                         struct extent_map_tree *em_tree;
1319                         em_tree = &BTRFS_I(inode)->extent_tree;
1320                         em = alloc_extent_map();
1321                         BUG_ON(!em); /* -ENOMEM */
1322                         em->start = cur_offset;
1323                         em->orig_start = found_key.offset - extent_offset;
1324                         em->len = num_bytes;
1325                         em->block_len = num_bytes;
1326                         em->block_start = disk_bytenr;
1327                         em->orig_block_len = disk_num_bytes;
1328                         em->ram_bytes = ram_bytes;
1329                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1330                         em->mod_start = em->start;
1331                         em->mod_len = em->len;
1332                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1333                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1334                         em->generation = -1;
1335                         while (1) {
1336                                 write_lock(&em_tree->lock);
1337                                 ret = add_extent_mapping(em_tree, em, 1);
1338                                 write_unlock(&em_tree->lock);
1339                                 if (ret != -EEXIST) {
1340                                         free_extent_map(em);
1341                                         break;
1342                                 }
1343                                 btrfs_drop_extent_cache(inode, em->start,
1344                                                 em->start + em->len - 1, 0);
1345                         }
1346                         type = BTRFS_ORDERED_PREALLOC;
1347                 } else {
1348                         type = BTRFS_ORDERED_NOCOW;
1349                 }
1350
1351                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1352                                                num_bytes, num_bytes, type);
1353                 BUG_ON(ret); /* -ENOMEM */
1354
1355                 if (root->root_key.objectid ==
1356                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1357                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1358                                                       num_bytes);
1359                         if (ret)
1360                                 goto error;
1361                 }
1362
1363                 extent_clear_unlock_delalloc(inode, cur_offset,
1364                                              cur_offset + num_bytes - 1,
1365                                              locked_page, EXTENT_LOCKED |
1366                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1367                                              PAGE_SET_PRIVATE2);
1368                 cur_offset = extent_end;
1369                 if (cur_offset > end)
1370                         break;
1371         }
1372         btrfs_release_path(path);
1373
1374         if (cur_offset <= end && cow_start == (u64)-1) {
1375                 cow_start = cur_offset;
1376                 cur_offset = end;
1377         }
1378
1379         if (cow_start != (u64)-1) {
1380                 ret = cow_file_range(inode, locked_page, cow_start, end,
1381                                      page_started, nr_written, 1);
1382                 if (ret)
1383                         goto error;
1384         }
1385
1386 error:
1387         err = btrfs_end_transaction(trans, root);
1388         if (!ret)
1389                 ret = err;
1390
1391         if (ret && cur_offset < end)
1392                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1393                                              locked_page, EXTENT_LOCKED |
1394                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1395                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1396                                              PAGE_CLEAR_DIRTY |
1397                                              PAGE_SET_WRITEBACK |
1398                                              PAGE_END_WRITEBACK);
1399         btrfs_free_path(path);
1400         return ret;
1401 }
1402
1403 /*
1404  * extent_io.c call back to do delayed allocation processing
1405  */
1406 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1407                               u64 start, u64 end, int *page_started,
1408                               unsigned long *nr_written)
1409 {
1410         int ret;
1411         struct btrfs_root *root = BTRFS_I(inode)->root;
1412
1413         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1414                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1415                                          page_started, 1, nr_written);
1416         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1417                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1418                                          page_started, 0, nr_written);
1419         } else if (!btrfs_test_opt(root, COMPRESS) &&
1420                    !(BTRFS_I(inode)->force_compress) &&
1421                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1422                 ret = cow_file_range(inode, locked_page, start, end,
1423                                       page_started, nr_written, 1);
1424         } else {
1425                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1426                         &BTRFS_I(inode)->runtime_flags);
1427                 ret = cow_file_range_async(inode, locked_page, start, end,
1428                                            page_started, nr_written);
1429         }
1430         return ret;
1431 }
1432
1433 static void btrfs_split_extent_hook(struct inode *inode,
1434                                     struct extent_state *orig, u64 split)
1435 {
1436         /* not delalloc, ignore it */
1437         if (!(orig->state & EXTENT_DELALLOC))
1438                 return;
1439
1440         spin_lock(&BTRFS_I(inode)->lock);
1441         BTRFS_I(inode)->outstanding_extents++;
1442         spin_unlock(&BTRFS_I(inode)->lock);
1443 }
1444
1445 /*
1446  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1447  * extents so we can keep track of new extents that are just merged onto old
1448  * extents, such as when we are doing sequential writes, so we can properly
1449  * account for the metadata space we'll need.
1450  */
1451 static void btrfs_merge_extent_hook(struct inode *inode,
1452                                     struct extent_state *new,
1453                                     struct extent_state *other)
1454 {
1455         /* not delalloc, ignore it */
1456         if (!(other->state & EXTENT_DELALLOC))
1457                 return;
1458
1459         spin_lock(&BTRFS_I(inode)->lock);
1460         BTRFS_I(inode)->outstanding_extents--;
1461         spin_unlock(&BTRFS_I(inode)->lock);
1462 }
1463
1464 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1465                                       struct inode *inode)
1466 {
1467         spin_lock(&root->delalloc_lock);
1468         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1470                               &root->delalloc_inodes);
1471                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1472                         &BTRFS_I(inode)->runtime_flags);
1473                 root->nr_delalloc_inodes++;
1474                 if (root->nr_delalloc_inodes == 1) {
1475                         spin_lock(&root->fs_info->delalloc_root_lock);
1476                         BUG_ON(!list_empty(&root->delalloc_root));
1477                         list_add_tail(&root->delalloc_root,
1478                                       &root->fs_info->delalloc_roots);
1479                         spin_unlock(&root->fs_info->delalloc_root_lock);
1480                 }
1481         }
1482         spin_unlock(&root->delalloc_lock);
1483 }
1484
1485 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1486                                      struct inode *inode)
1487 {
1488         spin_lock(&root->delalloc_lock);
1489         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1490                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1491                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1492                           &BTRFS_I(inode)->runtime_flags);
1493                 root->nr_delalloc_inodes--;
1494                 if (!root->nr_delalloc_inodes) {
1495                         spin_lock(&root->fs_info->delalloc_root_lock);
1496                         BUG_ON(list_empty(&root->delalloc_root));
1497                         list_del_init(&root->delalloc_root);
1498                         spin_unlock(&root->fs_info->delalloc_root_lock);
1499                 }
1500         }
1501         spin_unlock(&root->delalloc_lock);
1502 }
1503
1504 /*
1505  * extent_io.c set_bit_hook, used to track delayed allocation
1506  * bytes in this file, and to maintain the list of inodes that
1507  * have pending delalloc work to be done.
1508  */
1509 static void btrfs_set_bit_hook(struct inode *inode,
1510                                struct extent_state *state, unsigned long *bits)
1511 {
1512
1513         /*
1514          * set_bit and clear bit hooks normally require _irqsave/restore
1515          * but in this case, we are only testing for the DELALLOC
1516          * bit, which is only set or cleared with irqs on
1517          */
1518         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1519                 struct btrfs_root *root = BTRFS_I(inode)->root;
1520                 u64 len = state->end + 1 - state->start;
1521                 bool do_list = !btrfs_is_free_space_inode(inode);
1522
1523                 if (*bits & EXTENT_FIRST_DELALLOC) {
1524                         *bits &= ~EXTENT_FIRST_DELALLOC;
1525                 } else {
1526                         spin_lock(&BTRFS_I(inode)->lock);
1527                         BTRFS_I(inode)->outstanding_extents++;
1528                         spin_unlock(&BTRFS_I(inode)->lock);
1529                 }
1530
1531                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1532                                      root->fs_info->delalloc_batch);
1533                 spin_lock(&BTRFS_I(inode)->lock);
1534                 BTRFS_I(inode)->delalloc_bytes += len;
1535                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1536                                          &BTRFS_I(inode)->runtime_flags))
1537                         btrfs_add_delalloc_inodes(root, inode);
1538                 spin_unlock(&BTRFS_I(inode)->lock);
1539         }
1540 }
1541
1542 /*
1543  * extent_io.c clear_bit_hook, see set_bit_hook for why
1544  */
1545 static void btrfs_clear_bit_hook(struct inode *inode,
1546                                  struct extent_state *state,
1547                                  unsigned long *bits)
1548 {
1549         /*
1550          * set_bit and clear bit hooks normally require _irqsave/restore
1551          * but in this case, we are only testing for the DELALLOC
1552          * bit, which is only set or cleared with irqs on
1553          */
1554         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1555                 struct btrfs_root *root = BTRFS_I(inode)->root;
1556                 u64 len = state->end + 1 - state->start;
1557                 bool do_list = !btrfs_is_free_space_inode(inode);
1558
1559                 if (*bits & EXTENT_FIRST_DELALLOC) {
1560                         *bits &= ~EXTENT_FIRST_DELALLOC;
1561                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1562                         spin_lock(&BTRFS_I(inode)->lock);
1563                         BTRFS_I(inode)->outstanding_extents--;
1564                         spin_unlock(&BTRFS_I(inode)->lock);
1565                 }
1566
1567                 /*
1568                  * We don't reserve metadata space for space cache inodes so we
1569                  * don't need to call dellalloc_release_metadata if there is an
1570                  * error.
1571                  */
1572                 if (*bits & EXTENT_DO_ACCOUNTING &&
1573                     root != root->fs_info->tree_root)
1574                         btrfs_delalloc_release_metadata(inode, len);
1575
1576                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1577                     && do_list && !(state->state & EXTENT_NORESERVE))
1578                         btrfs_free_reserved_data_space(inode, len);
1579
1580                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1581                                      root->fs_info->delalloc_batch);
1582                 spin_lock(&BTRFS_I(inode)->lock);
1583                 BTRFS_I(inode)->delalloc_bytes -= len;
1584                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1585                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1586                              &BTRFS_I(inode)->runtime_flags))
1587                         btrfs_del_delalloc_inode(root, inode);
1588                 spin_unlock(&BTRFS_I(inode)->lock);
1589         }
1590 }
1591
1592 /*
1593  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1594  * we don't create bios that span stripes or chunks
1595  */
1596 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1597                          size_t size, struct bio *bio,
1598                          unsigned long bio_flags)
1599 {
1600         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1601         u64 logical = (u64)bio->bi_sector << 9;
1602         u64 length = 0;
1603         u64 map_length;
1604         int ret;
1605
1606         if (bio_flags & EXTENT_BIO_COMPRESSED)
1607                 return 0;
1608
1609         length = bio->bi_size;
1610         map_length = length;
1611         ret = btrfs_map_block(root->fs_info, rw, logical,
1612                               &map_length, NULL, 0);
1613         /* Will always return 0 with map_multi == NULL */
1614         BUG_ON(ret < 0);
1615         if (map_length < length + size)
1616                 return 1;
1617         return 0;
1618 }
1619
1620 /*
1621  * in order to insert checksums into the metadata in large chunks,
1622  * we wait until bio submission time.   All the pages in the bio are
1623  * checksummed and sums are attached onto the ordered extent record.
1624  *
1625  * At IO completion time the cums attached on the ordered extent record
1626  * are inserted into the btree
1627  */
1628 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1629                                     struct bio *bio, int mirror_num,
1630                                     unsigned long bio_flags,
1631                                     u64 bio_offset)
1632 {
1633         struct btrfs_root *root = BTRFS_I(inode)->root;
1634         int ret = 0;
1635
1636         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1637         BUG_ON(ret); /* -ENOMEM */
1638         return 0;
1639 }
1640
1641 /*
1642  * in order to insert checksums into the metadata in large chunks,
1643  * we wait until bio submission time.   All the pages in the bio are
1644  * checksummed and sums are attached onto the ordered extent record.
1645  *
1646  * At IO completion time the cums attached on the ordered extent record
1647  * are inserted into the btree
1648  */
1649 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1650                           int mirror_num, unsigned long bio_flags,
1651                           u64 bio_offset)
1652 {
1653         struct btrfs_root *root = BTRFS_I(inode)->root;
1654         int ret;
1655
1656         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1657         if (ret)
1658                 bio_endio(bio, ret);
1659         return ret;
1660 }
1661
1662 /*
1663  * extent_io.c submission hook. This does the right thing for csum calculation
1664  * on write, or reading the csums from the tree before a read
1665  */
1666 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1667                           int mirror_num, unsigned long bio_flags,
1668                           u64 bio_offset)
1669 {
1670         struct btrfs_root *root = BTRFS_I(inode)->root;
1671         int ret = 0;
1672         int skip_sum;
1673         int metadata = 0;
1674         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1675
1676         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1677
1678         if (btrfs_is_free_space_inode(inode))
1679                 metadata = 2;
1680
1681         if (!(rw & REQ_WRITE)) {
1682                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1683                 if (ret)
1684                         goto out;
1685
1686                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1687                         ret = btrfs_submit_compressed_read(inode, bio,
1688                                                            mirror_num,
1689                                                            bio_flags);
1690                         goto out;
1691                 } else if (!skip_sum) {
1692                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1693                         if (ret)
1694                                 goto out;
1695                 }
1696                 goto mapit;
1697         } else if (async && !skip_sum) {
1698                 /* csum items have already been cloned */
1699                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1700                         goto mapit;
1701                 /* we're doing a write, do the async checksumming */
1702                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1703                                    inode, rw, bio, mirror_num,
1704                                    bio_flags, bio_offset,
1705                                    __btrfs_submit_bio_start,
1706                                    __btrfs_submit_bio_done);
1707                 goto out;
1708         } else if (!skip_sum) {
1709                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1710                 if (ret)
1711                         goto out;
1712         }
1713
1714 mapit:
1715         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1716
1717 out:
1718         if (ret < 0)
1719                 bio_endio(bio, ret);
1720         return ret;
1721 }
1722
1723 /*
1724  * given a list of ordered sums record them in the inode.  This happens
1725  * at IO completion time based on sums calculated at bio submission time.
1726  */
1727 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1728                              struct inode *inode, u64 file_offset,
1729                              struct list_head *list)
1730 {
1731         struct btrfs_ordered_sum *sum;
1732
1733         list_for_each_entry(sum, list, list) {
1734                 trans->adding_csums = 1;
1735                 btrfs_csum_file_blocks(trans,
1736                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1737                 trans->adding_csums = 0;
1738         }
1739         return 0;
1740 }
1741
1742 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1743                               struct extent_state **cached_state)
1744 {
1745         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1746         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1747                                    cached_state, GFP_NOFS);
1748 }
1749
1750 /* see btrfs_writepage_start_hook for details on why this is required */
1751 struct btrfs_writepage_fixup {
1752         struct page *page;
1753         struct btrfs_work work;
1754 };
1755
1756 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1757 {
1758         struct btrfs_writepage_fixup *fixup;
1759         struct btrfs_ordered_extent *ordered;
1760         struct extent_state *cached_state = NULL;
1761         struct page *page;
1762         struct inode *inode;
1763         u64 page_start;
1764         u64 page_end;
1765         int ret;
1766
1767         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1768         page = fixup->page;
1769 again:
1770         lock_page(page);
1771         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1772                 ClearPageChecked(page);
1773                 goto out_page;
1774         }
1775
1776         inode = page->mapping->host;
1777         page_start = page_offset(page);
1778         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1779
1780         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1781                          &cached_state);
1782
1783         /* already ordered? We're done */
1784         if (PagePrivate2(page))
1785                 goto out;
1786
1787         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1788         if (ordered) {
1789                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1790                                      page_end, &cached_state, GFP_NOFS);
1791                 unlock_page(page);
1792                 btrfs_start_ordered_extent(inode, ordered, 1);
1793                 btrfs_put_ordered_extent(ordered);
1794                 goto again;
1795         }
1796
1797         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1798         if (ret) {
1799                 mapping_set_error(page->mapping, ret);
1800                 end_extent_writepage(page, ret, page_start, page_end);
1801                 ClearPageChecked(page);
1802                 goto out;
1803          }
1804
1805         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1806         ClearPageChecked(page);
1807         set_page_dirty(page);
1808 out:
1809         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1810                              &cached_state, GFP_NOFS);
1811 out_page:
1812         unlock_page(page);
1813         page_cache_release(page);
1814         kfree(fixup);
1815 }
1816
1817 /*
1818  * There are a few paths in the higher layers of the kernel that directly
1819  * set the page dirty bit without asking the filesystem if it is a
1820  * good idea.  This causes problems because we want to make sure COW
1821  * properly happens and the data=ordered rules are followed.
1822  *
1823  * In our case any range that doesn't have the ORDERED bit set
1824  * hasn't been properly setup for IO.  We kick off an async process
1825  * to fix it up.  The async helper will wait for ordered extents, set
1826  * the delalloc bit and make it safe to write the page.
1827  */
1828 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1829 {
1830         struct inode *inode = page->mapping->host;
1831         struct btrfs_writepage_fixup *fixup;
1832         struct btrfs_root *root = BTRFS_I(inode)->root;
1833
1834         /* this page is properly in the ordered list */
1835         if (TestClearPagePrivate2(page))
1836                 return 0;
1837
1838         if (PageChecked(page))
1839                 return -EAGAIN;
1840
1841         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1842         if (!fixup)
1843                 return -EAGAIN;
1844
1845         SetPageChecked(page);
1846         page_cache_get(page);
1847         fixup->work.func = btrfs_writepage_fixup_worker;
1848         fixup->page = page;
1849         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1850         return -EBUSY;
1851 }
1852
1853 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1854                                        struct inode *inode, u64 file_pos,
1855                                        u64 disk_bytenr, u64 disk_num_bytes,
1856                                        u64 num_bytes, u64 ram_bytes,
1857                                        u8 compression, u8 encryption,
1858                                        u16 other_encoding, int extent_type)
1859 {
1860         struct btrfs_root *root = BTRFS_I(inode)->root;
1861         struct btrfs_file_extent_item *fi;
1862         struct btrfs_path *path;
1863         struct extent_buffer *leaf;
1864         struct btrfs_key ins;
1865         int extent_inserted = 0;
1866         int ret;
1867
1868         path = btrfs_alloc_path();
1869         if (!path)
1870                 return -ENOMEM;
1871
1872         /*
1873          * we may be replacing one extent in the tree with another.
1874          * The new extent is pinned in the extent map, and we don't want
1875          * to drop it from the cache until it is completely in the btree.
1876          *
1877          * So, tell btrfs_drop_extents to leave this extent in the cache.
1878          * the caller is expected to unpin it and allow it to be merged
1879          * with the others.
1880          */
1881         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1882                                    file_pos + num_bytes, NULL, 0,
1883                                    1, sizeof(*fi), &extent_inserted);
1884         if (ret)
1885                 goto out;
1886
1887         if (!extent_inserted) {
1888                 ins.objectid = btrfs_ino(inode);
1889                 ins.offset = file_pos;
1890                 ins.type = BTRFS_EXTENT_DATA_KEY;
1891
1892                 path->leave_spinning = 1;
1893                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1894                                               sizeof(*fi));
1895                 if (ret)
1896                         goto out;
1897         }
1898         leaf = path->nodes[0];
1899         fi = btrfs_item_ptr(leaf, path->slots[0],
1900                             struct btrfs_file_extent_item);
1901         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1902         btrfs_set_file_extent_type(leaf, fi, extent_type);
1903         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1904         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1905         btrfs_set_file_extent_offset(leaf, fi, 0);
1906         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1907         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1908         btrfs_set_file_extent_compression(leaf, fi, compression);
1909         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1910         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1911
1912         btrfs_mark_buffer_dirty(leaf);
1913         btrfs_release_path(path);
1914
1915         inode_add_bytes(inode, num_bytes);
1916
1917         ins.objectid = disk_bytenr;
1918         ins.offset = disk_num_bytes;
1919         ins.type = BTRFS_EXTENT_ITEM_KEY;
1920         ret = btrfs_alloc_reserved_file_extent(trans, root,
1921                                         root->root_key.objectid,
1922                                         btrfs_ino(inode), file_pos, &ins);
1923 out:
1924         btrfs_free_path(path);
1925
1926         return ret;
1927 }
1928
1929 /* snapshot-aware defrag */
1930 struct sa_defrag_extent_backref {
1931         struct rb_node node;
1932         struct old_sa_defrag_extent *old;
1933         u64 root_id;
1934         u64 inum;
1935         u64 file_pos;
1936         u64 extent_offset;
1937         u64 num_bytes;
1938         u64 generation;
1939 };
1940
1941 struct old_sa_defrag_extent {
1942         struct list_head list;
1943         struct new_sa_defrag_extent *new;
1944
1945         u64 extent_offset;
1946         u64 bytenr;
1947         u64 offset;
1948         u64 len;
1949         int count;
1950 };
1951
1952 struct new_sa_defrag_extent {
1953         struct rb_root root;
1954         struct list_head head;
1955         struct btrfs_path *path;
1956         struct inode *inode;
1957         u64 file_pos;
1958         u64 len;
1959         u64 bytenr;
1960         u64 disk_len;
1961         u8 compress_type;
1962 };
1963
1964 static int backref_comp(struct sa_defrag_extent_backref *b1,
1965                         struct sa_defrag_extent_backref *b2)
1966 {
1967         if (b1->root_id < b2->root_id)
1968                 return -1;
1969         else if (b1->root_id > b2->root_id)
1970                 return 1;
1971
1972         if (b1->inum < b2->inum)
1973                 return -1;
1974         else if (b1->inum > b2->inum)
1975                 return 1;
1976
1977         if (b1->file_pos < b2->file_pos)
1978                 return -1;
1979         else if (b1->file_pos > b2->file_pos)
1980                 return 1;
1981
1982         /*
1983          * [------------------------------] ===> (a range of space)
1984          *     |<--->|   |<---->| =============> (fs/file tree A)
1985          * |<---------------------------->| ===> (fs/file tree B)
1986          *
1987          * A range of space can refer to two file extents in one tree while
1988          * refer to only one file extent in another tree.
1989          *
1990          * So we may process a disk offset more than one time(two extents in A)
1991          * and locate at the same extent(one extent in B), then insert two same
1992          * backrefs(both refer to the extent in B).
1993          */
1994         return 0;
1995 }
1996
1997 static void backref_insert(struct rb_root *root,
1998                            struct sa_defrag_extent_backref *backref)
1999 {
2000         struct rb_node **p = &root->rb_node;
2001         struct rb_node *parent = NULL;
2002         struct sa_defrag_extent_backref *entry;
2003         int ret;
2004
2005         while (*p) {
2006                 parent = *p;
2007                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2008
2009                 ret = backref_comp(backref, entry);
2010                 if (ret < 0)
2011                         p = &(*p)->rb_left;
2012                 else
2013                         p = &(*p)->rb_right;
2014         }
2015
2016         rb_link_node(&backref->node, parent, p);
2017         rb_insert_color(&backref->node, root);
2018 }
2019
2020 /*
2021  * Note the backref might has changed, and in this case we just return 0.
2022  */
2023 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2024                                        void *ctx)
2025 {
2026         struct btrfs_file_extent_item *extent;
2027         struct btrfs_fs_info *fs_info;
2028         struct old_sa_defrag_extent *old = ctx;
2029         struct new_sa_defrag_extent *new = old->new;
2030         struct btrfs_path *path = new->path;
2031         struct btrfs_key key;
2032         struct btrfs_root *root;
2033         struct sa_defrag_extent_backref *backref;
2034         struct extent_buffer *leaf;
2035         struct inode *inode = new->inode;
2036         int slot;
2037         int ret;
2038         u64 extent_offset;
2039         u64 num_bytes;
2040
2041         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2042             inum == btrfs_ino(inode))
2043                 return 0;
2044
2045         key.objectid = root_id;
2046         key.type = BTRFS_ROOT_ITEM_KEY;
2047         key.offset = (u64)-1;
2048
2049         fs_info = BTRFS_I(inode)->root->fs_info;
2050         root = btrfs_read_fs_root_no_name(fs_info, &key);
2051         if (IS_ERR(root)) {
2052                 if (PTR_ERR(root) == -ENOENT)
2053                         return 0;
2054                 WARN_ON(1);
2055                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2056                          inum, offset, root_id);
2057                 return PTR_ERR(root);
2058         }
2059
2060         key.objectid = inum;
2061         key.type = BTRFS_EXTENT_DATA_KEY;
2062         if (offset > (u64)-1 << 32)
2063                 key.offset = 0;
2064         else
2065                 key.offset = offset;
2066
2067         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2068         if (WARN_ON(ret < 0))
2069                 return ret;
2070         ret = 0;
2071
2072         while (1) {
2073                 cond_resched();
2074
2075                 leaf = path->nodes[0];
2076                 slot = path->slots[0];
2077
2078                 if (slot >= btrfs_header_nritems(leaf)) {
2079                         ret = btrfs_next_leaf(root, path);
2080                         if (ret < 0) {
2081                                 goto out;
2082                         } else if (ret > 0) {
2083                                 ret = 0;
2084                                 goto out;
2085                         }
2086                         continue;
2087                 }
2088
2089                 path->slots[0]++;
2090
2091                 btrfs_item_key_to_cpu(leaf, &key, slot);
2092
2093                 if (key.objectid > inum)
2094                         goto out;
2095
2096                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2097                         continue;
2098
2099                 extent = btrfs_item_ptr(leaf, slot,
2100                                         struct btrfs_file_extent_item);
2101
2102                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2103                         continue;
2104
2105                 /*
2106                  * 'offset' refers to the exact key.offset,
2107                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2108                  * (key.offset - extent_offset).
2109                  */
2110                 if (key.offset != offset)
2111                         continue;
2112
2113                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2114                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2115
2116                 if (extent_offset >= old->extent_offset + old->offset +
2117                     old->len || extent_offset + num_bytes <=
2118                     old->extent_offset + old->offset)
2119                         continue;
2120                 break;
2121         }
2122
2123         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2124         if (!backref) {
2125                 ret = -ENOENT;
2126                 goto out;
2127         }
2128
2129         backref->root_id = root_id;
2130         backref->inum = inum;
2131         backref->file_pos = offset;
2132         backref->num_bytes = num_bytes;
2133         backref->extent_offset = extent_offset;
2134         backref->generation = btrfs_file_extent_generation(leaf, extent);
2135         backref->old = old;
2136         backref_insert(&new->root, backref);
2137         old->count++;
2138 out:
2139         btrfs_release_path(path);
2140         WARN_ON(ret);
2141         return ret;
2142 }
2143
2144 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2145                                    struct new_sa_defrag_extent *new)
2146 {
2147         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2148         struct old_sa_defrag_extent *old, *tmp;
2149         int ret;
2150
2151         new->path = path;
2152
2153         list_for_each_entry_safe(old, tmp, &new->head, list) {
2154                 ret = iterate_inodes_from_logical(old->bytenr +
2155                                                   old->extent_offset, fs_info,
2156                                                   path, record_one_backref,
2157                                                   old);
2158                 if (ret < 0 && ret != -ENOENT)
2159                         return false;
2160
2161                 /* no backref to be processed for this extent */
2162                 if (!old->count) {
2163                         list_del(&old->list);
2164                         kfree(old);
2165                 }
2166         }
2167
2168         if (list_empty(&new->head))
2169                 return false;
2170
2171         return true;
2172 }
2173
2174 static int relink_is_mergable(struct extent_buffer *leaf,
2175                               struct btrfs_file_extent_item *fi,
2176                               struct new_sa_defrag_extent *new)
2177 {
2178         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2179                 return 0;
2180
2181         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2182                 return 0;
2183
2184         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2185                 return 0;
2186
2187         if (btrfs_file_extent_encryption(leaf, fi) ||
2188             btrfs_file_extent_other_encoding(leaf, fi))
2189                 return 0;
2190
2191         return 1;
2192 }
2193
2194 /*
2195  * Note the backref might has changed, and in this case we just return 0.
2196  */
2197 static noinline int relink_extent_backref(struct btrfs_path *path,
2198                                  struct sa_defrag_extent_backref *prev,
2199                                  struct sa_defrag_extent_backref *backref)
2200 {
2201         struct btrfs_file_extent_item *extent;
2202         struct btrfs_file_extent_item *item;
2203         struct btrfs_ordered_extent *ordered;
2204         struct btrfs_trans_handle *trans;
2205         struct btrfs_fs_info *fs_info;
2206         struct btrfs_root *root;
2207         struct btrfs_key key;
2208         struct extent_buffer *leaf;
2209         struct old_sa_defrag_extent *old = backref->old;
2210         struct new_sa_defrag_extent *new = old->new;
2211         struct inode *src_inode = new->inode;
2212         struct inode *inode;
2213         struct extent_state *cached = NULL;
2214         int ret = 0;
2215         u64 start;
2216         u64 len;
2217         u64 lock_start;
2218         u64 lock_end;
2219         bool merge = false;
2220         int index;
2221
2222         if (prev && prev->root_id == backref->root_id &&
2223             prev->inum == backref->inum &&
2224             prev->file_pos + prev->num_bytes == backref->file_pos)
2225                 merge = true;
2226
2227         /* step 1: get root */
2228         key.objectid = backref->root_id;
2229         key.type = BTRFS_ROOT_ITEM_KEY;
2230         key.offset = (u64)-1;
2231
2232         fs_info = BTRFS_I(src_inode)->root->fs_info;
2233         index = srcu_read_lock(&fs_info->subvol_srcu);
2234
2235         root = btrfs_read_fs_root_no_name(fs_info, &key);
2236         if (IS_ERR(root)) {
2237                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2238                 if (PTR_ERR(root) == -ENOENT)
2239                         return 0;
2240                 return PTR_ERR(root);
2241         }
2242
2243         /* step 2: get inode */
2244         key.objectid = backref->inum;
2245         key.type = BTRFS_INODE_ITEM_KEY;
2246         key.offset = 0;
2247
2248         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2249         if (IS_ERR(inode)) {
2250                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2251                 return 0;
2252         }
2253
2254         srcu_read_unlock(&fs_info->subvol_srcu, index);
2255
2256         /* step 3: relink backref */
2257         lock_start = backref->file_pos;
2258         lock_end = backref->file_pos + backref->num_bytes - 1;
2259         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2260                          0, &cached);
2261
2262         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2263         if (ordered) {
2264                 btrfs_put_ordered_extent(ordered);
2265                 goto out_unlock;
2266         }
2267
2268         trans = btrfs_join_transaction(root);
2269         if (IS_ERR(trans)) {
2270                 ret = PTR_ERR(trans);
2271                 goto out_unlock;
2272         }
2273
2274         key.objectid = backref->inum;
2275         key.type = BTRFS_EXTENT_DATA_KEY;
2276         key.offset = backref->file_pos;
2277
2278         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2279         if (ret < 0) {
2280                 goto out_free_path;
2281         } else if (ret > 0) {
2282                 ret = 0;
2283                 goto out_free_path;
2284         }
2285
2286         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2287                                 struct btrfs_file_extent_item);
2288
2289         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2290             backref->generation)
2291                 goto out_free_path;
2292
2293         btrfs_release_path(path);
2294
2295         start = backref->file_pos;
2296         if (backref->extent_offset < old->extent_offset + old->offset)
2297                 start += old->extent_offset + old->offset -
2298                          backref->extent_offset;
2299
2300         len = min(backref->extent_offset + backref->num_bytes,
2301                   old->extent_offset + old->offset + old->len);
2302         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2303
2304         ret = btrfs_drop_extents(trans, root, inode, start,
2305                                  start + len, 1);
2306         if (ret)
2307                 goto out_free_path;
2308 again:
2309         key.objectid = btrfs_ino(inode);
2310         key.type = BTRFS_EXTENT_DATA_KEY;
2311         key.offset = start;
2312
2313         path->leave_spinning = 1;
2314         if (merge) {
2315                 struct btrfs_file_extent_item *fi;
2316                 u64 extent_len;
2317                 struct btrfs_key found_key;
2318
2319                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2320                 if (ret < 0)
2321                         goto out_free_path;
2322
2323                 path->slots[0]--;
2324                 leaf = path->nodes[0];
2325                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2326
2327                 fi = btrfs_item_ptr(leaf, path->slots[0],
2328                                     struct btrfs_file_extent_item);
2329                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2330
2331                 if (extent_len + found_key.offset == start &&
2332                     relink_is_mergable(leaf, fi, new)) {
2333                         btrfs_set_file_extent_num_bytes(leaf, fi,
2334                                                         extent_len + len);
2335                         btrfs_mark_buffer_dirty(leaf);
2336                         inode_add_bytes(inode, len);
2337
2338                         ret = 1;
2339                         goto out_free_path;
2340                 } else {
2341                         merge = false;
2342                         btrfs_release_path(path);
2343                         goto again;
2344                 }
2345         }
2346
2347         ret = btrfs_insert_empty_item(trans, root, path, &key,
2348                                         sizeof(*extent));
2349         if (ret) {
2350                 btrfs_abort_transaction(trans, root, ret);
2351                 goto out_free_path;
2352         }
2353
2354         leaf = path->nodes[0];
2355         item = btrfs_item_ptr(leaf, path->slots[0],
2356                                 struct btrfs_file_extent_item);
2357         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2358         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2359         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2360         btrfs_set_file_extent_num_bytes(leaf, item, len);
2361         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2362         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2363         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2364         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2365         btrfs_set_file_extent_encryption(leaf, item, 0);
2366         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2367
2368         btrfs_mark_buffer_dirty(leaf);
2369         inode_add_bytes(inode, len);
2370         btrfs_release_path(path);
2371
2372         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2373                         new->disk_len, 0,
2374                         backref->root_id, backref->inum,
2375                         new->file_pos, 0);      /* start - extent_offset */
2376         if (ret) {
2377                 btrfs_abort_transaction(trans, root, ret);
2378                 goto out_free_path;
2379         }
2380
2381         ret = 1;
2382 out_free_path:
2383         btrfs_release_path(path);
2384         path->leave_spinning = 0;
2385         btrfs_end_transaction(trans, root);
2386 out_unlock:
2387         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2388                              &cached, GFP_NOFS);
2389         iput(inode);
2390         return ret;
2391 }
2392
2393 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2394 {
2395         struct old_sa_defrag_extent *old, *tmp;
2396
2397         if (!new)
2398                 return;
2399
2400         list_for_each_entry_safe(old, tmp, &new->head, list) {
2401                 list_del(&old->list);
2402                 kfree(old);
2403         }
2404         kfree(new);
2405 }
2406
2407 static void relink_file_extents(struct new_sa_defrag_extent *new)
2408 {
2409         struct btrfs_path *path;
2410         struct sa_defrag_extent_backref *backref;
2411         struct sa_defrag_extent_backref *prev = NULL;
2412         struct inode *inode;
2413         struct btrfs_root *root;
2414         struct rb_node *node;
2415         int ret;
2416
2417         inode = new->inode;
2418         root = BTRFS_I(inode)->root;
2419
2420         path = btrfs_alloc_path();
2421         if (!path)
2422                 return;
2423
2424         if (!record_extent_backrefs(path, new)) {
2425                 btrfs_free_path(path);
2426                 goto out;
2427         }
2428         btrfs_release_path(path);
2429
2430         while (1) {
2431                 node = rb_first(&new->root);
2432                 if (!node)
2433                         break;
2434                 rb_erase(node, &new->root);
2435
2436                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2437
2438                 ret = relink_extent_backref(path, prev, backref);
2439                 WARN_ON(ret < 0);
2440
2441                 kfree(prev);
2442
2443                 if (ret == 1)
2444                         prev = backref;
2445                 else
2446                         prev = NULL;
2447                 cond_resched();
2448         }
2449         kfree(prev);
2450
2451         btrfs_free_path(path);
2452 out:
2453         free_sa_defrag_extent(new);
2454
2455         atomic_dec(&root->fs_info->defrag_running);
2456         wake_up(&root->fs_info->transaction_wait);
2457 }
2458
2459 static struct new_sa_defrag_extent *
2460 record_old_file_extents(struct inode *inode,
2461                         struct btrfs_ordered_extent *ordered)
2462 {
2463         struct btrfs_root *root = BTRFS_I(inode)->root;
2464         struct btrfs_path *path;
2465         struct btrfs_key key;
2466         struct old_sa_defrag_extent *old;
2467         struct new_sa_defrag_extent *new;
2468         int ret;
2469
2470         new = kmalloc(sizeof(*new), GFP_NOFS);
2471         if (!new)
2472                 return NULL;
2473
2474         new->inode = inode;
2475         new->file_pos = ordered->file_offset;
2476         new->len = ordered->len;
2477         new->bytenr = ordered->start;
2478         new->disk_len = ordered->disk_len;
2479         new->compress_type = ordered->compress_type;
2480         new->root = RB_ROOT;
2481         INIT_LIST_HEAD(&new->head);
2482
2483         path = btrfs_alloc_path();
2484         if (!path)
2485                 goto out_kfree;
2486
2487         key.objectid = btrfs_ino(inode);
2488         key.type = BTRFS_EXTENT_DATA_KEY;
2489         key.offset = new->file_pos;
2490
2491         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2492         if (ret < 0)
2493                 goto out_free_path;
2494         if (ret > 0 && path->slots[0] > 0)
2495                 path->slots[0]--;
2496
2497         /* find out all the old extents for the file range */
2498         while (1) {
2499                 struct btrfs_file_extent_item *extent;
2500                 struct extent_buffer *l;
2501                 int slot;
2502                 u64 num_bytes;
2503                 u64 offset;
2504                 u64 end;
2505                 u64 disk_bytenr;
2506                 u64 extent_offset;
2507
2508                 l = path->nodes[0];
2509                 slot = path->slots[0];
2510
2511                 if (slot >= btrfs_header_nritems(l)) {
2512                         ret = btrfs_next_leaf(root, path);
2513                         if (ret < 0)
2514                                 goto out_free_path;
2515                         else if (ret > 0)
2516                                 break;
2517                         continue;
2518                 }
2519
2520                 btrfs_item_key_to_cpu(l, &key, slot);
2521
2522                 if (key.objectid != btrfs_ino(inode))
2523                         break;
2524                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2525                         break;
2526                 if (key.offset >= new->file_pos + new->len)
2527                         break;
2528
2529                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2530
2531                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2532                 if (key.offset + num_bytes < new->file_pos)
2533                         goto next;
2534
2535                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2536                 if (!disk_bytenr)
2537                         goto next;
2538
2539                 extent_offset = btrfs_file_extent_offset(l, extent);
2540
2541                 old = kmalloc(sizeof(*old), GFP_NOFS);
2542                 if (!old)
2543                         goto out_free_path;
2544
2545                 offset = max(new->file_pos, key.offset);
2546                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2547
2548                 old->bytenr = disk_bytenr;
2549                 old->extent_offset = extent_offset;
2550                 old->offset = offset - key.offset;
2551                 old->len = end - offset;
2552                 old->new = new;
2553                 old->count = 0;
2554                 list_add_tail(&old->list, &new->head);
2555 next:
2556                 path->slots[0]++;
2557                 cond_resched();
2558         }
2559
2560         btrfs_free_path(path);
2561         atomic_inc(&root->fs_info->defrag_running);
2562
2563         return new;
2564
2565 out_free_path:
2566         btrfs_free_path(path);
2567 out_kfree:
2568         free_sa_defrag_extent(new);
2569         return NULL;
2570 }
2571
2572 /* as ordered data IO finishes, this gets called so we can finish
2573  * an ordered extent if the range of bytes in the file it covers are
2574  * fully written.
2575  */
2576 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2577 {
2578         struct inode *inode = ordered_extent->inode;
2579         struct btrfs_root *root = BTRFS_I(inode)->root;
2580         struct btrfs_trans_handle *trans = NULL;
2581         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2582         struct extent_state *cached_state = NULL;
2583         struct new_sa_defrag_extent *new = NULL;
2584         int compress_type = 0;
2585         int ret = 0;
2586         u64 logical_len = ordered_extent->len;
2587         bool nolock;
2588         bool truncated = false;
2589
2590         nolock = btrfs_is_free_space_inode(inode);
2591
2592         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2593                 ret = -EIO;
2594                 goto out;
2595         }
2596
2597         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2598                 truncated = true;
2599                 logical_len = ordered_extent->truncated_len;
2600                 /* Truncated the entire extent, don't bother adding */
2601                 if (!logical_len)
2602                         goto out;
2603         }
2604
2605         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2606                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2607                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2608                 if (nolock)
2609                         trans = btrfs_join_transaction_nolock(root);
2610                 else
2611                         trans = btrfs_join_transaction(root);
2612                 if (IS_ERR(trans)) {
2613                         ret = PTR_ERR(trans);
2614                         trans = NULL;
2615                         goto out;
2616                 }
2617                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2618                 ret = btrfs_update_inode_fallback(trans, root, inode);
2619                 if (ret) /* -ENOMEM or corruption */
2620                         btrfs_abort_transaction(trans, root, ret);
2621                 goto out;
2622         }
2623
2624         lock_extent_bits(io_tree, ordered_extent->file_offset,
2625                          ordered_extent->file_offset + ordered_extent->len - 1,
2626                          0, &cached_state);
2627
2628         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2629                         ordered_extent->file_offset + ordered_extent->len - 1,
2630                         EXTENT_DEFRAG, 1, cached_state);
2631         if (ret) {
2632                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2633                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2634                         /* the inode is shared */
2635                         new = record_old_file_extents(inode, ordered_extent);
2636
2637                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2638                         ordered_extent->file_offset + ordered_extent->len - 1,
2639                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2640         }
2641
2642         if (nolock)
2643                 trans = btrfs_join_transaction_nolock(root);
2644         else
2645                 trans = btrfs_join_transaction(root);
2646         if (IS_ERR(trans)) {
2647                 ret = PTR_ERR(trans);
2648                 trans = NULL;
2649                 goto out_unlock;
2650         }
2651         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2652
2653         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2654                 compress_type = ordered_extent->compress_type;
2655         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2656                 BUG_ON(compress_type);
2657                 ret = btrfs_mark_extent_written(trans, inode,
2658                                                 ordered_extent->file_offset,
2659                                                 ordered_extent->file_offset +
2660                                                 logical_len);
2661         } else {
2662                 BUG_ON(root == root->fs_info->tree_root);
2663                 ret = insert_reserved_file_extent(trans, inode,
2664                                                 ordered_extent->file_offset,
2665                                                 ordered_extent->start,
2666                                                 ordered_extent->disk_len,
2667                                                 logical_len, logical_len,
2668                                                 compress_type, 0, 0,
2669                                                 BTRFS_FILE_EXTENT_REG);
2670         }
2671         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2672                            ordered_extent->file_offset, ordered_extent->len,
2673                            trans->transid);
2674         if (ret < 0) {
2675                 btrfs_abort_transaction(trans, root, ret);
2676                 goto out_unlock;
2677         }
2678
2679         add_pending_csums(trans, inode, ordered_extent->file_offset,
2680                           &ordered_extent->list);
2681
2682         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2683         ret = btrfs_update_inode_fallback(trans, root, inode);
2684         if (ret) { /* -ENOMEM or corruption */
2685                 btrfs_abort_transaction(trans, root, ret);
2686                 goto out_unlock;
2687         }
2688         ret = 0;
2689 out_unlock:
2690         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2691                              ordered_extent->file_offset +
2692                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2693 out:
2694         if (root != root->fs_info->tree_root)
2695                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2696         if (trans)
2697                 btrfs_end_transaction(trans, root);
2698
2699         if (ret || truncated) {
2700                 u64 start, end;
2701
2702                 if (truncated)
2703                         start = ordered_extent->file_offset + logical_len;
2704                 else
2705                         start = ordered_extent->file_offset;
2706                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2707                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2708
2709                 /* Drop the cache for the part of the extent we didn't write. */
2710                 btrfs_drop_extent_cache(inode, start, end, 0);
2711
2712                 /*
2713                  * If the ordered extent had an IOERR or something else went
2714                  * wrong we need to return the space for this ordered extent
2715                  * back to the allocator.  We only free the extent in the
2716                  * truncated case if we didn't write out the extent at all.
2717                  */
2718                 if ((ret || !logical_len) &&
2719                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2720                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2721                         btrfs_free_reserved_extent(root, ordered_extent->start,
2722                                                    ordered_extent->disk_len);
2723         }
2724
2725
2726         /*
2727          * This needs to be done to make sure anybody waiting knows we are done
2728          * updating everything for this ordered extent.
2729          */
2730         btrfs_remove_ordered_extent(inode, ordered_extent);
2731
2732         /* for snapshot-aware defrag */
2733         if (new) {
2734                 if (ret) {
2735                         free_sa_defrag_extent(new);
2736                         atomic_dec(&root->fs_info->defrag_running);
2737                 } else {
2738                         relink_file_extents(new);
2739                 }
2740         }
2741
2742         /* once for us */
2743         btrfs_put_ordered_extent(ordered_extent);
2744         /* once for the tree */
2745         btrfs_put_ordered_extent(ordered_extent);
2746
2747         return ret;
2748 }
2749
2750 static void finish_ordered_fn(struct btrfs_work *work)
2751 {
2752         struct btrfs_ordered_extent *ordered_extent;
2753         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2754         btrfs_finish_ordered_io(ordered_extent);
2755 }
2756
2757 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2758                                 struct extent_state *state, int uptodate)
2759 {
2760         struct inode *inode = page->mapping->host;
2761         struct btrfs_root *root = BTRFS_I(inode)->root;
2762         struct btrfs_ordered_extent *ordered_extent = NULL;
2763         struct btrfs_workers *workers;
2764
2765         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2766
2767         ClearPagePrivate2(page);
2768         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2769                                             end - start + 1, uptodate))
2770                 return 0;
2771
2772         ordered_extent->work.func = finish_ordered_fn;
2773         ordered_extent->work.flags = 0;
2774
2775         if (btrfs_is_free_space_inode(inode))
2776                 workers = &root->fs_info->endio_freespace_worker;
2777         else
2778                 workers = &root->fs_info->endio_write_workers;
2779         btrfs_queue_worker(workers, &ordered_extent->work);
2780
2781         return 0;
2782 }
2783
2784 /*
2785  * when reads are done, we need to check csums to verify the data is correct
2786  * if there's a match, we allow the bio to finish.  If not, the code in
2787  * extent_io.c will try to find good copies for us.
2788  */
2789 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2790                                       u64 phy_offset, struct page *page,
2791                                       u64 start, u64 end, int mirror)
2792 {
2793         size_t offset = start - page_offset(page);
2794         struct inode *inode = page->mapping->host;
2795         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2796         char *kaddr;
2797         struct btrfs_root *root = BTRFS_I(inode)->root;
2798         u32 csum_expected;
2799         u32 csum = ~(u32)0;
2800         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2801                                       DEFAULT_RATELIMIT_BURST);
2802
2803         if (PageChecked(page)) {
2804                 ClearPageChecked(page);
2805                 goto good;
2806         }
2807
2808         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2809                 goto good;
2810
2811         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2812             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2813                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2814                                   GFP_NOFS);
2815                 return 0;
2816         }
2817
2818         phy_offset >>= inode->i_sb->s_blocksize_bits;
2819         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2820
2821         kaddr = kmap_atomic(page);
2822         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2823         btrfs_csum_final(csum, (char *)&csum);
2824         if (csum != csum_expected)
2825                 goto zeroit;
2826
2827         kunmap_atomic(kaddr);
2828 good:
2829         return 0;
2830
2831 zeroit:
2832         if (__ratelimit(&_rs))
2833                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2834                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2835         memset(kaddr + offset, 1, end - start + 1);
2836         flush_dcache_page(page);
2837         kunmap_atomic(kaddr);
2838         if (csum_expected == 0)
2839                 return 0;
2840         return -EIO;
2841 }
2842
2843 struct delayed_iput {
2844         struct list_head list;
2845         struct inode *inode;
2846 };
2847
2848 /* JDM: If this is fs-wide, why can't we add a pointer to
2849  * btrfs_inode instead and avoid the allocation? */
2850 void btrfs_add_delayed_iput(struct inode *inode)
2851 {
2852         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2853         struct delayed_iput *delayed;
2854
2855         if (atomic_add_unless(&inode->i_count, -1, 1))
2856                 return;
2857
2858         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2859         delayed->inode = inode;
2860
2861         spin_lock(&fs_info->delayed_iput_lock);
2862         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2863         spin_unlock(&fs_info->delayed_iput_lock);
2864 }
2865
2866 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2867 {
2868         LIST_HEAD(list);
2869         struct btrfs_fs_info *fs_info = root->fs_info;
2870         struct delayed_iput *delayed;
2871         int empty;
2872
2873         spin_lock(&fs_info->delayed_iput_lock);
2874         empty = list_empty(&fs_info->delayed_iputs);
2875         spin_unlock(&fs_info->delayed_iput_lock);
2876         if (empty)
2877                 return;
2878
2879         spin_lock(&fs_info->delayed_iput_lock);
2880         list_splice_init(&fs_info->delayed_iputs, &list);
2881         spin_unlock(&fs_info->delayed_iput_lock);
2882
2883         while (!list_empty(&list)) {
2884                 delayed = list_entry(list.next, struct delayed_iput, list);
2885                 list_del(&delayed->list);
2886                 iput(delayed->inode);
2887                 kfree(delayed);
2888         }
2889 }
2890
2891 /*
2892  * This is called in transaction commit time. If there are no orphan
2893  * files in the subvolume, it removes orphan item and frees block_rsv
2894  * structure.
2895  */
2896 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2897                               struct btrfs_root *root)
2898 {
2899         struct btrfs_block_rsv *block_rsv;
2900         int ret;
2901
2902         if (atomic_read(&root->orphan_inodes) ||
2903             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2904                 return;
2905
2906         spin_lock(&root->orphan_lock);
2907         if (atomic_read(&root->orphan_inodes)) {
2908                 spin_unlock(&root->orphan_lock);
2909                 return;
2910         }
2911
2912         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2913                 spin_unlock(&root->orphan_lock);
2914                 return;
2915         }
2916
2917         block_rsv = root->orphan_block_rsv;
2918         root->orphan_block_rsv = NULL;
2919         spin_unlock(&root->orphan_lock);
2920
2921         if (root->orphan_item_inserted &&
2922             btrfs_root_refs(&root->root_item) > 0) {
2923                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2924                                             root->root_key.objectid);
2925                 if (ret)
2926                         btrfs_abort_transaction(trans, root, ret);
2927                 else
2928                         root->orphan_item_inserted = 0;
2929         }
2930
2931         if (block_rsv) {
2932                 WARN_ON(block_rsv->size > 0);
2933                 btrfs_free_block_rsv(root, block_rsv);
2934         }
2935 }
2936
2937 /*
2938  * This creates an orphan entry for the given inode in case something goes
2939  * wrong in the middle of an unlink/truncate.
2940  *
2941  * NOTE: caller of this function should reserve 5 units of metadata for
2942  *       this function.
2943  */
2944 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2945 {
2946         struct btrfs_root *root = BTRFS_I(inode)->root;
2947         struct btrfs_block_rsv *block_rsv = NULL;
2948         int reserve = 0;
2949         int insert = 0;
2950         int ret;
2951
2952         if (!root->orphan_block_rsv) {
2953                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2954                 if (!block_rsv)
2955                         return -ENOMEM;
2956         }
2957
2958         spin_lock(&root->orphan_lock);
2959         if (!root->orphan_block_rsv) {
2960                 root->orphan_block_rsv = block_rsv;
2961         } else if (block_rsv) {
2962                 btrfs_free_block_rsv(root, block_rsv);
2963                 block_rsv = NULL;
2964         }
2965
2966         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2967                               &BTRFS_I(inode)->runtime_flags)) {
2968 #if 0
2969                 /*
2970                  * For proper ENOSPC handling, we should do orphan
2971                  * cleanup when mounting. But this introduces backward
2972                  * compatibility issue.
2973                  */
2974                 if (!xchg(&root->orphan_item_inserted, 1))
2975                         insert = 2;
2976                 else
2977                         insert = 1;
2978 #endif
2979                 insert = 1;
2980                 atomic_inc(&root->orphan_inodes);
2981         }
2982
2983         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2984                               &BTRFS_I(inode)->runtime_flags))
2985                 reserve = 1;
2986         spin_unlock(&root->orphan_lock);
2987
2988         /* grab metadata reservation from transaction handle */
2989         if (reserve) {
2990                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2991                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2992         }
2993
2994         /* insert an orphan item to track this unlinked/truncated file */
2995         if (insert >= 1) {
2996                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2997                 if (ret) {
2998                         atomic_dec(&root->orphan_inodes);
2999                         if (reserve) {
3000                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3001                                           &BTRFS_I(inode)->runtime_flags);
3002                                 btrfs_orphan_release_metadata(inode);
3003                         }
3004                         if (ret != -EEXIST) {
3005                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3006                                           &BTRFS_I(inode)->runtime_flags);
3007                                 btrfs_abort_transaction(trans, root, ret);
3008                                 return ret;
3009                         }
3010                 }
3011                 ret = 0;
3012         }
3013
3014         /* insert an orphan item to track subvolume contains orphan files */
3015         if (insert >= 2) {
3016                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3017                                                root->root_key.objectid);
3018                 if (ret && ret != -EEXIST) {
3019                         btrfs_abort_transaction(trans, root, ret);
3020                         return ret;
3021                 }
3022         }
3023         return 0;
3024 }
3025
3026 /*
3027  * We have done the truncate/delete so we can go ahead and remove the orphan
3028  * item for this particular inode.
3029  */
3030 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3031                             struct inode *inode)
3032 {
3033         struct btrfs_root *root = BTRFS_I(inode)->root;
3034         int delete_item = 0;
3035         int release_rsv = 0;
3036         int ret = 0;
3037
3038         spin_lock(&root->orphan_lock);
3039         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3040                                &BTRFS_I(inode)->runtime_flags))
3041                 delete_item = 1;
3042
3043         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3044                                &BTRFS_I(inode)->runtime_flags))
3045                 release_rsv = 1;
3046         spin_unlock(&root->orphan_lock);
3047
3048         if (delete_item) {
3049                 atomic_dec(&root->orphan_inodes);
3050                 if (trans)
3051                         ret = btrfs_del_orphan_item(trans, root,
3052                                                     btrfs_ino(inode));
3053         }
3054
3055         if (release_rsv)
3056                 btrfs_orphan_release_metadata(inode);
3057
3058         return ret;
3059 }
3060
3061 /*
3062  * this cleans up any orphans that may be left on the list from the last use
3063  * of this root.
3064  */
3065 int btrfs_orphan_cleanup(struct btrfs_root *root)
3066 {
3067         struct btrfs_path *path;
3068         struct extent_buffer *leaf;
3069         struct btrfs_key key, found_key;
3070         struct btrfs_trans_handle *trans;
3071         struct inode *inode;
3072         u64 last_objectid = 0;
3073         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3074
3075         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3076                 return 0;
3077
3078         path = btrfs_alloc_path();
3079         if (!path) {
3080                 ret = -ENOMEM;
3081                 goto out;
3082         }
3083         path->reada = -1;
3084
3085         key.objectid = BTRFS_ORPHAN_OBJECTID;
3086         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3087         key.offset = (u64)-1;
3088
3089         while (1) {
3090                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3091                 if (ret < 0)
3092                         goto out;
3093
3094                 /*
3095                  * if ret == 0 means we found what we were searching for, which
3096                  * is weird, but possible, so only screw with path if we didn't
3097                  * find the key and see if we have stuff that matches
3098                  */
3099                 if (ret > 0) {
3100                         ret = 0;
3101                         if (path->slots[0] == 0)
3102                                 break;
3103                         path->slots[0]--;
3104                 }
3105
3106                 /* pull out the item */
3107                 leaf = path->nodes[0];
3108                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3109
3110                 /* make sure the item matches what we want */
3111                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3112                         break;
3113                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3114                         break;
3115
3116                 /* release the path since we're done with it */
3117                 btrfs_release_path(path);
3118
3119                 /*
3120                  * this is where we are basically btrfs_lookup, without the
3121                  * crossing root thing.  we store the inode number in the
3122                  * offset of the orphan item.
3123                  */
3124
3125                 if (found_key.offset == last_objectid) {
3126                         btrfs_err(root->fs_info,
3127                                 "Error removing orphan entry, stopping orphan cleanup");
3128                         ret = -EINVAL;
3129                         goto out;
3130                 }
3131
3132                 last_objectid = found_key.offset;
3133
3134                 found_key.objectid = found_key.offset;
3135                 found_key.type = BTRFS_INODE_ITEM_KEY;
3136                 found_key.offset = 0;
3137                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3138                 ret = PTR_ERR_OR_ZERO(inode);
3139                 if (ret && ret != -ESTALE)
3140                         goto out;
3141
3142                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3143                         struct btrfs_root *dead_root;
3144                         struct btrfs_fs_info *fs_info = root->fs_info;
3145                         int is_dead_root = 0;
3146
3147                         /*
3148                          * this is an orphan in the tree root. Currently these
3149                          * could come from 2 sources:
3150                          *  a) a snapshot deletion in progress
3151                          *  b) a free space cache inode
3152                          * We need to distinguish those two, as the snapshot
3153                          * orphan must not get deleted.
3154                          * find_dead_roots already ran before us, so if this
3155                          * is a snapshot deletion, we should find the root
3156                          * in the dead_roots list
3157                          */
3158                         spin_lock(&fs_info->trans_lock);
3159                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3160                                             root_list) {
3161                                 if (dead_root->root_key.objectid ==
3162                                     found_key.objectid) {
3163                                         is_dead_root = 1;
3164                                         break;
3165                                 }
3166                         }
3167                         spin_unlock(&fs_info->trans_lock);
3168                         if (is_dead_root) {
3169                                 /* prevent this orphan from being found again */
3170                                 key.offset = found_key.objectid - 1;
3171                                 continue;
3172                         }
3173                 }
3174                 /*
3175                  * Inode is already gone but the orphan item is still there,
3176                  * kill the orphan item.
3177                  */
3178                 if (ret == -ESTALE) {
3179                         trans = btrfs_start_transaction(root, 1);
3180                         if (IS_ERR(trans)) {
3181                                 ret = PTR_ERR(trans);
3182                                 goto out;
3183                         }
3184                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3185                                 found_key.objectid);
3186                         ret = btrfs_del_orphan_item(trans, root,
3187                                                     found_key.objectid);
3188                         btrfs_end_transaction(trans, root);
3189                         if (ret)
3190                                 goto out;
3191                         continue;
3192                 }
3193
3194                 /*
3195                  * add this inode to the orphan list so btrfs_orphan_del does
3196                  * the proper thing when we hit it
3197                  */
3198                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3199                         &BTRFS_I(inode)->runtime_flags);
3200                 atomic_inc(&root->orphan_inodes);
3201
3202                 /* if we have links, this was a truncate, lets do that */
3203                 if (inode->i_nlink) {
3204                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3205                                 iput(inode);
3206                                 continue;
3207                         }
3208                         nr_truncate++;
3209
3210                         /* 1 for the orphan item deletion. */
3211                         trans = btrfs_start_transaction(root, 1);
3212                         if (IS_ERR(trans)) {
3213                                 iput(inode);
3214                                 ret = PTR_ERR(trans);
3215                                 goto out;
3216                         }
3217                         ret = btrfs_orphan_add(trans, inode);
3218                         btrfs_end_transaction(trans, root);
3219                         if (ret) {
3220                                 iput(inode);
3221                                 goto out;
3222                         }
3223
3224                         ret = btrfs_truncate(inode);
3225                         if (ret)
3226                                 btrfs_orphan_del(NULL, inode);
3227                 } else {
3228                         nr_unlink++;
3229                 }
3230
3231                 /* this will do delete_inode and everything for us */
3232                 iput(inode);
3233                 if (ret)
3234                         goto out;
3235         }
3236         /* release the path since we're done with it */
3237         btrfs_release_path(path);
3238
3239         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3240
3241         if (root->orphan_block_rsv)
3242                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3243                                         (u64)-1);
3244
3245         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3246                 trans = btrfs_join_transaction(root);
3247                 if (!IS_ERR(trans))
3248                         btrfs_end_transaction(trans, root);
3249         }
3250
3251         if (nr_unlink)
3252                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3253         if (nr_truncate)
3254                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3255
3256 out:
3257         if (ret)
3258                 btrfs_crit(root->fs_info,
3259                         "could not do orphan cleanup %d", ret);
3260         btrfs_free_path(path);
3261         return ret;
3262 }
3263
3264 /*
3265  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3266  * don't find any xattrs, we know there can't be any acls.
3267  *
3268  * slot is the slot the inode is in, objectid is the objectid of the inode
3269  */
3270 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3271                                           int slot, u64 objectid,
3272                                           int *first_xattr_slot)
3273 {
3274         u32 nritems = btrfs_header_nritems(leaf);
3275         struct btrfs_key found_key;
3276         static u64 xattr_access = 0;
3277         static u64 xattr_default = 0;
3278         int scanned = 0;
3279
3280         if (!xattr_access) {
3281                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3282                                         strlen(POSIX_ACL_XATTR_ACCESS));
3283                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3284                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3285         }
3286
3287         slot++;
3288         *first_xattr_slot = -1;
3289         while (slot < nritems) {
3290                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3291
3292                 /* we found a different objectid, there must not be acls */
3293                 if (found_key.objectid != objectid)
3294                         return 0;
3295
3296                 /* we found an xattr, assume we've got an acl */
3297                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3298                         if (*first_xattr_slot == -1)
3299                                 *first_xattr_slot = slot;
3300                         if (found_key.offset == xattr_access ||
3301                             found_key.offset == xattr_default)
3302                                 return 1;
3303                 }
3304
3305                 /*
3306                  * we found a key greater than an xattr key, there can't
3307                  * be any acls later on
3308                  */
3309                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3310                         return 0;
3311
3312                 slot++;
3313                 scanned++;
3314
3315                 /*
3316                  * it goes inode, inode backrefs, xattrs, extents,
3317                  * so if there are a ton of hard links to an inode there can
3318                  * be a lot of backrefs.  Don't waste time searching too hard,
3319                  * this is just an optimization
3320                  */
3321                 if (scanned >= 8)
3322                         break;
3323         }
3324         /* we hit the end of the leaf before we found an xattr or
3325          * something larger than an xattr.  We have to assume the inode
3326          * has acls
3327          */
3328         if (*first_xattr_slot == -1)
3329                 *first_xattr_slot = slot;
3330         return 1;
3331 }
3332
3333 /*
3334  * read an inode from the btree into the in-memory inode
3335  */
3336 static void btrfs_read_locked_inode(struct inode *inode)
3337 {
3338         struct btrfs_path *path;
3339         struct extent_buffer *leaf;
3340         struct btrfs_inode_item *inode_item;
3341         struct btrfs_timespec *tspec;
3342         struct btrfs_root *root = BTRFS_I(inode)->root;
3343         struct btrfs_key location;
3344         unsigned long ptr;
3345         int maybe_acls;
3346         u32 rdev;
3347         int ret;
3348         bool filled = false;
3349         int first_xattr_slot;
3350
3351         ret = btrfs_fill_inode(inode, &rdev);
3352         if (!ret)
3353                 filled = true;
3354
3355         path = btrfs_alloc_path();
3356         if (!path)
3357                 goto make_bad;
3358
3359         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3360
3361         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3362         if (ret)
3363                 goto make_bad;
3364
3365         leaf = path->nodes[0];
3366
3367         if (filled)
3368                 goto cache_index;
3369
3370         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3371                                     struct btrfs_inode_item);
3372         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3373         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3374         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3375         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3376         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3377
3378         tspec = btrfs_inode_atime(inode_item);
3379         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3380         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3381
3382         tspec = btrfs_inode_mtime(inode_item);
3383         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3384         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3385
3386         tspec = btrfs_inode_ctime(inode_item);
3387         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3388         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3389
3390         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3391         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3392         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3393
3394         /*
3395          * If we were modified in the current generation and evicted from memory
3396          * and then re-read we need to do a full sync since we don't have any
3397          * idea about which extents were modified before we were evicted from
3398          * cache.
3399          */
3400         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3401                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3402                         &BTRFS_I(inode)->runtime_flags);
3403
3404         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3405         inode->i_generation = BTRFS_I(inode)->generation;
3406         inode->i_rdev = 0;
3407         rdev = btrfs_inode_rdev(leaf, inode_item);
3408
3409         BTRFS_I(inode)->index_cnt = (u64)-1;
3410         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3411
3412 cache_index:
3413         path->slots[0]++;
3414         if (inode->i_nlink != 1 ||
3415             path->slots[0] >= btrfs_header_nritems(leaf))
3416                 goto cache_acl;
3417
3418         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3419         if (location.objectid != btrfs_ino(inode))
3420                 goto cache_acl;
3421
3422         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3423         if (location.type == BTRFS_INODE_REF_KEY) {
3424                 struct btrfs_inode_ref *ref;
3425
3426                 ref = (struct btrfs_inode_ref *)ptr;
3427                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3428         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3429                 struct btrfs_inode_extref *extref;
3430
3431                 extref = (struct btrfs_inode_extref *)ptr;
3432                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3433                                                                      extref);
3434         }
3435 cache_acl:
3436         /*
3437          * try to precache a NULL acl entry for files that don't have
3438          * any xattrs or acls
3439          */
3440         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3441                                            btrfs_ino(inode), &first_xattr_slot);
3442         if (first_xattr_slot != -1) {
3443                 path->slots[0] = first_xattr_slot;
3444                 ret = btrfs_load_inode_props(inode, path);
3445                 if (ret)
3446                         btrfs_err(root->fs_info,
3447                                   "error loading props for ino %llu (root %llu): %d\n",
3448                                   btrfs_ino(inode),
3449                                   root->root_key.objectid, ret);
3450         }
3451         btrfs_free_path(path);
3452
3453         if (!maybe_acls)
3454                 cache_no_acl(inode);
3455
3456         switch (inode->i_mode & S_IFMT) {
3457         case S_IFREG:
3458                 inode->i_mapping->a_ops = &btrfs_aops;
3459                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3460                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3461                 inode->i_fop = &btrfs_file_operations;
3462                 inode->i_op = &btrfs_file_inode_operations;
3463                 break;
3464         case S_IFDIR:
3465                 inode->i_fop = &btrfs_dir_file_operations;
3466                 if (root == root->fs_info->tree_root)
3467                         inode->i_op = &btrfs_dir_ro_inode_operations;
3468                 else
3469                         inode->i_op = &btrfs_dir_inode_operations;
3470                 break;
3471         case S_IFLNK:
3472                 inode->i_op = &btrfs_symlink_inode_operations;
3473                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3474                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3475                 break;
3476         default:
3477                 inode->i_op = &btrfs_special_inode_operations;
3478                 init_special_inode(inode, inode->i_mode, rdev);
3479                 break;
3480         }
3481
3482         btrfs_update_iflags(inode);
3483         return;
3484
3485 make_bad:
3486         btrfs_free_path(path);
3487         make_bad_inode(inode);
3488 }
3489
3490 /*
3491  * given a leaf and an inode, copy the inode fields into the leaf
3492  */
3493 static void fill_inode_item(struct btrfs_trans_handle *trans,
3494                             struct extent_buffer *leaf,
3495                             struct btrfs_inode_item *item,
3496                             struct inode *inode)
3497 {
3498         struct btrfs_map_token token;
3499
3500         btrfs_init_map_token(&token);
3501
3502         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3503         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3504         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3505                                    &token);
3506         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3507         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3508
3509         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3510                                      inode->i_atime.tv_sec, &token);
3511         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3512                                       inode->i_atime.tv_nsec, &token);
3513
3514         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3515                                      inode->i_mtime.tv_sec, &token);
3516         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3517                                       inode->i_mtime.tv_nsec, &token);
3518
3519         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3520                                      inode->i_ctime.tv_sec, &token);
3521         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3522                                       inode->i_ctime.tv_nsec, &token);
3523
3524         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3525                                      &token);
3526         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3527                                          &token);
3528         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3529         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3530         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3531         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3532         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3533 }
3534
3535 /*
3536  * copy everything in the in-memory inode into the btree.
3537  */
3538 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3539                                 struct btrfs_root *root, struct inode *inode)
3540 {
3541         struct btrfs_inode_item *inode_item;
3542         struct btrfs_path *path;
3543         struct extent_buffer *leaf;
3544         int ret;
3545
3546         path = btrfs_alloc_path();
3547         if (!path)
3548                 return -ENOMEM;
3549
3550         path->leave_spinning = 1;
3551         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3552                                  1);
3553         if (ret) {
3554                 if (ret > 0)
3555                         ret = -ENOENT;
3556                 goto failed;
3557         }
3558
3559         leaf = path->nodes[0];
3560         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3561                                     struct btrfs_inode_item);
3562
3563         fill_inode_item(trans, leaf, inode_item, inode);
3564         btrfs_mark_buffer_dirty(leaf);
3565         btrfs_set_inode_last_trans(trans, inode);
3566         ret = 0;
3567 failed:
3568         btrfs_free_path(path);
3569         return ret;
3570 }
3571
3572 /*
3573  * copy everything in the in-memory inode into the btree.
3574  */
3575 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3576                                 struct btrfs_root *root, struct inode *inode)
3577 {
3578         int ret;
3579
3580         /*
3581          * If the inode is a free space inode, we can deadlock during commit
3582          * if we put it into the delayed code.
3583          *
3584          * The data relocation inode should also be directly updated
3585          * without delay
3586          */
3587         if (!btrfs_is_free_space_inode(inode)
3588             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3589                 btrfs_update_root_times(trans, root);
3590
3591                 ret = btrfs_delayed_update_inode(trans, root, inode);
3592                 if (!ret)
3593                         btrfs_set_inode_last_trans(trans, inode);
3594                 return ret;
3595         }
3596
3597         return btrfs_update_inode_item(trans, root, inode);
3598 }
3599
3600 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3601                                          struct btrfs_root *root,
3602                                          struct inode *inode)
3603 {
3604         int ret;
3605
3606         ret = btrfs_update_inode(trans, root, inode);
3607         if (ret == -ENOSPC)
3608                 return btrfs_update_inode_item(trans, root, inode);
3609         return ret;
3610 }
3611
3612 /*
3613  * unlink helper that gets used here in inode.c and in the tree logging
3614  * recovery code.  It remove a link in a directory with a given name, and
3615  * also drops the back refs in the inode to the directory
3616  */
3617 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3618                                 struct btrfs_root *root,
3619                                 struct inode *dir, struct inode *inode,
3620                                 const char *name, int name_len)
3621 {
3622         struct btrfs_path *path;
3623         int ret = 0;
3624         struct extent_buffer *leaf;
3625         struct btrfs_dir_item *di;
3626         struct btrfs_key key;
3627         u64 index;
3628         u64 ino = btrfs_ino(inode);
3629         u64 dir_ino = btrfs_ino(dir);
3630
3631         path = btrfs_alloc_path();
3632         if (!path) {
3633                 ret = -ENOMEM;
3634                 goto out;
3635         }
3636
3637         path->leave_spinning = 1;
3638         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3639                                     name, name_len, -1);
3640         if (IS_ERR(di)) {
3641                 ret = PTR_ERR(di);
3642                 goto err;
3643         }
3644         if (!di) {
3645                 ret = -ENOENT;
3646                 goto err;
3647         }
3648         leaf = path->nodes[0];
3649         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3650         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3651         if (ret)
3652                 goto err;
3653         btrfs_release_path(path);
3654
3655         /*
3656          * If we don't have dir index, we have to get it by looking up
3657          * the inode ref, since we get the inode ref, remove it directly,
3658          * it is unnecessary to do delayed deletion.
3659          *
3660          * But if we have dir index, needn't search inode ref to get it.
3661          * Since the inode ref is close to the inode item, it is better
3662          * that we delay to delete it, and just do this deletion when
3663          * we update the inode item.
3664          */
3665         if (BTRFS_I(inode)->dir_index) {
3666                 ret = btrfs_delayed_delete_inode_ref(inode);
3667                 if (!ret) {
3668                         index = BTRFS_I(inode)->dir_index;
3669                         goto skip_backref;
3670                 }
3671         }
3672
3673         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3674                                   dir_ino, &index);
3675         if (ret) {
3676                 btrfs_info(root->fs_info,
3677                         "failed to delete reference to %.*s, inode %llu parent %llu",
3678                         name_len, name, ino, dir_ino);
3679                 btrfs_abort_transaction(trans, root, ret);
3680                 goto err;
3681         }
3682 skip_backref:
3683         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3684         if (ret) {
3685                 btrfs_abort_transaction(trans, root, ret);
3686                 goto err;
3687         }
3688
3689         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3690                                          inode, dir_ino);
3691         if (ret != 0 && ret != -ENOENT) {
3692                 btrfs_abort_transaction(trans, root, ret);
3693                 goto err;
3694         }
3695
3696         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3697                                            dir, index);
3698         if (ret == -ENOENT)
3699                 ret = 0;
3700         else if (ret)
3701                 btrfs_abort_transaction(trans, root, ret);
3702 err:
3703         btrfs_free_path(path);
3704         if (ret)
3705                 goto out;
3706
3707         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3708         inode_inc_iversion(inode);
3709         inode_inc_iversion(dir);
3710         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3711         ret = btrfs_update_inode(trans, root, dir);
3712 out:
3713         return ret;
3714 }
3715
3716 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3717                        struct btrfs_root *root,
3718                        struct inode *dir, struct inode *inode,
3719                        const char *name, int name_len)
3720 {
3721         int ret;
3722         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3723         if (!ret) {
3724                 drop_nlink(inode);
3725                 ret = btrfs_update_inode(trans, root, inode);
3726         }
3727         return ret;
3728 }
3729
3730 /*
3731  * helper to start transaction for unlink and rmdir.
3732  *
3733  * unlink and rmdir are special in btrfs, they do not always free space, so
3734  * if we cannot make our reservations the normal way try and see if there is
3735  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3736  * allow the unlink to occur.
3737  */
3738 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3739 {
3740         struct btrfs_trans_handle *trans;
3741         struct btrfs_root *root = BTRFS_I(dir)->root;
3742         int ret;
3743
3744         /*
3745          * 1 for the possible orphan item
3746          * 1 for the dir item
3747          * 1 for the dir index
3748          * 1 for the inode ref
3749          * 1 for the inode
3750          */
3751         trans = btrfs_start_transaction(root, 5);
3752         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3753                 return trans;
3754
3755         if (PTR_ERR(trans) == -ENOSPC) {
3756                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3757
3758                 trans = btrfs_start_transaction(root, 0);
3759                 if (IS_ERR(trans))
3760                         return trans;
3761                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3762                                                &root->fs_info->trans_block_rsv,
3763                                                num_bytes, 5);
3764                 if (ret) {
3765                         btrfs_end_transaction(trans, root);
3766                         return ERR_PTR(ret);
3767                 }
3768                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3769                 trans->bytes_reserved = num_bytes;
3770         }
3771         return trans;
3772 }
3773
3774 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3775 {
3776         struct btrfs_root *root = BTRFS_I(dir)->root;
3777         struct btrfs_trans_handle *trans;
3778         struct inode *inode = dentry->d_inode;
3779         int ret;
3780
3781         trans = __unlink_start_trans(dir);
3782         if (IS_ERR(trans))
3783                 return PTR_ERR(trans);
3784
3785         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3786
3787         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3788                                  dentry->d_name.name, dentry->d_name.len);
3789         if (ret)
3790                 goto out;
3791
3792         if (inode->i_nlink == 0) {
3793                 ret = btrfs_orphan_add(trans, inode);
3794                 if (ret)
3795                         goto out;
3796         }
3797
3798 out:
3799         btrfs_end_transaction(trans, root);
3800         btrfs_btree_balance_dirty(root);
3801         return ret;
3802 }
3803
3804 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3805                         struct btrfs_root *root,
3806                         struct inode *dir, u64 objectid,
3807                         const char *name, int name_len)
3808 {
3809         struct btrfs_path *path;
3810         struct extent_buffer *leaf;
3811         struct btrfs_dir_item *di;
3812         struct btrfs_key key;
3813         u64 index;
3814         int ret;
3815         u64 dir_ino = btrfs_ino(dir);
3816
3817         path = btrfs_alloc_path();
3818         if (!path)
3819                 return -ENOMEM;
3820
3821         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3822                                    name, name_len, -1);
3823         if (IS_ERR_OR_NULL(di)) {
3824                 if (!di)
3825                         ret = -ENOENT;
3826                 else
3827                         ret = PTR_ERR(di);
3828                 goto out;
3829         }
3830
3831         leaf = path->nodes[0];
3832         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3833         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3834         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3835         if (ret) {
3836                 btrfs_abort_transaction(trans, root, ret);
3837                 goto out;
3838         }
3839         btrfs_release_path(path);
3840
3841         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3842                                  objectid, root->root_key.objectid,
3843                                  dir_ino, &index, name, name_len);
3844         if (ret < 0) {
3845                 if (ret != -ENOENT) {
3846                         btrfs_abort_transaction(trans, root, ret);
3847                         goto out;
3848                 }
3849                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3850                                                  name, name_len);
3851                 if (IS_ERR_OR_NULL(di)) {
3852                         if (!di)
3853                                 ret = -ENOENT;
3854                         else
3855                                 ret = PTR_ERR(di);
3856                         btrfs_abort_transaction(trans, root, ret);
3857                         goto out;
3858                 }
3859
3860                 leaf = path->nodes[0];
3861                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3862                 btrfs_release_path(path);
3863                 index = key.offset;
3864         }
3865         btrfs_release_path(path);
3866
3867         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3868         if (ret) {
3869                 btrfs_abort_transaction(trans, root, ret);
3870                 goto out;
3871         }
3872
3873         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3874         inode_inc_iversion(dir);
3875         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3876         ret = btrfs_update_inode_fallback(trans, root, dir);
3877         if (ret)
3878                 btrfs_abort_transaction(trans, root, ret);
3879 out:
3880         btrfs_free_path(path);
3881         return ret;
3882 }
3883
3884 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3885 {
3886         struct inode *inode = dentry->d_inode;
3887         int err = 0;
3888         struct btrfs_root *root = BTRFS_I(dir)->root;
3889         struct btrfs_trans_handle *trans;
3890
3891         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3892                 return -ENOTEMPTY;
3893         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3894                 return -EPERM;
3895
3896         trans = __unlink_start_trans(dir);
3897         if (IS_ERR(trans))
3898                 return PTR_ERR(trans);
3899
3900         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3901                 err = btrfs_unlink_subvol(trans, root, dir,
3902                                           BTRFS_I(inode)->location.objectid,
3903                                           dentry->d_name.name,
3904                                           dentry->d_name.len);
3905                 goto out;
3906         }
3907
3908         err = btrfs_orphan_add(trans, inode);
3909         if (err)
3910                 goto out;
3911
3912         /* now the directory is empty */
3913         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3914                                  dentry->d_name.name, dentry->d_name.len);
3915         if (!err)
3916                 btrfs_i_size_write(inode, 0);
3917 out:
3918         btrfs_end_transaction(trans, root);
3919         btrfs_btree_balance_dirty(root);
3920
3921         return err;
3922 }
3923
3924 /*
3925  * this can truncate away extent items, csum items and directory items.
3926  * It starts at a high offset and removes keys until it can't find
3927  * any higher than new_size
3928  *
3929  * csum items that cross the new i_size are truncated to the new size
3930  * as well.
3931  *
3932  * min_type is the minimum key type to truncate down to.  If set to 0, this
3933  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3934  */
3935 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3936                                struct btrfs_root *root,
3937                                struct inode *inode,
3938                                u64 new_size, u32 min_type)
3939 {
3940         struct btrfs_path *path;
3941         struct extent_buffer *leaf;
3942         struct btrfs_file_extent_item *fi;
3943         struct btrfs_key key;
3944         struct btrfs_key found_key;
3945         u64 extent_start = 0;
3946         u64 extent_num_bytes = 0;
3947         u64 extent_offset = 0;
3948         u64 item_end = 0;
3949         u64 last_size = (u64)-1;
3950         u32 found_type = (u8)-1;
3951         int found_extent;
3952         int del_item;
3953         int pending_del_nr = 0;
3954         int pending_del_slot = 0;
3955         int extent_type = -1;
3956         int ret;
3957         int err = 0;
3958         u64 ino = btrfs_ino(inode);
3959
3960         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3961
3962         path = btrfs_alloc_path();
3963         if (!path)
3964                 return -ENOMEM;
3965         path->reada = -1;
3966
3967         /*
3968          * We want to drop from the next block forward in case this new size is
3969          * not block aligned since we will be keeping the last block of the
3970          * extent just the way it is.
3971          */
3972         if (root->ref_cows || root == root->fs_info->tree_root)
3973                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3974                                         root->sectorsize), (u64)-1, 0);
3975
3976         /*
3977          * This function is also used to drop the items in the log tree before
3978          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3979          * it is used to drop the loged items. So we shouldn't kill the delayed
3980          * items.
3981          */
3982         if (min_type == 0 && root == BTRFS_I(inode)->root)
3983                 btrfs_kill_delayed_inode_items(inode);
3984
3985         key.objectid = ino;
3986         key.offset = (u64)-1;
3987         key.type = (u8)-1;
3988
3989 search_again:
3990         path->leave_spinning = 1;
3991         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3992         if (ret < 0) {
3993                 err = ret;
3994                 goto out;
3995         }
3996
3997         if (ret > 0) {
3998                 /* there are no items in the tree for us to truncate, we're
3999                  * done
4000                  */
4001                 if (path->slots[0] == 0)
4002                         goto out;
4003                 path->slots[0]--;
4004         }
4005
4006         while (1) {
4007                 fi = NULL;
4008                 leaf = path->nodes[0];
4009                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4010                 found_type = btrfs_key_type(&found_key);
4011
4012                 if (found_key.objectid != ino)
4013                         break;
4014
4015                 if (found_type < min_type)
4016                         break;
4017
4018                 item_end = found_key.offset;
4019                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4020                         fi = btrfs_item_ptr(leaf, path->slots[0],
4021                                             struct btrfs_file_extent_item);
4022                         extent_type = btrfs_file_extent_type(leaf, fi);
4023                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4024                                 item_end +=
4025                                     btrfs_file_extent_num_bytes(leaf, fi);
4026                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4027                                 item_end += btrfs_file_extent_inline_len(leaf,
4028                                                          path->slots[0], fi);
4029                         }
4030                         item_end--;
4031                 }
4032                 if (found_type > min_type) {
4033                         del_item = 1;
4034                 } else {
4035                         if (item_end < new_size)
4036                                 break;
4037                         if (found_key.offset >= new_size)
4038                                 del_item = 1;
4039                         else
4040                                 del_item = 0;
4041                 }
4042                 found_extent = 0;
4043                 /* FIXME, shrink the extent if the ref count is only 1 */
4044                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4045                         goto delete;
4046
4047                 if (del_item)
4048                         last_size = found_key.offset;
4049                 else
4050                         last_size = new_size;
4051
4052                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4053                         u64 num_dec;
4054                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4055                         if (!del_item) {
4056                                 u64 orig_num_bytes =
4057                                         btrfs_file_extent_num_bytes(leaf, fi);
4058                                 extent_num_bytes = ALIGN(new_size -
4059                                                 found_key.offset,
4060                                                 root->sectorsize);
4061                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4062                                                          extent_num_bytes);
4063                                 num_dec = (orig_num_bytes -
4064                                            extent_num_bytes);
4065                                 if (root->ref_cows && extent_start != 0)
4066                                         inode_sub_bytes(inode, num_dec);
4067                                 btrfs_mark_buffer_dirty(leaf);
4068                         } else {
4069                                 extent_num_bytes =
4070                                         btrfs_file_extent_disk_num_bytes(leaf,
4071                                                                          fi);
4072                                 extent_offset = found_key.offset -
4073                                         btrfs_file_extent_offset(leaf, fi);
4074
4075                                 /* FIXME blocksize != 4096 */
4076                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4077                                 if (extent_start != 0) {
4078                                         found_extent = 1;
4079                                         if (root->ref_cows)
4080                                                 inode_sub_bytes(inode, num_dec);
4081                                 }
4082                         }
4083                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4084                         /*
4085                          * we can't truncate inline items that have had
4086                          * special encodings
4087                          */
4088                         if (!del_item &&
4089                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4090                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4091                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4092                                 u32 size = new_size - found_key.offset;
4093
4094                                 if (root->ref_cows) {
4095                                         inode_sub_bytes(inode, item_end + 1 -
4096                                                         new_size);
4097                                 }
4098
4099                                 /*
4100                                  * update the ram bytes to properly reflect
4101                                  * the new size of our item
4102                                  */
4103                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4104                                 size =
4105                                     btrfs_file_extent_calc_inline_size(size);
4106                                 btrfs_truncate_item(root, path, size, 1);
4107                         } else if (root->ref_cows) {
4108                                 inode_sub_bytes(inode, item_end + 1 -
4109                                                 found_key.offset);
4110                         }
4111                 }
4112 delete:
4113                 if (del_item) {
4114                         if (!pending_del_nr) {
4115                                 /* no pending yet, add ourselves */
4116                                 pending_del_slot = path->slots[0];
4117                                 pending_del_nr = 1;
4118                         } else if (pending_del_nr &&
4119                                    path->slots[0] + 1 == pending_del_slot) {
4120                                 /* hop on the pending chunk */
4121                                 pending_del_nr++;
4122                                 pending_del_slot = path->slots[0];
4123                         } else {
4124                                 BUG();
4125                         }
4126                 } else {
4127                         break;
4128                 }
4129                 if (found_extent && (root->ref_cows ||
4130                                      root == root->fs_info->tree_root)) {
4131                         btrfs_set_path_blocking(path);
4132                         ret = btrfs_free_extent(trans, root, extent_start,
4133                                                 extent_num_bytes, 0,
4134                                                 btrfs_header_owner(leaf),
4135                                                 ino, extent_offset, 0);
4136                         BUG_ON(ret);
4137                 }
4138
4139                 if (found_type == BTRFS_INODE_ITEM_KEY)
4140                         break;
4141
4142                 if (path->slots[0] == 0 ||
4143                     path->slots[0] != pending_del_slot) {
4144                         if (pending_del_nr) {
4145                                 ret = btrfs_del_items(trans, root, path,
4146                                                 pending_del_slot,
4147                                                 pending_del_nr);
4148                                 if (ret) {
4149                                         btrfs_abort_transaction(trans,
4150                                                                 root, ret);
4151                                         goto error;
4152                                 }
4153                                 pending_del_nr = 0;
4154                         }
4155                         btrfs_release_path(path);
4156                         goto search_again;
4157                 } else {
4158                         path->slots[0]--;
4159                 }
4160         }
4161 out:
4162         if (pending_del_nr) {
4163                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4164                                       pending_del_nr);
4165                 if (ret)
4166                         btrfs_abort_transaction(trans, root, ret);
4167         }
4168 error:
4169         if (last_size != (u64)-1)
4170                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4171         btrfs_free_path(path);
4172         return err;
4173 }
4174
4175 /*
4176  * btrfs_truncate_page - read, zero a chunk and write a page
4177  * @inode - inode that we're zeroing
4178  * @from - the offset to start zeroing
4179  * @len - the length to zero, 0 to zero the entire range respective to the
4180  *      offset
4181  * @front - zero up to the offset instead of from the offset on
4182  *
4183  * This will find the page for the "from" offset and cow the page and zero the
4184  * part we want to zero.  This is used with truncate and hole punching.
4185  */
4186 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4187                         int front)
4188 {
4189         struct address_space *mapping = inode->i_mapping;
4190         struct btrfs_root *root = BTRFS_I(inode)->root;
4191         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4192         struct btrfs_ordered_extent *ordered;
4193         struct extent_state *cached_state = NULL;
4194         char *kaddr;
4195         u32 blocksize = root->sectorsize;
4196         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4197         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4198         struct page *page;
4199         gfp_t mask = btrfs_alloc_write_mask(mapping);
4200         int ret = 0;
4201         u64 page_start;
4202         u64 page_end;
4203
4204         if ((offset & (blocksize - 1)) == 0 &&
4205             (!len || ((len & (blocksize - 1)) == 0)))
4206                 goto out;
4207         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4208         if (ret)
4209                 goto out;
4210
4211 again:
4212         page = find_or_create_page(mapping, index, mask);
4213         if (!page) {
4214                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4215                 ret = -ENOMEM;
4216                 goto out;
4217         }
4218
4219         page_start = page_offset(page);
4220         page_end = page_start + PAGE_CACHE_SIZE - 1;
4221
4222         if (!PageUptodate(page)) {
4223                 ret = btrfs_readpage(NULL, page);
4224                 lock_page(page);
4225                 if (page->mapping != mapping) {
4226                         unlock_page(page);
4227                         page_cache_release(page);
4228                         goto again;
4229                 }
4230                 if (!PageUptodate(page)) {
4231                         ret = -EIO;
4232                         goto out_unlock;
4233                 }
4234         }
4235         wait_on_page_writeback(page);
4236
4237         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4238         set_page_extent_mapped(page);
4239
4240         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4241         if (ordered) {
4242                 unlock_extent_cached(io_tree, page_start, page_end,
4243                                      &cached_state, GFP_NOFS);
4244                 unlock_page(page);
4245                 page_cache_release(page);
4246                 btrfs_start_ordered_extent(inode, ordered, 1);
4247                 btrfs_put_ordered_extent(ordered);
4248                 goto again;
4249         }
4250
4251         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4252                           EXTENT_DIRTY | EXTENT_DELALLOC |
4253                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4254                           0, 0, &cached_state, GFP_NOFS);
4255
4256         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4257                                         &cached_state);
4258         if (ret) {
4259                 unlock_extent_cached(io_tree, page_start, page_end,
4260                                      &cached_state, GFP_NOFS);
4261                 goto out_unlock;
4262         }
4263
4264         if (offset != PAGE_CACHE_SIZE) {
4265                 if (!len)
4266                         len = PAGE_CACHE_SIZE - offset;
4267                 kaddr = kmap(page);
4268                 if (front)
4269                         memset(kaddr, 0, offset);
4270                 else
4271                         memset(kaddr + offset, 0, len);
4272                 flush_dcache_page(page);
4273                 kunmap(page);
4274         }
4275         ClearPageChecked(page);
4276         set_page_dirty(page);
4277         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4278                              GFP_NOFS);
4279
4280 out_unlock:
4281         if (ret)
4282                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4283         unlock_page(page);
4284         page_cache_release(page);
4285 out:
4286         return ret;
4287 }
4288
4289 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4290                              u64 offset, u64 len)
4291 {
4292         struct btrfs_trans_handle *trans;
4293         int ret;
4294
4295         /*
4296          * Still need to make sure the inode looks like it's been updated so
4297          * that any holes get logged if we fsync.
4298          */
4299         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4300                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4301                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4302                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4303                 return 0;
4304         }
4305
4306         /*
4307          * 1 - for the one we're dropping
4308          * 1 - for the one we're adding
4309          * 1 - for updating the inode.
4310          */
4311         trans = btrfs_start_transaction(root, 3);
4312         if (IS_ERR(trans))
4313                 return PTR_ERR(trans);
4314
4315         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4316         if (ret) {
4317                 btrfs_abort_transaction(trans, root, ret);
4318                 btrfs_end_transaction(trans, root);
4319                 return ret;
4320         }
4321
4322         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4323                                        0, 0, len, 0, len, 0, 0, 0);
4324         if (ret)
4325                 btrfs_abort_transaction(trans, root, ret);
4326         else
4327                 btrfs_update_inode(trans, root, inode);
4328         btrfs_end_transaction(trans, root);
4329         return ret;
4330 }
4331
4332 /*
4333  * This function puts in dummy file extents for the area we're creating a hole
4334  * for.  So if we are truncating this file to a larger size we need to insert
4335  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4336  * the range between oldsize and size
4337  */
4338 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4339 {
4340         struct btrfs_root *root = BTRFS_I(inode)->root;
4341         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4342         struct extent_map *em = NULL;
4343         struct extent_state *cached_state = NULL;
4344         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4345         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4346         u64 block_end = ALIGN(size, root->sectorsize);
4347         u64 last_byte;
4348         u64 cur_offset;
4349         u64 hole_size;
4350         int err = 0;
4351
4352         /*
4353          * If our size started in the middle of a page we need to zero out the
4354          * rest of the page before we expand the i_size, otherwise we could
4355          * expose stale data.
4356          */
4357         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4358         if (err)
4359                 return err;
4360
4361         if (size <= hole_start)
4362                 return 0;
4363
4364         while (1) {
4365                 struct btrfs_ordered_extent *ordered;
4366
4367                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4368                                  &cached_state);
4369                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4370                                                      block_end - hole_start);
4371                 if (!ordered)
4372                         break;
4373                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4374                                      &cached_state, GFP_NOFS);
4375                 btrfs_start_ordered_extent(inode, ordered, 1);
4376                 btrfs_put_ordered_extent(ordered);
4377         }
4378
4379         cur_offset = hole_start;
4380         while (1) {
4381                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4382                                 block_end - cur_offset, 0);
4383                 if (IS_ERR(em)) {
4384                         err = PTR_ERR(em);
4385                         em = NULL;
4386                         break;
4387                 }
4388                 last_byte = min(extent_map_end(em), block_end);
4389                 last_byte = ALIGN(last_byte , root->sectorsize);
4390                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4391                         struct extent_map *hole_em;
4392                         hole_size = last_byte - cur_offset;
4393
4394                         err = maybe_insert_hole(root, inode, cur_offset,
4395                                                 hole_size);
4396                         if (err)
4397                                 break;
4398                         btrfs_drop_extent_cache(inode, cur_offset,
4399                                                 cur_offset + hole_size - 1, 0);
4400                         hole_em = alloc_extent_map();
4401                         if (!hole_em) {
4402                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4403                                         &BTRFS_I(inode)->runtime_flags);
4404                                 goto next;
4405                         }
4406                         hole_em->start = cur_offset;
4407                         hole_em->len = hole_size;
4408                         hole_em->orig_start = cur_offset;
4409
4410                         hole_em->block_start = EXTENT_MAP_HOLE;
4411                         hole_em->block_len = 0;
4412                         hole_em->orig_block_len = 0;
4413                         hole_em->ram_bytes = hole_size;
4414                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4415                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4416                         hole_em->generation = root->fs_info->generation;
4417
4418                         while (1) {
4419                                 write_lock(&em_tree->lock);
4420                                 err = add_extent_mapping(em_tree, hole_em, 1);
4421                                 write_unlock(&em_tree->lock);
4422                                 if (err != -EEXIST)
4423                                         break;
4424                                 btrfs_drop_extent_cache(inode, cur_offset,
4425                                                         cur_offset +
4426                                                         hole_size - 1, 0);
4427                         }
4428                         free_extent_map(hole_em);
4429                 }
4430 next:
4431                 free_extent_map(em);
4432                 em = NULL;
4433                 cur_offset = last_byte;
4434                 if (cur_offset >= block_end)
4435                         break;
4436         }
4437         free_extent_map(em);
4438         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4439                              GFP_NOFS);
4440         return err;
4441 }
4442
4443 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4444 {
4445         struct btrfs_root *root = BTRFS_I(inode)->root;
4446         struct btrfs_trans_handle *trans;
4447         loff_t oldsize = i_size_read(inode);
4448         loff_t newsize = attr->ia_size;
4449         int mask = attr->ia_valid;
4450         int ret;
4451
4452         /*
4453          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4454          * special case where we need to update the times despite not having
4455          * these flags set.  For all other operations the VFS set these flags
4456          * explicitly if it wants a timestamp update.
4457          */
4458         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4459                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4460
4461         if (newsize > oldsize) {
4462                 truncate_pagecache(inode, newsize);
4463                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4464                 if (ret)
4465                         return ret;
4466
4467                 trans = btrfs_start_transaction(root, 1);
4468                 if (IS_ERR(trans))
4469                         return PTR_ERR(trans);
4470
4471                 i_size_write(inode, newsize);
4472                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4473                 ret = btrfs_update_inode(trans, root, inode);
4474                 btrfs_end_transaction(trans, root);
4475         } else {
4476
4477                 /*
4478                  * We're truncating a file that used to have good data down to
4479                  * zero. Make sure it gets into the ordered flush list so that
4480                  * any new writes get down to disk quickly.
4481                  */
4482                 if (newsize == 0)
4483                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4484                                 &BTRFS_I(inode)->runtime_flags);
4485
4486                 /*
4487                  * 1 for the orphan item we're going to add
4488                  * 1 for the orphan item deletion.
4489                  */
4490                 trans = btrfs_start_transaction(root, 2);
4491                 if (IS_ERR(trans))
4492                         return PTR_ERR(trans);
4493
4494                 /*
4495                  * We need to do this in case we fail at _any_ point during the
4496                  * actual truncate.  Once we do the truncate_setsize we could
4497                  * invalidate pages which forces any outstanding ordered io to
4498                  * be instantly completed which will give us extents that need
4499                  * to be truncated.  If we fail to get an orphan inode down we
4500                  * could have left over extents that were never meant to live,
4501                  * so we need to garuntee from this point on that everything
4502                  * will be consistent.
4503                  */
4504                 ret = btrfs_orphan_add(trans, inode);
4505                 btrfs_end_transaction(trans, root);
4506                 if (ret)
4507                         return ret;
4508
4509                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4510                 truncate_setsize(inode, newsize);
4511
4512                 /* Disable nonlocked read DIO to avoid the end less truncate */
4513                 btrfs_inode_block_unlocked_dio(inode);
4514                 inode_dio_wait(inode);
4515                 btrfs_inode_resume_unlocked_dio(inode);
4516
4517                 ret = btrfs_truncate(inode);
4518                 if (ret && inode->i_nlink) {
4519                         int err;
4520
4521                         /*
4522                          * failed to truncate, disk_i_size is only adjusted down
4523                          * as we remove extents, so it should represent the true
4524                          * size of the inode, so reset the in memory size and
4525                          * delete our orphan entry.
4526                          */
4527                         trans = btrfs_join_transaction(root);
4528                         if (IS_ERR(trans)) {
4529                                 btrfs_orphan_del(NULL, inode);
4530                                 return ret;
4531                         }
4532                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4533                         err = btrfs_orphan_del(trans, inode);
4534                         if (err)
4535                                 btrfs_abort_transaction(trans, root, err);
4536                         btrfs_end_transaction(trans, root);
4537                 }
4538         }
4539
4540         return ret;
4541 }
4542
4543 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4544 {
4545         struct inode *inode = dentry->d_inode;
4546         struct btrfs_root *root = BTRFS_I(inode)->root;
4547         int err;
4548
4549         if (btrfs_root_readonly(root))
4550                 return -EROFS;
4551
4552         err = inode_change_ok(inode, attr);
4553         if (err)
4554                 return err;
4555
4556         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4557                 err = btrfs_setsize(inode, attr);
4558                 if (err)
4559                         return err;
4560         }
4561
4562         if (attr->ia_valid) {
4563                 setattr_copy(inode, attr);
4564                 inode_inc_iversion(inode);
4565                 err = btrfs_dirty_inode(inode);
4566
4567                 if (!err && attr->ia_valid & ATTR_MODE)
4568                         err = btrfs_acl_chmod(inode);
4569         }
4570
4571         return err;
4572 }
4573
4574 /*
4575  * While truncating the inode pages during eviction, we get the VFS calling
4576  * btrfs_invalidatepage() against each page of the inode. This is slow because
4577  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4578  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4579  * extent_state structures over and over, wasting lots of time.
4580  *
4581  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4582  * those expensive operations on a per page basis and do only the ordered io
4583  * finishing, while we release here the extent_map and extent_state structures,
4584  * without the excessive merging and splitting.
4585  */
4586 static void evict_inode_truncate_pages(struct inode *inode)
4587 {
4588         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4589         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4590         struct rb_node *node;
4591
4592         ASSERT(inode->i_state & I_FREEING);
4593         truncate_inode_pages(&inode->i_data, 0);
4594
4595         write_lock(&map_tree->lock);
4596         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4597                 struct extent_map *em;
4598
4599                 node = rb_first(&map_tree->map);
4600                 em = rb_entry(node, struct extent_map, rb_node);
4601                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4602                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4603                 remove_extent_mapping(map_tree, em);
4604                 free_extent_map(em);
4605         }
4606         write_unlock(&map_tree->lock);
4607
4608         spin_lock(&io_tree->lock);
4609         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4610                 struct extent_state *state;
4611                 struct extent_state *cached_state = NULL;
4612
4613                 node = rb_first(&io_tree->state);
4614                 state = rb_entry(node, struct extent_state, rb_node);
4615                 atomic_inc(&state->refs);
4616                 spin_unlock(&io_tree->lock);
4617
4618                 lock_extent_bits(io_tree, state->start, state->end,
4619                                  0, &cached_state);
4620                 clear_extent_bit(io_tree, state->start, state->end,
4621                                  EXTENT_LOCKED | EXTENT_DIRTY |
4622                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4623                                  EXTENT_DEFRAG, 1, 1,
4624                                  &cached_state, GFP_NOFS);
4625                 free_extent_state(state);
4626
4627                 spin_lock(&io_tree->lock);
4628         }
4629         spin_unlock(&io_tree->lock);
4630 }
4631
4632 void btrfs_evict_inode(struct inode *inode)
4633 {
4634         struct btrfs_trans_handle *trans;
4635         struct btrfs_root *root = BTRFS_I(inode)->root;
4636         struct btrfs_block_rsv *rsv, *global_rsv;
4637         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4638         int ret;
4639
4640         trace_btrfs_inode_evict(inode);
4641
4642         evict_inode_truncate_pages(inode);
4643
4644         if (inode->i_nlink &&
4645             ((btrfs_root_refs(&root->root_item) != 0 &&
4646               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4647              btrfs_is_free_space_inode(inode)))
4648                 goto no_delete;
4649
4650         if (is_bad_inode(inode)) {
4651                 btrfs_orphan_del(NULL, inode);
4652                 goto no_delete;
4653         }
4654         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4655         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4656
4657         if (root->fs_info->log_root_recovering) {
4658                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4659                                  &BTRFS_I(inode)->runtime_flags));
4660                 goto no_delete;
4661         }
4662
4663         if (inode->i_nlink > 0) {
4664                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4665                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4666                 goto no_delete;
4667         }
4668
4669         ret = btrfs_commit_inode_delayed_inode(inode);
4670         if (ret) {
4671                 btrfs_orphan_del(NULL, inode);
4672                 goto no_delete;
4673         }
4674
4675         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4676         if (!rsv) {
4677                 btrfs_orphan_del(NULL, inode);
4678                 goto no_delete;
4679         }
4680         rsv->size = min_size;
4681         rsv->failfast = 1;
4682         global_rsv = &root->fs_info->global_block_rsv;
4683
4684         btrfs_i_size_write(inode, 0);
4685
4686         /*
4687          * This is a bit simpler than btrfs_truncate since we've already
4688          * reserved our space for our orphan item in the unlink, so we just
4689          * need to reserve some slack space in case we add bytes and update
4690          * inode item when doing the truncate.
4691          */
4692         while (1) {
4693                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4694                                              BTRFS_RESERVE_FLUSH_LIMIT);
4695
4696                 /*
4697                  * Try and steal from the global reserve since we will
4698                  * likely not use this space anyway, we want to try as
4699                  * hard as possible to get this to work.
4700                  */
4701                 if (ret)
4702                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4703
4704                 if (ret) {
4705                         btrfs_warn(root->fs_info,
4706                                 "Could not get space for a delete, will truncate on mount %d",
4707                                 ret);
4708                         btrfs_orphan_del(NULL, inode);
4709                         btrfs_free_block_rsv(root, rsv);
4710                         goto no_delete;
4711                 }
4712
4713                 trans = btrfs_join_transaction(root);
4714                 if (IS_ERR(trans)) {
4715                         btrfs_orphan_del(NULL, inode);
4716                         btrfs_free_block_rsv(root, rsv);
4717                         goto no_delete;
4718                 }
4719
4720                 trans->block_rsv = rsv;
4721
4722                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4723                 if (ret != -ENOSPC)
4724                         break;
4725
4726                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4727                 btrfs_end_transaction(trans, root);
4728                 trans = NULL;
4729                 btrfs_btree_balance_dirty(root);
4730         }
4731
4732         btrfs_free_block_rsv(root, rsv);
4733
4734         /*
4735          * Errors here aren't a big deal, it just means we leave orphan items
4736          * in the tree.  They will be cleaned up on the next mount.
4737          */
4738         if (ret == 0) {
4739                 trans->block_rsv = root->orphan_block_rsv;
4740                 btrfs_orphan_del(trans, inode);
4741         } else {
4742                 btrfs_orphan_del(NULL, inode);
4743         }
4744
4745         trans->block_rsv = &root->fs_info->trans_block_rsv;
4746         if (!(root == root->fs_info->tree_root ||
4747               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4748                 btrfs_return_ino(root, btrfs_ino(inode));
4749
4750         btrfs_end_transaction(trans, root);
4751         btrfs_btree_balance_dirty(root);
4752 no_delete:
4753         btrfs_remove_delayed_node(inode);
4754         clear_inode(inode);
4755         return;
4756 }
4757
4758 /*
4759  * this returns the key found in the dir entry in the location pointer.
4760  * If no dir entries were found, location->objectid is 0.
4761  */
4762 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4763                                struct btrfs_key *location)
4764 {
4765         const char *name = dentry->d_name.name;
4766         int namelen = dentry->d_name.len;
4767         struct btrfs_dir_item *di;
4768         struct btrfs_path *path;
4769         struct btrfs_root *root = BTRFS_I(dir)->root;
4770         int ret = 0;
4771
4772         path = btrfs_alloc_path();
4773         if (!path)
4774                 return -ENOMEM;
4775
4776         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4777                                     namelen, 0);
4778         if (IS_ERR(di))
4779                 ret = PTR_ERR(di);
4780
4781         if (IS_ERR_OR_NULL(di))
4782                 goto out_err;
4783
4784         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4785 out:
4786         btrfs_free_path(path);
4787         return ret;
4788 out_err:
4789         location->objectid = 0;
4790         goto out;
4791 }
4792
4793 /*
4794  * when we hit a tree root in a directory, the btrfs part of the inode
4795  * needs to be changed to reflect the root directory of the tree root.  This
4796  * is kind of like crossing a mount point.
4797  */
4798 static int fixup_tree_root_location(struct btrfs_root *root,
4799                                     struct inode *dir,
4800                                     struct dentry *dentry,
4801                                     struct btrfs_key *location,
4802                                     struct btrfs_root **sub_root)
4803 {
4804         struct btrfs_path *path;
4805         struct btrfs_root *new_root;
4806         struct btrfs_root_ref *ref;
4807         struct extent_buffer *leaf;
4808         int ret;
4809         int err = 0;
4810
4811         path = btrfs_alloc_path();
4812         if (!path) {
4813                 err = -ENOMEM;
4814                 goto out;
4815         }
4816
4817         err = -ENOENT;
4818         ret = btrfs_find_item(root->fs_info->tree_root, path,
4819                                 BTRFS_I(dir)->root->root_key.objectid,
4820                                 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4821         if (ret) {
4822                 if (ret < 0)
4823                         err = ret;
4824                 goto out;
4825         }
4826
4827         leaf = path->nodes[0];
4828         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4829         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4830             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4831                 goto out;
4832
4833         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4834                                    (unsigned long)(ref + 1),
4835                                    dentry->d_name.len);
4836         if (ret)
4837                 goto out;
4838
4839         btrfs_release_path(path);
4840
4841         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4842         if (IS_ERR(new_root)) {
4843                 err = PTR_ERR(new_root);
4844                 goto out;
4845         }
4846
4847         *sub_root = new_root;
4848         location->objectid = btrfs_root_dirid(&new_root->root_item);
4849         location->type = BTRFS_INODE_ITEM_KEY;
4850         location->offset = 0;
4851         err = 0;
4852 out:
4853         btrfs_free_path(path);
4854         return err;
4855 }
4856
4857 static void inode_tree_add(struct inode *inode)
4858 {
4859         struct btrfs_root *root = BTRFS_I(inode)->root;
4860         struct btrfs_inode *entry;
4861         struct rb_node **p;
4862         struct rb_node *parent;
4863         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4864         u64 ino = btrfs_ino(inode);
4865
4866         if (inode_unhashed(inode))
4867                 return;
4868         parent = NULL;
4869         spin_lock(&root->inode_lock);
4870         p = &root->inode_tree.rb_node;
4871         while (*p) {
4872                 parent = *p;
4873                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4874
4875                 if (ino < btrfs_ino(&entry->vfs_inode))
4876                         p = &parent->rb_left;
4877                 else if (ino > btrfs_ino(&entry->vfs_inode))
4878                         p = &parent->rb_right;
4879                 else {
4880                         WARN_ON(!(entry->vfs_inode.i_state &
4881                                   (I_WILL_FREE | I_FREEING)));
4882                         rb_replace_node(parent, new, &root->inode_tree);
4883                         RB_CLEAR_NODE(parent);
4884                         spin_unlock(&root->inode_lock);
4885                         return;
4886                 }
4887         }
4888         rb_link_node(new, parent, p);
4889         rb_insert_color(new, &root->inode_tree);
4890         spin_unlock(&root->inode_lock);
4891 }
4892
4893 static void inode_tree_del(struct inode *inode)
4894 {
4895         struct btrfs_root *root = BTRFS_I(inode)->root;
4896         int empty = 0;
4897
4898         spin_lock(&root->inode_lock);
4899         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4900                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4901                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4902                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4903         }
4904         spin_unlock(&root->inode_lock);
4905
4906         if (empty && btrfs_root_refs(&root->root_item) == 0) {
4907                 synchronize_srcu(&root->fs_info->subvol_srcu);
4908                 spin_lock(&root->inode_lock);
4909                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4910                 spin_unlock(&root->inode_lock);
4911                 if (empty)
4912                         btrfs_add_dead_root(root);
4913         }
4914 }
4915
4916 void btrfs_invalidate_inodes(struct btrfs_root *root)
4917 {
4918         struct rb_node *node;
4919         struct rb_node *prev;
4920         struct btrfs_inode *entry;
4921         struct inode *inode;
4922         u64 objectid = 0;
4923
4924         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4925
4926         spin_lock(&root->inode_lock);
4927 again:
4928         node = root->inode_tree.rb_node;
4929         prev = NULL;
4930         while (node) {
4931                 prev = node;
4932                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4933
4934                 if (objectid < btrfs_ino(&entry->vfs_inode))
4935                         node = node->rb_left;
4936                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4937                         node = node->rb_right;
4938                 else
4939                         break;
4940         }
4941         if (!node) {
4942                 while (prev) {
4943                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4944                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4945                                 node = prev;
4946                                 break;
4947                         }
4948                         prev = rb_next(prev);
4949                 }
4950         }
4951         while (node) {
4952                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4953                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4954                 inode = igrab(&entry->vfs_inode);
4955                 if (inode) {
4956                         spin_unlock(&root->inode_lock);
4957                         if (atomic_read(&inode->i_count) > 1)
4958                                 d_prune_aliases(inode);
4959                         /*
4960                          * btrfs_drop_inode will have it removed from
4961                          * the inode cache when its usage count
4962                          * hits zero.
4963                          */
4964                         iput(inode);
4965                         cond_resched();
4966                         spin_lock(&root->inode_lock);
4967                         goto again;
4968                 }
4969
4970                 if (cond_resched_lock(&root->inode_lock))
4971                         goto again;
4972
4973                 node = rb_next(node);
4974         }
4975         spin_unlock(&root->inode_lock);
4976 }
4977
4978 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4979 {
4980         struct btrfs_iget_args *args = p;
4981         inode->i_ino = args->location->objectid;
4982         memcpy(&BTRFS_I(inode)->location, args->location,
4983                sizeof(*args->location));
4984         BTRFS_I(inode)->root = args->root;
4985         return 0;
4986 }
4987
4988 static int btrfs_find_actor(struct inode *inode, void *opaque)
4989 {
4990         struct btrfs_iget_args *args = opaque;
4991         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
4992                 args->root == BTRFS_I(inode)->root;
4993 }
4994
4995 static struct inode *btrfs_iget_locked(struct super_block *s,
4996                                        struct btrfs_key *location,
4997                                        struct btrfs_root *root)
4998 {
4999         struct inode *inode;
5000         struct btrfs_iget_args args;
5001         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5002
5003         args.location = location;
5004         args.root = root;
5005
5006         inode = iget5_locked(s, hashval, btrfs_find_actor,
5007                              btrfs_init_locked_inode,
5008                              (void *)&args);
5009         return inode;
5010 }
5011
5012 /* Get an inode object given its location and corresponding root.
5013  * Returns in *is_new if the inode was read from disk
5014  */
5015 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5016                          struct btrfs_root *root, int *new)
5017 {
5018         struct inode *inode;
5019
5020         inode = btrfs_iget_locked(s, location, root);
5021         if (!inode)
5022                 return ERR_PTR(-ENOMEM);
5023
5024         if (inode->i_state & I_NEW) {
5025                 btrfs_read_locked_inode(inode);
5026                 if (!is_bad_inode(inode)) {
5027                         inode_tree_add(inode);
5028                         unlock_new_inode(inode);
5029                         if (new)
5030                                 *new = 1;
5031                 } else {
5032                         unlock_new_inode(inode);
5033                         iput(inode);
5034                         inode = ERR_PTR(-ESTALE);
5035                 }
5036         }
5037
5038         return inode;
5039 }
5040
5041 static struct inode *new_simple_dir(struct super_block *s,
5042                                     struct btrfs_key *key,
5043                                     struct btrfs_root *root)
5044 {
5045         struct inode *inode = new_inode(s);
5046
5047         if (!inode)
5048                 return ERR_PTR(-ENOMEM);
5049
5050         BTRFS_I(inode)->root = root;
5051         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5052         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5053
5054         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5055         inode->i_op = &btrfs_dir_ro_inode_operations;
5056         inode->i_fop = &simple_dir_operations;
5057         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5058         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5059
5060         return inode;
5061 }
5062
5063 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5064 {
5065         struct inode *inode;
5066         struct btrfs_root *root = BTRFS_I(dir)->root;
5067         struct btrfs_root *sub_root = root;
5068         struct btrfs_key location;
5069         int index;
5070         int ret = 0;
5071
5072         if (dentry->d_name.len > BTRFS_NAME_LEN)
5073                 return ERR_PTR(-ENAMETOOLONG);
5074
5075         ret = btrfs_inode_by_name(dir, dentry, &location);
5076         if (ret < 0)
5077                 return ERR_PTR(ret);
5078
5079         if (location.objectid == 0)
5080                 return ERR_PTR(-ENOENT);
5081
5082         if (location.type == BTRFS_INODE_ITEM_KEY) {
5083                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5084                 return inode;
5085         }
5086
5087         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5088
5089         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5090         ret = fixup_tree_root_location(root, dir, dentry,
5091                                        &location, &sub_root);
5092         if (ret < 0) {
5093                 if (ret != -ENOENT)
5094                         inode = ERR_PTR(ret);
5095                 else
5096                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5097         } else {
5098                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5099         }
5100         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5101
5102         if (!IS_ERR(inode) && root != sub_root) {
5103                 down_read(&root->fs_info->cleanup_work_sem);
5104                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5105                         ret = btrfs_orphan_cleanup(sub_root);
5106                 up_read(&root->fs_info->cleanup_work_sem);
5107                 if (ret) {
5108                         iput(inode);
5109                         inode = ERR_PTR(ret);
5110                 }
5111         }
5112
5113         return inode;
5114 }
5115
5116 static int btrfs_dentry_delete(const struct dentry *dentry)
5117 {
5118         struct btrfs_root *root;
5119         struct inode *inode = dentry->d_inode;
5120
5121         if (!inode && !IS_ROOT(dentry))
5122                 inode = dentry->d_parent->d_inode;
5123
5124         if (inode) {
5125                 root = BTRFS_I(inode)->root;
5126                 if (btrfs_root_refs(&root->root_item) == 0)
5127                         return 1;
5128
5129                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5130                         return 1;
5131         }
5132         return 0;
5133 }
5134
5135 static void btrfs_dentry_release(struct dentry *dentry)
5136 {
5137         if (dentry->d_fsdata)
5138                 kfree(dentry->d_fsdata);
5139 }
5140
5141 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5142                                    unsigned int flags)
5143 {
5144         struct inode *inode;
5145
5146         inode = btrfs_lookup_dentry(dir, dentry);
5147         if (IS_ERR(inode)) {
5148                 if (PTR_ERR(inode) == -ENOENT)
5149                         inode = NULL;
5150                 else
5151                         return ERR_CAST(inode);
5152         }
5153
5154         return d_materialise_unique(dentry, inode);
5155 }
5156
5157 unsigned char btrfs_filetype_table[] = {
5158         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5159 };
5160
5161 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5162 {
5163         struct inode *inode = file_inode(file);
5164         struct btrfs_root *root = BTRFS_I(inode)->root;
5165         struct btrfs_item *item;
5166         struct btrfs_dir_item *di;
5167         struct btrfs_key key;
5168         struct btrfs_key found_key;
5169         struct btrfs_path *path;
5170         struct list_head ins_list;
5171         struct list_head del_list;
5172         int ret;
5173         struct extent_buffer *leaf;
5174         int slot;
5175         unsigned char d_type;
5176         int over = 0;
5177         u32 di_cur;
5178         u32 di_total;
5179         u32 di_len;
5180         int key_type = BTRFS_DIR_INDEX_KEY;
5181         char tmp_name[32];
5182         char *name_ptr;
5183         int name_len;
5184         int is_curr = 0;        /* ctx->pos points to the current index? */
5185
5186         /* FIXME, use a real flag for deciding about the key type */
5187         if (root->fs_info->tree_root == root)
5188                 key_type = BTRFS_DIR_ITEM_KEY;
5189
5190         if (!dir_emit_dots(file, ctx))
5191                 return 0;
5192
5193         path = btrfs_alloc_path();
5194         if (!path)
5195                 return -ENOMEM;
5196
5197         path->reada = 1;
5198
5199         if (key_type == BTRFS_DIR_INDEX_KEY) {
5200                 INIT_LIST_HEAD(&ins_list);
5201                 INIT_LIST_HEAD(&del_list);
5202                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5203         }
5204
5205         btrfs_set_key_type(&key, key_type);
5206         key.offset = ctx->pos;
5207         key.objectid = btrfs_ino(inode);
5208
5209         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5210         if (ret < 0)
5211                 goto err;
5212
5213         while (1) {
5214                 leaf = path->nodes[0];
5215                 slot = path->slots[0];
5216                 if (slot >= btrfs_header_nritems(leaf)) {
5217                         ret = btrfs_next_leaf(root, path);
5218                         if (ret < 0)
5219                                 goto err;
5220                         else if (ret > 0)
5221                                 break;
5222                         continue;
5223                 }
5224
5225                 item = btrfs_item_nr(slot);
5226                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5227
5228                 if (found_key.objectid != key.objectid)
5229                         break;
5230                 if (btrfs_key_type(&found_key) != key_type)
5231                         break;
5232                 if (found_key.offset < ctx->pos)
5233                         goto next;
5234                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5235                     btrfs_should_delete_dir_index(&del_list,
5236                                                   found_key.offset))
5237                         goto next;
5238
5239                 ctx->pos = found_key.offset;
5240                 is_curr = 1;
5241
5242                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5243                 di_cur = 0;
5244                 di_total = btrfs_item_size(leaf, item);
5245
5246                 while (di_cur < di_total) {
5247                         struct btrfs_key location;
5248
5249                         if (verify_dir_item(root, leaf, di))
5250                                 break;
5251
5252                         name_len = btrfs_dir_name_len(leaf, di);
5253                         if (name_len <= sizeof(tmp_name)) {
5254                                 name_ptr = tmp_name;
5255                         } else {
5256                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5257                                 if (!name_ptr) {
5258                                         ret = -ENOMEM;
5259                                         goto err;
5260                                 }
5261                         }
5262                         read_extent_buffer(leaf, name_ptr,
5263                                            (unsigned long)(di + 1), name_len);
5264
5265                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5266                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5267
5268
5269                         /* is this a reference to our own snapshot? If so
5270                          * skip it.
5271                          *
5272                          * In contrast to old kernels, we insert the snapshot's
5273                          * dir item and dir index after it has been created, so
5274                          * we won't find a reference to our own snapshot. We
5275                          * still keep the following code for backward
5276                          * compatibility.
5277                          */
5278                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5279                             location.objectid == root->root_key.objectid) {
5280                                 over = 0;
5281                                 goto skip;
5282                         }
5283                         over = !dir_emit(ctx, name_ptr, name_len,
5284                                        location.objectid, d_type);
5285
5286 skip:
5287                         if (name_ptr != tmp_name)
5288                                 kfree(name_ptr);
5289
5290                         if (over)
5291                                 goto nopos;
5292                         di_len = btrfs_dir_name_len(leaf, di) +
5293                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5294                         di_cur += di_len;
5295                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5296                 }
5297 next:
5298                 path->slots[0]++;
5299         }
5300
5301         if (key_type == BTRFS_DIR_INDEX_KEY) {
5302                 if (is_curr)
5303                         ctx->pos++;
5304                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5305                 if (ret)
5306                         goto nopos;
5307         }
5308
5309         /* Reached end of directory/root. Bump pos past the last item. */
5310         ctx->pos++;
5311
5312         /*
5313          * Stop new entries from being returned after we return the last
5314          * entry.
5315          *
5316          * New directory entries are assigned a strictly increasing
5317          * offset.  This means that new entries created during readdir
5318          * are *guaranteed* to be seen in the future by that readdir.
5319          * This has broken buggy programs which operate on names as
5320          * they're returned by readdir.  Until we re-use freed offsets
5321          * we have this hack to stop new entries from being returned
5322          * under the assumption that they'll never reach this huge
5323          * offset.
5324          *
5325          * This is being careful not to overflow 32bit loff_t unless the
5326          * last entry requires it because doing so has broken 32bit apps
5327          * in the past.
5328          */
5329         if (key_type == BTRFS_DIR_INDEX_KEY) {
5330                 if (ctx->pos >= INT_MAX)
5331                         ctx->pos = LLONG_MAX;
5332                 else
5333                         ctx->pos = INT_MAX;
5334         }
5335 nopos:
5336         ret = 0;
5337 err:
5338         if (key_type == BTRFS_DIR_INDEX_KEY)
5339                 btrfs_put_delayed_items(&ins_list, &del_list);
5340         btrfs_free_path(path);
5341         return ret;
5342 }
5343
5344 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5345 {
5346         struct btrfs_root *root = BTRFS_I(inode)->root;
5347         struct btrfs_trans_handle *trans;
5348         int ret = 0;
5349         bool nolock = false;
5350
5351         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5352                 return 0;
5353
5354         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5355                 nolock = true;
5356
5357         if (wbc->sync_mode == WB_SYNC_ALL) {
5358                 if (nolock)
5359                         trans = btrfs_join_transaction_nolock(root);
5360                 else
5361                         trans = btrfs_join_transaction(root);
5362                 if (IS_ERR(trans))
5363                         return PTR_ERR(trans);
5364                 ret = btrfs_commit_transaction(trans, root);
5365         }
5366         return ret;
5367 }
5368
5369 /*
5370  * This is somewhat expensive, updating the tree every time the
5371  * inode changes.  But, it is most likely to find the inode in cache.
5372  * FIXME, needs more benchmarking...there are no reasons other than performance
5373  * to keep or drop this code.
5374  */
5375 static int btrfs_dirty_inode(struct inode *inode)
5376 {
5377         struct btrfs_root *root = BTRFS_I(inode)->root;
5378         struct btrfs_trans_handle *trans;
5379         int ret;
5380
5381         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5382                 return 0;
5383
5384         trans = btrfs_join_transaction(root);
5385         if (IS_ERR(trans))
5386                 return PTR_ERR(trans);
5387
5388         ret = btrfs_update_inode(trans, root, inode);
5389         if (ret && ret == -ENOSPC) {
5390                 /* whoops, lets try again with the full transaction */
5391                 btrfs_end_transaction(trans, root);
5392                 trans = btrfs_start_transaction(root, 1);
5393                 if (IS_ERR(trans))
5394                         return PTR_ERR(trans);
5395
5396                 ret = btrfs_update_inode(trans, root, inode);
5397         }
5398         btrfs_end_transaction(trans, root);
5399         if (BTRFS_I(inode)->delayed_node)
5400                 btrfs_balance_delayed_items(root);
5401
5402         return ret;
5403 }
5404
5405 /*
5406  * This is a copy of file_update_time.  We need this so we can return error on
5407  * ENOSPC for updating the inode in the case of file write and mmap writes.
5408  */
5409 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5410                              int flags)
5411 {
5412         struct btrfs_root *root = BTRFS_I(inode)->root;
5413
5414         if (btrfs_root_readonly(root))
5415                 return -EROFS;
5416
5417         if (flags & S_VERSION)
5418                 inode_inc_iversion(inode);
5419         if (flags & S_CTIME)
5420                 inode->i_ctime = *now;
5421         if (flags & S_MTIME)
5422                 inode->i_mtime = *now;
5423         if (flags & S_ATIME)
5424                 inode->i_atime = *now;
5425         return btrfs_dirty_inode(inode);
5426 }
5427
5428 /*
5429  * find the highest existing sequence number in a directory
5430  * and then set the in-memory index_cnt variable to reflect
5431  * free sequence numbers
5432  */
5433 static int btrfs_set_inode_index_count(struct inode *inode)
5434 {
5435         struct btrfs_root *root = BTRFS_I(inode)->root;
5436         struct btrfs_key key, found_key;
5437         struct btrfs_path *path;
5438         struct extent_buffer *leaf;
5439         int ret;
5440
5441         key.objectid = btrfs_ino(inode);
5442         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5443         key.offset = (u64)-1;
5444
5445         path = btrfs_alloc_path();
5446         if (!path)
5447                 return -ENOMEM;
5448
5449         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5450         if (ret < 0)
5451                 goto out;
5452         /* FIXME: we should be able to handle this */
5453         if (ret == 0)
5454                 goto out;
5455         ret = 0;
5456
5457         /*
5458          * MAGIC NUMBER EXPLANATION:
5459          * since we search a directory based on f_pos we have to start at 2
5460          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5461          * else has to start at 2
5462          */
5463         if (path->slots[0] == 0) {
5464                 BTRFS_I(inode)->index_cnt = 2;
5465                 goto out;
5466         }
5467
5468         path->slots[0]--;
5469
5470         leaf = path->nodes[0];
5471         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5472
5473         if (found_key.objectid != btrfs_ino(inode) ||
5474             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5475                 BTRFS_I(inode)->index_cnt = 2;
5476                 goto out;
5477         }
5478
5479         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5480 out:
5481         btrfs_free_path(path);
5482         return ret;
5483 }
5484
5485 /*
5486  * helper to find a free sequence number in a given directory.  This current
5487  * code is very simple, later versions will do smarter things in the btree
5488  */
5489 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5490 {
5491         int ret = 0;
5492
5493         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5494                 ret = btrfs_inode_delayed_dir_index_count(dir);
5495                 if (ret) {
5496                         ret = btrfs_set_inode_index_count(dir);
5497                         if (ret)
5498                                 return ret;
5499                 }
5500         }
5501
5502         *index = BTRFS_I(dir)->index_cnt;
5503         BTRFS_I(dir)->index_cnt++;
5504
5505         return ret;
5506 }
5507
5508 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5509                                      struct btrfs_root *root,
5510                                      struct inode *dir,
5511                                      const char *name, int name_len,
5512                                      u64 ref_objectid, u64 objectid,
5513                                      umode_t mode, u64 *index)
5514 {
5515         struct inode *inode;
5516         struct btrfs_inode_item *inode_item;
5517         struct btrfs_key *location;
5518         struct btrfs_path *path;
5519         struct btrfs_inode_ref *ref;
5520         struct btrfs_key key[2];
5521         u32 sizes[2];
5522         unsigned long ptr;
5523         int ret;
5524
5525         path = btrfs_alloc_path();
5526         if (!path)
5527                 return ERR_PTR(-ENOMEM);
5528
5529         inode = new_inode(root->fs_info->sb);
5530         if (!inode) {
5531                 btrfs_free_path(path);
5532                 return ERR_PTR(-ENOMEM);
5533         }
5534
5535         /*
5536          * we have to initialize this early, so we can reclaim the inode
5537          * number if we fail afterwards in this function.
5538          */
5539         inode->i_ino = objectid;
5540
5541         if (dir) {
5542                 trace_btrfs_inode_request(dir);
5543
5544                 ret = btrfs_set_inode_index(dir, index);
5545                 if (ret) {
5546                         btrfs_free_path(path);
5547                         iput(inode);
5548                         return ERR_PTR(ret);
5549                 }
5550         }
5551         /*
5552          * index_cnt is ignored for everything but a dir,
5553          * btrfs_get_inode_index_count has an explanation for the magic
5554          * number
5555          */
5556         BTRFS_I(inode)->index_cnt = 2;
5557         BTRFS_I(inode)->dir_index = *index;
5558         BTRFS_I(inode)->root = root;
5559         BTRFS_I(inode)->generation = trans->transid;
5560         inode->i_generation = BTRFS_I(inode)->generation;
5561
5562         /*
5563          * We could have gotten an inode number from somebody who was fsynced
5564          * and then removed in this same transaction, so let's just set full
5565          * sync since it will be a full sync anyway and this will blow away the
5566          * old info in the log.
5567          */
5568         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5569
5570         key[0].objectid = objectid;
5571         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5572         key[0].offset = 0;
5573
5574         /*
5575          * Start new inodes with an inode_ref. This is slightly more
5576          * efficient for small numbers of hard links since they will
5577          * be packed into one item. Extended refs will kick in if we
5578          * add more hard links than can fit in the ref item.
5579          */
5580         key[1].objectid = objectid;
5581         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5582         key[1].offset = ref_objectid;
5583
5584         sizes[0] = sizeof(struct btrfs_inode_item);
5585         sizes[1] = name_len + sizeof(*ref);
5586
5587         path->leave_spinning = 1;
5588         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5589         if (ret != 0)
5590                 goto fail;
5591
5592         inode_init_owner(inode, dir, mode);
5593         inode_set_bytes(inode, 0);
5594         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5595         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5596                                   struct btrfs_inode_item);
5597         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5598                              sizeof(*inode_item));
5599         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5600
5601         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5602                              struct btrfs_inode_ref);
5603         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5604         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5605         ptr = (unsigned long)(ref + 1);
5606         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5607
5608         btrfs_mark_buffer_dirty(path->nodes[0]);
5609         btrfs_free_path(path);
5610
5611         location = &BTRFS_I(inode)->location;
5612         location->objectid = objectid;
5613         location->offset = 0;
5614         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5615
5616         btrfs_inherit_iflags(inode, dir);
5617
5618         if (S_ISREG(mode)) {
5619                 if (btrfs_test_opt(root, NODATASUM))
5620                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5621                 if (btrfs_test_opt(root, NODATACOW))
5622                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5623                                 BTRFS_INODE_NODATASUM;
5624         }
5625
5626         btrfs_insert_inode_hash(inode);
5627         inode_tree_add(inode);
5628
5629         trace_btrfs_inode_new(inode);
5630         btrfs_set_inode_last_trans(trans, inode);
5631
5632         btrfs_update_root_times(trans, root);
5633
5634         ret = btrfs_inode_inherit_props(trans, inode, dir);
5635         if (ret)
5636                 btrfs_err(root->fs_info,
5637                           "error inheriting props for ino %llu (root %llu): %d",
5638                           btrfs_ino(inode), root->root_key.objectid, ret);
5639
5640         return inode;
5641 fail:
5642         if (dir)
5643                 BTRFS_I(dir)->index_cnt--;
5644         btrfs_free_path(path);
5645         iput(inode);
5646         return ERR_PTR(ret);
5647 }
5648
5649 static inline u8 btrfs_inode_type(struct inode *inode)
5650 {
5651         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5652 }
5653
5654 /*
5655  * utility function to add 'inode' into 'parent_inode' with
5656  * a give name and a given sequence number.
5657  * if 'add_backref' is true, also insert a backref from the
5658  * inode to the parent directory.
5659  */
5660 int btrfs_add_link(struct btrfs_trans_handle *trans,
5661                    struct inode *parent_inode, struct inode *inode,
5662                    const char *name, int name_len, int add_backref, u64 index)
5663 {
5664         int ret = 0;
5665         struct btrfs_key key;
5666         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5667         u64 ino = btrfs_ino(inode);
5668         u64 parent_ino = btrfs_ino(parent_inode);
5669
5670         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5671                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5672         } else {
5673                 key.objectid = ino;
5674                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5675                 key.offset = 0;
5676         }
5677
5678         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5679                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5680                                          key.objectid, root->root_key.objectid,
5681                                          parent_ino, index, name, name_len);
5682         } else if (add_backref) {
5683                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5684                                              parent_ino, index);
5685         }
5686
5687         /* Nothing to clean up yet */
5688         if (ret)
5689                 return ret;
5690
5691         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5692                                     parent_inode, &key,
5693                                     btrfs_inode_type(inode), index);
5694         if (ret == -EEXIST || ret == -EOVERFLOW)
5695                 goto fail_dir_item;
5696         else if (ret) {
5697                 btrfs_abort_transaction(trans, root, ret);
5698                 return ret;
5699         }
5700
5701         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5702                            name_len * 2);
5703         inode_inc_iversion(parent_inode);
5704         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5705         ret = btrfs_update_inode(trans, root, parent_inode);
5706         if (ret)
5707                 btrfs_abort_transaction(trans, root, ret);
5708         return ret;
5709
5710 fail_dir_item:
5711         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5712                 u64 local_index;
5713                 int err;
5714                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5715                                  key.objectid, root->root_key.objectid,
5716                                  parent_ino, &local_index, name, name_len);
5717
5718         } else if (add_backref) {
5719                 u64 local_index;
5720                 int err;
5721
5722                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5723                                           ino, parent_ino, &local_index);
5724         }
5725         return ret;
5726 }
5727
5728 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5729                             struct inode *dir, struct dentry *dentry,
5730                             struct inode *inode, int backref, u64 index)
5731 {
5732         int err = btrfs_add_link(trans, dir, inode,
5733                                  dentry->d_name.name, dentry->d_name.len,
5734                                  backref, index);
5735         if (err > 0)
5736                 err = -EEXIST;
5737         return err;
5738 }
5739
5740 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5741                         umode_t mode, dev_t rdev)
5742 {
5743         struct btrfs_trans_handle *trans;
5744         struct btrfs_root *root = BTRFS_I(dir)->root;
5745         struct inode *inode = NULL;
5746         int err;
5747         int drop_inode = 0;
5748         u64 objectid;
5749         u64 index = 0;
5750
5751         if (!new_valid_dev(rdev))
5752                 return -EINVAL;
5753
5754         /*
5755          * 2 for inode item and ref
5756          * 2 for dir items
5757          * 1 for xattr if selinux is on
5758          */
5759         trans = btrfs_start_transaction(root, 5);
5760         if (IS_ERR(trans))
5761                 return PTR_ERR(trans);
5762
5763         err = btrfs_find_free_ino(root, &objectid);
5764         if (err)
5765                 goto out_unlock;
5766
5767         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5768                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5769                                 mode, &index);
5770         if (IS_ERR(inode)) {
5771                 err = PTR_ERR(inode);
5772                 goto out_unlock;
5773         }
5774
5775         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5776         if (err) {
5777                 drop_inode = 1;
5778                 goto out_unlock;
5779         }
5780
5781         /*
5782         * If the active LSM wants to access the inode during
5783         * d_instantiate it needs these. Smack checks to see
5784         * if the filesystem supports xattrs by looking at the
5785         * ops vector.
5786         */
5787
5788         inode->i_op = &btrfs_special_inode_operations;
5789         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5790         if (err)
5791                 drop_inode = 1;
5792         else {
5793                 init_special_inode(inode, inode->i_mode, rdev);
5794                 btrfs_update_inode(trans, root, inode);
5795                 d_instantiate(dentry, inode);
5796         }
5797 out_unlock:
5798         btrfs_end_transaction(trans, root);
5799         btrfs_balance_delayed_items(root);
5800         btrfs_btree_balance_dirty(root);
5801         if (drop_inode) {
5802                 inode_dec_link_count(inode);
5803                 iput(inode);
5804         }
5805         return err;
5806 }
5807
5808 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5809                         umode_t mode, bool excl)
5810 {
5811         struct btrfs_trans_handle *trans;
5812         struct btrfs_root *root = BTRFS_I(dir)->root;
5813         struct inode *inode = NULL;
5814         int drop_inode_on_err = 0;
5815         int err;
5816         u64 objectid;
5817         u64 index = 0;
5818
5819         /*
5820          * 2 for inode item and ref
5821          * 2 for dir items
5822          * 1 for xattr if selinux is on
5823          */
5824         trans = btrfs_start_transaction(root, 5);
5825         if (IS_ERR(trans))
5826                 return PTR_ERR(trans);
5827
5828         err = btrfs_find_free_ino(root, &objectid);
5829         if (err)
5830                 goto out_unlock;
5831
5832         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5833                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5834                                 mode, &index);
5835         if (IS_ERR(inode)) {
5836                 err = PTR_ERR(inode);
5837                 goto out_unlock;
5838         }
5839         drop_inode_on_err = 1;
5840
5841         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5842         if (err)
5843                 goto out_unlock;
5844
5845         err = btrfs_update_inode(trans, root, inode);
5846         if (err)
5847                 goto out_unlock;
5848
5849         /*
5850         * If the active LSM wants to access the inode during
5851         * d_instantiate it needs these. Smack checks to see
5852         * if the filesystem supports xattrs by looking at the
5853         * ops vector.
5854         */
5855         inode->i_fop = &btrfs_file_operations;
5856         inode->i_op = &btrfs_file_inode_operations;
5857
5858         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5859         if (err)
5860                 goto out_unlock;
5861
5862         inode->i_mapping->a_ops = &btrfs_aops;
5863         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5864         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5865         d_instantiate(dentry, inode);
5866
5867 out_unlock:
5868         btrfs_end_transaction(trans, root);
5869         if (err && drop_inode_on_err) {
5870                 inode_dec_link_count(inode);
5871                 iput(inode);
5872         }
5873         btrfs_balance_delayed_items(root);
5874         btrfs_btree_balance_dirty(root);
5875         return err;
5876 }
5877
5878 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5879                       struct dentry *dentry)
5880 {
5881         struct btrfs_trans_handle *trans;
5882         struct btrfs_root *root = BTRFS_I(dir)->root;
5883         struct inode *inode = old_dentry->d_inode;
5884         u64 index;
5885         int err;
5886         int drop_inode = 0;
5887
5888         /* do not allow sys_link's with other subvols of the same device */
5889         if (root->objectid != BTRFS_I(inode)->root->objectid)
5890                 return -EXDEV;
5891
5892         if (inode->i_nlink >= BTRFS_LINK_MAX)
5893                 return -EMLINK;
5894
5895         err = btrfs_set_inode_index(dir, &index);
5896         if (err)
5897                 goto fail;
5898
5899         /*
5900          * 2 items for inode and inode ref
5901          * 2 items for dir items
5902          * 1 item for parent inode
5903          */
5904         trans = btrfs_start_transaction(root, 5);
5905         if (IS_ERR(trans)) {
5906                 err = PTR_ERR(trans);
5907                 goto fail;
5908         }
5909
5910         /* There are several dir indexes for this inode, clear the cache. */
5911         BTRFS_I(inode)->dir_index = 0ULL;
5912         inc_nlink(inode);
5913         inode_inc_iversion(inode);
5914         inode->i_ctime = CURRENT_TIME;
5915         ihold(inode);
5916         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5917
5918         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5919
5920         if (err) {
5921                 drop_inode = 1;
5922         } else {
5923                 struct dentry *parent = dentry->d_parent;
5924                 err = btrfs_update_inode(trans, root, inode);
5925                 if (err)
5926                         goto fail;
5927                 d_instantiate(dentry, inode);
5928                 btrfs_log_new_name(trans, inode, NULL, parent);
5929         }
5930
5931         btrfs_end_transaction(trans, root);
5932         btrfs_balance_delayed_items(root);
5933 fail:
5934         if (drop_inode) {
5935                 inode_dec_link_count(inode);
5936                 iput(inode);
5937         }
5938         btrfs_btree_balance_dirty(root);
5939         return err;
5940 }
5941
5942 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5943 {
5944         struct inode *inode = NULL;
5945         struct btrfs_trans_handle *trans;
5946         struct btrfs_root *root = BTRFS_I(dir)->root;
5947         int err = 0;
5948         int drop_on_err = 0;
5949         u64 objectid = 0;
5950         u64 index = 0;
5951
5952         /*
5953          * 2 items for inode and ref
5954          * 2 items for dir items
5955          * 1 for xattr if selinux is on
5956          */
5957         trans = btrfs_start_transaction(root, 5);
5958         if (IS_ERR(trans))
5959                 return PTR_ERR(trans);
5960
5961         err = btrfs_find_free_ino(root, &objectid);
5962         if (err)
5963                 goto out_fail;
5964
5965         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5966                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5967                                 S_IFDIR | mode, &index);
5968         if (IS_ERR(inode)) {
5969                 err = PTR_ERR(inode);
5970                 goto out_fail;
5971         }
5972
5973         drop_on_err = 1;
5974
5975         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5976         if (err)
5977                 goto out_fail;
5978
5979         inode->i_op = &btrfs_dir_inode_operations;
5980         inode->i_fop = &btrfs_dir_file_operations;
5981
5982         btrfs_i_size_write(inode, 0);
5983         err = btrfs_update_inode(trans, root, inode);
5984         if (err)
5985                 goto out_fail;
5986
5987         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5988                              dentry->d_name.len, 0, index);
5989         if (err)
5990                 goto out_fail;
5991
5992         d_instantiate(dentry, inode);
5993         drop_on_err = 0;
5994
5995 out_fail:
5996         btrfs_end_transaction(trans, root);
5997         if (drop_on_err)
5998                 iput(inode);
5999         btrfs_balance_delayed_items(root);
6000         btrfs_btree_balance_dirty(root);
6001         return err;
6002 }
6003
6004 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6005  * and an extent that you want to insert, deal with overlap and insert
6006  * the new extent into the tree.
6007  */
6008 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6009                                 struct extent_map *existing,
6010                                 struct extent_map *em,
6011                                 u64 map_start, u64 map_len)
6012 {
6013         u64 start_diff;
6014
6015         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6016         start_diff = map_start - em->start;
6017         em->start = map_start;
6018         em->len = map_len;
6019         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6020             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6021                 em->block_start += start_diff;
6022                 em->block_len -= start_diff;
6023         }
6024         return add_extent_mapping(em_tree, em, 0);
6025 }
6026
6027 static noinline int uncompress_inline(struct btrfs_path *path,
6028                                       struct inode *inode, struct page *page,
6029                                       size_t pg_offset, u64 extent_offset,
6030                                       struct btrfs_file_extent_item *item)
6031 {
6032         int ret;
6033         struct extent_buffer *leaf = path->nodes[0];
6034         char *tmp;
6035         size_t max_size;
6036         unsigned long inline_size;
6037         unsigned long ptr;
6038         int compress_type;
6039
6040         WARN_ON(pg_offset != 0);
6041         compress_type = btrfs_file_extent_compression(leaf, item);
6042         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6043         inline_size = btrfs_file_extent_inline_item_len(leaf,
6044                                         btrfs_item_nr(path->slots[0]));
6045         tmp = kmalloc(inline_size, GFP_NOFS);
6046         if (!tmp)
6047                 return -ENOMEM;
6048         ptr = btrfs_file_extent_inline_start(item);
6049
6050         read_extent_buffer(leaf, tmp, ptr, inline_size);
6051
6052         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6053         ret = btrfs_decompress(compress_type, tmp, page,
6054                                extent_offset, inline_size, max_size);
6055         if (ret) {
6056                 char *kaddr = kmap_atomic(page);
6057                 unsigned long copy_size = min_t(u64,
6058                                   PAGE_CACHE_SIZE - pg_offset,
6059                                   max_size - extent_offset);
6060                 memset(kaddr + pg_offset, 0, copy_size);
6061                 kunmap_atomic(kaddr);
6062         }
6063         kfree(tmp);
6064         return 0;
6065 }
6066
6067 /*
6068  * a bit scary, this does extent mapping from logical file offset to the disk.
6069  * the ugly parts come from merging extents from the disk with the in-ram
6070  * representation.  This gets more complex because of the data=ordered code,
6071  * where the in-ram extents might be locked pending data=ordered completion.
6072  *
6073  * This also copies inline extents directly into the page.
6074  */
6075
6076 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6077                                     size_t pg_offset, u64 start, u64 len,
6078                                     int create)
6079 {
6080         int ret;
6081         int err = 0;
6082         u64 bytenr;
6083         u64 extent_start = 0;
6084         u64 extent_end = 0;
6085         u64 objectid = btrfs_ino(inode);
6086         u32 found_type;
6087         struct btrfs_path *path = NULL;
6088         struct btrfs_root *root = BTRFS_I(inode)->root;
6089         struct btrfs_file_extent_item *item;
6090         struct extent_buffer *leaf;
6091         struct btrfs_key found_key;
6092         struct extent_map *em = NULL;
6093         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6094         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6095         struct btrfs_trans_handle *trans = NULL;
6096         int compress_type;
6097
6098 again:
6099         read_lock(&em_tree->lock);
6100         em = lookup_extent_mapping(em_tree, start, len);
6101         if (em)
6102                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6103         read_unlock(&em_tree->lock);
6104
6105         if (em) {
6106                 if (em->start > start || em->start + em->len <= start)
6107                         free_extent_map(em);
6108                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6109                         free_extent_map(em);
6110                 else
6111                         goto out;
6112         }
6113         em = alloc_extent_map();
6114         if (!em) {
6115                 err = -ENOMEM;
6116                 goto out;
6117         }
6118         em->bdev = root->fs_info->fs_devices->latest_bdev;
6119         em->start = EXTENT_MAP_HOLE;
6120         em->orig_start = EXTENT_MAP_HOLE;
6121         em->len = (u64)-1;
6122         em->block_len = (u64)-1;
6123
6124         if (!path) {
6125                 path = btrfs_alloc_path();
6126                 if (!path) {
6127                         err = -ENOMEM;
6128                         goto out;
6129                 }
6130                 /*
6131                  * Chances are we'll be called again, so go ahead and do
6132                  * readahead
6133                  */
6134                 path->reada = 1;
6135         }
6136
6137         ret = btrfs_lookup_file_extent(trans, root, path,
6138                                        objectid, start, trans != NULL);
6139         if (ret < 0) {
6140                 err = ret;
6141                 goto out;
6142         }
6143
6144         if (ret != 0) {
6145                 if (path->slots[0] == 0)
6146                         goto not_found;
6147                 path->slots[0]--;
6148         }
6149
6150         leaf = path->nodes[0];
6151         item = btrfs_item_ptr(leaf, path->slots[0],
6152                               struct btrfs_file_extent_item);
6153         /* are we inside the extent that was found? */
6154         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6155         found_type = btrfs_key_type(&found_key);
6156         if (found_key.objectid != objectid ||
6157             found_type != BTRFS_EXTENT_DATA_KEY) {
6158                 /*
6159                  * If we backup past the first extent we want to move forward
6160                  * and see if there is an extent in front of us, otherwise we'll
6161                  * say there is a hole for our whole search range which can
6162                  * cause problems.
6163                  */
6164                 extent_end = start;
6165                 goto next;
6166         }
6167
6168         found_type = btrfs_file_extent_type(leaf, item);
6169         extent_start = found_key.offset;
6170         compress_type = btrfs_file_extent_compression(leaf, item);
6171         if (found_type == BTRFS_FILE_EXTENT_REG ||
6172             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6173                 extent_end = extent_start +
6174                        btrfs_file_extent_num_bytes(leaf, item);
6175         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6176                 size_t size;
6177                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6178                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6179         }
6180 next:
6181         if (start >= extent_end) {
6182                 path->slots[0]++;
6183                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6184                         ret = btrfs_next_leaf(root, path);
6185                         if (ret < 0) {
6186                                 err = ret;
6187                                 goto out;
6188                         }
6189                         if (ret > 0)
6190                                 goto not_found;
6191                         leaf = path->nodes[0];
6192                 }
6193                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6194                 if (found_key.objectid != objectid ||
6195                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6196                         goto not_found;
6197                 if (start + len <= found_key.offset)
6198                         goto not_found;
6199                 em->start = start;
6200                 em->orig_start = start;
6201                 em->len = found_key.offset - start;
6202                 goto not_found_em;
6203         }
6204
6205         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6206         if (found_type == BTRFS_FILE_EXTENT_REG ||
6207             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6208                 em->start = extent_start;
6209                 em->len = extent_end - extent_start;
6210                 em->orig_start = extent_start -
6211                                  btrfs_file_extent_offset(leaf, item);
6212                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6213                                                                       item);
6214                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6215                 if (bytenr == 0) {
6216                         em->block_start = EXTENT_MAP_HOLE;
6217                         goto insert;
6218                 }
6219                 if (compress_type != BTRFS_COMPRESS_NONE) {
6220                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6221                         em->compress_type = compress_type;
6222                         em->block_start = bytenr;
6223                         em->block_len = em->orig_block_len;
6224                 } else {
6225                         bytenr += btrfs_file_extent_offset(leaf, item);
6226                         em->block_start = bytenr;
6227                         em->block_len = em->len;
6228                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6229                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6230                 }
6231                 goto insert;
6232         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6233                 unsigned long ptr;
6234                 char *map;
6235                 size_t size;
6236                 size_t extent_offset;
6237                 size_t copy_size;
6238
6239                 em->block_start = EXTENT_MAP_INLINE;
6240                 if (!page || create) {
6241                         em->start = extent_start;
6242                         em->len = extent_end - extent_start;
6243                         goto out;
6244                 }
6245
6246                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6247                 extent_offset = page_offset(page) + pg_offset - extent_start;
6248                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6249                                 size - extent_offset);
6250                 em->start = extent_start + extent_offset;
6251                 em->len = ALIGN(copy_size, root->sectorsize);
6252                 em->orig_block_len = em->len;
6253                 em->orig_start = em->start;
6254                 if (compress_type) {
6255                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6256                         em->compress_type = compress_type;
6257                 }
6258                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6259                 if (create == 0 && !PageUptodate(page)) {
6260                         if (btrfs_file_extent_compression(leaf, item) !=
6261                             BTRFS_COMPRESS_NONE) {
6262                                 ret = uncompress_inline(path, inode, page,
6263                                                         pg_offset,
6264                                                         extent_offset, item);
6265                                 BUG_ON(ret); /* -ENOMEM */
6266                         } else {
6267                                 map = kmap(page);
6268                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6269                                                    copy_size);
6270                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6271                                         memset(map + pg_offset + copy_size, 0,
6272                                                PAGE_CACHE_SIZE - pg_offset -
6273                                                copy_size);
6274                                 }
6275                                 kunmap(page);
6276                         }
6277                         flush_dcache_page(page);
6278                 } else if (create && PageUptodate(page)) {
6279                         BUG();
6280                         if (!trans) {
6281                                 kunmap(page);
6282                                 free_extent_map(em);
6283                                 em = NULL;
6284
6285                                 btrfs_release_path(path);
6286                                 trans = btrfs_join_transaction(root);
6287
6288                                 if (IS_ERR(trans))
6289                                         return ERR_CAST(trans);
6290                                 goto again;
6291                         }
6292                         map = kmap(page);
6293                         write_extent_buffer(leaf, map + pg_offset, ptr,
6294                                             copy_size);
6295                         kunmap(page);
6296                         btrfs_mark_buffer_dirty(leaf);
6297                 }
6298                 set_extent_uptodate(io_tree, em->start,
6299                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6300                 goto insert;
6301         } else {
6302                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6303         }
6304 not_found:
6305         em->start = start;
6306         em->orig_start = start;
6307         em->len = len;
6308 not_found_em:
6309         em->block_start = EXTENT_MAP_HOLE;
6310         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6311 insert:
6312         btrfs_release_path(path);
6313         if (em->start > start || extent_map_end(em) <= start) {
6314                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6315                         em->start, em->len, start, len);
6316                 err = -EIO;
6317                 goto out;
6318         }
6319
6320         err = 0;
6321         write_lock(&em_tree->lock);
6322         ret = add_extent_mapping(em_tree, em, 0);
6323         /* it is possible that someone inserted the extent into the tree
6324          * while we had the lock dropped.  It is also possible that
6325          * an overlapping map exists in the tree
6326          */
6327         if (ret == -EEXIST) {
6328                 struct extent_map *existing;
6329
6330                 ret = 0;
6331
6332                 existing = lookup_extent_mapping(em_tree, start, len);
6333                 if (existing && (existing->start > start ||
6334                     existing->start + existing->len <= start)) {
6335                         free_extent_map(existing);
6336                         existing = NULL;
6337                 }
6338                 if (!existing) {
6339                         existing = lookup_extent_mapping(em_tree, em->start,
6340                                                          em->len);
6341                         if (existing) {
6342                                 err = merge_extent_mapping(em_tree, existing,
6343                                                            em, start,
6344                                                            root->sectorsize);
6345                                 free_extent_map(existing);
6346                                 if (err) {
6347                                         free_extent_map(em);
6348                                         em = NULL;
6349                                 }
6350                         } else {
6351                                 err = -EIO;
6352                                 free_extent_map(em);
6353                                 em = NULL;
6354                         }
6355                 } else {
6356                         free_extent_map(em);
6357                         em = existing;
6358                         err = 0;
6359                 }
6360         }
6361         write_unlock(&em_tree->lock);
6362 out:
6363
6364         trace_btrfs_get_extent(root, em);
6365
6366         if (path)
6367                 btrfs_free_path(path);
6368         if (trans) {
6369                 ret = btrfs_end_transaction(trans, root);
6370                 if (!err)
6371                         err = ret;
6372         }
6373         if (err) {
6374                 free_extent_map(em);
6375                 return ERR_PTR(err);
6376         }
6377         BUG_ON(!em); /* Error is always set */
6378         return em;
6379 }
6380
6381 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6382                                            size_t pg_offset, u64 start, u64 len,
6383                                            int create)
6384 {
6385         struct extent_map *em;
6386         struct extent_map *hole_em = NULL;
6387         u64 range_start = start;
6388         u64 end;
6389         u64 found;
6390         u64 found_end;
6391         int err = 0;
6392
6393         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6394         if (IS_ERR(em))
6395                 return em;
6396         if (em) {
6397                 /*
6398                  * if our em maps to
6399                  * -  a hole or
6400                  * -  a pre-alloc extent,
6401                  * there might actually be delalloc bytes behind it.
6402                  */
6403                 if (em->block_start != EXTENT_MAP_HOLE &&
6404                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6405                         return em;
6406                 else
6407                         hole_em = em;
6408         }
6409
6410         /* check to see if we've wrapped (len == -1 or similar) */
6411         end = start + len;
6412         if (end < start)
6413                 end = (u64)-1;
6414         else
6415                 end -= 1;
6416
6417         em = NULL;
6418
6419         /* ok, we didn't find anything, lets look for delalloc */
6420         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6421                                  end, len, EXTENT_DELALLOC, 1);
6422         found_end = range_start + found;
6423         if (found_end < range_start)
6424                 found_end = (u64)-1;
6425
6426         /*
6427          * we didn't find anything useful, return
6428          * the original results from get_extent()
6429          */
6430         if (range_start > end || found_end <= start) {
6431                 em = hole_em;
6432                 hole_em = NULL;
6433                 goto out;
6434         }
6435
6436         /* adjust the range_start to make sure it doesn't
6437          * go backwards from the start they passed in
6438          */
6439         range_start = max(start, range_start);
6440         found = found_end - range_start;
6441
6442         if (found > 0) {
6443                 u64 hole_start = start;
6444                 u64 hole_len = len;
6445
6446                 em = alloc_extent_map();
6447                 if (!em) {
6448                         err = -ENOMEM;
6449                         goto out;
6450                 }
6451                 /*
6452                  * when btrfs_get_extent can't find anything it
6453                  * returns one huge hole
6454                  *
6455                  * make sure what it found really fits our range, and
6456                  * adjust to make sure it is based on the start from
6457                  * the caller
6458                  */
6459                 if (hole_em) {
6460                         u64 calc_end = extent_map_end(hole_em);
6461
6462                         if (calc_end <= start || (hole_em->start > end)) {
6463                                 free_extent_map(hole_em);
6464                                 hole_em = NULL;
6465                         } else {
6466                                 hole_start = max(hole_em->start, start);
6467                                 hole_len = calc_end - hole_start;
6468                         }
6469                 }
6470                 em->bdev = NULL;
6471                 if (hole_em && range_start > hole_start) {
6472                         /* our hole starts before our delalloc, so we
6473                          * have to return just the parts of the hole
6474                          * that go until  the delalloc starts
6475                          */
6476                         em->len = min(hole_len,
6477                                       range_start - hole_start);
6478                         em->start = hole_start;
6479                         em->orig_start = hole_start;
6480                         /*
6481                          * don't adjust block start at all,
6482                          * it is fixed at EXTENT_MAP_HOLE
6483                          */
6484                         em->block_start = hole_em->block_start;
6485                         em->block_len = hole_len;
6486                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6487                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6488                 } else {
6489                         em->start = range_start;
6490                         em->len = found;
6491                         em->orig_start = range_start;
6492                         em->block_start = EXTENT_MAP_DELALLOC;
6493                         em->block_len = found;
6494                 }
6495         } else if (hole_em) {
6496                 return hole_em;
6497         }
6498 out:
6499
6500         free_extent_map(hole_em);
6501         if (err) {
6502                 free_extent_map(em);
6503                 return ERR_PTR(err);
6504         }
6505         return em;
6506 }
6507
6508 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6509                                                   u64 start, u64 len)
6510 {
6511         struct btrfs_root *root = BTRFS_I(inode)->root;
6512         struct extent_map *em;
6513         struct btrfs_key ins;
6514         u64 alloc_hint;
6515         int ret;
6516
6517         alloc_hint = get_extent_allocation_hint(inode, start, len);
6518         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6519                                    alloc_hint, &ins, 1);
6520         if (ret)
6521                 return ERR_PTR(ret);
6522
6523         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6524                               ins.offset, ins.offset, ins.offset, 0);
6525         if (IS_ERR(em)) {
6526                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6527                 return em;
6528         }
6529
6530         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6531                                            ins.offset, ins.offset, 0);
6532         if (ret) {
6533                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6534                 free_extent_map(em);
6535                 return ERR_PTR(ret);
6536         }
6537
6538         return em;
6539 }
6540
6541 /*
6542  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6543  * block must be cow'd
6544  */
6545 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6546                               u64 *orig_start, u64 *orig_block_len,
6547                               u64 *ram_bytes)
6548 {
6549         struct btrfs_trans_handle *trans;
6550         struct btrfs_path *path;
6551         int ret;
6552         struct extent_buffer *leaf;
6553         struct btrfs_root *root = BTRFS_I(inode)->root;
6554         struct btrfs_file_extent_item *fi;
6555         struct btrfs_key key;
6556         u64 disk_bytenr;
6557         u64 backref_offset;
6558         u64 extent_end;
6559         u64 num_bytes;
6560         int slot;
6561         int found_type;
6562         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6563
6564         path = btrfs_alloc_path();
6565         if (!path)
6566                 return -ENOMEM;
6567
6568         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6569                                        offset, 0);
6570         if (ret < 0)
6571                 goto out;
6572
6573         slot = path->slots[0];
6574         if (ret == 1) {
6575                 if (slot == 0) {
6576                         /* can't find the item, must cow */
6577                         ret = 0;
6578                         goto out;
6579                 }
6580                 slot--;
6581         }
6582         ret = 0;
6583         leaf = path->nodes[0];
6584         btrfs_item_key_to_cpu(leaf, &key, slot);
6585         if (key.objectid != btrfs_ino(inode) ||
6586             key.type != BTRFS_EXTENT_DATA_KEY) {
6587                 /* not our file or wrong item type, must cow */
6588                 goto out;
6589         }
6590
6591         if (key.offset > offset) {
6592                 /* Wrong offset, must cow */
6593                 goto out;
6594         }
6595
6596         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6597         found_type = btrfs_file_extent_type(leaf, fi);
6598         if (found_type != BTRFS_FILE_EXTENT_REG &&
6599             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6600                 /* not a regular extent, must cow */
6601                 goto out;
6602         }
6603
6604         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6605                 goto out;
6606
6607         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6608         if (extent_end <= offset)
6609                 goto out;
6610
6611         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6612         if (disk_bytenr == 0)
6613                 goto out;
6614
6615         if (btrfs_file_extent_compression(leaf, fi) ||
6616             btrfs_file_extent_encryption(leaf, fi) ||
6617             btrfs_file_extent_other_encoding(leaf, fi))
6618                 goto out;
6619
6620         backref_offset = btrfs_file_extent_offset(leaf, fi);
6621
6622         if (orig_start) {
6623                 *orig_start = key.offset - backref_offset;
6624                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6625                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6626         }
6627
6628         if (btrfs_extent_readonly(root, disk_bytenr))
6629                 goto out;
6630         btrfs_release_path(path);
6631
6632         /*
6633          * look for other files referencing this extent, if we
6634          * find any we must cow
6635          */
6636         trans = btrfs_join_transaction(root);
6637         if (IS_ERR(trans)) {
6638                 ret = 0;
6639                 goto out;
6640         }
6641
6642         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6643                                     key.offset - backref_offset, disk_bytenr);
6644         btrfs_end_transaction(trans, root);
6645         if (ret) {
6646                 ret = 0;
6647                 goto out;
6648         }
6649
6650         /*
6651          * adjust disk_bytenr and num_bytes to cover just the bytes
6652          * in this extent we are about to write.  If there
6653          * are any csums in that range we have to cow in order
6654          * to keep the csums correct
6655          */
6656         disk_bytenr += backref_offset;
6657         disk_bytenr += offset - key.offset;
6658         num_bytes = min(offset + *len, extent_end) - offset;
6659         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6660                                 goto out;
6661         /*
6662          * all of the above have passed, it is safe to overwrite this extent
6663          * without cow
6664          */
6665         *len = num_bytes;
6666         ret = 1;
6667 out:
6668         btrfs_free_path(path);
6669         return ret;
6670 }
6671
6672 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6673                               struct extent_state **cached_state, int writing)
6674 {
6675         struct btrfs_ordered_extent *ordered;
6676         int ret = 0;
6677
6678         while (1) {
6679                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6680                                  0, cached_state);
6681                 /*
6682                  * We're concerned with the entire range that we're going to be
6683                  * doing DIO to, so we need to make sure theres no ordered
6684                  * extents in this range.
6685                  */
6686                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6687                                                      lockend - lockstart + 1);
6688
6689                 /*
6690                  * We need to make sure there are no buffered pages in this
6691                  * range either, we could have raced between the invalidate in
6692                  * generic_file_direct_write and locking the extent.  The
6693                  * invalidate needs to happen so that reads after a write do not
6694                  * get stale data.
6695                  */
6696                 if (!ordered && (!writing ||
6697                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6698                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6699                                     *cached_state)))
6700                         break;
6701
6702                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6703                                      cached_state, GFP_NOFS);
6704
6705                 if (ordered) {
6706                         btrfs_start_ordered_extent(inode, ordered, 1);
6707                         btrfs_put_ordered_extent(ordered);
6708                 } else {
6709                         /* Screw you mmap */
6710                         ret = filemap_write_and_wait_range(inode->i_mapping,
6711                                                            lockstart,
6712                                                            lockend);
6713                         if (ret)
6714                                 break;
6715
6716                         /*
6717                          * If we found a page that couldn't be invalidated just
6718                          * fall back to buffered.
6719                          */
6720                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6721                                         lockstart >> PAGE_CACHE_SHIFT,
6722                                         lockend >> PAGE_CACHE_SHIFT);
6723                         if (ret)
6724                                 break;
6725                 }
6726
6727                 cond_resched();
6728         }
6729
6730         return ret;
6731 }
6732
6733 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6734                                            u64 len, u64 orig_start,
6735                                            u64 block_start, u64 block_len,
6736                                            u64 orig_block_len, u64 ram_bytes,
6737                                            int type)
6738 {
6739         struct extent_map_tree *em_tree;
6740         struct extent_map *em;
6741         struct btrfs_root *root = BTRFS_I(inode)->root;
6742         int ret;
6743
6744         em_tree = &BTRFS_I(inode)->extent_tree;
6745         em = alloc_extent_map();
6746         if (!em)
6747                 return ERR_PTR(-ENOMEM);
6748
6749         em->start = start;
6750         em->orig_start = orig_start;
6751         em->mod_start = start;
6752         em->mod_len = len;
6753         em->len = len;
6754         em->block_len = block_len;
6755         em->block_start = block_start;
6756         em->bdev = root->fs_info->fs_devices->latest_bdev;
6757         em->orig_block_len = orig_block_len;
6758         em->ram_bytes = ram_bytes;
6759         em->generation = -1;
6760         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6761         if (type == BTRFS_ORDERED_PREALLOC)
6762                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6763
6764         do {
6765                 btrfs_drop_extent_cache(inode, em->start,
6766                                 em->start + em->len - 1, 0);
6767                 write_lock(&em_tree->lock);
6768                 ret = add_extent_mapping(em_tree, em, 1);
6769                 write_unlock(&em_tree->lock);
6770         } while (ret == -EEXIST);
6771
6772         if (ret) {
6773                 free_extent_map(em);
6774                 return ERR_PTR(ret);
6775         }
6776
6777         return em;
6778 }
6779
6780
6781 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6782                                    struct buffer_head *bh_result, int create)
6783 {
6784         struct extent_map *em;
6785         struct btrfs_root *root = BTRFS_I(inode)->root;
6786         struct extent_state *cached_state = NULL;
6787         u64 start = iblock << inode->i_blkbits;
6788         u64 lockstart, lockend;
6789         u64 len = bh_result->b_size;
6790         int unlock_bits = EXTENT_LOCKED;
6791         int ret = 0;
6792
6793         if (create)
6794                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6795         else
6796                 len = min_t(u64, len, root->sectorsize);
6797
6798         lockstart = start;
6799         lockend = start + len - 1;
6800
6801         /*
6802          * If this errors out it's because we couldn't invalidate pagecache for
6803          * this range and we need to fallback to buffered.
6804          */
6805         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6806                 return -ENOTBLK;
6807
6808         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6809         if (IS_ERR(em)) {
6810                 ret = PTR_ERR(em);
6811                 goto unlock_err;
6812         }
6813
6814         /*
6815          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6816          * io.  INLINE is special, and we could probably kludge it in here, but
6817          * it's still buffered so for safety lets just fall back to the generic
6818          * buffered path.
6819          *
6820          * For COMPRESSED we _have_ to read the entire extent in so we can
6821          * decompress it, so there will be buffering required no matter what we
6822          * do, so go ahead and fallback to buffered.
6823          *
6824          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6825          * to buffered IO.  Don't blame me, this is the price we pay for using
6826          * the generic code.
6827          */
6828         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6829             em->block_start == EXTENT_MAP_INLINE) {
6830                 free_extent_map(em);
6831                 ret = -ENOTBLK;
6832                 goto unlock_err;
6833         }
6834
6835         /* Just a good old fashioned hole, return */
6836         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6837                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6838                 free_extent_map(em);
6839                 goto unlock_err;
6840         }
6841
6842         /*
6843          * We don't allocate a new extent in the following cases
6844          *
6845          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6846          * existing extent.
6847          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6848          * just use the extent.
6849          *
6850          */
6851         if (!create) {
6852                 len = min(len, em->len - (start - em->start));
6853                 lockstart = start + len;
6854                 goto unlock;
6855         }
6856
6857         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6858             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6859              em->block_start != EXTENT_MAP_HOLE)) {
6860                 int type;
6861                 int ret;
6862                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6863
6864                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6865                         type = BTRFS_ORDERED_PREALLOC;
6866                 else
6867                         type = BTRFS_ORDERED_NOCOW;
6868                 len = min(len, em->len - (start - em->start));
6869                 block_start = em->block_start + (start - em->start);
6870
6871                 if (can_nocow_extent(inode, start, &len, &orig_start,
6872                                      &orig_block_len, &ram_bytes) == 1) {
6873                         if (type == BTRFS_ORDERED_PREALLOC) {
6874                                 free_extent_map(em);
6875                                 em = create_pinned_em(inode, start, len,
6876                                                        orig_start,
6877                                                        block_start, len,
6878                                                        orig_block_len,
6879                                                        ram_bytes, type);
6880                                 if (IS_ERR(em))
6881                                         goto unlock_err;
6882                         }
6883
6884                         ret = btrfs_add_ordered_extent_dio(inode, start,
6885                                            block_start, len, len, type);
6886                         if (ret) {
6887                                 free_extent_map(em);
6888                                 goto unlock_err;
6889                         }
6890                         goto unlock;
6891                 }
6892         }
6893
6894         /*
6895          * this will cow the extent, reset the len in case we changed
6896          * it above
6897          */
6898         len = bh_result->b_size;
6899         free_extent_map(em);
6900         em = btrfs_new_extent_direct(inode, start, len);
6901         if (IS_ERR(em)) {
6902                 ret = PTR_ERR(em);
6903                 goto unlock_err;
6904         }
6905         len = min(len, em->len - (start - em->start));
6906 unlock:
6907         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6908                 inode->i_blkbits;
6909         bh_result->b_size = len;
6910         bh_result->b_bdev = em->bdev;
6911         set_buffer_mapped(bh_result);
6912         if (create) {
6913                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6914                         set_buffer_new(bh_result);
6915
6916                 /*
6917                  * Need to update the i_size under the extent lock so buffered
6918                  * readers will get the updated i_size when we unlock.
6919                  */
6920                 if (start + len > i_size_read(inode))
6921                         i_size_write(inode, start + len);
6922
6923                 spin_lock(&BTRFS_I(inode)->lock);
6924                 BTRFS_I(inode)->outstanding_extents++;
6925                 spin_unlock(&BTRFS_I(inode)->lock);
6926
6927                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6928                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6929                                      &cached_state, GFP_NOFS);
6930                 BUG_ON(ret);
6931         }
6932
6933         /*
6934          * In the case of write we need to clear and unlock the entire range,
6935          * in the case of read we need to unlock only the end area that we
6936          * aren't using if there is any left over space.
6937          */
6938         if (lockstart < lockend) {
6939                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6940                                  lockend, unlock_bits, 1, 0,
6941                                  &cached_state, GFP_NOFS);
6942         } else {
6943                 free_extent_state(cached_state);
6944         }
6945
6946         free_extent_map(em);
6947
6948         return 0;
6949
6950 unlock_err:
6951         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6952                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6953         return ret;
6954 }
6955
6956 static void btrfs_endio_direct_read(struct bio *bio, int err)
6957 {
6958         struct btrfs_dio_private *dip = bio->bi_private;
6959         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6960         struct bio_vec *bvec = bio->bi_io_vec;
6961         struct inode *inode = dip->inode;
6962         struct btrfs_root *root = BTRFS_I(inode)->root;
6963         struct bio *dio_bio;
6964         u32 *csums = (u32 *)dip->csum;
6965         int index = 0;
6966         u64 start;
6967
6968         start = dip->logical_offset;
6969         do {
6970                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6971                         struct page *page = bvec->bv_page;
6972                         char *kaddr;
6973                         u32 csum = ~(u32)0;
6974                         unsigned long flags;
6975
6976                         local_irq_save(flags);
6977                         kaddr = kmap_atomic(page);
6978                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6979                                                csum, bvec->bv_len);
6980                         btrfs_csum_final(csum, (char *)&csum);
6981                         kunmap_atomic(kaddr);
6982                         local_irq_restore(flags);
6983
6984                         flush_dcache_page(bvec->bv_page);
6985                         if (csum != csums[index]) {
6986                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6987                                           btrfs_ino(inode), start, csum,
6988                                           csums[index]);
6989                                 err = -EIO;
6990                         }
6991                 }
6992
6993                 start += bvec->bv_len;
6994                 bvec++;
6995                 index++;
6996         } while (bvec <= bvec_end);
6997
6998         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6999                       dip->logical_offset + dip->bytes - 1);
7000         dio_bio = dip->dio_bio;
7001
7002         kfree(dip);
7003
7004         /* If we had a csum failure make sure to clear the uptodate flag */
7005         if (err)
7006                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7007         dio_end_io(dio_bio, err);
7008         bio_put(bio);
7009 }
7010
7011 static void btrfs_endio_direct_write(struct bio *bio, int err)
7012 {
7013         struct btrfs_dio_private *dip = bio->bi_private;
7014         struct inode *inode = dip->inode;
7015         struct btrfs_root *root = BTRFS_I(inode)->root;
7016         struct btrfs_ordered_extent *ordered = NULL;
7017         u64 ordered_offset = dip->logical_offset;
7018         u64 ordered_bytes = dip->bytes;
7019         struct bio *dio_bio;
7020         int ret;
7021
7022         if (err)
7023                 goto out_done;
7024 again:
7025         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7026                                                    &ordered_offset,
7027                                                    ordered_bytes, !err);
7028         if (!ret)
7029                 goto out_test;
7030
7031         ordered->work.func = finish_ordered_fn;
7032         ordered->work.flags = 0;
7033         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7034                            &ordered->work);
7035 out_test:
7036         /*
7037          * our bio might span multiple ordered extents.  If we haven't
7038          * completed the accounting for the whole dio, go back and try again
7039          */
7040         if (ordered_offset < dip->logical_offset + dip->bytes) {
7041                 ordered_bytes = dip->logical_offset + dip->bytes -
7042                         ordered_offset;
7043                 ordered = NULL;
7044                 goto again;
7045         }
7046 out_done:
7047         dio_bio = dip->dio_bio;
7048
7049         kfree(dip);
7050
7051         /* If we had an error make sure to clear the uptodate flag */
7052         if (err)
7053                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7054         dio_end_io(dio_bio, err);
7055         bio_put(bio);
7056 }
7057
7058 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7059                                     struct bio *bio, int mirror_num,
7060                                     unsigned long bio_flags, u64 offset)
7061 {
7062         int ret;
7063         struct btrfs_root *root = BTRFS_I(inode)->root;
7064         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7065         BUG_ON(ret); /* -ENOMEM */
7066         return 0;
7067 }
7068
7069 static void btrfs_end_dio_bio(struct bio *bio, int err)
7070 {
7071         struct btrfs_dio_private *dip = bio->bi_private;
7072
7073         if (err) {
7074                 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7075                           "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7076                       btrfs_ino(dip->inode), bio->bi_rw,
7077                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7078                 dip->errors = 1;
7079
7080                 /*
7081                  * before atomic variable goto zero, we must make sure
7082                  * dip->errors is perceived to be set.
7083                  */
7084                 smp_mb__before_atomic_dec();
7085         }
7086
7087         /* if there are more bios still pending for this dio, just exit */
7088         if (!atomic_dec_and_test(&dip->pending_bios))
7089                 goto out;
7090
7091         if (dip->errors) {
7092                 bio_io_error(dip->orig_bio);
7093         } else {
7094                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7095                 bio_endio(dip->orig_bio, 0);
7096         }
7097 out:
7098         bio_put(bio);
7099 }
7100
7101 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7102                                        u64 first_sector, gfp_t gfp_flags)
7103 {
7104         int nr_vecs = bio_get_nr_vecs(bdev);
7105         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7106 }
7107
7108 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7109                                          int rw, u64 file_offset, int skip_sum,
7110                                          int async_submit)
7111 {
7112         struct btrfs_dio_private *dip = bio->bi_private;
7113         int write = rw & REQ_WRITE;
7114         struct btrfs_root *root = BTRFS_I(inode)->root;
7115         int ret;
7116
7117         if (async_submit)
7118                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7119
7120         bio_get(bio);
7121
7122         if (!write) {
7123                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7124                 if (ret)
7125                         goto err;
7126         }
7127
7128         if (skip_sum)
7129                 goto map;
7130
7131         if (write && async_submit) {
7132                 ret = btrfs_wq_submit_bio(root->fs_info,
7133                                    inode, rw, bio, 0, 0,
7134                                    file_offset,
7135                                    __btrfs_submit_bio_start_direct_io,
7136                                    __btrfs_submit_bio_done);
7137                 goto err;
7138         } else if (write) {
7139                 /*
7140                  * If we aren't doing async submit, calculate the csum of the
7141                  * bio now.
7142                  */
7143                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7144                 if (ret)
7145                         goto err;
7146         } else if (!skip_sum) {
7147                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7148                                                 file_offset);
7149                 if (ret)
7150                         goto err;
7151         }
7152
7153 map:
7154         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7155 err:
7156         bio_put(bio);
7157         return ret;
7158 }
7159
7160 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7161                                     int skip_sum)
7162 {
7163         struct inode *inode = dip->inode;
7164         struct btrfs_root *root = BTRFS_I(inode)->root;
7165         struct bio *bio;
7166         struct bio *orig_bio = dip->orig_bio;
7167         struct bio_vec *bvec = orig_bio->bi_io_vec;
7168         u64 start_sector = orig_bio->bi_sector;
7169         u64 file_offset = dip->logical_offset;
7170         u64 submit_len = 0;
7171         u64 map_length;
7172         int nr_pages = 0;
7173         int ret = 0;
7174         int async_submit = 0;
7175
7176         map_length = orig_bio->bi_size;
7177         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7178                               &map_length, NULL, 0);
7179         if (ret) {
7180                 bio_put(orig_bio);
7181                 return -EIO;
7182         }
7183
7184         if (map_length >= orig_bio->bi_size) {
7185                 bio = orig_bio;
7186                 goto submit;
7187         }
7188
7189         /* async crcs make it difficult to collect full stripe writes. */
7190         if (btrfs_get_alloc_profile(root, 1) &
7191             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7192                 async_submit = 0;
7193         else
7194                 async_submit = 1;
7195
7196         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7197         if (!bio)
7198                 return -ENOMEM;
7199         bio->bi_private = dip;
7200         bio->bi_end_io = btrfs_end_dio_bio;
7201         atomic_inc(&dip->pending_bios);
7202
7203         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7204                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7205                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7206                                  bvec->bv_offset) < bvec->bv_len)) {
7207                         /*
7208                          * inc the count before we submit the bio so
7209                          * we know the end IO handler won't happen before
7210                          * we inc the count. Otherwise, the dip might get freed
7211                          * before we're done setting it up
7212                          */
7213                         atomic_inc(&dip->pending_bios);
7214                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7215                                                      file_offset, skip_sum,
7216                                                      async_submit);
7217                         if (ret) {
7218                                 bio_put(bio);
7219                                 atomic_dec(&dip->pending_bios);
7220                                 goto out_err;
7221                         }
7222
7223                         start_sector += submit_len >> 9;
7224                         file_offset += submit_len;
7225
7226                         submit_len = 0;
7227                         nr_pages = 0;
7228
7229                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7230                                                   start_sector, GFP_NOFS);
7231                         if (!bio)
7232                                 goto out_err;
7233                         bio->bi_private = dip;
7234                         bio->bi_end_io = btrfs_end_dio_bio;
7235
7236                         map_length = orig_bio->bi_size;
7237                         ret = btrfs_map_block(root->fs_info, rw,
7238                                               start_sector << 9,
7239                                               &map_length, NULL, 0);
7240                         if (ret) {
7241                                 bio_put(bio);
7242                                 goto out_err;
7243                         }
7244                 } else {
7245                         submit_len += bvec->bv_len;
7246                         nr_pages++;
7247                         bvec++;
7248                 }
7249         }
7250
7251 submit:
7252         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7253                                      async_submit);
7254         if (!ret)
7255                 return 0;
7256
7257         bio_put(bio);
7258 out_err:
7259         dip->errors = 1;
7260         /*
7261          * before atomic variable goto zero, we must
7262          * make sure dip->errors is perceived to be set.
7263          */
7264         smp_mb__before_atomic_dec();
7265         if (atomic_dec_and_test(&dip->pending_bios))
7266                 bio_io_error(dip->orig_bio);
7267
7268         /* bio_end_io() will handle error, so we needn't return it */
7269         return 0;
7270 }
7271
7272 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7273                                 struct inode *inode, loff_t file_offset)
7274 {
7275         struct btrfs_root *root = BTRFS_I(inode)->root;
7276         struct btrfs_dio_private *dip;
7277         struct bio *io_bio;
7278         int skip_sum;
7279         int sum_len;
7280         int write = rw & REQ_WRITE;
7281         int ret = 0;
7282         u16 csum_size;
7283
7284         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7285
7286         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7287         if (!io_bio) {
7288                 ret = -ENOMEM;
7289                 goto free_ordered;
7290         }
7291
7292         if (!skip_sum && !write) {
7293                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7294                 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7295                 sum_len *= csum_size;
7296         } else {
7297                 sum_len = 0;
7298         }
7299
7300         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7301         if (!dip) {
7302                 ret = -ENOMEM;
7303                 goto free_io_bio;
7304         }
7305
7306         dip->private = dio_bio->bi_private;
7307         dip->inode = inode;
7308         dip->logical_offset = file_offset;
7309         dip->bytes = dio_bio->bi_size;
7310         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7311         io_bio->bi_private = dip;
7312         dip->errors = 0;
7313         dip->orig_bio = io_bio;
7314         dip->dio_bio = dio_bio;
7315         atomic_set(&dip->pending_bios, 0);
7316
7317         if (write)
7318                 io_bio->bi_end_io = btrfs_endio_direct_write;
7319         else
7320                 io_bio->bi_end_io = btrfs_endio_direct_read;
7321
7322         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7323         if (!ret)
7324                 return;
7325
7326 free_io_bio:
7327         bio_put(io_bio);
7328
7329 free_ordered:
7330         /*
7331          * If this is a write, we need to clean up the reserved space and kill
7332          * the ordered extent.
7333          */
7334         if (write) {
7335                 struct btrfs_ordered_extent *ordered;
7336                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7337                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7338                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7339                         btrfs_free_reserved_extent(root, ordered->start,
7340                                                    ordered->disk_len);
7341                 btrfs_put_ordered_extent(ordered);
7342                 btrfs_put_ordered_extent(ordered);
7343         }
7344         bio_endio(dio_bio, ret);
7345 }
7346
7347 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7348                         const struct iovec *iov, loff_t offset,
7349                         unsigned long nr_segs)
7350 {
7351         int seg;
7352         int i;
7353         size_t size;
7354         unsigned long addr;
7355         unsigned blocksize_mask = root->sectorsize - 1;
7356         ssize_t retval = -EINVAL;
7357         loff_t end = offset;
7358
7359         if (offset & blocksize_mask)
7360                 goto out;
7361
7362         /* Check the memory alignment.  Blocks cannot straddle pages */
7363         for (seg = 0; seg < nr_segs; seg++) {
7364                 addr = (unsigned long)iov[seg].iov_base;
7365                 size = iov[seg].iov_len;
7366                 end += size;
7367                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7368                         goto out;
7369
7370                 /* If this is a write we don't need to check anymore */
7371                 if (rw & WRITE)
7372                         continue;
7373
7374                 /*
7375                  * Check to make sure we don't have duplicate iov_base's in this
7376                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7377                  * when reading back.
7378                  */
7379                 for (i = seg + 1; i < nr_segs; i++) {
7380                         if (iov[seg].iov_base == iov[i].iov_base)
7381                                 goto out;
7382                 }
7383         }
7384         retval = 0;
7385 out:
7386         return retval;
7387 }
7388
7389 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7390                         const struct iovec *iov, loff_t offset,
7391                         unsigned long nr_segs)
7392 {
7393         struct file *file = iocb->ki_filp;
7394         struct inode *inode = file->f_mapping->host;
7395         size_t count = 0;
7396         int flags = 0;
7397         bool wakeup = true;
7398         bool relock = false;
7399         ssize_t ret;
7400
7401         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7402                             offset, nr_segs))
7403                 return 0;
7404
7405         atomic_inc(&inode->i_dio_count);
7406         smp_mb__after_atomic_inc();
7407
7408         /*
7409          * The generic stuff only does filemap_write_and_wait_range, which isn't
7410          * enough if we've written compressed pages to this area, so we need to
7411          * call btrfs_wait_ordered_range to make absolutely sure that any
7412          * outstanding dirty pages are on disk.
7413          */
7414         count = iov_length(iov, nr_segs);
7415         ret = btrfs_wait_ordered_range(inode, offset, count);
7416         if (ret)
7417                 return ret;
7418
7419         if (rw & WRITE) {
7420                 /*
7421                  * If the write DIO is beyond the EOF, we need update
7422                  * the isize, but it is protected by i_mutex. So we can
7423                  * not unlock the i_mutex at this case.
7424                  */
7425                 if (offset + count <= inode->i_size) {
7426                         mutex_unlock(&inode->i_mutex);
7427                         relock = true;
7428                 }
7429                 ret = btrfs_delalloc_reserve_space(inode, count);
7430                 if (ret)
7431                         goto out;
7432         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7433                                      &BTRFS_I(inode)->runtime_flags))) {
7434                 inode_dio_done(inode);
7435                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7436                 wakeup = false;
7437         }
7438
7439         ret = __blockdev_direct_IO(rw, iocb, inode,
7440                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7441                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7442                         btrfs_submit_direct, flags);
7443         if (rw & WRITE) {
7444                 if (ret < 0 && ret != -EIOCBQUEUED)
7445                         btrfs_delalloc_release_space(inode, count);
7446                 else if (ret >= 0 && (size_t)ret < count)
7447                         btrfs_delalloc_release_space(inode,
7448                                                      count - (size_t)ret);
7449                 else
7450                         btrfs_delalloc_release_metadata(inode, 0);
7451         }
7452 out:
7453         if (wakeup)
7454                 inode_dio_done(inode);
7455         if (relock)
7456                 mutex_lock(&inode->i_mutex);
7457
7458         return ret;
7459 }
7460
7461 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7462
7463 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7464                 __u64 start, __u64 len)
7465 {
7466         int     ret;
7467
7468         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7469         if (ret)
7470                 return ret;
7471
7472         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7473 }
7474
7475 int btrfs_readpage(struct file *file, struct page *page)
7476 {
7477         struct extent_io_tree *tree;
7478         tree = &BTRFS_I(page->mapping->host)->io_tree;
7479         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7480 }
7481
7482 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7483 {
7484         struct extent_io_tree *tree;
7485
7486
7487         if (current->flags & PF_MEMALLOC) {
7488                 redirty_page_for_writepage(wbc, page);
7489                 unlock_page(page);
7490                 return 0;
7491         }
7492         tree = &BTRFS_I(page->mapping->host)->io_tree;
7493         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7494 }
7495
7496 static int btrfs_writepages(struct address_space *mapping,
7497                             struct writeback_control *wbc)
7498 {
7499         struct extent_io_tree *tree;
7500
7501         tree = &BTRFS_I(mapping->host)->io_tree;
7502         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7503 }
7504
7505 static int
7506 btrfs_readpages(struct file *file, struct address_space *mapping,
7507                 struct list_head *pages, unsigned nr_pages)
7508 {
7509         struct extent_io_tree *tree;
7510         tree = &BTRFS_I(mapping->host)->io_tree;
7511         return extent_readpages(tree, mapping, pages, nr_pages,
7512                                 btrfs_get_extent);
7513 }
7514 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7515 {
7516         struct extent_io_tree *tree;
7517         struct extent_map_tree *map;
7518         int ret;
7519
7520         tree = &BTRFS_I(page->mapping->host)->io_tree;
7521         map = &BTRFS_I(page->mapping->host)->extent_tree;
7522         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7523         if (ret == 1) {
7524                 ClearPagePrivate(page);
7525                 set_page_private(page, 0);
7526                 page_cache_release(page);
7527         }
7528         return ret;
7529 }
7530
7531 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7532 {
7533         if (PageWriteback(page) || PageDirty(page))
7534                 return 0;
7535         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7536 }
7537
7538 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7539                                  unsigned int length)
7540 {
7541         struct inode *inode = page->mapping->host;
7542         struct extent_io_tree *tree;
7543         struct btrfs_ordered_extent *ordered;
7544         struct extent_state *cached_state = NULL;
7545         u64 page_start = page_offset(page);
7546         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7547         int inode_evicting = inode->i_state & I_FREEING;
7548
7549         /*
7550          * we have the page locked, so new writeback can't start,
7551          * and the dirty bit won't be cleared while we are here.
7552          *
7553          * Wait for IO on this page so that we can safely clear
7554          * the PagePrivate2 bit and do ordered accounting
7555          */
7556         wait_on_page_writeback(page);
7557
7558         tree = &BTRFS_I(inode)->io_tree;
7559         if (offset) {
7560                 btrfs_releasepage(page, GFP_NOFS);
7561                 return;
7562         }
7563
7564         if (!inode_evicting)
7565                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7566         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7567         if (ordered) {
7568                 /*
7569                  * IO on this page will never be started, so we need
7570                  * to account for any ordered extents now
7571                  */
7572                 if (!inode_evicting)
7573                         clear_extent_bit(tree, page_start, page_end,
7574                                          EXTENT_DIRTY | EXTENT_DELALLOC |
7575                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7576                                          EXTENT_DEFRAG, 1, 0, &cached_state,
7577                                          GFP_NOFS);
7578                 /*
7579                  * whoever cleared the private bit is responsible
7580                  * for the finish_ordered_io
7581                  */
7582                 if (TestClearPagePrivate2(page)) {
7583                         struct btrfs_ordered_inode_tree *tree;
7584                         u64 new_len;
7585
7586                         tree = &BTRFS_I(inode)->ordered_tree;
7587
7588                         spin_lock_irq(&tree->lock);
7589                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7590                         new_len = page_start - ordered->file_offset;
7591                         if (new_len < ordered->truncated_len)
7592                                 ordered->truncated_len = new_len;
7593                         spin_unlock_irq(&tree->lock);
7594
7595                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7596                                                            page_start,
7597                                                            PAGE_CACHE_SIZE, 1))
7598                                 btrfs_finish_ordered_io(ordered);
7599                 }
7600                 btrfs_put_ordered_extent(ordered);
7601                 if (!inode_evicting) {
7602                         cached_state = NULL;
7603                         lock_extent_bits(tree, page_start, page_end, 0,
7604                                          &cached_state);
7605                 }
7606         }
7607
7608         if (!inode_evicting) {
7609                 clear_extent_bit(tree, page_start, page_end,
7610                                  EXTENT_LOCKED | EXTENT_DIRTY |
7611                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7612                                  EXTENT_DEFRAG, 1, 1,
7613                                  &cached_state, GFP_NOFS);
7614
7615                 __btrfs_releasepage(page, GFP_NOFS);
7616         }
7617
7618         ClearPageChecked(page);
7619         if (PagePrivate(page)) {
7620                 ClearPagePrivate(page);
7621                 set_page_private(page, 0);
7622                 page_cache_release(page);
7623         }
7624 }
7625
7626 /*
7627  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7628  * called from a page fault handler when a page is first dirtied. Hence we must
7629  * be careful to check for EOF conditions here. We set the page up correctly
7630  * for a written page which means we get ENOSPC checking when writing into
7631  * holes and correct delalloc and unwritten extent mapping on filesystems that
7632  * support these features.
7633  *
7634  * We are not allowed to take the i_mutex here so we have to play games to
7635  * protect against truncate races as the page could now be beyond EOF.  Because
7636  * vmtruncate() writes the inode size before removing pages, once we have the
7637  * page lock we can determine safely if the page is beyond EOF. If it is not
7638  * beyond EOF, then the page is guaranteed safe against truncation until we
7639  * unlock the page.
7640  */
7641 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7642 {
7643         struct page *page = vmf->page;
7644         struct inode *inode = file_inode(vma->vm_file);
7645         struct btrfs_root *root = BTRFS_I(inode)->root;
7646         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7647         struct btrfs_ordered_extent *ordered;
7648         struct extent_state *cached_state = NULL;
7649         char *kaddr;
7650         unsigned long zero_start;
7651         loff_t size;
7652         int ret;
7653         int reserved = 0;
7654         u64 page_start;
7655         u64 page_end;
7656
7657         sb_start_pagefault(inode->i_sb);
7658         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7659         if (!ret) {
7660                 ret = file_update_time(vma->vm_file);
7661                 reserved = 1;
7662         }
7663         if (ret) {
7664                 if (ret == -ENOMEM)
7665                         ret = VM_FAULT_OOM;
7666                 else /* -ENOSPC, -EIO, etc */
7667                         ret = VM_FAULT_SIGBUS;
7668                 if (reserved)
7669                         goto out;
7670                 goto out_noreserve;
7671         }
7672
7673         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7674 again:
7675         lock_page(page);
7676         size = i_size_read(inode);
7677         page_start = page_offset(page);
7678         page_end = page_start + PAGE_CACHE_SIZE - 1;
7679
7680         if ((page->mapping != inode->i_mapping) ||
7681             (page_start >= size)) {
7682                 /* page got truncated out from underneath us */
7683                 goto out_unlock;
7684         }
7685         wait_on_page_writeback(page);
7686
7687         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7688         set_page_extent_mapped(page);
7689
7690         /*
7691          * we can't set the delalloc bits if there are pending ordered
7692          * extents.  Drop our locks and wait for them to finish
7693          */
7694         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7695         if (ordered) {
7696                 unlock_extent_cached(io_tree, page_start, page_end,
7697                                      &cached_state, GFP_NOFS);
7698                 unlock_page(page);
7699                 btrfs_start_ordered_extent(inode, ordered, 1);
7700                 btrfs_put_ordered_extent(ordered);
7701                 goto again;
7702         }
7703
7704         /*
7705          * XXX - page_mkwrite gets called every time the page is dirtied, even
7706          * if it was already dirty, so for space accounting reasons we need to
7707          * clear any delalloc bits for the range we are fixing to save.  There
7708          * is probably a better way to do this, but for now keep consistent with
7709          * prepare_pages in the normal write path.
7710          */
7711         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7712                           EXTENT_DIRTY | EXTENT_DELALLOC |
7713                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7714                           0, 0, &cached_state, GFP_NOFS);
7715
7716         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7717                                         &cached_state);
7718         if (ret) {
7719                 unlock_extent_cached(io_tree, page_start, page_end,
7720                                      &cached_state, GFP_NOFS);
7721                 ret = VM_FAULT_SIGBUS;
7722                 goto out_unlock;
7723         }
7724         ret = 0;
7725
7726         /* page is wholly or partially inside EOF */
7727         if (page_start + PAGE_CACHE_SIZE > size)
7728                 zero_start = size & ~PAGE_CACHE_MASK;
7729         else
7730                 zero_start = PAGE_CACHE_SIZE;
7731
7732         if (zero_start != PAGE_CACHE_SIZE) {
7733                 kaddr = kmap(page);
7734                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7735                 flush_dcache_page(page);
7736                 kunmap(page);
7737         }
7738         ClearPageChecked(page);
7739         set_page_dirty(page);
7740         SetPageUptodate(page);
7741
7742         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7743         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7744         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7745
7746         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7747
7748 out_unlock:
7749         if (!ret) {
7750                 sb_end_pagefault(inode->i_sb);
7751                 return VM_FAULT_LOCKED;
7752         }
7753         unlock_page(page);
7754 out:
7755         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7756 out_noreserve:
7757         sb_end_pagefault(inode->i_sb);
7758         return ret;
7759 }
7760
7761 static int btrfs_truncate(struct inode *inode)
7762 {
7763         struct btrfs_root *root = BTRFS_I(inode)->root;
7764         struct btrfs_block_rsv *rsv;
7765         int ret = 0;
7766         int err = 0;
7767         struct btrfs_trans_handle *trans;
7768         u64 mask = root->sectorsize - 1;
7769         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7770
7771         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7772                                        (u64)-1);
7773         if (ret)
7774                 return ret;
7775
7776         /*
7777          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7778          * 3 things going on here
7779          *
7780          * 1) We need to reserve space for our orphan item and the space to
7781          * delete our orphan item.  Lord knows we don't want to have a dangling
7782          * orphan item because we didn't reserve space to remove it.
7783          *
7784          * 2) We need to reserve space to update our inode.
7785          *
7786          * 3) We need to have something to cache all the space that is going to
7787          * be free'd up by the truncate operation, but also have some slack
7788          * space reserved in case it uses space during the truncate (thank you
7789          * very much snapshotting).
7790          *
7791          * And we need these to all be seperate.  The fact is we can use alot of
7792          * space doing the truncate, and we have no earthly idea how much space
7793          * we will use, so we need the truncate reservation to be seperate so it
7794          * doesn't end up using space reserved for updating the inode or
7795          * removing the orphan item.  We also need to be able to stop the
7796          * transaction and start a new one, which means we need to be able to
7797          * update the inode several times, and we have no idea of knowing how
7798          * many times that will be, so we can't just reserve 1 item for the
7799          * entirety of the opration, so that has to be done seperately as well.
7800          * Then there is the orphan item, which does indeed need to be held on
7801          * to for the whole operation, and we need nobody to touch this reserved
7802          * space except the orphan code.
7803          *
7804          * So that leaves us with
7805          *
7806          * 1) root->orphan_block_rsv - for the orphan deletion.
7807          * 2) rsv - for the truncate reservation, which we will steal from the
7808          * transaction reservation.
7809          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7810          * updating the inode.
7811          */
7812         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7813         if (!rsv)
7814                 return -ENOMEM;
7815         rsv->size = min_size;
7816         rsv->failfast = 1;
7817
7818         /*
7819          * 1 for the truncate slack space
7820          * 1 for updating the inode.
7821          */
7822         trans = btrfs_start_transaction(root, 2);
7823         if (IS_ERR(trans)) {
7824                 err = PTR_ERR(trans);
7825                 goto out;
7826         }
7827
7828         /* Migrate the slack space for the truncate to our reserve */
7829         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7830                                       min_size);
7831         BUG_ON(ret);
7832
7833         /*
7834          * setattr is responsible for setting the ordered_data_close flag,
7835          * but that is only tested during the last file release.  That
7836          * could happen well after the next commit, leaving a great big
7837          * window where new writes may get lost if someone chooses to write
7838          * to this file after truncating to zero
7839          *
7840          * The inode doesn't have any dirty data here, and so if we commit
7841          * this is a noop.  If someone immediately starts writing to the inode
7842          * it is very likely we'll catch some of their writes in this
7843          * transaction, and the commit will find this file on the ordered
7844          * data list with good things to send down.
7845          *
7846          * This is a best effort solution, there is still a window where
7847          * using truncate to replace the contents of the file will
7848          * end up with a zero length file after a crash.
7849          */
7850         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7851                                            &BTRFS_I(inode)->runtime_flags))
7852                 btrfs_add_ordered_operation(trans, root, inode);
7853
7854         /*
7855          * So if we truncate and then write and fsync we normally would just
7856          * write the extents that changed, which is a problem if we need to
7857          * first truncate that entire inode.  So set this flag so we write out
7858          * all of the extents in the inode to the sync log so we're completely
7859          * safe.
7860          */
7861         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7862         trans->block_rsv = rsv;
7863
7864         while (1) {
7865                 ret = btrfs_truncate_inode_items(trans, root, inode,
7866                                                  inode->i_size,
7867                                                  BTRFS_EXTENT_DATA_KEY);
7868                 if (ret != -ENOSPC) {
7869                         err = ret;
7870                         break;
7871                 }
7872
7873                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7874                 ret = btrfs_update_inode(trans, root, inode);
7875                 if (ret) {
7876                         err = ret;
7877                         break;
7878                 }
7879
7880                 btrfs_end_transaction(trans, root);
7881                 btrfs_btree_balance_dirty(root);
7882
7883                 trans = btrfs_start_transaction(root, 2);
7884                 if (IS_ERR(trans)) {
7885                         ret = err = PTR_ERR(trans);
7886                         trans = NULL;
7887                         break;
7888                 }
7889
7890                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7891                                               rsv, min_size);
7892                 BUG_ON(ret);    /* shouldn't happen */
7893                 trans->block_rsv = rsv;
7894         }
7895
7896         if (ret == 0 && inode->i_nlink > 0) {
7897                 trans->block_rsv = root->orphan_block_rsv;
7898                 ret = btrfs_orphan_del(trans, inode);
7899                 if (ret)
7900                         err = ret;
7901         }
7902
7903         if (trans) {
7904                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7905                 ret = btrfs_update_inode(trans, root, inode);
7906                 if (ret && !err)
7907                         err = ret;
7908
7909                 ret = btrfs_end_transaction(trans, root);
7910                 btrfs_btree_balance_dirty(root);
7911         }
7912
7913 out:
7914         btrfs_free_block_rsv(root, rsv);
7915
7916         if (ret && !err)
7917                 err = ret;
7918
7919         return err;
7920 }
7921
7922 /*
7923  * create a new subvolume directory/inode (helper for the ioctl).
7924  */
7925 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7926                              struct btrfs_root *new_root,
7927                              struct btrfs_root *parent_root,
7928                              u64 new_dirid)
7929 {
7930         struct inode *inode;
7931         int err;
7932         u64 index = 0;
7933
7934         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7935                                 new_dirid, new_dirid,
7936                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7937                                 &index);
7938         if (IS_ERR(inode))
7939                 return PTR_ERR(inode);
7940         inode->i_op = &btrfs_dir_inode_operations;
7941         inode->i_fop = &btrfs_dir_file_operations;
7942
7943         set_nlink(inode, 1);
7944         btrfs_i_size_write(inode, 0);
7945
7946         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
7947         if (err)
7948                 btrfs_err(new_root->fs_info,
7949                           "error inheriting subvolume %llu properties: %d\n",
7950                           new_root->root_key.objectid, err);
7951
7952         err = btrfs_update_inode(trans, new_root, inode);
7953
7954         iput(inode);
7955         return err;
7956 }
7957
7958 struct inode *btrfs_alloc_inode(struct super_block *sb)
7959 {
7960         struct btrfs_inode *ei;
7961         struct inode *inode;
7962
7963         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7964         if (!ei)
7965                 return NULL;
7966
7967         ei->root = NULL;
7968         ei->generation = 0;
7969         ei->last_trans = 0;
7970         ei->last_sub_trans = 0;
7971         ei->logged_trans = 0;
7972         ei->delalloc_bytes = 0;
7973         ei->disk_i_size = 0;
7974         ei->flags = 0;
7975         ei->csum_bytes = 0;
7976         ei->index_cnt = (u64)-1;
7977         ei->dir_index = 0;
7978         ei->last_unlink_trans = 0;
7979         ei->last_log_commit = 0;
7980
7981         spin_lock_init(&ei->lock);
7982         ei->outstanding_extents = 0;
7983         ei->reserved_extents = 0;
7984
7985         ei->runtime_flags = 0;
7986         ei->force_compress = BTRFS_COMPRESS_NONE;
7987
7988         ei->delayed_node = NULL;
7989
7990         inode = &ei->vfs_inode;
7991         extent_map_tree_init(&ei->extent_tree);
7992         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7993         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7994         ei->io_tree.track_uptodate = 1;
7995         ei->io_failure_tree.track_uptodate = 1;
7996         atomic_set(&ei->sync_writers, 0);
7997         mutex_init(&ei->log_mutex);
7998         mutex_init(&ei->delalloc_mutex);
7999         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8000         INIT_LIST_HEAD(&ei->delalloc_inodes);
8001         INIT_LIST_HEAD(&ei->ordered_operations);
8002         RB_CLEAR_NODE(&ei->rb_node);
8003
8004         return inode;
8005 }
8006
8007 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8008 void btrfs_test_destroy_inode(struct inode *inode)
8009 {
8010         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8011         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8012 }
8013 #endif
8014
8015 static void btrfs_i_callback(struct rcu_head *head)
8016 {
8017         struct inode *inode = container_of(head, struct inode, i_rcu);
8018         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8019 }
8020
8021 void btrfs_destroy_inode(struct inode *inode)
8022 {
8023         struct btrfs_ordered_extent *ordered;
8024         struct btrfs_root *root = BTRFS_I(inode)->root;
8025
8026         WARN_ON(!hlist_empty(&inode->i_dentry));
8027         WARN_ON(inode->i_data.nrpages);
8028         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8029         WARN_ON(BTRFS_I(inode)->reserved_extents);
8030         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8031         WARN_ON(BTRFS_I(inode)->csum_bytes);
8032
8033         /*
8034          * This can happen where we create an inode, but somebody else also
8035          * created the same inode and we need to destroy the one we already
8036          * created.
8037          */
8038         if (!root)
8039                 goto free;
8040
8041         /*
8042          * Make sure we're properly removed from the ordered operation
8043          * lists.
8044          */
8045         smp_mb();
8046         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8047                 spin_lock(&root->fs_info->ordered_root_lock);
8048                 list_del_init(&BTRFS_I(inode)->ordered_operations);
8049                 spin_unlock(&root->fs_info->ordered_root_lock);
8050         }
8051
8052         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8053                      &BTRFS_I(inode)->runtime_flags)) {
8054                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8055                         btrfs_ino(inode));
8056                 atomic_dec(&root->orphan_inodes);
8057         }
8058
8059         while (1) {
8060                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8061                 if (!ordered)
8062                         break;
8063                 else {
8064                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8065                                 ordered->file_offset, ordered->len);
8066                         btrfs_remove_ordered_extent(inode, ordered);
8067                         btrfs_put_ordered_extent(ordered);
8068                         btrfs_put_ordered_extent(ordered);
8069                 }
8070         }
8071         inode_tree_del(inode);
8072         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8073 free:
8074         call_rcu(&inode->i_rcu, btrfs_i_callback);
8075 }
8076
8077 int btrfs_drop_inode(struct inode *inode)
8078 {
8079         struct btrfs_root *root = BTRFS_I(inode)->root;
8080
8081         if (root == NULL)
8082                 return 1;
8083
8084         /* the snap/subvol tree is on deleting */
8085         if (btrfs_root_refs(&root->root_item) == 0)
8086                 return 1;
8087         else
8088                 return generic_drop_inode(inode);
8089 }
8090
8091 static void init_once(void *foo)
8092 {
8093         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8094
8095         inode_init_once(&ei->vfs_inode);
8096 }
8097
8098 void btrfs_destroy_cachep(void)
8099 {
8100         /*
8101          * Make sure all delayed rcu free inodes are flushed before we
8102          * destroy cache.
8103          */
8104         rcu_barrier();
8105         if (btrfs_inode_cachep)
8106                 kmem_cache_destroy(btrfs_inode_cachep);
8107         if (btrfs_trans_handle_cachep)
8108                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8109         if (btrfs_transaction_cachep)
8110                 kmem_cache_destroy(btrfs_transaction_cachep);
8111         if (btrfs_path_cachep)
8112                 kmem_cache_destroy(btrfs_path_cachep);
8113         if (btrfs_free_space_cachep)
8114                 kmem_cache_destroy(btrfs_free_space_cachep);
8115         if (btrfs_delalloc_work_cachep)
8116                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8117 }
8118
8119 int btrfs_init_cachep(void)
8120 {
8121         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8122                         sizeof(struct btrfs_inode), 0,
8123                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8124         if (!btrfs_inode_cachep)
8125                 goto fail;
8126
8127         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8128                         sizeof(struct btrfs_trans_handle), 0,
8129                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8130         if (!btrfs_trans_handle_cachep)
8131                 goto fail;
8132
8133         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8134                         sizeof(struct btrfs_transaction), 0,
8135                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8136         if (!btrfs_transaction_cachep)
8137                 goto fail;
8138
8139         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8140                         sizeof(struct btrfs_path), 0,
8141                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8142         if (!btrfs_path_cachep)
8143                 goto fail;
8144
8145         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8146                         sizeof(struct btrfs_free_space), 0,
8147                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8148         if (!btrfs_free_space_cachep)
8149                 goto fail;
8150
8151         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8152                         sizeof(struct btrfs_delalloc_work), 0,
8153                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8154                         NULL);
8155         if (!btrfs_delalloc_work_cachep)
8156                 goto fail;
8157
8158         return 0;
8159 fail:
8160         btrfs_destroy_cachep();
8161         return -ENOMEM;
8162 }
8163
8164 static int btrfs_getattr(struct vfsmount *mnt,
8165                          struct dentry *dentry, struct kstat *stat)
8166 {
8167         u64 delalloc_bytes;
8168         struct inode *inode = dentry->d_inode;
8169         u32 blocksize = inode->i_sb->s_blocksize;
8170
8171         generic_fillattr(inode, stat);
8172         stat->dev = BTRFS_I(inode)->root->anon_dev;
8173         stat->blksize = PAGE_CACHE_SIZE;
8174
8175         spin_lock(&BTRFS_I(inode)->lock);
8176         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8177         spin_unlock(&BTRFS_I(inode)->lock);
8178         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8179                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8180         return 0;
8181 }
8182
8183 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8184                            struct inode *new_dir, struct dentry *new_dentry)
8185 {
8186         struct btrfs_trans_handle *trans;
8187         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8188         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8189         struct inode *new_inode = new_dentry->d_inode;
8190         struct inode *old_inode = old_dentry->d_inode;
8191         struct timespec ctime = CURRENT_TIME;
8192         u64 index = 0;
8193         u64 root_objectid;
8194         int ret;
8195         u64 old_ino = btrfs_ino(old_inode);
8196
8197         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8198                 return -EPERM;
8199
8200         /* we only allow rename subvolume link between subvolumes */
8201         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8202                 return -EXDEV;
8203
8204         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8205             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8206                 return -ENOTEMPTY;
8207
8208         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8209             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8210                 return -ENOTEMPTY;
8211
8212
8213         /* check for collisions, even if the  name isn't there */
8214         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8215                              new_dentry->d_name.name,
8216                              new_dentry->d_name.len);
8217
8218         if (ret) {
8219                 if (ret == -EEXIST) {
8220                         /* we shouldn't get
8221                          * eexist without a new_inode */
8222                         if (WARN_ON(!new_inode)) {
8223                                 return ret;
8224                         }
8225                 } else {
8226                         /* maybe -EOVERFLOW */
8227                         return ret;
8228                 }
8229         }
8230         ret = 0;
8231
8232         /*
8233          * we're using rename to replace one file with another.
8234          * and the replacement file is large.  Start IO on it now so
8235          * we don't add too much work to the end of the transaction
8236          */
8237         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8238             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8239                 filemap_flush(old_inode->i_mapping);
8240
8241         /* close the racy window with snapshot create/destroy ioctl */
8242         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8243                 down_read(&root->fs_info->subvol_sem);
8244         /*
8245          * We want to reserve the absolute worst case amount of items.  So if
8246          * both inodes are subvols and we need to unlink them then that would
8247          * require 4 item modifications, but if they are both normal inodes it
8248          * would require 5 item modifications, so we'll assume their normal
8249          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8250          * should cover the worst case number of items we'll modify.
8251          */
8252         trans = btrfs_start_transaction(root, 11);
8253         if (IS_ERR(trans)) {
8254                 ret = PTR_ERR(trans);
8255                 goto out_notrans;
8256         }
8257
8258         if (dest != root)
8259                 btrfs_record_root_in_trans(trans, dest);
8260
8261         ret = btrfs_set_inode_index(new_dir, &index);
8262         if (ret)
8263                 goto out_fail;
8264
8265         BTRFS_I(old_inode)->dir_index = 0ULL;
8266         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8267                 /* force full log commit if subvolume involved. */
8268                 root->fs_info->last_trans_log_full_commit = trans->transid;
8269         } else {
8270                 ret = btrfs_insert_inode_ref(trans, dest,
8271                                              new_dentry->d_name.name,
8272                                              new_dentry->d_name.len,
8273                                              old_ino,
8274                                              btrfs_ino(new_dir), index);
8275                 if (ret)
8276                         goto out_fail;
8277                 /*
8278                  * this is an ugly little race, but the rename is required
8279                  * to make sure that if we crash, the inode is either at the
8280                  * old name or the new one.  pinning the log transaction lets
8281                  * us make sure we don't allow a log commit to come in after
8282                  * we unlink the name but before we add the new name back in.
8283                  */
8284                 btrfs_pin_log_trans(root);
8285         }
8286         /*
8287          * make sure the inode gets flushed if it is replacing
8288          * something.
8289          */
8290         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8291                 btrfs_add_ordered_operation(trans, root, old_inode);
8292
8293         inode_inc_iversion(old_dir);
8294         inode_inc_iversion(new_dir);
8295         inode_inc_iversion(old_inode);
8296         old_dir->i_ctime = old_dir->i_mtime = ctime;
8297         new_dir->i_ctime = new_dir->i_mtime = ctime;
8298         old_inode->i_ctime = ctime;
8299
8300         if (old_dentry->d_parent != new_dentry->d_parent)
8301                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8302
8303         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8304                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8305                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8306                                         old_dentry->d_name.name,
8307                                         old_dentry->d_name.len);
8308         } else {
8309                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8310                                         old_dentry->d_inode,
8311                                         old_dentry->d_name.name,
8312                                         old_dentry->d_name.len);
8313                 if (!ret)
8314                         ret = btrfs_update_inode(trans, root, old_inode);
8315         }
8316         if (ret) {
8317                 btrfs_abort_transaction(trans, root, ret);
8318                 goto out_fail;
8319         }
8320
8321         if (new_inode) {
8322                 inode_inc_iversion(new_inode);
8323                 new_inode->i_ctime = CURRENT_TIME;
8324                 if (unlikely(btrfs_ino(new_inode) ==
8325                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8326                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8327                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8328                                                 root_objectid,
8329                                                 new_dentry->d_name.name,
8330                                                 new_dentry->d_name.len);
8331                         BUG_ON(new_inode->i_nlink == 0);
8332                 } else {
8333                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8334                                                  new_dentry->d_inode,
8335                                                  new_dentry->d_name.name,
8336                                                  new_dentry->d_name.len);
8337                 }
8338                 if (!ret && new_inode->i_nlink == 0)
8339                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8340                 if (ret) {
8341                         btrfs_abort_transaction(trans, root, ret);
8342                         goto out_fail;
8343                 }
8344         }
8345
8346         ret = btrfs_add_link(trans, new_dir, old_inode,
8347                              new_dentry->d_name.name,
8348                              new_dentry->d_name.len, 0, index);
8349         if (ret) {
8350                 btrfs_abort_transaction(trans, root, ret);
8351                 goto out_fail;
8352         }
8353
8354         if (old_inode->i_nlink == 1)
8355                 BTRFS_I(old_inode)->dir_index = index;
8356
8357         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8358                 struct dentry *parent = new_dentry->d_parent;
8359                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8360                 btrfs_end_log_trans(root);
8361         }
8362 out_fail:
8363         btrfs_end_transaction(trans, root);
8364 out_notrans:
8365         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8366                 up_read(&root->fs_info->subvol_sem);
8367
8368         return ret;
8369 }
8370
8371 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8372 {
8373         struct btrfs_delalloc_work *delalloc_work;
8374         struct inode *inode;
8375
8376         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8377                                      work);
8378         inode = delalloc_work->inode;
8379         if (delalloc_work->wait) {
8380                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8381         } else {
8382                 filemap_flush(inode->i_mapping);
8383                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8384                              &BTRFS_I(inode)->runtime_flags))
8385                         filemap_flush(inode->i_mapping);
8386         }
8387
8388         if (delalloc_work->delay_iput)
8389                 btrfs_add_delayed_iput(inode);
8390         else
8391                 iput(inode);
8392         complete(&delalloc_work->completion);
8393 }
8394
8395 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8396                                                     int wait, int delay_iput)
8397 {
8398         struct btrfs_delalloc_work *work;
8399
8400         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8401         if (!work)
8402                 return NULL;
8403
8404         init_completion(&work->completion);
8405         INIT_LIST_HEAD(&work->list);
8406         work->inode = inode;
8407         work->wait = wait;
8408         work->delay_iput = delay_iput;
8409         work->work.func = btrfs_run_delalloc_work;
8410
8411         return work;
8412 }
8413
8414 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8415 {
8416         wait_for_completion(&work->completion);
8417         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8418 }
8419
8420 /*
8421  * some fairly slow code that needs optimization. This walks the list
8422  * of all the inodes with pending delalloc and forces them to disk.
8423  */
8424 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8425 {
8426         struct btrfs_inode *binode;
8427         struct inode *inode;
8428         struct btrfs_delalloc_work *work, *next;
8429         struct list_head works;
8430         struct list_head splice;
8431         int ret = 0;
8432
8433         INIT_LIST_HEAD(&works);
8434         INIT_LIST_HEAD(&splice);
8435
8436         spin_lock(&root->delalloc_lock);
8437         list_splice_init(&root->delalloc_inodes, &splice);
8438         while (!list_empty(&splice)) {
8439                 binode = list_entry(splice.next, struct btrfs_inode,
8440                                     delalloc_inodes);
8441
8442                 list_move_tail(&binode->delalloc_inodes,
8443                                &root->delalloc_inodes);
8444                 inode = igrab(&binode->vfs_inode);
8445                 if (!inode) {
8446                         cond_resched_lock(&root->delalloc_lock);
8447                         continue;
8448                 }
8449                 spin_unlock(&root->delalloc_lock);
8450
8451                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8452                 if (unlikely(!work)) {
8453                         if (delay_iput)
8454                                 btrfs_add_delayed_iput(inode);
8455                         else
8456                                 iput(inode);
8457                         ret = -ENOMEM;
8458                         goto out;
8459                 }
8460                 list_add_tail(&work->list, &works);
8461                 btrfs_queue_worker(&root->fs_info->flush_workers,
8462                                    &work->work);
8463
8464                 cond_resched();
8465                 spin_lock(&root->delalloc_lock);
8466         }
8467         spin_unlock(&root->delalloc_lock);
8468
8469         list_for_each_entry_safe(work, next, &works, list) {
8470                 list_del_init(&work->list);
8471                 btrfs_wait_and_free_delalloc_work(work);
8472         }
8473         return 0;
8474 out:
8475         list_for_each_entry_safe(work, next, &works, list) {
8476                 list_del_init(&work->list);
8477                 btrfs_wait_and_free_delalloc_work(work);
8478         }
8479
8480         if (!list_empty_careful(&splice)) {
8481                 spin_lock(&root->delalloc_lock);
8482                 list_splice_tail(&splice, &root->delalloc_inodes);
8483                 spin_unlock(&root->delalloc_lock);
8484         }
8485         return ret;
8486 }
8487
8488 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8489 {
8490         int ret;
8491
8492         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8493                 return -EROFS;
8494
8495         ret = __start_delalloc_inodes(root, delay_iput);
8496         /*
8497          * the filemap_flush will queue IO into the worker threads, but
8498          * we have to make sure the IO is actually started and that
8499          * ordered extents get created before we return
8500          */
8501         atomic_inc(&root->fs_info->async_submit_draining);
8502         while (atomic_read(&root->fs_info->nr_async_submits) ||
8503               atomic_read(&root->fs_info->async_delalloc_pages)) {
8504                 wait_event(root->fs_info->async_submit_wait,
8505                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8506                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8507         }
8508         atomic_dec(&root->fs_info->async_submit_draining);
8509         return ret;
8510 }
8511
8512 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8513 {
8514         struct btrfs_root *root;
8515         struct list_head splice;
8516         int ret;
8517
8518         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8519                 return -EROFS;
8520
8521         INIT_LIST_HEAD(&splice);
8522
8523         spin_lock(&fs_info->delalloc_root_lock);
8524         list_splice_init(&fs_info->delalloc_roots, &splice);
8525         while (!list_empty(&splice)) {
8526                 root = list_first_entry(&splice, struct btrfs_root,
8527                                         delalloc_root);
8528                 root = btrfs_grab_fs_root(root);
8529                 BUG_ON(!root);
8530                 list_move_tail(&root->delalloc_root,
8531                                &fs_info->delalloc_roots);
8532                 spin_unlock(&fs_info->delalloc_root_lock);
8533
8534                 ret = __start_delalloc_inodes(root, delay_iput);
8535                 btrfs_put_fs_root(root);
8536                 if (ret)
8537                         goto out;
8538
8539                 spin_lock(&fs_info->delalloc_root_lock);
8540         }
8541         spin_unlock(&fs_info->delalloc_root_lock);
8542
8543         atomic_inc(&fs_info->async_submit_draining);
8544         while (atomic_read(&fs_info->nr_async_submits) ||
8545               atomic_read(&fs_info->async_delalloc_pages)) {
8546                 wait_event(fs_info->async_submit_wait,
8547                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8548                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8549         }
8550         atomic_dec(&fs_info->async_submit_draining);
8551         return 0;
8552 out:
8553         if (!list_empty_careful(&splice)) {
8554                 spin_lock(&fs_info->delalloc_root_lock);
8555                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8556                 spin_unlock(&fs_info->delalloc_root_lock);
8557         }
8558         return ret;
8559 }
8560
8561 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8562                          const char *symname)
8563 {
8564         struct btrfs_trans_handle *trans;
8565         struct btrfs_root *root = BTRFS_I(dir)->root;
8566         struct btrfs_path *path;
8567         struct btrfs_key key;
8568         struct inode *inode = NULL;
8569         int err;
8570         int drop_inode = 0;
8571         u64 objectid;
8572         u64 index = 0;
8573         int name_len;
8574         int datasize;
8575         unsigned long ptr;
8576         struct btrfs_file_extent_item *ei;
8577         struct extent_buffer *leaf;
8578
8579         name_len = strlen(symname);
8580         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8581                 return -ENAMETOOLONG;
8582
8583         /*
8584          * 2 items for inode item and ref
8585          * 2 items for dir items
8586          * 1 item for xattr if selinux is on
8587          */
8588         trans = btrfs_start_transaction(root, 5);
8589         if (IS_ERR(trans))
8590                 return PTR_ERR(trans);
8591
8592         err = btrfs_find_free_ino(root, &objectid);
8593         if (err)
8594                 goto out_unlock;
8595
8596         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8597                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8598                                 S_IFLNK|S_IRWXUGO, &index);
8599         if (IS_ERR(inode)) {
8600                 err = PTR_ERR(inode);
8601                 goto out_unlock;
8602         }
8603
8604         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8605         if (err) {
8606                 drop_inode = 1;
8607                 goto out_unlock;
8608         }
8609
8610         /*
8611         * If the active LSM wants to access the inode during
8612         * d_instantiate it needs these. Smack checks to see
8613         * if the filesystem supports xattrs by looking at the
8614         * ops vector.
8615         */
8616         inode->i_fop = &btrfs_file_operations;
8617         inode->i_op = &btrfs_file_inode_operations;
8618
8619         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8620         if (err)
8621                 drop_inode = 1;
8622         else {
8623                 inode->i_mapping->a_ops = &btrfs_aops;
8624                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8625                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8626         }
8627         if (drop_inode)
8628                 goto out_unlock;
8629
8630         path = btrfs_alloc_path();
8631         if (!path) {
8632                 err = -ENOMEM;
8633                 drop_inode = 1;
8634                 goto out_unlock;
8635         }
8636         key.objectid = btrfs_ino(inode);
8637         key.offset = 0;
8638         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8639         datasize = btrfs_file_extent_calc_inline_size(name_len);
8640         err = btrfs_insert_empty_item(trans, root, path, &key,
8641                                       datasize);
8642         if (err) {
8643                 drop_inode = 1;
8644                 btrfs_free_path(path);
8645                 goto out_unlock;
8646         }
8647         leaf = path->nodes[0];
8648         ei = btrfs_item_ptr(leaf, path->slots[0],
8649                             struct btrfs_file_extent_item);
8650         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8651         btrfs_set_file_extent_type(leaf, ei,
8652                                    BTRFS_FILE_EXTENT_INLINE);
8653         btrfs_set_file_extent_encryption(leaf, ei, 0);
8654         btrfs_set_file_extent_compression(leaf, ei, 0);
8655         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8656         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8657
8658         ptr = btrfs_file_extent_inline_start(ei);
8659         write_extent_buffer(leaf, symname, ptr, name_len);
8660         btrfs_mark_buffer_dirty(leaf);
8661         btrfs_free_path(path);
8662
8663         inode->i_op = &btrfs_symlink_inode_operations;
8664         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8665         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8666         inode_set_bytes(inode, name_len);
8667         btrfs_i_size_write(inode, name_len);
8668         err = btrfs_update_inode(trans, root, inode);
8669         if (err)
8670                 drop_inode = 1;
8671
8672 out_unlock:
8673         if (!err)
8674                 d_instantiate(dentry, inode);
8675         btrfs_end_transaction(trans, root);
8676         if (drop_inode) {
8677                 inode_dec_link_count(inode);
8678                 iput(inode);
8679         }
8680         btrfs_btree_balance_dirty(root);
8681         return err;
8682 }
8683
8684 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8685                                        u64 start, u64 num_bytes, u64 min_size,
8686                                        loff_t actual_len, u64 *alloc_hint,
8687                                        struct btrfs_trans_handle *trans)
8688 {
8689         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8690         struct extent_map *em;
8691         struct btrfs_root *root = BTRFS_I(inode)->root;
8692         struct btrfs_key ins;
8693         u64 cur_offset = start;
8694         u64 i_size;
8695         u64 cur_bytes;
8696         int ret = 0;
8697         bool own_trans = true;
8698
8699         if (trans)
8700                 own_trans = false;
8701         while (num_bytes > 0) {
8702                 if (own_trans) {
8703                         trans = btrfs_start_transaction(root, 3);
8704                         if (IS_ERR(trans)) {
8705                                 ret = PTR_ERR(trans);
8706                                 break;
8707                         }
8708                 }
8709
8710                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8711                 cur_bytes = max(cur_bytes, min_size);
8712                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8713                                            *alloc_hint, &ins, 1);
8714                 if (ret) {
8715                         if (own_trans)
8716                                 btrfs_end_transaction(trans, root);
8717                         break;
8718                 }
8719
8720                 ret = insert_reserved_file_extent(trans, inode,
8721                                                   cur_offset, ins.objectid,
8722                                                   ins.offset, ins.offset,
8723                                                   ins.offset, 0, 0, 0,
8724                                                   BTRFS_FILE_EXTENT_PREALLOC);
8725                 if (ret) {
8726                         btrfs_free_reserved_extent(root, ins.objectid,
8727                                                    ins.offset);
8728                         btrfs_abort_transaction(trans, root, ret);
8729                         if (own_trans)
8730                                 btrfs_end_transaction(trans, root);
8731                         break;
8732                 }
8733                 btrfs_drop_extent_cache(inode, cur_offset,
8734                                         cur_offset + ins.offset -1, 0);
8735
8736                 em = alloc_extent_map();
8737                 if (!em) {
8738                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8739                                 &BTRFS_I(inode)->runtime_flags);
8740                         goto next;
8741                 }
8742
8743                 em->start = cur_offset;
8744                 em->orig_start = cur_offset;
8745                 em->len = ins.offset;
8746                 em->block_start = ins.objectid;
8747                 em->block_len = ins.offset;
8748                 em->orig_block_len = ins.offset;
8749                 em->ram_bytes = ins.offset;
8750                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8751                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8752                 em->generation = trans->transid;
8753
8754                 while (1) {
8755                         write_lock(&em_tree->lock);
8756                         ret = add_extent_mapping(em_tree, em, 1);
8757                         write_unlock(&em_tree->lock);
8758                         if (ret != -EEXIST)
8759                                 break;
8760                         btrfs_drop_extent_cache(inode, cur_offset,
8761                                                 cur_offset + ins.offset - 1,
8762                                                 0);
8763                 }
8764                 free_extent_map(em);
8765 next:
8766                 num_bytes -= ins.offset;
8767                 cur_offset += ins.offset;
8768                 *alloc_hint = ins.objectid + ins.offset;
8769
8770                 inode_inc_iversion(inode);
8771                 inode->i_ctime = CURRENT_TIME;
8772                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8773                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8774                     (actual_len > inode->i_size) &&
8775                     (cur_offset > inode->i_size)) {
8776                         if (cur_offset > actual_len)
8777                                 i_size = actual_len;
8778                         else
8779                                 i_size = cur_offset;
8780                         i_size_write(inode, i_size);
8781                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8782                 }
8783
8784                 ret = btrfs_update_inode(trans, root, inode);
8785
8786                 if (ret) {
8787                         btrfs_abort_transaction(trans, root, ret);
8788                         if (own_trans)
8789                                 btrfs_end_transaction(trans, root);
8790                         break;
8791                 }
8792
8793                 if (own_trans)
8794                         btrfs_end_transaction(trans, root);
8795         }
8796         return ret;
8797 }
8798
8799 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8800                               u64 start, u64 num_bytes, u64 min_size,
8801                               loff_t actual_len, u64 *alloc_hint)
8802 {
8803         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8804                                            min_size, actual_len, alloc_hint,
8805                                            NULL);
8806 }
8807
8808 int btrfs_prealloc_file_range_trans(struct inode *inode,
8809                                     struct btrfs_trans_handle *trans, int mode,
8810                                     u64 start, u64 num_bytes, u64 min_size,
8811                                     loff_t actual_len, u64 *alloc_hint)
8812 {
8813         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8814                                            min_size, actual_len, alloc_hint, trans);
8815 }
8816
8817 static int btrfs_set_page_dirty(struct page *page)
8818 {
8819         return __set_page_dirty_nobuffers(page);
8820 }
8821
8822 static int btrfs_permission(struct inode *inode, int mask)
8823 {
8824         struct btrfs_root *root = BTRFS_I(inode)->root;
8825         umode_t mode = inode->i_mode;
8826
8827         if (mask & MAY_WRITE &&
8828             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8829                 if (btrfs_root_readonly(root))
8830                         return -EROFS;
8831                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8832                         return -EACCES;
8833         }
8834         return generic_permission(inode, mask);
8835 }
8836
8837 static const struct inode_operations btrfs_dir_inode_operations = {
8838         .getattr        = btrfs_getattr,
8839         .lookup         = btrfs_lookup,
8840         .create         = btrfs_create,
8841         .unlink         = btrfs_unlink,
8842         .link           = btrfs_link,
8843         .mkdir          = btrfs_mkdir,
8844         .rmdir          = btrfs_rmdir,
8845         .rename         = btrfs_rename,
8846         .symlink        = btrfs_symlink,
8847         .setattr        = btrfs_setattr,
8848         .mknod          = btrfs_mknod,
8849         .setxattr       = btrfs_setxattr,
8850         .getxattr       = btrfs_getxattr,
8851         .listxattr      = btrfs_listxattr,
8852         .removexattr    = btrfs_removexattr,
8853         .permission     = btrfs_permission,
8854         .get_acl        = btrfs_get_acl,
8855         .update_time    = btrfs_update_time,
8856 };
8857 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8858         .lookup         = btrfs_lookup,
8859         .permission     = btrfs_permission,
8860         .get_acl        = btrfs_get_acl,
8861         .update_time    = btrfs_update_time,
8862 };
8863
8864 static const struct file_operations btrfs_dir_file_operations = {
8865         .llseek         = generic_file_llseek,
8866         .read           = generic_read_dir,
8867         .iterate        = btrfs_real_readdir,
8868         .unlocked_ioctl = btrfs_ioctl,
8869 #ifdef CONFIG_COMPAT
8870         .compat_ioctl   = btrfs_ioctl,
8871 #endif
8872         .release        = btrfs_release_file,
8873         .fsync          = btrfs_sync_file,
8874 };
8875
8876 static struct extent_io_ops btrfs_extent_io_ops = {
8877         .fill_delalloc = run_delalloc_range,
8878         .submit_bio_hook = btrfs_submit_bio_hook,
8879         .merge_bio_hook = btrfs_merge_bio_hook,
8880         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8881         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8882         .writepage_start_hook = btrfs_writepage_start_hook,
8883         .set_bit_hook = btrfs_set_bit_hook,
8884         .clear_bit_hook = btrfs_clear_bit_hook,
8885         .merge_extent_hook = btrfs_merge_extent_hook,
8886         .split_extent_hook = btrfs_split_extent_hook,
8887 };
8888
8889 /*
8890  * btrfs doesn't support the bmap operation because swapfiles
8891  * use bmap to make a mapping of extents in the file.  They assume
8892  * these extents won't change over the life of the file and they
8893  * use the bmap result to do IO directly to the drive.
8894  *
8895  * the btrfs bmap call would return logical addresses that aren't
8896  * suitable for IO and they also will change frequently as COW
8897  * operations happen.  So, swapfile + btrfs == corruption.
8898  *
8899  * For now we're avoiding this by dropping bmap.
8900  */
8901 static const struct address_space_operations btrfs_aops = {
8902         .readpage       = btrfs_readpage,
8903         .writepage      = btrfs_writepage,
8904         .writepages     = btrfs_writepages,
8905         .readpages      = btrfs_readpages,
8906         .direct_IO      = btrfs_direct_IO,
8907         .invalidatepage = btrfs_invalidatepage,
8908         .releasepage    = btrfs_releasepage,
8909         .set_page_dirty = btrfs_set_page_dirty,
8910         .error_remove_page = generic_error_remove_page,
8911 };
8912
8913 static const struct address_space_operations btrfs_symlink_aops = {
8914         .readpage       = btrfs_readpage,
8915         .writepage      = btrfs_writepage,
8916         .invalidatepage = btrfs_invalidatepage,
8917         .releasepage    = btrfs_releasepage,
8918 };
8919
8920 static const struct inode_operations btrfs_file_inode_operations = {
8921         .getattr        = btrfs_getattr,
8922         .setattr        = btrfs_setattr,
8923         .setxattr       = btrfs_setxattr,
8924         .getxattr       = btrfs_getxattr,
8925         .listxattr      = btrfs_listxattr,
8926         .removexattr    = btrfs_removexattr,
8927         .permission     = btrfs_permission,
8928         .fiemap         = btrfs_fiemap,
8929         .get_acl        = btrfs_get_acl,
8930         .update_time    = btrfs_update_time,
8931 };
8932 static const struct inode_operations btrfs_special_inode_operations = {
8933         .getattr        = btrfs_getattr,
8934         .setattr        = btrfs_setattr,
8935         .permission     = btrfs_permission,
8936         .setxattr       = btrfs_setxattr,
8937         .getxattr       = btrfs_getxattr,
8938         .listxattr      = btrfs_listxattr,
8939         .removexattr    = btrfs_removexattr,
8940         .get_acl        = btrfs_get_acl,
8941         .update_time    = btrfs_update_time,
8942 };
8943 static const struct inode_operations btrfs_symlink_inode_operations = {
8944         .readlink       = generic_readlink,
8945         .follow_link    = page_follow_link_light,
8946         .put_link       = page_put_link,
8947         .getattr        = btrfs_getattr,
8948         .setattr        = btrfs_setattr,
8949         .permission     = btrfs_permission,
8950         .setxattr       = btrfs_setxattr,
8951         .getxattr       = btrfs_getxattr,
8952         .listxattr      = btrfs_listxattr,
8953         .removexattr    = btrfs_removexattr,
8954         .get_acl        = btrfs_get_acl,
8955         .update_time    = btrfs_update_time,
8956 };
8957
8958 const struct dentry_operations btrfs_dentry_operations = {
8959         .d_delete       = btrfs_dentry_delete,
8960         .d_release      = btrfs_dentry_release,
8961 };