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 "compat.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
  61 struct btrfs_iget_args {
  62         u64 ino;
  63         struct btrfs_root *root;
  64 };
  65
  66 static const struct inode_operations btrfs_dir_inode_operations;
  67 static const struct inode_operations btrfs_symlink_inode_operations;
  68 static const struct inode_operations btrfs_dir_ro_inode_operations;
  69 static const struct inode_operations btrfs_special_inode_operations;
  70 static const struct inode_operations btrfs_file_inode_operations;
  71 static const struct address_space_operations btrfs_aops;
  72 static const struct address_space_operations btrfs_symlink_aops;
  73 static const struct file_operations btrfs_dir_file_operations;
  74 static struct extent_io_ops btrfs_extent_io_ops;
  75
  76 static struct kmem_cache *btrfs_inode_cachep;
  77 static struct kmem_cache *btrfs_delalloc_work_cachep;
  78 struct kmem_cache *btrfs_trans_handle_cachep;
  79 struct kmem_cache *btrfs_transaction_cachep;
  80 struct kmem_cache *btrfs_path_cachep;
  81 struct kmem_cache *btrfs_free_space_cachep;
  82
  83 #define S_SHIFT 12
  84 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  85         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  86         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  87         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  88         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  89         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  90         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  91         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  92 };
  93
  94 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  95 static int btrfs_truncate(struct inode *inode);
  96 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  97 static noinline int cow_file_range(struct inode *inode,
  98                                    struct page *locked_page,
  99                                    u64 start, u64 end, int *page_started,
 100                                    unsigned long *nr_written, int unlock);
 101 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 102                                            u64 len, u64 orig_start,
 103                                            u64 block_start, u64 block_len,
 104                                            u64 orig_block_len, u64 ram_bytes,
 105                                            int type);
 106
 107 static int btrfs_dirty_inode(struct inode *inode);
 108
 109 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 110                                      struct inode *inode,  struct inode *dir,
 111                                      const struct qstr *qstr)
 112 {
 113         int err;
 114
 115         err = btrfs_init_acl(trans, inode, dir);
 116         if (!err)
 117                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 118         return err;
 119 }
 120
 121 /*
 122  * this does all the hard work for inserting an inline extent into
 123  * the btree.  The caller should have done a btrfs_drop_extents so that
 124  * no overlapping inline items exist in the btree
 125  */
 126 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 127                                 struct btrfs_root *root, struct inode *inode,
 128                                 u64 start, size_t size, size_t compressed_size,
 129                                 int compress_type,
 130                                 struct page **compressed_pages)
 131 {
 132         struct btrfs_key key;
 133         struct btrfs_path *path;
 134         struct extent_buffer *leaf;
 135         struct page *page = NULL;
 136         char *kaddr;
 137         unsigned long ptr;
 138         struct btrfs_file_extent_item *ei;
 139         int err = 0;
 140         int ret;
 141         size_t cur_size = size;
 142         size_t datasize;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         path = btrfs_alloc_path();
 149         if (!path)
 150                 return -ENOMEM;
 151
 152         path->leave_spinning = 1;
 153
 154         key.objectid = btrfs_ino(inode);
 155         key.offset = start;
 156         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 158
 159         inode_add_bytes(inode, size);
 160         ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                       datasize);
 162         if (ret) {
 163                 err = ret;
 164                 goto fail;
 165         }
 166         leaf = path->nodes[0];
 167         ei = btrfs_item_ptr(leaf, path->slots[0],
 168                             struct btrfs_file_extent_item);
 169         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 170         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 171         btrfs_set_file_extent_encryption(leaf, ei, 0);
 172         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 173         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 174         ptr = btrfs_file_extent_inline_start(ei);
 175
 176         if (compress_type != BTRFS_COMPRESS_NONE) {
 177                 struct page *cpage;
 178                 int i = 0;
 179                 while (compressed_size > 0) {
 180                         cpage = compressed_pages[i];
 181                         cur_size = min_t(unsigned long, compressed_size,
 182                                        PAGE_CACHE_SIZE);
 183
 184                         kaddr = kmap_atomic(cpage);
 185                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 186                         kunmap_atomic(kaddr);
 187
 188                         i++;
 189                         ptr += cur_size;
 190                         compressed_size -= cur_size;
 191                 }
 192                 btrfs_set_file_extent_compression(leaf, ei,
 193                                                   compress_type);
 194         } else {
 195                 page = find_get_page(inode->i_mapping,
 196                                      start >> PAGE_CACHE_SHIFT);
 197                 btrfs_set_file_extent_compression(leaf, ei, 0);
 198                 kaddr = kmap_atomic(page);
 199                 offset = start & (PAGE_CACHE_SIZE - 1);
 200                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 201                 kunmap_atomic(kaddr);
 202                 page_cache_release(page);
 203         }
 204         btrfs_mark_buffer_dirty(leaf);
 205         btrfs_free_path(path);
 206
 207         /*
 208          * we're an inline extent, so nobody can
 209          * extend the file past i_size without locking
 210          * a page we already have locked.
 211          *
 212          * We must do any isize and inode updates
 213          * before we unlock the pages.  Otherwise we
 214          * could end up racing with unlink.
 215          */
 216         BTRFS_I(inode)->disk_i_size = inode->i_size;
 217         ret = btrfs_update_inode(trans, root, inode);
 218
 219         return ret;
 220 fail:
 221         btrfs_free_path(path);
 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_trans_handle *trans,
 232                                  struct btrfs_root *root,
 233                                  struct inode *inode, u64 start, u64 end,
 234                                  size_t compressed_size, int compress_type,
 235                                  struct page **compressed_pages)
 236 {
 237         u64 isize = i_size_read(inode);
 238         u64 actual_end = min(end + 1, isize);
 239         u64 inline_len = actual_end - start;
 240         u64 aligned_end = ALIGN(end, root->sectorsize);
 241         u64 data_len = inline_len;
 242         int ret;
 243
 244         if (compressed_size)
 245                 data_len = compressed_size;
 246
 247         if (start > 0 ||
 248             actual_end >= PAGE_CACHE_SIZE ||
 249             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 250             (!compressed_size &&
 251             (actual_end & (root->sectorsize - 1)) == 0) ||
 252             end + 1 < isize ||
 253             data_len > root->fs_info->max_inline) {
 254                 return 1;
 255         }
 256
 257         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 258         if (ret)
 259                 return ret;
 260
 261         if (isize > actual_end)
 262                 inline_len = min_t(u64, isize, actual_end);
 263         ret = insert_inline_extent(trans, root, inode, start,
 264                                    inline_len, compressed_size,
 265                                    compress_type, compressed_pages);
 266         if (ret && ret != -ENOSPC) {
 267                 btrfs_abort_transaction(trans, root, ret);
 268                 return ret;
 269         } else if (ret == -ENOSPC) {
 270                 return 1;
 271         }
 272
 273         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 274         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 275         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 276         return 0;
 277 }
 278
 279 struct async_extent {
 280         u64 start;
 281         u64 ram_size;
 282         u64 compressed_size;
 283         struct page **pages;
 284         unsigned long nr_pages;
 285         int compress_type;
 286         struct list_head list;
 287 };
 288
 289 struct async_cow {
 290         struct inode *inode;
 291         struct btrfs_root *root;
 292         struct page *locked_page;
 293         u64 start;
 294         u64 end;
 295         struct list_head extents;
 296         struct btrfs_work work;
 297 };
 298
 299 static noinline int add_async_extent(struct async_cow *cow,
 300                                      u64 start, u64 ram_size,
 301                                      u64 compressed_size,
 302                                      struct page **pages,
 303                                      unsigned long nr_pages,
 304                                      int compress_type)
 305 {
 306         struct async_extent *async_extent;
 307
 308         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 309         BUG_ON(!async_extent); /* -ENOMEM */
 310         async_extent->start = start;
 311         async_extent->ram_size = ram_size;
 312         async_extent->compressed_size = compressed_size;
 313         async_extent->pages = pages;
 314         async_extent->nr_pages = nr_pages;
 315         async_extent->compress_type = compress_type;
 316         list_add_tail(&async_extent->list, &cow->extents);
 317         return 0;
 318 }
 319
 320 /*
 321  * we create compressed extents in two phases.  The first
 322  * phase compresses a range of pages that have already been
 323  * locked (both pages and state bits are locked).
 324  *
 325  * This is done inside an ordered work queue, and the compression
 326  * is spread across many cpus.  The actual IO submission is step
 327  * two, and the ordered work queue takes care of making sure that
 328  * happens in the same order things were put onto the queue by
 329  * writepages and friends.
 330  *
 331  * If this code finds it can't get good compression, it puts an
 332  * entry onto the work queue to write the uncompressed bytes.  This
 333  * makes sure that both compressed inodes and uncompressed inodes
 334  * are written in the same order that the flusher thread sent them
 335  * down.
 336  */
 337 static noinline int compress_file_range(struct inode *inode,
 338                                         struct page *locked_page,
 339                                         u64 start, u64 end,
 340                                         struct async_cow *async_cow,
 341                                         int *num_added)
 342 {
 343         struct btrfs_root *root = BTRFS_I(inode)->root;
 344         struct btrfs_trans_handle *trans;
 345         u64 num_bytes;
 346         u64 blocksize = root->sectorsize;
 347         u64 actual_end;
 348         u64 isize = i_size_read(inode);
 349         int ret = 0;
 350         struct page **pages = NULL;
 351         unsigned long nr_pages;
 352         unsigned long nr_pages_ret = 0;
 353         unsigned long total_compressed = 0;
 354         unsigned long total_in = 0;
 355         unsigned long max_compressed = 128 * 1024;
 356         unsigned long max_uncompressed = 128 * 1024;
 357         int i;
 358         int will_compress;
 359         int compress_type = root->fs_info->compress_type;
 360         int redirty = 0;
 361
 362         /* if this is a small write inside eof, kick off a defrag */
 363         if ((end - start + 1) < 16 * 1024 &&
 364             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 365                 btrfs_add_inode_defrag(NULL, inode);
 366
 367         actual_end = min_t(u64, isize, end + 1);
 368 again:
 369         will_compress = 0;
 370         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 371         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 372
 373         /*
 374          * we don't want to send crud past the end of i_size through
 375          * compression, that's just a waste of CPU time.  So, if the
 376          * end of the file is before the start of our current
 377          * requested range of bytes, we bail out to the uncompressed
 378          * cleanup code that can deal with all of this.
 379          *
 380          * It isn't really the fastest way to fix things, but this is a
 381          * very uncommon corner.
 382          */
 383         if (actual_end <= start)
 384                 goto cleanup_and_bail_uncompressed;
 385
 386         total_compressed = actual_end - start;
 387
 388         /* we want to make sure that amount of ram required to uncompress
 389          * an extent is reasonable, so we limit the total size in ram
 390          * of a compressed extent to 128k.  This is a crucial number
 391          * because it also controls how easily we can spread reads across
 392          * cpus for decompression.
 393          *
 394          * We also want to make sure the amount of IO required to do
 395          * a random read is reasonably small, so we limit the size of
 396          * a compressed extent to 128k.
 397          */
 398         total_compressed = min(total_compressed, max_uncompressed);
 399         num_bytes = ALIGN(end - start + 1, blocksize);
 400         num_bytes = max(blocksize,  num_bytes);
 401         total_in = 0;
 402         ret = 0;
 403
 404         /*
 405          * we do compression for mount -o compress and when the
 406          * inode has not been flagged as nocompress.  This flag can
 407          * change at any time if we discover bad compression ratios.
 408          */
 409         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 410             (btrfs_test_opt(root, COMPRESS) ||
 411              (BTRFS_I(inode)->force_compress) ||
 412              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 413                 WARN_ON(pages);
 414                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 415                 if (!pages) {
 416                         /* just bail out to the uncompressed code */
 417                         goto cont;
 418                 }
 419
 420                 if (BTRFS_I(inode)->force_compress)
 421                         compress_type = BTRFS_I(inode)->force_compress;
 422
 423                 /*
 424                  * we need to call clear_page_dirty_for_io on each
 425                  * page in the range.  Otherwise applications with the file
 426                  * mmap'd can wander in and change the page contents while
 427                  * we are compressing them.
 428                  *
 429                  * If the compression fails for any reason, we set the pages
 430                  * dirty again later on.
 431                  */
 432                 extent_range_clear_dirty_for_io(inode, start, end);
 433                 redirty = 1;
 434                 ret = btrfs_compress_pages(compress_type,
 435                                            inode->i_mapping, start,
 436                                            total_compressed, pages,
 437                                            nr_pages, &nr_pages_ret,
 438                                            &total_in,
 439                                            &total_compressed,
 440                                            max_compressed);
 441
 442                 if (!ret) {
 443                         unsigned long offset = total_compressed &
 444                                 (PAGE_CACHE_SIZE - 1);
 445                         struct page *page = pages[nr_pages_ret - 1];
 446                         char *kaddr;
 447
 448                         /* zero the tail end of the last page, we might be
 449                          * sending it down to disk
 450                          */
 451                         if (offset) {
 452                                 kaddr = kmap_atomic(page);
 453                                 memset(kaddr + offset, 0,
 454                                        PAGE_CACHE_SIZE - offset);
 455                                 kunmap_atomic(kaddr);
 456                         }
 457                         will_compress = 1;
 458                 }
 459         }
 460 cont:
 461         if (start == 0) {
 462                 trans = btrfs_join_transaction(root);
 463                 if (IS_ERR(trans)) {
 464                         ret = PTR_ERR(trans);
 465                         trans = NULL;
 466                         goto cleanup_and_out;
 467                 }
 468                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 469
 470                 /* lets try to make an inline extent */
 471                 if (ret || total_in < (actual_end - start)) {
 472                         /* we didn't compress the entire range, try
 473                          * to make an uncompressed inline extent.
 474                          */
 475                         ret = cow_file_range_inline(trans, root, inode,
 476                                                     start, end, 0, 0, NULL);
 477                 } else {
 478                         /* try making a compressed inline extent */
 479                         ret = cow_file_range_inline(trans, root, inode,
 480                                                     start, end,
 481                                                     total_compressed,
 482                                                     compress_type, pages);
 483                 }
 484                 if (ret <= 0) {
 485                         /*
 486                          * inline extent creation worked or returned error,
 487                          * we don't need to create any more async work items.
 488                          * Unlock and free up our temp pages.
 489                          */
 490                         extent_clear_unlock_delalloc(inode,
 491                              &BTRFS_I(inode)->io_tree,
 492                              start, end, NULL,
 493                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 494                              EXTENT_CLEAR_DELALLOC |
 495                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 496
 497                         btrfs_end_transaction(trans, root);
 498                         goto free_pages_out;
 499                 }
 500                 btrfs_end_transaction(trans, root);
 501         }
 502
 503         if (will_compress) {
 504                 /*
 505                  * we aren't doing an inline extent round the compressed size
 506                  * up to a block size boundary so the allocator does sane
 507                  * things
 508                  */
 509                 total_compressed = ALIGN(total_compressed, blocksize);
 510
 511                 /*
 512                  * one last check to make sure the compression is really a
 513                  * win, compare the page count read with the blocks on disk
 514                  */
 515                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 516                 if (total_compressed >= total_in) {
 517                         will_compress = 0;
 518                 } else {
 519                         num_bytes = total_in;
 520                 }
 521         }
 522         if (!will_compress && pages) {
 523                 /*
 524                  * the compression code ran but failed to make things smaller,
 525                  * free any pages it allocated and our page pointer array
 526                  */
 527                 for (i = 0; i < nr_pages_ret; i++) {
 528                         WARN_ON(pages[i]->mapping);
 529                         page_cache_release(pages[i]);
 530                 }
 531                 kfree(pages);
 532                 pages = NULL;
 533                 total_compressed = 0;
 534                 nr_pages_ret = 0;
 535
 536                 /* flag the file so we don't compress in the future */
 537                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 538                     !(BTRFS_I(inode)->force_compress)) {
 539                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 540                 }
 541         }
 542         if (will_compress) {
 543                 *num_added += 1;
 544
 545                 /* the async work queues will take care of doing actual
 546                  * allocation on disk for these compressed pages,
 547                  * and will submit them to the elevator.
 548                  */
 549                 add_async_extent(async_cow, start, num_bytes,
 550                                  total_compressed, pages, nr_pages_ret,
 551                                  compress_type);
 552
 553                 if (start + num_bytes < end) {
 554                         start += num_bytes;
 555                         pages = NULL;
 556                         cond_resched();
 557                         goto again;
 558                 }
 559         } else {
 560 cleanup_and_bail_uncompressed:
 561                 /*
 562                  * No compression, but we still need to write the pages in
 563                  * the file we've been given so far.  redirty the locked
 564                  * page if it corresponds to our extent and set things up
 565                  * for the async work queue to run cow_file_range to do
 566                  * the normal delalloc dance
 567                  */
 568                 if (page_offset(locked_page) >= start &&
 569                     page_offset(locked_page) <= end) {
 570                         __set_page_dirty_nobuffers(locked_page);
 571                         /* unlocked later on in the async handlers */
 572                 }
 573                 if (redirty)
 574                         extent_range_redirty_for_io(inode, start, end);
 575                 add_async_extent(async_cow, start, end - start + 1,
 576                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 577                 *num_added += 1;
 578         }
 579
 580 out:
 581         return ret;
 582
 583 free_pages_out:
 584         for (i = 0; i < nr_pages_ret; i++) {
 585                 WARN_ON(pages[i]->mapping);
 586                 page_cache_release(pages[i]);
 587         }
 588         kfree(pages);
 589
 590         goto out;
 591
 592 cleanup_and_out:
 593         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 594                                      start, end, NULL,
 595                                      EXTENT_CLEAR_UNLOCK_PAGE |
 596                                      EXTENT_CLEAR_DIRTY |
 597                                      EXTENT_CLEAR_DELALLOC |
 598                                      EXTENT_SET_WRITEBACK |
 599                                      EXTENT_END_WRITEBACK);
 600         if (!trans || IS_ERR(trans))
 601                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 602         else
 603                 btrfs_abort_transaction(trans, root, ret);
 604         goto free_pages_out;
 605 }
 606
 607 /*
 608  * phase two of compressed writeback.  This is the ordered portion
 609  * of the code, which only gets called in the order the work was
 610  * queued.  We walk all the async extents created by compress_file_range
 611  * and send them down to the disk.
 612  */
 613 static noinline int submit_compressed_extents(struct inode *inode,
 614                                               struct async_cow *async_cow)
 615 {
 616         struct async_extent *async_extent;
 617         u64 alloc_hint = 0;
 618         struct btrfs_trans_handle *trans;
 619         struct btrfs_key ins;
 620         struct extent_map *em;
 621         struct btrfs_root *root = BTRFS_I(inode)->root;
 622         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 623         struct extent_io_tree *io_tree;
 624         int ret = 0;
 625
 626         if (list_empty(&async_cow->extents))
 627                 return 0;
 628
 629 again:
 630         while (!list_empty(&async_cow->extents)) {
 631                 async_extent = list_entry(async_cow->extents.next,
 632                                           struct async_extent, list);
 633                 list_del(&async_extent->list);
 634
 635                 io_tree = &BTRFS_I(inode)->io_tree;
 636
 637 retry:
 638                 /* did the compression code fall back to uncompressed IO? */
 639                 if (!async_extent->pages) {
 640                         int page_started = 0;
 641                         unsigned long nr_written = 0;
 642
 643                         lock_extent(io_tree, async_extent->start,
 644                                          async_extent->start +
 645                                          async_extent->ram_size - 1);
 646
 647                         /* allocate blocks */
 648                         ret = cow_file_range(inode, async_cow->locked_page,
 649                                              async_extent->start,
 650                                              async_extent->start +
 651                                              async_extent->ram_size - 1,
 652                                              &page_started, &nr_written, 0);
 653
 654                         /* JDM XXX */
 655
 656                         /*
 657                          * if page_started, cow_file_range inserted an
 658                          * inline extent and took care of all the unlocking
 659                          * and IO for us.  Otherwise, we need to submit
 660                          * all those pages down to the drive.
 661                          */
 662                         if (!page_started && !ret)
 663                                 extent_write_locked_range(io_tree,
 664                                                   inode, async_extent->start,
 665                                                   async_extent->start +
 666                                                   async_extent->ram_size - 1,
 667                                                   btrfs_get_extent,
 668                                                   WB_SYNC_ALL);
 669                         else if (ret)
 670                                 unlock_page(async_cow->locked_page);
 671                         kfree(async_extent);
 672                         cond_resched();
 673                         continue;
 674                 }
 675
 676                 lock_extent(io_tree, async_extent->start,
 677                             async_extent->start + async_extent->ram_size - 1);
 678
 679                 trans = btrfs_join_transaction(root);
 680                 if (IS_ERR(trans)) {
 681                         ret = PTR_ERR(trans);
 682                 } else {
 683                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 684                         ret = btrfs_reserve_extent(trans, root,
 685                                            async_extent->compressed_size,
 686                                            async_extent->compressed_size,
 687                                            0, alloc_hint, &ins, 1);
 688                         if (ret && ret != -ENOSPC)
 689                                 btrfs_abort_transaction(trans, root, ret);
 690                         btrfs_end_transaction(trans, root);
 691                 }
 692
 693                 if (ret) {
 694                         int i;
 695
 696                         for (i = 0; i < async_extent->nr_pages; i++) {
 697                                 WARN_ON(async_extent->pages[i]->mapping);
 698                                 page_cache_release(async_extent->pages[i]);
 699                         }
 700                         kfree(async_extent->pages);
 701                         async_extent->nr_pages = 0;
 702                         async_extent->pages = NULL;
 703
 704                         if (ret == -ENOSPC)
 705                                 goto retry;
 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,
 768                                 &BTRFS_I(inode)->io_tree,
 769                                 async_extent->start,
 770                                 async_extent->start +
 771                                 async_extent->ram_size - 1,
 772                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 773                                 EXTENT_CLEAR_UNLOCK |
 774                                 EXTENT_CLEAR_DELALLOC |
 775                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 776
 777                 ret = btrfs_submit_compressed_write(inode,
 778                                     async_extent->start,
 779                                     async_extent->ram_size,
 780                                     ins.objectid,
 781                                     ins.offset, async_extent->pages,
 782                                     async_extent->nr_pages);
 783                 alloc_hint = ins.objectid + ins.offset;
 784                 kfree(async_extent);
 785                 if (ret)
 786                         goto out;
 787                 cond_resched();
 788         }
 789         ret = 0;
 790 out:
 791         return ret;
 792 out_free_reserve:
 793         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 794 out_free:
 795         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 796                                      async_extent->start,
 797                                      async_extent->start +
 798                                      async_extent->ram_size - 1,
 799                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 800                                      EXTENT_CLEAR_UNLOCK |
 801                                      EXTENT_CLEAR_DELALLOC |
 802                                      EXTENT_CLEAR_DIRTY |
 803                                      EXTENT_SET_WRITEBACK |
 804                                      EXTENT_END_WRITEBACK);
 805         kfree(async_extent);
 806         goto again;
 807 }
 808
 809 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 810                                       u64 num_bytes)
 811 {
 812         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 813         struct extent_map *em;
 814         u64 alloc_hint = 0;
 815
 816         read_lock(&em_tree->lock);
 817         em = search_extent_mapping(em_tree, start, num_bytes);
 818         if (em) {
 819                 /*
 820                  * if block start isn't an actual block number then find the
 821                  * first block in this inode and use that as a hint.  If that
 822                  * block is also bogus then just don't worry about it.
 823                  */
 824                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 825                         free_extent_map(em);
 826                         em = search_extent_mapping(em_tree, 0, 0);
 827                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 828                                 alloc_hint = em->block_start;
 829                         if (em)
 830                                 free_extent_map(em);
 831                 } else {
 832                         alloc_hint = em->block_start;
 833                         free_extent_map(em);
 834                 }
 835         }
 836         read_unlock(&em_tree->lock);
 837
 838         return alloc_hint;
 839 }
 840
 841 /*
 842  * when extent_io.c finds a delayed allocation range in the file,
 843  * the call backs end up in this code.  The basic idea is to
 844  * allocate extents on disk for the range, and create ordered data structs
 845  * in ram to track those extents.
 846  *
 847  * locked_page is the page that writepage had locked already.  We use
 848  * it to make sure we don't do extra locks or unlocks.
 849  *
 850  * *page_started is set to one if we unlock locked_page and do everything
 851  * required to start IO on it.  It may be clean and already done with
 852  * IO when we return.
 853  */
 854 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 855                                      struct inode *inode,
 856                                      struct btrfs_root *root,
 857                                      struct page *locked_page,
 858                                      u64 start, u64 end, int *page_started,
 859                                      unsigned long *nr_written,
 860                                      int unlock)
 861 {
 862         u64 alloc_hint = 0;
 863         u64 num_bytes;
 864         unsigned long ram_size;
 865         u64 disk_num_bytes;
 866         u64 cur_alloc_size;
 867         u64 blocksize = root->sectorsize;
 868         struct btrfs_key ins;
 869         struct extent_map *em;
 870         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 871         int ret = 0;
 872
 873         BUG_ON(btrfs_is_free_space_inode(inode));
 874
 875         num_bytes = ALIGN(end - start + 1, blocksize);
 876         num_bytes = max(blocksize,  num_bytes);
 877         disk_num_bytes = num_bytes;
 878
 879         /* if this is a small write inside eof, kick off defrag */
 880         if (num_bytes < 64 * 1024 &&
 881             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 882                 btrfs_add_inode_defrag(trans, inode);
 883
 884         if (start == 0) {
 885                 /* lets try to make an inline extent */
 886                 ret = cow_file_range_inline(trans, root, inode,
 887                                             start, end, 0, 0, NULL);
 888                 if (ret == 0) {
 889                         extent_clear_unlock_delalloc(inode,
 890                                      &BTRFS_I(inode)->io_tree,
 891                                      start, end, NULL,
 892                                      EXTENT_CLEAR_UNLOCK_PAGE |
 893                                      EXTENT_CLEAR_UNLOCK |
 894                                      EXTENT_CLEAR_DELALLOC |
 895                                      EXTENT_CLEAR_DIRTY |
 896                                      EXTENT_SET_WRITEBACK |
 897                                      EXTENT_END_WRITEBACK);
 898
 899                         *nr_written = *nr_written +
 900                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 901                         *page_started = 1;
 902                         goto out;
 903                 } else if (ret < 0) {
 904                         btrfs_abort_transaction(trans, root, ret);
 905                         goto out_unlock;
 906                 }
 907         }
 908
 909         BUG_ON(disk_num_bytes >
 910                btrfs_super_total_bytes(root->fs_info->super_copy));
 911
 912         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 913         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 914
 915         while (disk_num_bytes > 0) {
 916                 unsigned long op;
 917
 918                 cur_alloc_size = disk_num_bytes;
 919                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 920                                            root->sectorsize, 0, alloc_hint,
 921                                            &ins, 1);
 922                 if (ret < 0) {
 923                         btrfs_abort_transaction(trans, root, ret);
 924                         goto out_unlock;
 925                 }
 926
 927                 em = alloc_extent_map();
 928                 if (!em) {
 929                         ret = -ENOMEM;
 930                         goto out_reserve;
 931                 }
 932                 em->start = start;
 933                 em->orig_start = em->start;
 934                 ram_size = ins.offset;
 935                 em->len = ins.offset;
 936                 em->mod_start = em->start;
 937                 em->mod_len = em->len;
 938
 939                 em->block_start = ins.objectid;
 940                 em->block_len = ins.offset;
 941                 em->orig_block_len = ins.offset;
 942                 em->ram_bytes = ram_size;
 943                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 944                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 945                 em->generation = -1;
 946
 947                 while (1) {
 948                         write_lock(&em_tree->lock);
 949                         ret = add_extent_mapping(em_tree, em, 1);
 950                         write_unlock(&em_tree->lock);
 951                         if (ret != -EEXIST) {
 952                                 free_extent_map(em);
 953                                 break;
 954                         }
 955                         btrfs_drop_extent_cache(inode, start,
 956                                                 start + ram_size - 1, 0);
 957                 }
 958                 if (ret)
 959                         goto out_reserve;
 960
 961                 cur_alloc_size = ins.offset;
 962                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 963                                                ram_size, cur_alloc_size, 0);
 964                 if (ret)
 965                         goto out_reserve;
 966
 967                 if (root->root_key.objectid ==
 968                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 969                         ret = btrfs_reloc_clone_csums(inode, start,
 970                                                       cur_alloc_size);
 971                         if (ret) {
 972                                 btrfs_abort_transaction(trans, root, ret);
 973                                 goto out_reserve;
 974                         }
 975                 }
 976
 977                 if (disk_num_bytes < cur_alloc_size)
 978                         break;
 979
 980                 /* we're not doing compressed IO, don't unlock the first
 981                  * page (which the caller expects to stay locked), don't
 982                  * clear any dirty bits and don't set any writeback bits
 983                  *
 984                  * Do set the Private2 bit so we know this page was properly
 985                  * setup for writepage
 986                  */
 987                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 988                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 989                         EXTENT_SET_PRIVATE2;
 990
 991                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 992                                              start, start + ram_size - 1,
 993                                              locked_page, op);
 994                 disk_num_bytes -= cur_alloc_size;
 995                 num_bytes -= cur_alloc_size;
 996                 alloc_hint = ins.objectid + ins.offset;
 997                 start += cur_alloc_size;
 998         }
 999 out:
1000         return ret;
1001
1002 out_reserve:
1003         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1004 out_unlock:
1005         extent_clear_unlock_delalloc(inode,
1006                      &BTRFS_I(inode)->io_tree,
1007                      start, end, locked_page,
1008                      EXTENT_CLEAR_UNLOCK_PAGE |
1009                      EXTENT_CLEAR_UNLOCK |
1010                      EXTENT_CLEAR_DELALLOC |
1011                      EXTENT_CLEAR_DIRTY |
1012                      EXTENT_SET_WRITEBACK |
1013                      EXTENT_END_WRITEBACK);
1014
1015         goto out;
1016 }
1017
1018 static noinline int cow_file_range(struct inode *inode,
1019                                    struct page *locked_page,
1020                                    u64 start, u64 end, int *page_started,
1021                                    unsigned long *nr_written,
1022                                    int unlock)
1023 {
1024         struct btrfs_trans_handle *trans;
1025         struct btrfs_root *root = BTRFS_I(inode)->root;
1026         int ret;
1027
1028         trans = btrfs_join_transaction(root);
1029         if (IS_ERR(trans)) {
1030                 extent_clear_unlock_delalloc(inode,
1031                              &BTRFS_I(inode)->io_tree,
1032                              start, end, locked_page,
1033                              EXTENT_CLEAR_UNLOCK_PAGE |
1034                              EXTENT_CLEAR_UNLOCK |
1035                              EXTENT_CLEAR_DELALLOC |
1036                              EXTENT_CLEAR_DIRTY |
1037                              EXTENT_SET_WRITEBACK |
1038                              EXTENT_END_WRITEBACK);
1039                 return PTR_ERR(trans);
1040         }
1041         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1042
1043         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1044                                page_started, nr_written, unlock);
1045
1046         btrfs_end_transaction(trans, root);
1047
1048         return ret;
1049 }
1050
1051 /*
1052  * work queue call back to started compression on a file and pages
1053  */
1054 static noinline void async_cow_start(struct btrfs_work *work)
1055 {
1056         struct async_cow *async_cow;
1057         int num_added = 0;
1058         async_cow = container_of(work, struct async_cow, work);
1059
1060         compress_file_range(async_cow->inode, async_cow->locked_page,
1061                             async_cow->start, async_cow->end, async_cow,
1062                             &num_added);
1063         if (num_added == 0) {
1064                 btrfs_add_delayed_iput(async_cow->inode);
1065                 async_cow->inode = NULL;
1066         }
1067 }
1068
1069 /*
1070  * work queue call back to submit previously compressed pages
1071  */
1072 static noinline void async_cow_submit(struct btrfs_work *work)
1073 {
1074         struct async_cow *async_cow;
1075         struct btrfs_root *root;
1076         unsigned long nr_pages;
1077
1078         async_cow = container_of(work, struct async_cow, work);
1079
1080         root = async_cow->root;
1081         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1082                 PAGE_CACHE_SHIFT;
1083
1084         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1085             5 * 1024 * 1024 &&
1086             waitqueue_active(&root->fs_info->async_submit_wait))
1087                 wake_up(&root->fs_info->async_submit_wait);
1088
1089         if (async_cow->inode)
1090                 submit_compressed_extents(async_cow->inode, async_cow);
1091 }
1092
1093 static noinline void async_cow_free(struct btrfs_work *work)
1094 {
1095         struct async_cow *async_cow;
1096         async_cow = container_of(work, struct async_cow, work);
1097         if (async_cow->inode)
1098                 btrfs_add_delayed_iput(async_cow->inode);
1099         kfree(async_cow);
1100 }
1101
1102 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1103                                 u64 start, u64 end, int *page_started,
1104                                 unsigned long *nr_written)
1105 {
1106         struct async_cow *async_cow;
1107         struct btrfs_root *root = BTRFS_I(inode)->root;
1108         unsigned long nr_pages;
1109         u64 cur_end;
1110         int limit = 10 * 1024 * 1024;
1111
1112         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1113                          1, 0, NULL, GFP_NOFS);
1114         while (start < end) {
1115                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1116                 BUG_ON(!async_cow); /* -ENOMEM */
1117                 async_cow->inode = igrab(inode);
1118                 async_cow->root = root;
1119                 async_cow->locked_page = locked_page;
1120                 async_cow->start = start;
1121
1122                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1123                         cur_end = end;
1124                 else
1125                         cur_end = min(end, start + 512 * 1024 - 1);
1126
1127                 async_cow->end = cur_end;
1128                 INIT_LIST_HEAD(&async_cow->extents);
1129
1130                 async_cow->work.func = async_cow_start;
1131                 async_cow->work.ordered_func = async_cow_submit;
1132                 async_cow->work.ordered_free = async_cow_free;
1133                 async_cow->work.flags = 0;
1134
1135                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1136                         PAGE_CACHE_SHIFT;
1137                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1138
1139                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1140                                    &async_cow->work);
1141
1142                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1143                         wait_event(root->fs_info->async_submit_wait,
1144                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1145                             limit));
1146                 }
1147
1148                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1149                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1150                         wait_event(root->fs_info->async_submit_wait,
1151                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1152                            0));
1153                 }
1154
1155                 *nr_written += nr_pages;
1156                 start = cur_end + 1;
1157         }
1158         *page_started = 1;
1159         return 0;
1160 }
1161
1162 static noinline int csum_exist_in_range(struct btrfs_root *root,
1163                                         u64 bytenr, u64 num_bytes)
1164 {
1165         int ret;
1166         struct btrfs_ordered_sum *sums;
1167         LIST_HEAD(list);
1168
1169         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1170                                        bytenr + num_bytes - 1, &list, 0);
1171         if (ret == 0 && list_empty(&list))
1172                 return 0;
1173
1174         while (!list_empty(&list)) {
1175                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1176                 list_del(&sums->list);
1177                 kfree(sums);
1178         }
1179         return 1;
1180 }
1181
1182 /*
1183  * when nowcow writeback call back.  This checks for snapshots or COW copies
1184  * of the extents that exist in the file, and COWs the file as required.
1185  *
1186  * If no cow copies or snapshots exist, we write directly to the existing
1187  * blocks on disk
1188  */
1189 static noinline int run_delalloc_nocow(struct inode *inode,
1190                                        struct page *locked_page,
1191                               u64 start, u64 end, int *page_started, int force,
1192                               unsigned long *nr_written)
1193 {
1194         struct btrfs_root *root = BTRFS_I(inode)->root;
1195         struct btrfs_trans_handle *trans;
1196         struct extent_buffer *leaf;
1197         struct btrfs_path *path;
1198         struct btrfs_file_extent_item *fi;
1199         struct btrfs_key found_key;
1200         u64 cow_start;
1201         u64 cur_offset;
1202         u64 extent_end;
1203         u64 extent_offset;
1204         u64 disk_bytenr;
1205         u64 num_bytes;
1206         u64 disk_num_bytes;
1207         u64 ram_bytes;
1208         int extent_type;
1209         int ret, err;
1210         int type;
1211         int nocow;
1212         int check_prev = 1;
1213         bool nolock;
1214         u64 ino = btrfs_ino(inode);
1215
1216         path = btrfs_alloc_path();
1217         if (!path) {
1218                 extent_clear_unlock_delalloc(inode,
1219                              &BTRFS_I(inode)->io_tree,
1220                              start, end, locked_page,
1221                              EXTENT_CLEAR_UNLOCK_PAGE |
1222                              EXTENT_CLEAR_UNLOCK |
1223                              EXTENT_CLEAR_DELALLOC |
1224                              EXTENT_CLEAR_DIRTY |
1225                              EXTENT_SET_WRITEBACK |
1226                              EXTENT_END_WRITEBACK);
1227                 return -ENOMEM;
1228         }
1229
1230         nolock = btrfs_is_free_space_inode(inode);
1231
1232         if (nolock)
1233                 trans = btrfs_join_transaction_nolock(root);
1234         else
1235                 trans = btrfs_join_transaction(root);
1236
1237         if (IS_ERR(trans)) {
1238                 extent_clear_unlock_delalloc(inode,
1239                              &BTRFS_I(inode)->io_tree,
1240                              start, end, locked_page,
1241                              EXTENT_CLEAR_UNLOCK_PAGE |
1242                              EXTENT_CLEAR_UNLOCK |
1243                              EXTENT_CLEAR_DELALLOC |
1244                              EXTENT_CLEAR_DIRTY |
1245                              EXTENT_SET_WRITEBACK |
1246                              EXTENT_END_WRITEBACK);
1247                 btrfs_free_path(path);
1248                 return PTR_ERR(trans);
1249         }
1250
1251         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1252
1253         cow_start = (u64)-1;
1254         cur_offset = start;
1255         while (1) {
1256                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1257                                                cur_offset, 0);
1258                 if (ret < 0) {
1259                         btrfs_abort_transaction(trans, root, ret);
1260                         goto error;
1261                 }
1262                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1263                         leaf = path->nodes[0];
1264                         btrfs_item_key_to_cpu(leaf, &found_key,
1265                                               path->slots[0] - 1);
1266                         if (found_key.objectid == ino &&
1267                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1268                                 path->slots[0]--;
1269                 }
1270                 check_prev = 0;
1271 next_slot:
1272                 leaf = path->nodes[0];
1273                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1274                         ret = btrfs_next_leaf(root, path);
1275                         if (ret < 0) {
1276                                 btrfs_abort_transaction(trans, root, ret);
1277                                 goto error;
1278                         }
1279                         if (ret > 0)
1280                                 break;
1281                         leaf = path->nodes[0];
1282                 }
1283
1284                 nocow = 0;
1285                 disk_bytenr = 0;
1286                 num_bytes = 0;
1287                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1288
1289                 if (found_key.objectid > ino ||
1290                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1291                     found_key.offset > end)
1292                         break;
1293
1294                 if (found_key.offset > cur_offset) {
1295                         extent_end = found_key.offset;
1296                         extent_type = 0;
1297                         goto out_check;
1298                 }
1299
1300                 fi = btrfs_item_ptr(leaf, path->slots[0],
1301                                     struct btrfs_file_extent_item);
1302                 extent_type = btrfs_file_extent_type(leaf, fi);
1303
1304                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1305                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1306                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1307                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1308                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1309                         extent_end = found_key.offset +
1310                                 btrfs_file_extent_num_bytes(leaf, fi);
1311                         disk_num_bytes =
1312                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1313                         if (extent_end <= start) {
1314                                 path->slots[0]++;
1315                                 goto next_slot;
1316                         }
1317                         if (disk_bytenr == 0)
1318                                 goto out_check;
1319                         if (btrfs_file_extent_compression(leaf, fi) ||
1320                             btrfs_file_extent_encryption(leaf, fi) ||
1321                             btrfs_file_extent_other_encoding(leaf, fi))
1322                                 goto out_check;
1323                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1324                                 goto out_check;
1325                         if (btrfs_extent_readonly(root, disk_bytenr))
1326                                 goto out_check;
1327                         if (btrfs_cross_ref_exist(trans, root, ino,
1328                                                   found_key.offset -
1329                                                   extent_offset, disk_bytenr))
1330                                 goto out_check;
1331                         disk_bytenr += extent_offset;
1332                         disk_bytenr += cur_offset - found_key.offset;
1333                         num_bytes = min(end + 1, extent_end) - cur_offset;
1334                         /*
1335                          * force cow if csum exists in the range.
1336                          * this ensure that csum for a given extent are
1337                          * either valid or do not exist.
1338                          */
1339                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1340                                 goto out_check;
1341                         nocow = 1;
1342                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1343                         extent_end = found_key.offset +
1344                                 btrfs_file_extent_inline_len(leaf, fi);
1345                         extent_end = ALIGN(extent_end, root->sectorsize);
1346                 } else {
1347                         BUG_ON(1);
1348                 }
1349 out_check:
1350                 if (extent_end <= start) {
1351                         path->slots[0]++;
1352                         goto next_slot;
1353                 }
1354                 if (!nocow) {
1355                         if (cow_start == (u64)-1)
1356                                 cow_start = cur_offset;
1357                         cur_offset = extent_end;
1358                         if (cur_offset > end)
1359                                 break;
1360                         path->slots[0]++;
1361                         goto next_slot;
1362                 }
1363
1364                 btrfs_release_path(path);
1365                 if (cow_start != (u64)-1) {
1366                         ret = __cow_file_range(trans, inode, root, locked_page,
1367                                                cow_start, found_key.offset - 1,
1368                                                page_started, nr_written, 1);
1369                         if (ret) {
1370                                 btrfs_abort_transaction(trans, root, ret);
1371                                 goto error;
1372                         }
1373                         cow_start = (u64)-1;
1374                 }
1375
1376                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1377                         struct extent_map *em;
1378                         struct extent_map_tree *em_tree;
1379                         em_tree = &BTRFS_I(inode)->extent_tree;
1380                         em = alloc_extent_map();
1381                         BUG_ON(!em); /* -ENOMEM */
1382                         em->start = cur_offset;
1383                         em->orig_start = found_key.offset - extent_offset;
1384                         em->len = num_bytes;
1385                         em->block_len = num_bytes;
1386                         em->block_start = disk_bytenr;
1387                         em->orig_block_len = disk_num_bytes;
1388                         em->ram_bytes = ram_bytes;
1389                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1390                         em->mod_start = em->start;
1391                         em->mod_len = em->len;
1392                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1393                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1394                         em->generation = -1;
1395                         while (1) {
1396                                 write_lock(&em_tree->lock);
1397                                 ret = add_extent_mapping(em_tree, em, 1);
1398                                 write_unlock(&em_tree->lock);
1399                                 if (ret != -EEXIST) {
1400                                         free_extent_map(em);
1401                                         break;
1402                                 }
1403                                 btrfs_drop_extent_cache(inode, em->start,
1404                                                 em->start + em->len - 1, 0);
1405                         }
1406                         type = BTRFS_ORDERED_PREALLOC;
1407                 } else {
1408                         type = BTRFS_ORDERED_NOCOW;
1409                 }
1410
1411                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1412                                                num_bytes, num_bytes, type);
1413                 BUG_ON(ret); /* -ENOMEM */
1414
1415                 if (root->root_key.objectid ==
1416                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1417                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1418                                                       num_bytes);
1419                         if (ret) {
1420                                 btrfs_abort_transaction(trans, root, ret);
1421                                 goto error;
1422                         }
1423                 }
1424
1425                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1426                                 cur_offset, cur_offset + num_bytes - 1,
1427                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1428                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1429                                 EXTENT_SET_PRIVATE2);
1430                 cur_offset = extent_end;
1431                 if (cur_offset > end)
1432                         break;
1433         }
1434         btrfs_release_path(path);
1435
1436         if (cur_offset <= end && cow_start == (u64)-1) {
1437                 cow_start = cur_offset;
1438                 cur_offset = end;
1439         }
1440
1441         if (cow_start != (u64)-1) {
1442                 ret = __cow_file_range(trans, inode, root, locked_page,
1443                                        cow_start, end,
1444                                        page_started, nr_written, 1);
1445                 if (ret) {
1446                         btrfs_abort_transaction(trans, root, ret);
1447                         goto error;
1448                 }
1449         }
1450
1451 error:
1452         err = btrfs_end_transaction(trans, root);
1453         if (!ret)
1454                 ret = err;
1455
1456         if (ret && cur_offset < end)
1457                 extent_clear_unlock_delalloc(inode,
1458                              &BTRFS_I(inode)->io_tree,
1459                              cur_offset, end, locked_page,
1460                              EXTENT_CLEAR_UNLOCK_PAGE |
1461                              EXTENT_CLEAR_UNLOCK |
1462                              EXTENT_CLEAR_DELALLOC |
1463                              EXTENT_CLEAR_DIRTY |
1464                              EXTENT_SET_WRITEBACK |
1465                              EXTENT_END_WRITEBACK);
1466
1467         btrfs_free_path(path);
1468         return ret;
1469 }
1470
1471 /*
1472  * extent_io.c call back to do delayed allocation processing
1473  */
1474 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1475                               u64 start, u64 end, int *page_started,
1476                               unsigned long *nr_written)
1477 {
1478         int ret;
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480
1481         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1482                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1483                                          page_started, 1, nr_written);
1484         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1485                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1486                                          page_started, 0, nr_written);
1487         } else if (!btrfs_test_opt(root, COMPRESS) &&
1488                    !(BTRFS_I(inode)->force_compress) &&
1489                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1490                 ret = cow_file_range(inode, locked_page, start, end,
1491                                       page_started, nr_written, 1);
1492         } else {
1493                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1494                         &BTRFS_I(inode)->runtime_flags);
1495                 ret = cow_file_range_async(inode, locked_page, start, end,
1496                                            page_started, nr_written);
1497         }
1498         return ret;
1499 }
1500
1501 static void btrfs_split_extent_hook(struct inode *inode,
1502                                     struct extent_state *orig, u64 split)
1503 {
1504         /* not delalloc, ignore it */
1505         if (!(orig->state & EXTENT_DELALLOC))
1506                 return;
1507
1508         spin_lock(&BTRFS_I(inode)->lock);
1509         BTRFS_I(inode)->outstanding_extents++;
1510         spin_unlock(&BTRFS_I(inode)->lock);
1511 }
1512
1513 /*
1514  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1515  * extents so we can keep track of new extents that are just merged onto old
1516  * extents, such as when we are doing sequential writes, so we can properly
1517  * account for the metadata space we'll need.
1518  */
1519 static void btrfs_merge_extent_hook(struct inode *inode,
1520                                     struct extent_state *new,
1521                                     struct extent_state *other)
1522 {
1523         /* not delalloc, ignore it */
1524         if (!(other->state & EXTENT_DELALLOC))
1525                 return;
1526
1527         spin_lock(&BTRFS_I(inode)->lock);
1528         BTRFS_I(inode)->outstanding_extents--;
1529         spin_unlock(&BTRFS_I(inode)->lock);
1530 }
1531
1532 /*
1533  * extent_io.c set_bit_hook, used to track delayed allocation
1534  * bytes in this file, and to maintain the list of inodes that
1535  * have pending delalloc work to be done.
1536  */
1537 static void btrfs_set_bit_hook(struct inode *inode,
1538                                struct extent_state *state, unsigned long *bits)
1539 {
1540
1541         /*
1542          * set_bit and clear bit hooks normally require _irqsave/restore
1543          * but in this case, we are only testing for the DELALLOC
1544          * bit, which is only set or cleared with irqs on
1545          */
1546         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1547                 struct btrfs_root *root = BTRFS_I(inode)->root;
1548                 u64 len = state->end + 1 - state->start;
1549                 bool do_list = !btrfs_is_free_space_inode(inode);
1550
1551                 if (*bits & EXTENT_FIRST_DELALLOC) {
1552                         *bits &= ~EXTENT_FIRST_DELALLOC;
1553                 } else {
1554                         spin_lock(&BTRFS_I(inode)->lock);
1555                         BTRFS_I(inode)->outstanding_extents++;
1556                         spin_unlock(&BTRFS_I(inode)->lock);
1557                 }
1558
1559                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1560                                      root->fs_info->delalloc_batch);
1561                 spin_lock(&BTRFS_I(inode)->lock);
1562                 BTRFS_I(inode)->delalloc_bytes += len;
1563                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1564                                          &BTRFS_I(inode)->runtime_flags)) {
1565                         spin_lock(&root->fs_info->delalloc_lock);
1566                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1567                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1568                                               &root->fs_info->delalloc_inodes);
1569                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1570                                         &BTRFS_I(inode)->runtime_flags);
1571                         }
1572                         spin_unlock(&root->fs_info->delalloc_lock);
1573                 }
1574                 spin_unlock(&BTRFS_I(inode)->lock);
1575         }
1576 }
1577
1578 /*
1579  * extent_io.c clear_bit_hook, see set_bit_hook for why
1580  */
1581 static void btrfs_clear_bit_hook(struct inode *inode,
1582                                  struct extent_state *state,
1583                                  unsigned long *bits)
1584 {
1585         /*
1586          * set_bit and clear bit hooks normally require _irqsave/restore
1587          * but in this case, we are only testing for the DELALLOC
1588          * bit, which is only set or cleared with irqs on
1589          */
1590         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1591                 struct btrfs_root *root = BTRFS_I(inode)->root;
1592                 u64 len = state->end + 1 - state->start;
1593                 bool do_list = !btrfs_is_free_space_inode(inode);
1594
1595                 if (*bits & EXTENT_FIRST_DELALLOC) {
1596                         *bits &= ~EXTENT_FIRST_DELALLOC;
1597                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1598                         spin_lock(&BTRFS_I(inode)->lock);
1599                         BTRFS_I(inode)->outstanding_extents--;
1600                         spin_unlock(&BTRFS_I(inode)->lock);
1601                 }
1602
1603                 if (*bits & EXTENT_DO_ACCOUNTING)
1604                         btrfs_delalloc_release_metadata(inode, len);
1605
1606                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1607                     && do_list)
1608                         btrfs_free_reserved_data_space(inode, len);
1609
1610                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1611                                      root->fs_info->delalloc_batch);
1612                 spin_lock(&BTRFS_I(inode)->lock);
1613                 BTRFS_I(inode)->delalloc_bytes -= len;
1614                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1615                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1616                              &BTRFS_I(inode)->runtime_flags)) {
1617                         spin_lock(&root->fs_info->delalloc_lock);
1618                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1619                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1620                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1621                                           &BTRFS_I(inode)->runtime_flags);
1622                         }
1623                         spin_unlock(&root->fs_info->delalloc_lock);
1624                 }
1625                 spin_unlock(&BTRFS_I(inode)->lock);
1626         }
1627 }
1628
1629 /*
1630  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1631  * we don't create bios that span stripes or chunks
1632  */
1633 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1634                          size_t size, struct bio *bio,
1635                          unsigned long bio_flags)
1636 {
1637         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1638         u64 logical = (u64)bio->bi_sector << 9;
1639         u64 length = 0;
1640         u64 map_length;
1641         int ret;
1642
1643         if (bio_flags & EXTENT_BIO_COMPRESSED)
1644                 return 0;
1645
1646         length = bio->bi_size;
1647         map_length = length;
1648         ret = btrfs_map_block(root->fs_info, rw, logical,
1649                               &map_length, NULL, 0);
1650         /* Will always return 0 with map_multi == NULL */
1651         BUG_ON(ret < 0);
1652         if (map_length < length + size)
1653                 return 1;
1654         return 0;
1655 }
1656
1657 /*
1658  * in order to insert checksums into the metadata in large chunks,
1659  * we wait until bio submission time.   All the pages in the bio are
1660  * checksummed and sums are attached onto the ordered extent record.
1661  *
1662  * At IO completion time the cums attached on the ordered extent record
1663  * are inserted into the btree
1664  */
1665 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1666                                     struct bio *bio, int mirror_num,
1667                                     unsigned long bio_flags,
1668                                     u64 bio_offset)
1669 {
1670         struct btrfs_root *root = BTRFS_I(inode)->root;
1671         int ret = 0;
1672
1673         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1674         BUG_ON(ret); /* -ENOMEM */
1675         return 0;
1676 }
1677
1678 /*
1679  * in order to insert checksums into the metadata in large chunks,
1680  * we wait until bio submission time.   All the pages in the bio are
1681  * checksummed and sums are attached onto the ordered extent record.
1682  *
1683  * At IO completion time the cums attached on the ordered extent record
1684  * are inserted into the btree
1685  */
1686 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1687                           int mirror_num, unsigned long bio_flags,
1688                           u64 bio_offset)
1689 {
1690         struct btrfs_root *root = BTRFS_I(inode)->root;
1691         int ret;
1692
1693         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1694         if (ret)
1695                 bio_endio(bio, ret);
1696         return ret;
1697 }
1698
1699 /*
1700  * extent_io.c submission hook. This does the right thing for csum calculation
1701  * on write, or reading the csums from the tree before a read
1702  */
1703 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1704                           int mirror_num, unsigned long bio_flags,
1705                           u64 bio_offset)
1706 {
1707         struct btrfs_root *root = BTRFS_I(inode)->root;
1708         int ret = 0;
1709         int skip_sum;
1710         int metadata = 0;
1711         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1712
1713         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1714
1715         if (btrfs_is_free_space_inode(inode))
1716                 metadata = 2;
1717
1718         if (!(rw & REQ_WRITE)) {
1719                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1720                 if (ret)
1721                         goto out;
1722
1723                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1724                         ret = btrfs_submit_compressed_read(inode, bio,
1725                                                            mirror_num,
1726                                                            bio_flags);
1727                         goto out;
1728                 } else if (!skip_sum) {
1729                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1730                         if (ret)
1731                                 goto out;
1732                 }
1733                 goto mapit;
1734         } else if (async && !skip_sum) {
1735                 /* csum items have already been cloned */
1736                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1737                         goto mapit;
1738                 /* we're doing a write, do the async checksumming */
1739                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1740                                    inode, rw, bio, mirror_num,
1741                                    bio_flags, bio_offset,
1742                                    __btrfs_submit_bio_start,
1743                                    __btrfs_submit_bio_done);
1744                 goto out;
1745         } else if (!skip_sum) {
1746                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1747                 if (ret)
1748                         goto out;
1749         }
1750
1751 mapit:
1752         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1753
1754 out:
1755         if (ret < 0)
1756                 bio_endio(bio, ret);
1757         return ret;
1758 }
1759
1760 /*
1761  * given a list of ordered sums record them in the inode.  This happens
1762  * at IO completion time based on sums calculated at bio submission time.
1763  */
1764 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1765                              struct inode *inode, u64 file_offset,
1766                              struct list_head *list)
1767 {
1768         struct btrfs_ordered_sum *sum;
1769
1770         list_for_each_entry(sum, list, list) {
1771                 trans->adding_csums = 1;
1772                 btrfs_csum_file_blocks(trans,
1773                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1774                 trans->adding_csums = 0;
1775         }
1776         return 0;
1777 }
1778
1779 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1780                               struct extent_state **cached_state)
1781 {
1782         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1783         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1784                                    cached_state, GFP_NOFS);
1785 }
1786
1787 /* see btrfs_writepage_start_hook for details on why this is required */
1788 struct btrfs_writepage_fixup {
1789         struct page *page;
1790         struct btrfs_work work;
1791 };
1792
1793 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1794 {
1795         struct btrfs_writepage_fixup *fixup;
1796         struct btrfs_ordered_extent *ordered;
1797         struct extent_state *cached_state = NULL;
1798         struct page *page;
1799         struct inode *inode;
1800         u64 page_start;
1801         u64 page_end;
1802         int ret;
1803
1804         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1805         page = fixup->page;
1806 again:
1807         lock_page(page);
1808         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1809                 ClearPageChecked(page);
1810                 goto out_page;
1811         }
1812
1813         inode = page->mapping->host;
1814         page_start = page_offset(page);
1815         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1816
1817         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1818                          &cached_state);
1819
1820         /* already ordered? We're done */
1821         if (PagePrivate2(page))
1822                 goto out;
1823
1824         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1825         if (ordered) {
1826                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1827                                      page_end, &cached_state, GFP_NOFS);
1828                 unlock_page(page);
1829                 btrfs_start_ordered_extent(inode, ordered, 1);
1830                 btrfs_put_ordered_extent(ordered);
1831                 goto again;
1832         }
1833
1834         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1835         if (ret) {
1836                 mapping_set_error(page->mapping, ret);
1837                 end_extent_writepage(page, ret, page_start, page_end);
1838                 ClearPageChecked(page);
1839                 goto out;
1840          }
1841
1842         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1843         ClearPageChecked(page);
1844         set_page_dirty(page);
1845 out:
1846         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1847                              &cached_state, GFP_NOFS);
1848 out_page:
1849         unlock_page(page);
1850         page_cache_release(page);
1851         kfree(fixup);
1852 }
1853
1854 /*
1855  * There are a few paths in the higher layers of the kernel that directly
1856  * set the page dirty bit without asking the filesystem if it is a
1857  * good idea.  This causes problems because we want to make sure COW
1858  * properly happens and the data=ordered rules are followed.
1859  *
1860  * In our case any range that doesn't have the ORDERED bit set
1861  * hasn't been properly setup for IO.  We kick off an async process
1862  * to fix it up.  The async helper will wait for ordered extents, set
1863  * the delalloc bit and make it safe to write the page.
1864  */
1865 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1866 {
1867         struct inode *inode = page->mapping->host;
1868         struct btrfs_writepage_fixup *fixup;
1869         struct btrfs_root *root = BTRFS_I(inode)->root;
1870
1871         /* this page is properly in the ordered list */
1872         if (TestClearPagePrivate2(page))
1873                 return 0;
1874
1875         if (PageChecked(page))
1876                 return -EAGAIN;
1877
1878         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1879         if (!fixup)
1880                 return -EAGAIN;
1881
1882         SetPageChecked(page);
1883         page_cache_get(page);
1884         fixup->work.func = btrfs_writepage_fixup_worker;
1885         fixup->page = page;
1886         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1887         return -EBUSY;
1888 }
1889
1890 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1891                                        struct inode *inode, u64 file_pos,
1892                                        u64 disk_bytenr, u64 disk_num_bytes,
1893                                        u64 num_bytes, u64 ram_bytes,
1894                                        u8 compression, u8 encryption,
1895                                        u16 other_encoding, int extent_type)
1896 {
1897         struct btrfs_root *root = BTRFS_I(inode)->root;
1898         struct btrfs_file_extent_item *fi;
1899         struct btrfs_path *path;
1900         struct extent_buffer *leaf;
1901         struct btrfs_key ins;
1902         int ret;
1903
1904         path = btrfs_alloc_path();
1905         if (!path)
1906                 return -ENOMEM;
1907
1908         path->leave_spinning = 1;
1909
1910         /*
1911          * we may be replacing one extent in the tree with another.
1912          * The new extent is pinned in the extent map, and we don't want
1913          * to drop it from the cache until it is completely in the btree.
1914          *
1915          * So, tell btrfs_drop_extents to leave this extent in the cache.
1916          * the caller is expected to unpin it and allow it to be merged
1917          * with the others.
1918          */
1919         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1920                                  file_pos + num_bytes, 0);
1921         if (ret)
1922                 goto out;
1923
1924         ins.objectid = btrfs_ino(inode);
1925         ins.offset = file_pos;
1926         ins.type = BTRFS_EXTENT_DATA_KEY;
1927         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1928         if (ret)
1929                 goto out;
1930         leaf = path->nodes[0];
1931         fi = btrfs_item_ptr(leaf, path->slots[0],
1932                             struct btrfs_file_extent_item);
1933         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1934         btrfs_set_file_extent_type(leaf, fi, extent_type);
1935         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1936         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1937         btrfs_set_file_extent_offset(leaf, fi, 0);
1938         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1939         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1940         btrfs_set_file_extent_compression(leaf, fi, compression);
1941         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1942         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1943
1944         btrfs_mark_buffer_dirty(leaf);
1945         btrfs_release_path(path);
1946
1947         inode_add_bytes(inode, num_bytes);
1948
1949         ins.objectid = disk_bytenr;
1950         ins.offset = disk_num_bytes;
1951         ins.type = BTRFS_EXTENT_ITEM_KEY;
1952         ret = btrfs_alloc_reserved_file_extent(trans, root,
1953                                         root->root_key.objectid,
1954                                         btrfs_ino(inode), file_pos, &ins);
1955 out:
1956         btrfs_free_path(path);
1957
1958         return ret;
1959 }
1960
1961 /* snapshot-aware defrag */
1962 struct sa_defrag_extent_backref {
1963         struct rb_node node;
1964         struct old_sa_defrag_extent *old;
1965         u64 root_id;
1966         u64 inum;
1967         u64 file_pos;
1968         u64 extent_offset;
1969         u64 num_bytes;
1970         u64 generation;
1971 };
1972
1973 struct old_sa_defrag_extent {
1974         struct list_head list;
1975         struct new_sa_defrag_extent *new;
1976
1977         u64 extent_offset;
1978         u64 bytenr;
1979         u64 offset;
1980         u64 len;
1981         int count;
1982 };
1983
1984 struct new_sa_defrag_extent {
1985         struct rb_root root;
1986         struct list_head head;
1987         struct btrfs_path *path;
1988         struct inode *inode;
1989         u64 file_pos;
1990         u64 len;
1991         u64 bytenr;
1992         u64 disk_len;
1993         u8 compress_type;
1994 };
1995
1996 static int backref_comp(struct sa_defrag_extent_backref *b1,
1997                         struct sa_defrag_extent_backref *b2)
1998 {
1999         if (b1->root_id < b2->root_id)
2000                 return -1;
2001         else if (b1->root_id > b2->root_id)
2002                 return 1;
2003
2004         if (b1->inum < b2->inum)
2005                 return -1;
2006         else if (b1->inum > b2->inum)
2007                 return 1;
2008
2009         if (b1->file_pos < b2->file_pos)
2010                 return -1;
2011         else if (b1->file_pos > b2->file_pos)
2012                 return 1;
2013
2014         /*
2015          * [------------------------------] ===> (a range of space)
2016          *     |<--->|   |<---->| =============> (fs/file tree A)
2017          * |<---------------------------->| ===> (fs/file tree B)
2018          *
2019          * A range of space can refer to two file extents in one tree while
2020          * refer to only one file extent in another tree.
2021          *
2022          * So we may process a disk offset more than one time(two extents in A)
2023          * and locate at the same extent(one extent in B), then insert two same
2024          * backrefs(both refer to the extent in B).
2025          */
2026         return 0;
2027 }
2028
2029 static void backref_insert(struct rb_root *root,
2030                            struct sa_defrag_extent_backref *backref)
2031 {
2032         struct rb_node **p = &root->rb_node;
2033         struct rb_node *parent = NULL;
2034         struct sa_defrag_extent_backref *entry;
2035         int ret;
2036
2037         while (*p) {
2038                 parent = *p;
2039                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2040
2041                 ret = backref_comp(backref, entry);
2042                 if (ret < 0)
2043                         p = &(*p)->rb_left;
2044                 else
2045                         p = &(*p)->rb_right;
2046         }
2047
2048         rb_link_node(&backref->node, parent, p);
2049         rb_insert_color(&backref->node, root);
2050 }
2051
2052 /*
2053  * Note the backref might has changed, and in this case we just return 0.
2054  */
2055 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2056                                        void *ctx)
2057 {
2058         struct btrfs_file_extent_item *extent;
2059         struct btrfs_fs_info *fs_info;
2060         struct old_sa_defrag_extent *old = ctx;
2061         struct new_sa_defrag_extent *new = old->new;
2062         struct btrfs_path *path = new->path;
2063         struct btrfs_key key;
2064         struct btrfs_root *root;
2065         struct sa_defrag_extent_backref *backref;
2066         struct extent_buffer *leaf;
2067         struct inode *inode = new->inode;
2068         int slot;
2069         int ret;
2070         u64 extent_offset;
2071         u64 num_bytes;
2072
2073         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2074             inum == btrfs_ino(inode))
2075                 return 0;
2076
2077         key.objectid = root_id;
2078         key.type = BTRFS_ROOT_ITEM_KEY;
2079         key.offset = (u64)-1;
2080
2081         fs_info = BTRFS_I(inode)->root->fs_info;
2082         root = btrfs_read_fs_root_no_name(fs_info, &key);
2083         if (IS_ERR(root)) {
2084                 if (PTR_ERR(root) == -ENOENT)
2085                         return 0;
2086                 WARN_ON(1);
2087                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2088                          inum, offset, root_id);
2089                 return PTR_ERR(root);
2090         }
2091
2092         key.objectid = inum;
2093         key.type = BTRFS_EXTENT_DATA_KEY;
2094         if (offset > (u64)-1 << 32)
2095                 key.offset = 0;
2096         else
2097                 key.offset = offset;
2098
2099         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2100         if (ret < 0) {
2101                 WARN_ON(1);
2102                 return ret;
2103         }
2104
2105         while (1) {
2106                 cond_resched();
2107
2108                 leaf = path->nodes[0];
2109                 slot = path->slots[0];
2110
2111                 if (slot >= btrfs_header_nritems(leaf)) {
2112                         ret = btrfs_next_leaf(root, path);
2113                         if (ret < 0) {
2114                                 goto out;
2115                         } else if (ret > 0) {
2116                                 ret = 0;
2117                                 goto out;
2118                         }
2119                         continue;
2120                 }
2121
2122                 path->slots[0]++;
2123
2124                 btrfs_item_key_to_cpu(leaf, &key, slot);
2125
2126                 if (key.objectid > inum)
2127                         goto out;
2128
2129                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2130                         continue;
2131
2132                 extent = btrfs_item_ptr(leaf, slot,
2133                                         struct btrfs_file_extent_item);
2134
2135                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2136                         continue;
2137
2138                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139                 if (key.offset - extent_offset != offset)
2140                         continue;
2141
2142                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2143                 if (extent_offset >= old->extent_offset + old->offset +
2144                     old->len || extent_offset + num_bytes <=
2145                     old->extent_offset + old->offset)
2146                         continue;
2147
2148                 break;
2149         }
2150
2151         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2152         if (!backref) {
2153                 ret = -ENOENT;
2154                 goto out;
2155         }
2156
2157         backref->root_id = root_id;
2158         backref->inum = inum;
2159         backref->file_pos = offset + extent_offset;
2160         backref->num_bytes = num_bytes;
2161         backref->extent_offset = extent_offset;
2162         backref->generation = btrfs_file_extent_generation(leaf, extent);
2163         backref->old = old;
2164         backref_insert(&new->root, backref);
2165         old->count++;
2166 out:
2167         btrfs_release_path(path);
2168         WARN_ON(ret);
2169         return ret;
2170 }
2171
2172 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2173                                    struct new_sa_defrag_extent *new)
2174 {
2175         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2176         struct old_sa_defrag_extent *old, *tmp;
2177         int ret;
2178
2179         new->path = path;
2180
2181         list_for_each_entry_safe(old, tmp, &new->head, list) {
2182                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2183                                                   path, record_one_backref,
2184                                                   old);
2185                 BUG_ON(ret < 0 && ret != -ENOENT);
2186
2187                 /* no backref to be processed for this extent */
2188                 if (!old->count) {
2189                         list_del(&old->list);
2190                         kfree(old);
2191                 }
2192         }
2193
2194         if (list_empty(&new->head))
2195                 return false;
2196
2197         return true;
2198 }
2199
2200 static int relink_is_mergable(struct extent_buffer *leaf,
2201                               struct btrfs_file_extent_item *fi,
2202                               u64 disk_bytenr)
2203 {
2204         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2205                 return 0;
2206
2207         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2208                 return 0;
2209
2210         if (btrfs_file_extent_compression(leaf, fi) ||
2211             btrfs_file_extent_encryption(leaf, fi) ||
2212             btrfs_file_extent_other_encoding(leaf, fi))
2213                 return 0;
2214
2215         return 1;
2216 }
2217
2218 /*
2219  * Note the backref might has changed, and in this case we just return 0.
2220  */
2221 static noinline int relink_extent_backref(struct btrfs_path *path,
2222                                  struct sa_defrag_extent_backref *prev,
2223                                  struct sa_defrag_extent_backref *backref)
2224 {
2225         struct btrfs_file_extent_item *extent;
2226         struct btrfs_file_extent_item *item;
2227         struct btrfs_ordered_extent *ordered;
2228         struct btrfs_trans_handle *trans;
2229         struct btrfs_fs_info *fs_info;
2230         struct btrfs_root *root;
2231         struct btrfs_key key;
2232         struct extent_buffer *leaf;
2233         struct old_sa_defrag_extent *old = backref->old;
2234         struct new_sa_defrag_extent *new = old->new;
2235         struct inode *src_inode = new->inode;
2236         struct inode *inode;
2237         struct extent_state *cached = NULL;
2238         int ret = 0;
2239         u64 start;
2240         u64 len;
2241         u64 lock_start;
2242         u64 lock_end;
2243         bool merge = false;
2244         int index;
2245
2246         if (prev && prev->root_id == backref->root_id &&
2247             prev->inum == backref->inum &&
2248             prev->file_pos + prev->num_bytes == backref->file_pos)
2249                 merge = true;
2250
2251         /* step 1: get root */
2252         key.objectid = backref->root_id;
2253         key.type = BTRFS_ROOT_ITEM_KEY;
2254         key.offset = (u64)-1;
2255
2256         fs_info = BTRFS_I(src_inode)->root->fs_info;
2257         index = srcu_read_lock(&fs_info->subvol_srcu);
2258
2259         root = btrfs_read_fs_root_no_name(fs_info, &key);
2260         if (IS_ERR(root)) {
2261                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2262                 if (PTR_ERR(root) == -ENOENT)
2263                         return 0;
2264                 return PTR_ERR(root);
2265         }
2266         if (btrfs_root_refs(&root->root_item) == 0) {
2267                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2268                 /* parse ENOENT to 0 */
2269                 return 0;
2270         }
2271
2272         /* step 2: get inode */
2273         key.objectid = backref->inum;
2274         key.type = BTRFS_INODE_ITEM_KEY;
2275         key.offset = 0;
2276
2277         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2278         if (IS_ERR(inode)) {
2279                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2280                 return 0;
2281         }
2282
2283         srcu_read_unlock(&fs_info->subvol_srcu, index);
2284
2285         /* step 3: relink backref */
2286         lock_start = backref->file_pos;
2287         lock_end = backref->file_pos + backref->num_bytes - 1;
2288         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2289                          0, &cached);
2290
2291         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2292         if (ordered) {
2293                 btrfs_put_ordered_extent(ordered);
2294                 goto out_unlock;
2295         }
2296
2297         trans = btrfs_join_transaction(root);
2298         if (IS_ERR(trans)) {
2299                 ret = PTR_ERR(trans);
2300                 goto out_unlock;
2301         }
2302
2303         key.objectid = backref->inum;
2304         key.type = BTRFS_EXTENT_DATA_KEY;
2305         key.offset = backref->file_pos;
2306
2307         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2308         if (ret < 0) {
2309                 goto out_free_path;
2310         } else if (ret > 0) {
2311                 ret = 0;
2312                 goto out_free_path;
2313         }
2314
2315         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2316                                 struct btrfs_file_extent_item);
2317
2318         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2319             backref->generation)
2320                 goto out_free_path;
2321
2322         btrfs_release_path(path);
2323
2324         start = backref->file_pos;
2325         if (backref->extent_offset < old->extent_offset + old->offset)
2326                 start += old->extent_offset + old->offset -
2327                          backref->extent_offset;
2328
2329         len = min(backref->extent_offset + backref->num_bytes,
2330                   old->extent_offset + old->offset + old->len);
2331         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2332
2333         ret = btrfs_drop_extents(trans, root, inode, start,
2334                                  start + len, 1);
2335         if (ret)
2336                 goto out_free_path;
2337 again:
2338         key.objectid = btrfs_ino(inode);
2339         key.type = BTRFS_EXTENT_DATA_KEY;
2340         key.offset = start;
2341
2342         path->leave_spinning = 1;
2343         if (merge) {
2344                 struct btrfs_file_extent_item *fi;
2345                 u64 extent_len;
2346                 struct btrfs_key found_key;
2347
2348                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2349                 if (ret < 0)
2350                         goto out_free_path;
2351
2352                 path->slots[0]--;
2353                 leaf = path->nodes[0];
2354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356                 fi = btrfs_item_ptr(leaf, path->slots[0],
2357                                     struct btrfs_file_extent_item);
2358                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2359
2360                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2361                     extent_len + found_key.offset == start) {
2362                         btrfs_set_file_extent_num_bytes(leaf, fi,
2363                                                         extent_len + len);
2364                         btrfs_mark_buffer_dirty(leaf);
2365                         inode_add_bytes(inode, len);
2366
2367                         ret = 1;
2368                         goto out_free_path;
2369                 } else {
2370                         merge = false;
2371                         btrfs_release_path(path);
2372                         goto again;
2373                 }
2374         }
2375
2376         ret = btrfs_insert_empty_item(trans, root, path, &key,
2377                                         sizeof(*extent));
2378         if (ret) {
2379                 btrfs_abort_transaction(trans, root, ret);
2380                 goto out_free_path;
2381         }
2382
2383         leaf = path->nodes[0];
2384         item = btrfs_item_ptr(leaf, path->slots[0],
2385                                 struct btrfs_file_extent_item);
2386         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2387         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2388         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2389         btrfs_set_file_extent_num_bytes(leaf, item, len);
2390         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2391         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2392         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2393         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2394         btrfs_set_file_extent_encryption(leaf, item, 0);
2395         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2396
2397         btrfs_mark_buffer_dirty(leaf);
2398         inode_add_bytes(inode, len);
2399         btrfs_release_path(path);
2400
2401         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2402                         new->disk_len, 0,
2403                         backref->root_id, backref->inum,
2404                         new->file_pos, 0);      /* start - extent_offset */
2405         if (ret) {
2406                 btrfs_abort_transaction(trans, root, ret);
2407                 goto out_free_path;
2408         }
2409
2410         ret = 1;
2411 out_free_path:
2412         btrfs_release_path(path);
2413         path->leave_spinning = 0;
2414         btrfs_end_transaction(trans, root);
2415 out_unlock:
2416         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2417                              &cached, GFP_NOFS);
2418         iput(inode);
2419         return ret;
2420 }
2421
2422 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2423 {
2424         struct old_sa_defrag_extent *old, *tmp;
2425
2426         if (!new)
2427                 return;
2428
2429         list_for_each_entry_safe(old, tmp, &new->head, list) {
2430                 list_del(&old->list);
2431                 kfree(old);
2432         }
2433         kfree(new);
2434 }
2435
2436 static void relink_file_extents(struct new_sa_defrag_extent *new)
2437 {
2438         struct btrfs_path *path;
2439         struct sa_defrag_extent_backref *backref;
2440         struct sa_defrag_extent_backref *prev = NULL;
2441         struct inode *inode;
2442         struct btrfs_root *root;
2443         struct rb_node *node;
2444         int ret;
2445
2446         inode = new->inode;
2447         root = BTRFS_I(inode)->root;
2448
2449         path = btrfs_alloc_path();
2450         if (!path)
2451                 return;
2452
2453         if (!record_extent_backrefs(path, new)) {
2454                 btrfs_free_path(path);
2455                 goto out;
2456         }
2457         btrfs_release_path(path);
2458
2459         while (1) {
2460                 node = rb_first(&new->root);
2461                 if (!node)
2462                         break;
2463                 rb_erase(node, &new->root);
2464
2465                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2466
2467                 ret = relink_extent_backref(path, prev, backref);
2468                 WARN_ON(ret < 0);
2469
2470                 kfree(prev);
2471
2472                 if (ret == 1)
2473                         prev = backref;
2474                 else
2475                         prev = NULL;
2476                 cond_resched();
2477         }
2478         kfree(prev);
2479
2480         btrfs_free_path(path);
2481 out:
2482         free_sa_defrag_extent(new);
2483
2484         atomic_dec(&root->fs_info->defrag_running);
2485         wake_up(&root->fs_info->transaction_wait);
2486 }
2487
2488 static struct new_sa_defrag_extent *
2489 record_old_file_extents(struct inode *inode,
2490                         struct btrfs_ordered_extent *ordered)
2491 {
2492         struct btrfs_root *root = BTRFS_I(inode)->root;
2493         struct btrfs_path *path;
2494         struct btrfs_key key;
2495         struct old_sa_defrag_extent *old;
2496         struct new_sa_defrag_extent *new;
2497         int ret;
2498
2499         new = kmalloc(sizeof(*new), GFP_NOFS);
2500         if (!new)
2501                 return NULL;
2502
2503         new->inode = inode;
2504         new->file_pos = ordered->file_offset;
2505         new->len = ordered->len;
2506         new->bytenr = ordered->start;
2507         new->disk_len = ordered->disk_len;
2508         new->compress_type = ordered->compress_type;
2509         new->root = RB_ROOT;
2510         INIT_LIST_HEAD(&new->head);
2511
2512         path = btrfs_alloc_path();
2513         if (!path)
2514                 goto out_kfree;
2515
2516         key.objectid = btrfs_ino(inode);
2517         key.type = BTRFS_EXTENT_DATA_KEY;
2518         key.offset = new->file_pos;
2519
2520         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2521         if (ret < 0)
2522                 goto out_free_path;
2523         if (ret > 0 && path->slots[0] > 0)
2524                 path->slots[0]--;
2525
2526         /* find out all the old extents for the file range */
2527         while (1) {
2528                 struct btrfs_file_extent_item *extent;
2529                 struct extent_buffer *l;
2530                 int slot;
2531                 u64 num_bytes;
2532                 u64 offset;
2533                 u64 end;
2534                 u64 disk_bytenr;
2535                 u64 extent_offset;
2536
2537                 l = path->nodes[0];
2538                 slot = path->slots[0];
2539
2540                 if (slot >= btrfs_header_nritems(l)) {
2541                         ret = btrfs_next_leaf(root, path);
2542                         if (ret < 0)
2543                                 goto out_free_path;
2544                         else if (ret > 0)
2545                                 break;
2546                         continue;
2547                 }
2548
2549                 btrfs_item_key_to_cpu(l, &key, slot);
2550
2551                 if (key.objectid != btrfs_ino(inode))
2552                         break;
2553                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2554                         break;
2555                 if (key.offset >= new->file_pos + new->len)
2556                         break;
2557
2558                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2559
2560                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2561                 if (key.offset + num_bytes < new->file_pos)
2562                         goto next;
2563
2564                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2565                 if (!disk_bytenr)
2566                         goto next;
2567
2568                 extent_offset = btrfs_file_extent_offset(l, extent);
2569
2570                 old = kmalloc(sizeof(*old), GFP_NOFS);
2571                 if (!old)
2572                         goto out_free_path;
2573
2574                 offset = max(new->file_pos, key.offset);
2575                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2576
2577                 old->bytenr = disk_bytenr;
2578                 old->extent_offset = extent_offset;
2579                 old->offset = offset - key.offset;
2580                 old->len = end - offset;
2581                 old->new = new;
2582                 old->count = 0;
2583                 list_add_tail(&old->list, &new->head);
2584 next:
2585                 path->slots[0]++;
2586                 cond_resched();
2587         }
2588
2589         btrfs_free_path(path);
2590         atomic_inc(&root->fs_info->defrag_running);
2591
2592         return new;
2593
2594 out_free_path:
2595         btrfs_free_path(path);
2596 out_kfree:
2597         free_sa_defrag_extent(new);
2598         return NULL;
2599 }
2600
2601 /*
2602  * helper function for btrfs_finish_ordered_io, this
2603  * just reads in some of the csum leaves to prime them into ram
2604  * before we start the transaction.  It limits the amount of btree
2605  * reads required while inside the transaction.
2606  */
2607 /* as ordered data IO finishes, this gets called so we can finish
2608  * an ordered extent if the range of bytes in the file it covers are
2609  * fully written.
2610  */
2611 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2612 {
2613         struct inode *inode = ordered_extent->inode;
2614         struct btrfs_root *root = BTRFS_I(inode)->root;
2615         struct btrfs_trans_handle *trans = NULL;
2616         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2617         struct extent_state *cached_state = NULL;
2618         struct new_sa_defrag_extent *new = NULL;
2619         int compress_type = 0;
2620         int ret;
2621         bool nolock;
2622
2623         nolock = btrfs_is_free_space_inode(inode);
2624
2625         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2626                 ret = -EIO;
2627                 goto out;
2628         }
2629
2630         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2631                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2632                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2633                 if (nolock)
2634                         trans = btrfs_join_transaction_nolock(root);
2635                 else
2636                         trans = btrfs_join_transaction(root);
2637                 if (IS_ERR(trans)) {
2638                         ret = PTR_ERR(trans);
2639                         trans = NULL;
2640                         goto out;
2641                 }
2642                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2643                 ret = btrfs_update_inode_fallback(trans, root, inode);
2644                 if (ret) /* -ENOMEM or corruption */
2645                         btrfs_abort_transaction(trans, root, ret);
2646                 goto out;
2647         }
2648
2649         lock_extent_bits(io_tree, ordered_extent->file_offset,
2650                          ordered_extent->file_offset + ordered_extent->len - 1,
2651                          0, &cached_state);
2652
2653         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2654                         ordered_extent->file_offset + ordered_extent->len - 1,
2655                         EXTENT_DEFRAG, 1, cached_state);
2656         if (ret) {
2657                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2658                 if (last_snapshot >= BTRFS_I(inode)->generation)
2659                         /* the inode is shared */
2660                         new = record_old_file_extents(inode, ordered_extent);
2661
2662                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2663                         ordered_extent->file_offset + ordered_extent->len - 1,
2664                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2665         }
2666
2667         if (nolock)
2668                 trans = btrfs_join_transaction_nolock(root);
2669         else
2670                 trans = btrfs_join_transaction(root);
2671         if (IS_ERR(trans)) {
2672                 ret = PTR_ERR(trans);
2673                 trans = NULL;
2674                 goto out_unlock;
2675         }
2676         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2677
2678         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2679                 compress_type = ordered_extent->compress_type;
2680         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2681                 BUG_ON(compress_type);
2682                 ret = btrfs_mark_extent_written(trans, inode,
2683                                                 ordered_extent->file_offset,
2684                                                 ordered_extent->file_offset +
2685                                                 ordered_extent->len);
2686         } else {
2687                 BUG_ON(root == root->fs_info->tree_root);
2688                 ret = insert_reserved_file_extent(trans, inode,
2689                                                 ordered_extent->file_offset,
2690                                                 ordered_extent->start,
2691                                                 ordered_extent->disk_len,
2692                                                 ordered_extent->len,
2693                                                 ordered_extent->len,
2694                                                 compress_type, 0, 0,
2695                                                 BTRFS_FILE_EXTENT_REG);
2696         }
2697         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2698                            ordered_extent->file_offset, ordered_extent->len,
2699                            trans->transid);
2700         if (ret < 0) {
2701                 btrfs_abort_transaction(trans, root, ret);
2702                 goto out_unlock;
2703         }
2704
2705         add_pending_csums(trans, inode, ordered_extent->file_offset,
2706                           &ordered_extent->list);
2707
2708         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2709         ret = btrfs_update_inode_fallback(trans, root, inode);
2710         if (ret) { /* -ENOMEM or corruption */
2711                 btrfs_abort_transaction(trans, root, ret);
2712                 goto out_unlock;
2713         }
2714         ret = 0;
2715 out_unlock:
2716         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2717                              ordered_extent->file_offset +
2718                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2719 out:
2720         if (root != root->fs_info->tree_root)
2721                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2722         if (trans)
2723                 btrfs_end_transaction(trans, root);
2724
2725         if (ret) {
2726                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2727                                       ordered_extent->file_offset +
2728                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2729
2730                 /*
2731                  * If the ordered extent had an IOERR or something else went
2732                  * wrong we need to return the space for this ordered extent
2733                  * back to the allocator.
2734                  */
2735                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2736                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2737                         btrfs_free_reserved_extent(root, ordered_extent->start,
2738                                                    ordered_extent->disk_len);
2739         }
2740
2741
2742         /*
2743          * This needs to be done to make sure anybody waiting knows we are done
2744          * updating everything for this ordered extent.
2745          */
2746         btrfs_remove_ordered_extent(inode, ordered_extent);
2747
2748         /* for snapshot-aware defrag */
2749         if (new) {
2750                 if (ret) {
2751                         free_sa_defrag_extent(new);
2752                         atomic_dec(&root->fs_info->defrag_running);
2753                 } else {
2754                         relink_file_extents(new);
2755                 }
2756         }
2757
2758         /* once for us */
2759         btrfs_put_ordered_extent(ordered_extent);
2760         /* once for the tree */
2761         btrfs_put_ordered_extent(ordered_extent);
2762
2763         return ret;
2764 }
2765
2766 static void finish_ordered_fn(struct btrfs_work *work)
2767 {
2768         struct btrfs_ordered_extent *ordered_extent;
2769         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2770         btrfs_finish_ordered_io(ordered_extent);
2771 }
2772
2773 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2774                                 struct extent_state *state, int uptodate)
2775 {
2776         struct inode *inode = page->mapping->host;
2777         struct btrfs_root *root = BTRFS_I(inode)->root;
2778         struct btrfs_ordered_extent *ordered_extent = NULL;
2779         struct btrfs_workers *workers;
2780
2781         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2782
2783         ClearPagePrivate2(page);
2784         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2785                                             end - start + 1, uptodate))
2786                 return 0;
2787
2788         ordered_extent->work.func = finish_ordered_fn;
2789         ordered_extent->work.flags = 0;
2790
2791         if (btrfs_is_free_space_inode(inode))
2792                 workers = &root->fs_info->endio_freespace_worker;
2793         else
2794                 workers = &root->fs_info->endio_write_workers;
2795         btrfs_queue_worker(workers, &ordered_extent->work);
2796
2797         return 0;
2798 }
2799
2800 /*
2801  * when reads are done, we need to check csums to verify the data is correct
2802  * if there's a match, we allow the bio to finish.  If not, the code in
2803  * extent_io.c will try to find good copies for us.
2804  */
2805 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2806                                struct extent_state *state, int mirror)
2807 {
2808         size_t offset = start - page_offset(page);
2809         struct inode *inode = page->mapping->host;
2810         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2811         char *kaddr;
2812         u64 private = ~(u32)0;
2813         int ret;
2814         struct btrfs_root *root = BTRFS_I(inode)->root;
2815         u32 csum = ~(u32)0;
2816         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2817                                       DEFAULT_RATELIMIT_BURST);
2818
2819         if (PageChecked(page)) {
2820                 ClearPageChecked(page);
2821                 goto good;
2822         }
2823
2824         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2825                 goto good;
2826
2827         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2828             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2829                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2830                                   GFP_NOFS);
2831                 return 0;
2832         }
2833
2834         if (state && state->start == start) {
2835                 private = state->private;
2836                 ret = 0;
2837         } else {
2838                 ret = get_state_private(io_tree, start, &private);
2839         }
2840         kaddr = kmap_atomic(page);
2841         if (ret)
2842                 goto zeroit;
2843
2844         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2845         btrfs_csum_final(csum, (char *)&csum);
2846         if (csum != private)
2847                 goto zeroit;
2848
2849         kunmap_atomic(kaddr);
2850 good:
2851         return 0;
2852
2853 zeroit:
2854         if (__ratelimit(&_rs))
2855                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2856                         (unsigned long long)btrfs_ino(page->mapping->host),
2857                         (unsigned long long)start, csum,
2858                         (unsigned long long)private);
2859         memset(kaddr + offset, 1, end - start + 1);
2860         flush_dcache_page(page);
2861         kunmap_atomic(kaddr);
2862         if (private == 0)
2863                 return 0;
2864         return -EIO;
2865 }
2866
2867 struct delayed_iput {
2868         struct list_head list;
2869         struct inode *inode;
2870 };
2871
2872 /* JDM: If this is fs-wide, why can't we add a pointer to
2873  * btrfs_inode instead and avoid the allocation? */
2874 void btrfs_add_delayed_iput(struct inode *inode)
2875 {
2876         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2877         struct delayed_iput *delayed;
2878
2879         if (atomic_add_unless(&inode->i_count, -1, 1))
2880                 return;
2881
2882         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2883         delayed->inode = inode;
2884
2885         spin_lock(&fs_info->delayed_iput_lock);
2886         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2887         spin_unlock(&fs_info->delayed_iput_lock);
2888 }
2889
2890 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2891 {
2892         LIST_HEAD(list);
2893         struct btrfs_fs_info *fs_info = root->fs_info;
2894         struct delayed_iput *delayed;
2895         int empty;
2896
2897         spin_lock(&fs_info->delayed_iput_lock);
2898         empty = list_empty(&fs_info->delayed_iputs);
2899         spin_unlock(&fs_info->delayed_iput_lock);
2900         if (empty)
2901                 return;
2902
2903         spin_lock(&fs_info->delayed_iput_lock);
2904         list_splice_init(&fs_info->delayed_iputs, &list);
2905         spin_unlock(&fs_info->delayed_iput_lock);
2906
2907         while (!list_empty(&list)) {
2908                 delayed = list_entry(list.next, struct delayed_iput, list);
2909                 list_del(&delayed->list);
2910                 iput(delayed->inode);
2911                 kfree(delayed);
2912         }
2913 }
2914
2915 /*
2916  * This is called in transaction commit time. If there are no orphan
2917  * files in the subvolume, it removes orphan item and frees block_rsv
2918  * structure.
2919  */
2920 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2921                               struct btrfs_root *root)
2922 {
2923         struct btrfs_block_rsv *block_rsv;
2924         int ret;
2925
2926         if (atomic_read(&root->orphan_inodes) ||
2927             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2928                 return;
2929
2930         spin_lock(&root->orphan_lock);
2931         if (atomic_read(&root->orphan_inodes)) {
2932                 spin_unlock(&root->orphan_lock);
2933                 return;
2934         }
2935
2936         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2937                 spin_unlock(&root->orphan_lock);
2938                 return;
2939         }
2940
2941         block_rsv = root->orphan_block_rsv;
2942         root->orphan_block_rsv = NULL;
2943         spin_unlock(&root->orphan_lock);
2944
2945         if (root->orphan_item_inserted &&
2946             btrfs_root_refs(&root->root_item) > 0) {
2947                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2948                                             root->root_key.objectid);
2949                 BUG_ON(ret);
2950                 root->orphan_item_inserted = 0;
2951         }
2952
2953         if (block_rsv) {
2954                 WARN_ON(block_rsv->size > 0);
2955                 btrfs_free_block_rsv(root, block_rsv);
2956         }
2957 }
2958
2959 /*
2960  * This creates an orphan entry for the given inode in case something goes
2961  * wrong in the middle of an unlink/truncate.
2962  *
2963  * NOTE: caller of this function should reserve 5 units of metadata for
2964  *       this function.
2965  */
2966 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2967 {
2968         struct btrfs_root *root = BTRFS_I(inode)->root;
2969         struct btrfs_block_rsv *block_rsv = NULL;
2970         int reserve = 0;
2971         int insert = 0;
2972         int ret;
2973
2974         if (!root->orphan_block_rsv) {
2975                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2976                 if (!block_rsv)
2977                         return -ENOMEM;
2978         }
2979
2980         spin_lock(&root->orphan_lock);
2981         if (!root->orphan_block_rsv) {
2982                 root->orphan_block_rsv = block_rsv;
2983         } else if (block_rsv) {
2984                 btrfs_free_block_rsv(root, block_rsv);
2985                 block_rsv = NULL;
2986         }
2987
2988         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2989                               &BTRFS_I(inode)->runtime_flags)) {
2990 #if 0
2991                 /*
2992                  * For proper ENOSPC handling, we should do orphan
2993                  * cleanup when mounting. But this introduces backward
2994                  * compatibility issue.
2995                  */
2996                 if (!xchg(&root->orphan_item_inserted, 1))
2997                         insert = 2;
2998                 else
2999                         insert = 1;
3000 #endif
3001                 insert = 1;
3002                 atomic_inc(&root->orphan_inodes);
3003         }
3004
3005         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3006                               &BTRFS_I(inode)->runtime_flags))
3007                 reserve = 1;
3008         spin_unlock(&root->orphan_lock);
3009
3010         /* grab metadata reservation from transaction handle */
3011         if (reserve) {
3012                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3013                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3014         }
3015
3016         /* insert an orphan item to track this unlinked/truncated file */
3017         if (insert >= 1) {
3018                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3019                 if (ret && ret != -EEXIST) {
3020                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3021                                   &BTRFS_I(inode)->runtime_flags);
3022                         btrfs_abort_transaction(trans, root, ret);
3023                         return ret;
3024                 }
3025                 ret = 0;
3026         }
3027
3028         /* insert an orphan item to track subvolume contains orphan files */
3029         if (insert >= 2) {
3030                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3031                                                root->root_key.objectid);
3032                 if (ret && ret != -EEXIST) {
3033                         btrfs_abort_transaction(trans, root, ret);
3034                         return ret;
3035                 }
3036         }
3037         return 0;
3038 }
3039
3040 /*
3041  * We have done the truncate/delete so we can go ahead and remove the orphan
3042  * item for this particular inode.
3043  */
3044 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3045                             struct inode *inode)
3046 {
3047         struct btrfs_root *root = BTRFS_I(inode)->root;
3048         int delete_item = 0;
3049         int release_rsv = 0;
3050         int ret = 0;
3051
3052         spin_lock(&root->orphan_lock);
3053         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3054                                &BTRFS_I(inode)->runtime_flags))
3055                 delete_item = 1;
3056
3057         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3058                                &BTRFS_I(inode)->runtime_flags))
3059                 release_rsv = 1;
3060         spin_unlock(&root->orphan_lock);
3061
3062         if (trans && delete_item) {
3063                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3064                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3065         }
3066
3067         if (release_rsv) {
3068                 btrfs_orphan_release_metadata(inode);
3069                 atomic_dec(&root->orphan_inodes);
3070         }
3071
3072         return 0;
3073 }
3074
3075 /*
3076  * this cleans up any orphans that may be left on the list from the last use
3077  * of this root.
3078  */
3079 int btrfs_orphan_cleanup(struct btrfs_root *root)
3080 {
3081         struct btrfs_path *path;
3082         struct extent_buffer *leaf;
3083         struct btrfs_key key, found_key;
3084         struct btrfs_trans_handle *trans;
3085         struct inode *inode;
3086         u64 last_objectid = 0;
3087         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3088
3089         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3090                 return 0;
3091
3092         path = btrfs_alloc_path();
3093         if (!path) {
3094                 ret = -ENOMEM;
3095                 goto out;
3096         }
3097         path->reada = -1;
3098
3099         key.objectid = BTRFS_ORPHAN_OBJECTID;
3100         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3101         key.offset = (u64)-1;
3102
3103         while (1) {
3104                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3105                 if (ret < 0)
3106                         goto out;
3107
3108                 /*
3109                  * if ret == 0 means we found what we were searching for, which
3110                  * is weird, but possible, so only screw with path if we didn't
3111                  * find the key and see if we have stuff that matches
3112                  */
3113                 if (ret > 0) {
3114                         ret = 0;
3115                         if (path->slots[0] == 0)
3116                                 break;
3117                         path->slots[0]--;
3118                 }
3119
3120                 /* pull out the item */
3121                 leaf = path->nodes[0];
3122                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3123
3124                 /* make sure the item matches what we want */
3125                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3126                         break;
3127                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3128                         break;
3129
3130                 /* release the path since we're done with it */
3131                 btrfs_release_path(path);
3132
3133                 /*
3134                  * this is where we are basically btrfs_lookup, without the
3135                  * crossing root thing.  we store the inode number in the
3136                  * offset of the orphan item.
3137                  */
3138
3139                 if (found_key.offset == last_objectid) {
3140                         btrfs_err(root->fs_info,
3141                                 "Error removing orphan entry, stopping orphan cleanup");
3142                         ret = -EINVAL;
3143                         goto out;
3144                 }
3145
3146                 last_objectid = found_key.offset;
3147
3148                 found_key.objectid = found_key.offset;
3149                 found_key.type = BTRFS_INODE_ITEM_KEY;
3150                 found_key.offset = 0;
3151                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3152                 ret = PTR_RET(inode);
3153                 if (ret && ret != -ESTALE)
3154                         goto out;
3155
3156                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3157                         struct btrfs_root *dead_root;
3158                         struct btrfs_fs_info *fs_info = root->fs_info;
3159                         int is_dead_root = 0;
3160
3161                         /*
3162                          * this is an orphan in the tree root. Currently these
3163                          * could come from 2 sources:
3164                          *  a) a snapshot deletion in progress
3165                          *  b) a free space cache inode
3166                          * We need to distinguish those two, as the snapshot
3167                          * orphan must not get deleted.
3168                          * find_dead_roots already ran before us, so if this
3169                          * is a snapshot deletion, we should find the root
3170                          * in the dead_roots list
3171                          */
3172                         spin_lock(&fs_info->trans_lock);
3173                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3174                                             root_list) {
3175                                 if (dead_root->root_key.objectid ==
3176                                     found_key.objectid) {
3177                                         is_dead_root = 1;
3178                                         break;
3179                                 }
3180                         }
3181                         spin_unlock(&fs_info->trans_lock);
3182                         if (is_dead_root) {
3183                                 /* prevent this orphan from being found again */
3184                                 key.offset = found_key.objectid - 1;
3185                                 continue;
3186                         }
3187                 }
3188                 /*
3189                  * Inode is already gone but the orphan item is still there,
3190                  * kill the orphan item.
3191                  */
3192                 if (ret == -ESTALE) {
3193                         trans = btrfs_start_transaction(root, 1);
3194                         if (IS_ERR(trans)) {
3195                                 ret = PTR_ERR(trans);
3196                                 goto out;
3197                         }
3198                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3199                                 found_key.objectid);
3200                         ret = btrfs_del_orphan_item(trans, root,
3201                                                     found_key.objectid);
3202                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3203                         btrfs_end_transaction(trans, root);
3204                         continue;
3205                 }
3206
3207                 /*
3208                  * add this inode to the orphan list so btrfs_orphan_del does
3209                  * the proper thing when we hit it
3210                  */
3211                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3212                         &BTRFS_I(inode)->runtime_flags);
3213                 atomic_inc(&root->orphan_inodes);
3214
3215                 /* if we have links, this was a truncate, lets do that */
3216                 if (inode->i_nlink) {
3217                         if (!S_ISREG(inode->i_mode)) {
3218                                 WARN_ON(1);
3219                                 iput(inode);
3220                                 continue;
3221                         }
3222                         nr_truncate++;
3223
3224                         /* 1 for the orphan item deletion. */
3225                         trans = btrfs_start_transaction(root, 1);
3226                         if (IS_ERR(trans)) {
3227                                 ret = PTR_ERR(trans);
3228                                 goto out;
3229                         }
3230                         ret = btrfs_orphan_add(trans, inode);
3231                         btrfs_end_transaction(trans, root);
3232                         if (ret)
3233                                 goto out;
3234
3235                         ret = btrfs_truncate(inode);
3236                         if (ret)
3237                                 btrfs_orphan_del(NULL, inode);
3238                 } else {
3239                         nr_unlink++;
3240                 }
3241
3242                 /* this will do delete_inode and everything for us */
3243                 iput(inode);
3244                 if (ret)
3245                         goto out;
3246         }
3247         /* release the path since we're done with it */
3248         btrfs_release_path(path);
3249
3250         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3251
3252         if (root->orphan_block_rsv)
3253                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3254                                         (u64)-1);
3255
3256         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3257                 trans = btrfs_join_transaction(root);
3258                 if (!IS_ERR(trans))
3259                         btrfs_end_transaction(trans, root);
3260         }
3261
3262         if (nr_unlink)
3263                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3264         if (nr_truncate)
3265                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3266
3267 out:
3268         if (ret)
3269                 btrfs_crit(root->fs_info,
3270                         "could not do orphan cleanup %d", ret);
3271         btrfs_free_path(path);
3272         return ret;
3273 }
3274
3275 /*
3276  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3277  * don't find any xattrs, we know there can't be any acls.
3278  *
3279  * slot is the slot the inode is in, objectid is the objectid of the inode
3280  */
3281 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3282                                           int slot, u64 objectid)
3283 {
3284         u32 nritems = btrfs_header_nritems(leaf);
3285         struct btrfs_key found_key;
3286         int scanned = 0;
3287
3288         slot++;
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                         return 1;
3299
3300                 /*
3301                  * we found a key greater than an xattr key, there can't
3302                  * be any acls later on
3303                  */
3304                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3305                         return 0;
3306
3307                 slot++;
3308                 scanned++;
3309
3310                 /*
3311                  * it goes inode, inode backrefs, xattrs, extents,
3312                  * so if there are a ton of hard links to an inode there can
3313                  * be a lot of backrefs.  Don't waste time searching too hard,
3314                  * this is just an optimization
3315                  */
3316                 if (scanned >= 8)
3317                         break;
3318         }
3319         /* we hit the end of the leaf before we found an xattr or
3320          * something larger than an xattr.  We have to assume the inode
3321          * has acls
3322          */
3323         return 1;
3324 }
3325
3326 /*
3327  * read an inode from the btree into the in-memory inode
3328  */
3329 static void btrfs_read_locked_inode(struct inode *inode)
3330 {
3331         struct btrfs_path *path;
3332         struct extent_buffer *leaf;
3333         struct btrfs_inode_item *inode_item;
3334         struct btrfs_timespec *tspec;
3335         struct btrfs_root *root = BTRFS_I(inode)->root;
3336         struct btrfs_key location;
3337         int maybe_acls;
3338         u32 rdev;
3339         int ret;
3340         bool filled = false;
3341
3342         ret = btrfs_fill_inode(inode, &rdev);
3343         if (!ret)
3344                 filled = true;
3345
3346         path = btrfs_alloc_path();
3347         if (!path)
3348                 goto make_bad;
3349
3350         path->leave_spinning = 1;
3351         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3352
3353         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3354         if (ret)
3355                 goto make_bad;
3356
3357         leaf = path->nodes[0];
3358
3359         if (filled)
3360                 goto cache_acl;
3361
3362         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3363                                     struct btrfs_inode_item);
3364         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3365         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3366         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3367         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3368         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3369
3370         tspec = btrfs_inode_atime(inode_item);
3371         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3372         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3373
3374         tspec = btrfs_inode_mtime(inode_item);
3375         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3376         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3377
3378         tspec = btrfs_inode_ctime(inode_item);
3379         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3380         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3381
3382         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3383         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3384         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3385
3386         /*
3387          * If we were modified in the current generation and evicted from memory
3388          * and then re-read we need to do a full sync since we don't have any
3389          * idea about which extents were modified before we were evicted from
3390          * cache.
3391          */
3392         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3393                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3394                         &BTRFS_I(inode)->runtime_flags);
3395
3396         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3397         inode->i_generation = BTRFS_I(inode)->generation;
3398         inode->i_rdev = 0;
3399         rdev = btrfs_inode_rdev(leaf, inode_item);
3400
3401         BTRFS_I(inode)->index_cnt = (u64)-1;
3402         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3403 cache_acl:
3404         /*
3405          * try to precache a NULL acl entry for files that don't have
3406          * any xattrs or acls
3407          */
3408         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3409                                            btrfs_ino(inode));
3410         if (!maybe_acls)
3411                 cache_no_acl(inode);
3412
3413         btrfs_free_path(path);
3414
3415         switch (inode->i_mode & S_IFMT) {
3416         case S_IFREG:
3417                 inode->i_mapping->a_ops = &btrfs_aops;
3418                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3419                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3420                 inode->i_fop = &btrfs_file_operations;
3421                 inode->i_op = &btrfs_file_inode_operations;
3422                 break;
3423         case S_IFDIR:
3424                 inode->i_fop = &btrfs_dir_file_operations;
3425                 if (root == root->fs_info->tree_root)
3426                         inode->i_op = &btrfs_dir_ro_inode_operations;
3427                 else
3428                         inode->i_op = &btrfs_dir_inode_operations;
3429                 break;
3430         case S_IFLNK:
3431                 inode->i_op = &btrfs_symlink_inode_operations;
3432                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3433                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3434                 break;
3435         default:
3436                 inode->i_op = &btrfs_special_inode_operations;
3437                 init_special_inode(inode, inode->i_mode, rdev);
3438                 break;
3439         }
3440
3441         btrfs_update_iflags(inode);
3442         return;
3443
3444 make_bad:
3445         btrfs_free_path(path);
3446         make_bad_inode(inode);
3447 }
3448
3449 /*
3450  * given a leaf and an inode, copy the inode fields into the leaf
3451  */
3452 static void fill_inode_item(struct btrfs_trans_handle *trans,
3453                             struct extent_buffer *leaf,
3454                             struct btrfs_inode_item *item,
3455                             struct inode *inode)
3456 {
3457         struct btrfs_map_token token;
3458
3459         btrfs_init_map_token(&token);
3460
3461         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3462         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3463         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3464                                    &token);
3465         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3466         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3467
3468         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3469                                      inode->i_atime.tv_sec, &token);
3470         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3471                                       inode->i_atime.tv_nsec, &token);
3472
3473         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3474                                      inode->i_mtime.tv_sec, &token);
3475         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3476                                       inode->i_mtime.tv_nsec, &token);
3477
3478         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3479                                      inode->i_ctime.tv_sec, &token);
3480         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3481                                       inode->i_ctime.tv_nsec, &token);
3482
3483         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3484                                      &token);
3485         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3486                                          &token);
3487         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3488         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3489         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3490         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3491         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3492 }
3493
3494 /*
3495  * copy everything in the in-memory inode into the btree.
3496  */
3497 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3498                                 struct btrfs_root *root, struct inode *inode)
3499 {
3500         struct btrfs_inode_item *inode_item;
3501         struct btrfs_path *path;
3502         struct extent_buffer *leaf;
3503         int ret;
3504
3505         path = btrfs_alloc_path();
3506         if (!path)
3507                 return -ENOMEM;
3508
3509         path->leave_spinning = 1;
3510         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3511                                  1);
3512         if (ret) {
3513                 if (ret > 0)
3514                         ret = -ENOENT;
3515                 goto failed;
3516         }
3517
3518         btrfs_unlock_up_safe(path, 1);
3519         leaf = path->nodes[0];
3520         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3521                                     struct btrfs_inode_item);
3522
3523         fill_inode_item(trans, leaf, inode_item, inode);
3524         btrfs_mark_buffer_dirty(leaf);
3525         btrfs_set_inode_last_trans(trans, inode);
3526         ret = 0;
3527 failed:
3528         btrfs_free_path(path);
3529         return ret;
3530 }
3531
3532 /*
3533  * copy everything in the in-memory inode into the btree.
3534  */
3535 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3536                                 struct btrfs_root *root, struct inode *inode)
3537 {
3538         int ret;
3539
3540         /*
3541          * If the inode is a free space inode, we can deadlock during commit
3542          * if we put it into the delayed code.
3543          *
3544          * The data relocation inode should also be directly updated
3545          * without delay
3546          */
3547         if (!btrfs_is_free_space_inode(inode)
3548             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3549                 btrfs_update_root_times(trans, root);
3550
3551                 ret = btrfs_delayed_update_inode(trans, root, inode);
3552                 if (!ret)
3553                         btrfs_set_inode_last_trans(trans, inode);
3554                 return ret;
3555         }
3556
3557         return btrfs_update_inode_item(trans, root, inode);
3558 }
3559
3560 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3561                                          struct btrfs_root *root,
3562                                          struct inode *inode)
3563 {
3564         int ret;
3565
3566         ret = btrfs_update_inode(trans, root, inode);
3567         if (ret == -ENOSPC)
3568                 return btrfs_update_inode_item(trans, root, inode);
3569         return ret;
3570 }
3571
3572 /*
3573  * unlink helper that gets used here in inode.c and in the tree logging
3574  * recovery code.  It remove a link in a directory with a given name, and
3575  * also drops the back refs in the inode to the directory
3576  */
3577 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3578                                 struct btrfs_root *root,
3579                                 struct inode *dir, struct inode *inode,
3580                                 const char *name, int name_len)
3581 {
3582         struct btrfs_path *path;
3583         int ret = 0;
3584         struct extent_buffer *leaf;
3585         struct btrfs_dir_item *di;
3586         struct btrfs_key key;
3587         u64 index;
3588         u64 ino = btrfs_ino(inode);
3589         u64 dir_ino = btrfs_ino(dir);
3590
3591         path = btrfs_alloc_path();
3592         if (!path) {
3593                 ret = -ENOMEM;
3594                 goto out;
3595         }
3596
3597         path->leave_spinning = 1;
3598         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3599                                     name, name_len, -1);
3600         if (IS_ERR(di)) {
3601                 ret = PTR_ERR(di);
3602                 goto err;
3603         }
3604         if (!di) {
3605                 ret = -ENOENT;
3606                 goto err;
3607         }
3608         leaf = path->nodes[0];
3609         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3610         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3611         if (ret)
3612                 goto err;
3613         btrfs_release_path(path);
3614
3615         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3616                                   dir_ino, &index);
3617         if (ret) {
3618                 btrfs_info(root->fs_info,
3619                         "failed to delete reference to %.*s, inode %llu parent %llu",
3620                         name_len, name,
3621                         (unsigned long long)ino, (unsigned long long)dir_ino);
3622                 btrfs_abort_transaction(trans, root, ret);
3623                 goto err;
3624         }
3625
3626         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3627         if (ret) {
3628                 btrfs_abort_transaction(trans, root, ret);
3629                 goto err;
3630         }
3631
3632         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3633                                          inode, dir_ino);
3634         if (ret != 0 && ret != -ENOENT) {
3635                 btrfs_abort_transaction(trans, root, ret);
3636                 goto err;
3637         }
3638
3639         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3640                                            dir, index);
3641         if (ret == -ENOENT)
3642                 ret = 0;
3643         else if (ret)
3644                 btrfs_abort_transaction(trans, root, ret);
3645 err:
3646         btrfs_free_path(path);
3647         if (ret)
3648                 goto out;
3649
3650         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3651         inode_inc_iversion(inode);
3652         inode_inc_iversion(dir);
3653         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3654         ret = btrfs_update_inode(trans, root, dir);
3655 out:
3656         return ret;
3657 }
3658
3659 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3660                        struct btrfs_root *root,
3661                        struct inode *dir, struct inode *inode,
3662                        const char *name, int name_len)
3663 {
3664         int ret;
3665         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3666         if (!ret) {
3667                 btrfs_drop_nlink(inode);
3668                 ret = btrfs_update_inode(trans, root, inode);
3669         }
3670         return ret;
3671 }
3672
3673
3674 /* helper to check if there is any shared block in the path */
3675 static int check_path_shared(struct btrfs_root *root,
3676                              struct btrfs_path *path)
3677 {
3678         struct extent_buffer *eb;
3679         int level;
3680         u64 refs = 1;
3681
3682         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3683                 int ret;
3684
3685                 if (!path->nodes[level])
3686                         break;
3687                 eb = path->nodes[level];
3688                 if (!btrfs_block_can_be_shared(root, eb))
3689                         continue;
3690                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3691                                                &refs, NULL);
3692                 if (refs > 1)
3693                         return 1;
3694         }
3695         return 0;
3696 }
3697
3698 /*
3699  * helper to start transaction for unlink and rmdir.
3700  *
3701  * unlink and rmdir are special in btrfs, they do not always free space.
3702  * so in enospc case, we should make sure they will free space before
3703  * allowing them to use the global metadata reservation.
3704  */
3705 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3706                                                        struct dentry *dentry)
3707 {
3708         struct btrfs_trans_handle *trans;
3709         struct btrfs_root *root = BTRFS_I(dir)->root;
3710         struct btrfs_path *path;
3711         struct btrfs_dir_item *di;
3712         struct inode *inode = dentry->d_inode;
3713         u64 index;
3714         int check_link = 1;
3715         int err = -ENOSPC;
3716         int ret;
3717         u64 ino = btrfs_ino(inode);
3718         u64 dir_ino = btrfs_ino(dir);
3719
3720         /*
3721          * 1 for the possible orphan item
3722          * 1 for the dir item
3723          * 1 for the dir index
3724          * 1 for the inode ref
3725          * 1 for the inode
3726          */
3727         trans = btrfs_start_transaction(root, 5);
3728         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3729                 return trans;
3730
3731         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3732                 return ERR_PTR(-ENOSPC);
3733
3734         /* check if there is someone else holds reference */
3735         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3736                 return ERR_PTR(-ENOSPC);
3737
3738         if (atomic_read(&inode->i_count) > 2)
3739                 return ERR_PTR(-ENOSPC);
3740
3741         if (xchg(&root->fs_info->enospc_unlink, 1))
3742                 return ERR_PTR(-ENOSPC);
3743
3744         path = btrfs_alloc_path();
3745         if (!path) {
3746                 root->fs_info->enospc_unlink = 0;
3747                 return ERR_PTR(-ENOMEM);
3748         }
3749
3750         /* 1 for the orphan item */
3751         trans = btrfs_start_transaction(root, 1);
3752         if (IS_ERR(trans)) {
3753                 btrfs_free_path(path);
3754                 root->fs_info->enospc_unlink = 0;
3755                 return trans;
3756         }
3757
3758         path->skip_locking = 1;
3759         path->search_commit_root = 1;
3760
3761         ret = btrfs_lookup_inode(trans, root, path,
3762                                 &BTRFS_I(dir)->location, 0);
3763         if (ret < 0) {
3764                 err = ret;
3765                 goto out;
3766         }
3767         if (ret == 0) {
3768                 if (check_path_shared(root, path))
3769                         goto out;
3770         } else {
3771                 check_link = 0;
3772         }
3773         btrfs_release_path(path);
3774
3775         ret = btrfs_lookup_inode(trans, root, path,
3776                                 &BTRFS_I(inode)->location, 0);
3777         if (ret < 0) {
3778                 err = ret;
3779                 goto out;
3780         }
3781         if (ret == 0) {
3782                 if (check_path_shared(root, path))
3783                         goto out;
3784         } else {
3785                 check_link = 0;
3786         }
3787         btrfs_release_path(path);
3788
3789         if (ret == 0 && S_ISREG(inode->i_mode)) {
3790                 ret = btrfs_lookup_file_extent(trans, root, path,
3791                                                ino, (u64)-1, 0);
3792                 if (ret < 0) {
3793                         err = ret;
3794                         goto out;
3795                 }
3796                 BUG_ON(ret == 0); /* Corruption */
3797                 if (check_path_shared(root, path))
3798                         goto out;
3799                 btrfs_release_path(path);
3800         }
3801
3802         if (!check_link) {
3803                 err = 0;
3804                 goto out;
3805         }
3806
3807         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3808                                 dentry->d_name.name, dentry->d_name.len, 0);
3809         if (IS_ERR(di)) {
3810                 err = PTR_ERR(di);
3811                 goto out;
3812         }
3813         if (di) {
3814                 if (check_path_shared(root, path))
3815                         goto out;
3816         } else {
3817                 err = 0;
3818                 goto out;
3819         }
3820         btrfs_release_path(path);
3821
3822         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3823                                         dentry->d_name.len, ino, dir_ino, 0,
3824                                         &index);
3825         if (ret) {
3826                 err = ret;
3827                 goto out;
3828         }
3829
3830         if (check_path_shared(root, path))
3831                 goto out;
3832
3833         btrfs_release_path(path);
3834
3835         /*
3836          * This is a commit root search, if we can lookup inode item and other
3837          * relative items in the commit root, it means the transaction of
3838          * dir/file creation has been committed, and the dir index item that we
3839          * delay to insert has also been inserted into the commit root. So
3840          * we needn't worry about the delayed insertion of the dir index item
3841          * here.
3842          */
3843         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3844                                 dentry->d_name.name, dentry->d_name.len, 0);
3845         if (IS_ERR(di)) {
3846                 err = PTR_ERR(di);
3847                 goto out;
3848         }
3849         BUG_ON(ret == -ENOENT);
3850         if (check_path_shared(root, path))
3851                 goto out;
3852
3853         err = 0;
3854 out:
3855         btrfs_free_path(path);
3856         /* Migrate the orphan reservation over */
3857         if (!err)
3858                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3859                                 &root->fs_info->global_block_rsv,
3860                                 trans->bytes_reserved);
3861
3862         if (err) {
3863                 btrfs_end_transaction(trans, root);
3864                 root->fs_info->enospc_unlink = 0;
3865                 return ERR_PTR(err);
3866         }
3867
3868         trans->block_rsv = &root->fs_info->global_block_rsv;
3869         return trans;
3870 }
3871
3872 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3873                                struct btrfs_root *root)
3874 {
3875         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3876                 btrfs_block_rsv_release(root, trans->block_rsv,
3877                                         trans->bytes_reserved);
3878                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3879                 BUG_ON(!root->fs_info->enospc_unlink);
3880                 root->fs_info->enospc_unlink = 0;
3881         }
3882         btrfs_end_transaction(trans, root);
3883 }
3884
3885 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3886 {
3887         struct btrfs_root *root = BTRFS_I(dir)->root;
3888         struct btrfs_trans_handle *trans;
3889         struct inode *inode = dentry->d_inode;
3890         int ret;
3891
3892         trans = __unlink_start_trans(dir, dentry);
3893         if (IS_ERR(trans))
3894                 return PTR_ERR(trans);
3895
3896         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3897
3898         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3899                                  dentry->d_name.name, dentry->d_name.len);
3900         if (ret)
3901                 goto out;
3902
3903         if (inode->i_nlink == 0) {
3904                 ret = btrfs_orphan_add(trans, inode);
3905                 if (ret)
3906                         goto out;
3907         }
3908
3909 out:
3910         __unlink_end_trans(trans, root);
3911         btrfs_btree_balance_dirty(root);
3912         return ret;
3913 }
3914
3915 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3916                         struct btrfs_root *root,
3917                         struct inode *dir, u64 objectid,
3918                         const char *name, int name_len)
3919 {
3920         struct btrfs_path *path;
3921         struct extent_buffer *leaf;
3922         struct btrfs_dir_item *di;
3923         struct btrfs_key key;
3924         u64 index;
3925         int ret;
3926         u64 dir_ino = btrfs_ino(dir);
3927
3928         path = btrfs_alloc_path();
3929         if (!path)
3930                 return -ENOMEM;
3931
3932         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3933                                    name, name_len, -1);
3934         if (IS_ERR_OR_NULL(di)) {
3935                 if (!di)
3936                         ret = -ENOENT;
3937                 else
3938                         ret = PTR_ERR(di);
3939                 goto out;
3940         }
3941
3942         leaf = path->nodes[0];
3943         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3944         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3945         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3946         if (ret) {
3947                 btrfs_abort_transaction(trans, root, ret);
3948                 goto out;
3949         }
3950         btrfs_release_path(path);
3951
3952         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3953                                  objectid, root->root_key.objectid,
3954                                  dir_ino, &index, name, name_len);
3955         if (ret < 0) {
3956                 if (ret != -ENOENT) {
3957                         btrfs_abort_transaction(trans, root, ret);
3958                         goto out;
3959                 }
3960                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3961                                                  name, name_len);
3962                 if (IS_ERR_OR_NULL(di)) {
3963                         if (!di)
3964                                 ret = -ENOENT;
3965                         else
3966                                 ret = PTR_ERR(di);
3967                         btrfs_abort_transaction(trans, root, ret);
3968                         goto out;
3969                 }
3970
3971                 leaf = path->nodes[0];
3972                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3973                 btrfs_release_path(path);
3974                 index = key.offset;
3975         }
3976         btrfs_release_path(path);
3977
3978         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3979         if (ret) {
3980                 btrfs_abort_transaction(trans, root, ret);
3981                 goto out;
3982         }
3983
3984         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3985         inode_inc_iversion(dir);
3986         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3987         ret = btrfs_update_inode_fallback(trans, root, dir);
3988         if (ret)
3989                 btrfs_abort_transaction(trans, root, ret);
3990 out:
3991         btrfs_free_path(path);
3992         return ret;
3993 }
3994
3995 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3996 {
3997         struct inode *inode = dentry->d_inode;
3998         int err = 0;
3999         struct btrfs_root *root = BTRFS_I(dir)->root;
4000         struct btrfs_trans_handle *trans;
4001
4002         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4003                 return -ENOTEMPTY;
4004         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4005                 return -EPERM;
4006
4007         trans = __unlink_start_trans(dir, dentry);
4008         if (IS_ERR(trans))
4009                 return PTR_ERR(trans);
4010
4011         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4012                 err = btrfs_unlink_subvol(trans, root, dir,
4013                                           BTRFS_I(inode)->location.objectid,
4014                                           dentry->d_name.name,
4015                                           dentry->d_name.len);
4016                 goto out;
4017         }
4018
4019         err = btrfs_orphan_add(trans, inode);
4020         if (err)
4021                 goto out;
4022
4023         /* now the directory is empty */
4024         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4025                                  dentry->d_name.name, dentry->d_name.len);
4026         if (!err)
4027                 btrfs_i_size_write(inode, 0);
4028 out:
4029         __unlink_end_trans(trans, root);
4030         btrfs_btree_balance_dirty(root);
4031
4032         return err;
4033 }
4034
4035 /*
4036  * this can truncate away extent items, csum items and directory items.
4037  * It starts at a high offset and removes keys until it can't find
4038  * any higher than new_size
4039  *
4040  * csum items that cross the new i_size are truncated to the new size
4041  * as well.
4042  *
4043  * min_type is the minimum key type to truncate down to.  If set to 0, this
4044  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4045  */
4046 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4047                                struct btrfs_root *root,
4048                                struct inode *inode,
4049                                u64 new_size, u32 min_type)
4050 {
4051         struct btrfs_path *path;
4052         struct extent_buffer *leaf;
4053         struct btrfs_file_extent_item *fi;
4054         struct btrfs_key key;
4055         struct btrfs_key found_key;
4056         u64 extent_start = 0;
4057         u64 extent_num_bytes = 0;
4058         u64 extent_offset = 0;
4059         u64 item_end = 0;
4060         u32 found_type = (u8)-1;
4061         int found_extent;
4062         int del_item;
4063         int pending_del_nr = 0;
4064         int pending_del_slot = 0;
4065         int extent_type = -1;
4066         int ret;
4067         int err = 0;
4068         u64 ino = btrfs_ino(inode);
4069
4070         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4071
4072         path = btrfs_alloc_path();
4073         if (!path)
4074                 return -ENOMEM;
4075         path->reada = -1;
4076
4077         /*
4078          * We want to drop from the next block forward in case this new size is
4079          * not block aligned since we will be keeping the last block of the
4080          * extent just the way it is.
4081          */
4082         if (root->ref_cows || root == root->fs_info->tree_root)
4083                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4084                                         root->sectorsize), (u64)-1, 0);
4085
4086         /*
4087          * This function is also used to drop the items in the log tree before
4088          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4089          * it is used to drop the loged items. So we shouldn't kill the delayed
4090          * items.
4091          */
4092         if (min_type == 0 && root == BTRFS_I(inode)->root)
4093                 btrfs_kill_delayed_inode_items(inode);
4094
4095         key.objectid = ino;
4096         key.offset = (u64)-1;
4097         key.type = (u8)-1;
4098
4099 search_again:
4100         path->leave_spinning = 1;
4101         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4102         if (ret < 0) {
4103                 err = ret;
4104                 goto out;
4105         }
4106
4107         if (ret > 0) {
4108                 /* there are no items in the tree for us to truncate, we're
4109                  * done
4110                  */
4111                 if (path->slots[0] == 0)
4112                         goto out;
4113                 path->slots[0]--;
4114         }
4115
4116         while (1) {
4117                 fi = NULL;
4118                 leaf = path->nodes[0];
4119                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4120                 found_type = btrfs_key_type(&found_key);
4121
4122                 if (found_key.objectid != ino)
4123                         break;
4124
4125                 if (found_type < min_type)
4126                         break;
4127
4128                 item_end = found_key.offset;
4129                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4130                         fi = btrfs_item_ptr(leaf, path->slots[0],
4131                                             struct btrfs_file_extent_item);
4132                         extent_type = btrfs_file_extent_type(leaf, fi);
4133                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4134                                 item_end +=
4135                                     btrfs_file_extent_num_bytes(leaf, fi);
4136                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4137                                 item_end += btrfs_file_extent_inline_len(leaf,
4138                                                                          fi);
4139                         }
4140                         item_end--;
4141                 }
4142                 if (found_type > min_type) {
4143                         del_item = 1;
4144                 } else {
4145                         if (item_end < new_size)
4146                                 break;
4147                         if (found_key.offset >= new_size)
4148                                 del_item = 1;
4149                         else
4150                                 del_item = 0;
4151                 }
4152                 found_extent = 0;
4153                 /* FIXME, shrink the extent if the ref count is only 1 */
4154                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4155                         goto delete;
4156
4157                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4158                         u64 num_dec;
4159                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4160                         if (!del_item) {
4161                                 u64 orig_num_bytes =
4162                                         btrfs_file_extent_num_bytes(leaf, fi);
4163                                 extent_num_bytes = ALIGN(new_size -
4164                                                 found_key.offset,
4165                                                 root->sectorsize);
4166                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4167                                                          extent_num_bytes);
4168                                 num_dec = (orig_num_bytes -
4169                                            extent_num_bytes);
4170                                 if (root->ref_cows && extent_start != 0)
4171                                         inode_sub_bytes(inode, num_dec);
4172                                 btrfs_mark_buffer_dirty(leaf);
4173                         } else {
4174                                 extent_num_bytes =
4175                                         btrfs_file_extent_disk_num_bytes(leaf,
4176                                                                          fi);
4177                                 extent_offset = found_key.offset -
4178                                         btrfs_file_extent_offset(leaf, fi);
4179
4180                                 /* FIXME blocksize != 4096 */
4181                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4182                                 if (extent_start != 0) {
4183                                         found_extent = 1;
4184                                         if (root->ref_cows)
4185                                                 inode_sub_bytes(inode, num_dec);
4186                                 }
4187                         }
4188                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4189                         /*
4190                          * we can't truncate inline items that have had
4191                          * special encodings
4192                          */
4193                         if (!del_item &&
4194                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4195                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4196                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4197                                 u32 size = new_size - found_key.offset;
4198
4199                                 if (root->ref_cows) {
4200                                         inode_sub_bytes(inode, item_end + 1 -
4201                                                         new_size);
4202                                 }
4203                                 size =
4204                                     btrfs_file_extent_calc_inline_size(size);
4205                                 btrfs_truncate_item(root, path, size, 1);
4206                         } else if (root->ref_cows) {
4207                                 inode_sub_bytes(inode, item_end + 1 -
4208                                                 found_key.offset);
4209                         }
4210                 }
4211 delete:
4212                 if (del_item) {
4213                         if (!pending_del_nr) {
4214                                 /* no pending yet, add ourselves */
4215                                 pending_del_slot = path->slots[0];
4216                                 pending_del_nr = 1;
4217                         } else if (pending_del_nr &&
4218                                    path->slots[0] + 1 == pending_del_slot) {
4219                                 /* hop on the pending chunk */
4220                                 pending_del_nr++;
4221                                 pending_del_slot = path->slots[0];
4222                         } else {
4223                                 BUG();
4224                         }
4225                 } else {
4226                         break;
4227                 }
4228                 if (found_extent && (root->ref_cows ||
4229                                      root == root->fs_info->tree_root)) {
4230                         btrfs_set_path_blocking(path);
4231                         ret = btrfs_free_extent(trans, root, extent_start,
4232                                                 extent_num_bytes, 0,
4233                                                 btrfs_header_owner(leaf),
4234                                                 ino, extent_offset, 0);
4235                         BUG_ON(ret);
4236                 }
4237
4238                 if (found_type == BTRFS_INODE_ITEM_KEY)
4239                         break;
4240
4241                 if (path->slots[0] == 0 ||
4242                     path->slots[0] != pending_del_slot) {
4243                         if (pending_del_nr) {
4244                                 ret = btrfs_del_items(trans, root, path,
4245                                                 pending_del_slot,
4246                                                 pending_del_nr);
4247                                 if (ret) {
4248                                         btrfs_abort_transaction(trans,
4249                                                                 root, ret);
4250                                         goto error;
4251                                 }
4252                                 pending_del_nr = 0;
4253                         }
4254                         btrfs_release_path(path);
4255                         goto search_again;
4256                 } else {
4257                         path->slots[0]--;
4258                 }
4259         }
4260 out:
4261         if (pending_del_nr) {
4262                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4263                                       pending_del_nr);
4264                 if (ret)
4265                         btrfs_abort_transaction(trans, root, ret);
4266         }
4267 error:
4268         btrfs_free_path(path);
4269         return err;
4270 }
4271
4272 /*
4273  * btrfs_truncate_page - read, zero a chunk and write a page
4274  * @inode - inode that we're zeroing
4275  * @from - the offset to start zeroing
4276  * @len - the length to zero, 0 to zero the entire range respective to the
4277  *      offset
4278  * @front - zero up to the offset instead of from the offset on
4279  *
4280  * This will find the page for the "from" offset and cow the page and zero the
4281  * part we want to zero.  This is used with truncate and hole punching.
4282  */
4283 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4284                         int front)
4285 {
4286         struct address_space *mapping = inode->i_mapping;
4287         struct btrfs_root *root = BTRFS_I(inode)->root;
4288         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4289         struct btrfs_ordered_extent *ordered;
4290         struct extent_state *cached_state = NULL;
4291         char *kaddr;
4292         u32 blocksize = root->sectorsize;
4293         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4294         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4295         struct page *page;
4296         gfp_t mask = btrfs_alloc_write_mask(mapping);
4297         int ret = 0;
4298         u64 page_start;
4299         u64 page_end;
4300
4301         if ((offset & (blocksize - 1)) == 0 &&
4302             (!len || ((len & (blocksize - 1)) == 0)))
4303                 goto out;
4304         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4305         if (ret)
4306                 goto out;
4307
4308 again:
4309         page = find_or_create_page(mapping, index, mask);
4310         if (!page) {
4311                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4312                 ret = -ENOMEM;
4313                 goto out;
4314         }
4315
4316         page_start = page_offset(page);
4317         page_end = page_start + PAGE_CACHE_SIZE - 1;
4318
4319         if (!PageUptodate(page)) {
4320                 ret = btrfs_readpage(NULL, page);
4321                 lock_page(page);
4322                 if (page->mapping != mapping) {
4323                         unlock_page(page);
4324                         page_cache_release(page);
4325                         goto again;
4326                 }
4327                 if (!PageUptodate(page)) {
4328                         ret = -EIO;
4329                         goto out_unlock;
4330                 }
4331         }
4332         wait_on_page_writeback(page);
4333
4334         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4335         set_page_extent_mapped(page);
4336
4337         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4338         if (ordered) {
4339                 unlock_extent_cached(io_tree, page_start, page_end,
4340                                      &cached_state, GFP_NOFS);
4341                 unlock_page(page);
4342                 page_cache_release(page);
4343                 btrfs_start_ordered_extent(inode, ordered, 1);
4344                 btrfs_put_ordered_extent(ordered);
4345                 goto again;
4346         }
4347
4348         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4349                           EXTENT_DIRTY | EXTENT_DELALLOC |
4350                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4351                           0, 0, &cached_state, GFP_NOFS);
4352
4353         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4354                                         &cached_state);
4355         if (ret) {
4356                 unlock_extent_cached(io_tree, page_start, page_end,
4357                                      &cached_state, GFP_NOFS);
4358                 goto out_unlock;
4359         }
4360
4361         if (offset != PAGE_CACHE_SIZE) {
4362                 if (!len)
4363                         len = PAGE_CACHE_SIZE - offset;
4364                 kaddr = kmap(page);
4365                 if (front)
4366                         memset(kaddr, 0, offset);
4367                 else
4368                         memset(kaddr + offset, 0, len);
4369                 flush_dcache_page(page);
4370                 kunmap(page);
4371         }
4372         ClearPageChecked(page);
4373         set_page_dirty(page);
4374         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4375                              GFP_NOFS);
4376
4377 out_unlock:
4378         if (ret)
4379                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4380         unlock_page(page);
4381         page_cache_release(page);
4382 out:
4383         return ret;
4384 }
4385
4386 /*
4387  * This function puts in dummy file extents for the area we're creating a hole
4388  * for.  So if we are truncating this file to a larger size we need to insert
4389  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4390  * the range between oldsize and size
4391  */
4392 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4393 {
4394         struct btrfs_trans_handle *trans;
4395         struct btrfs_root *root = BTRFS_I(inode)->root;
4396         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4397         struct extent_map *em = NULL;
4398         struct extent_state *cached_state = NULL;
4399         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4400         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4401         u64 block_end = ALIGN(size, root->sectorsize);
4402         u64 last_byte;
4403         u64 cur_offset;
4404         u64 hole_size;
4405         int err = 0;
4406
4407         if (size <= hole_start)
4408                 return 0;
4409
4410         while (1) {
4411                 struct btrfs_ordered_extent *ordered;
4412                 btrfs_wait_ordered_range(inode, hole_start,
4413                                          block_end - hole_start);
4414                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4415                                  &cached_state);
4416                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4417                 if (!ordered)
4418                         break;
4419                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4420                                      &cached_state, GFP_NOFS);
4421                 btrfs_put_ordered_extent(ordered);
4422         }
4423
4424         cur_offset = hole_start;
4425         while (1) {
4426                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4427                                 block_end - cur_offset, 0);
4428                 if (IS_ERR(em)) {
4429                         err = PTR_ERR(em);
4430                         em = NULL;
4431                         break;
4432                 }
4433                 last_byte = min(extent_map_end(em), block_end);
4434                 last_byte = ALIGN(last_byte , root->sectorsize);
4435                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4436                         struct extent_map *hole_em;
4437                         hole_size = last_byte - cur_offset;
4438
4439                         trans = btrfs_start_transaction(root, 3);
4440                         if (IS_ERR(trans)) {
4441                                 err = PTR_ERR(trans);
4442                                 break;
4443                         }
4444
4445                         err = btrfs_drop_extents(trans, root, inode,
4446                                                  cur_offset,
4447                                                  cur_offset + hole_size, 1);
4448                         if (err) {
4449                                 btrfs_abort_transaction(trans, root, err);
4450                                 btrfs_end_transaction(trans, root);
4451                                 break;
4452                         }
4453
4454                         err = btrfs_insert_file_extent(trans, root,
4455                                         btrfs_ino(inode), cur_offset, 0,
4456                                         0, hole_size, 0, hole_size,
4457                                         0, 0, 0);
4458                         if (err) {
4459                                 btrfs_abort_transaction(trans, root, err);
4460                                 btrfs_end_transaction(trans, root);
4461                                 break;
4462                         }
4463
4464                         btrfs_drop_extent_cache(inode, cur_offset,
4465                                                 cur_offset + hole_size - 1, 0);
4466                         hole_em = alloc_extent_map();
4467                         if (!hole_em) {
4468                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4469                                         &BTRFS_I(inode)->runtime_flags);
4470                                 goto next;
4471                         }
4472                         hole_em->start = cur_offset;
4473                         hole_em->len = hole_size;
4474                         hole_em->orig_start = cur_offset;
4475
4476                         hole_em->block_start = EXTENT_MAP_HOLE;
4477                         hole_em->block_len = 0;
4478                         hole_em->orig_block_len = 0;
4479                         hole_em->ram_bytes = hole_size;
4480                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4481                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4482                         hole_em->generation = trans->transid;
4483
4484                         while (1) {
4485                                 write_lock(&em_tree->lock);
4486                                 err = add_extent_mapping(em_tree, hole_em, 1);
4487                                 write_unlock(&em_tree->lock);
4488                                 if (err != -EEXIST)
4489                                         break;
4490                                 btrfs_drop_extent_cache(inode, cur_offset,
4491                                                         cur_offset +
4492                                                         hole_size - 1, 0);
4493                         }
4494                         free_extent_map(hole_em);
4495 next:
4496                         btrfs_update_inode(trans, root, inode);
4497                         btrfs_end_transaction(trans, root);
4498                 }
4499                 free_extent_map(em);
4500                 em = NULL;
4501                 cur_offset = last_byte;
4502                 if (cur_offset >= block_end)
4503                         break;
4504         }
4505
4506         free_extent_map(em);
4507         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4508                              GFP_NOFS);
4509         return err;
4510 }
4511
4512 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4513 {
4514         struct btrfs_root *root = BTRFS_I(inode)->root;
4515         struct btrfs_trans_handle *trans;
4516         loff_t oldsize = i_size_read(inode);
4517         loff_t newsize = attr->ia_size;
4518         int mask = attr->ia_valid;
4519         int ret;
4520
4521         if (newsize == oldsize)
4522                 return 0;
4523
4524         /*
4525          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4526          * special case where we need to update the times despite not having
4527          * these flags set.  For all other operations the VFS set these flags
4528          * explicitly if it wants a timestamp update.
4529          */
4530         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4531                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4532
4533         if (newsize > oldsize) {
4534                 truncate_pagecache(inode, oldsize, newsize);
4535                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4536                 if (ret)
4537                         return ret;
4538
4539                 trans = btrfs_start_transaction(root, 1);
4540                 if (IS_ERR(trans))
4541                         return PTR_ERR(trans);
4542
4543                 i_size_write(inode, newsize);
4544                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4545                 ret = btrfs_update_inode(trans, root, inode);
4546                 btrfs_end_transaction(trans, root);
4547         } else {
4548
4549                 /*
4550                  * We're truncating a file that used to have good data down to
4551                  * zero. Make sure it gets into the ordered flush list so that
4552                  * any new writes get down to disk quickly.
4553                  */
4554                 if (newsize == 0)
4555                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4556                                 &BTRFS_I(inode)->runtime_flags);
4557
4558                 /*
4559                  * 1 for the orphan item we're going to add
4560                  * 1 for the orphan item deletion.
4561                  */
4562                 trans = btrfs_start_transaction(root, 2);
4563                 if (IS_ERR(trans))
4564                         return PTR_ERR(trans);
4565
4566                 /*
4567                  * We need to do this in case we fail at _any_ point during the
4568                  * actual truncate.  Once we do the truncate_setsize we could
4569                  * invalidate pages which forces any outstanding ordered io to
4570                  * be instantly completed which will give us extents that need
4571                  * to be truncated.  If we fail to get an orphan inode down we
4572                  * could have left over extents that were never meant to live,
4573                  * so we need to garuntee from this point on that everything
4574                  * will be consistent.
4575                  */
4576                 ret = btrfs_orphan_add(trans, inode);
4577                 btrfs_end_transaction(trans, root);
4578                 if (ret)
4579                         return ret;
4580
4581                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4582                 truncate_setsize(inode, newsize);
4583
4584                 /* Disable nonlocked read DIO to avoid the end less truncate */
4585                 btrfs_inode_block_unlocked_dio(inode);
4586                 inode_dio_wait(inode);
4587                 btrfs_inode_resume_unlocked_dio(inode);
4588
4589                 ret = btrfs_truncate(inode);
4590                 if (ret && inode->i_nlink)
4591                         btrfs_orphan_del(NULL, inode);
4592         }
4593
4594         return ret;
4595 }
4596
4597 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4598 {
4599         struct inode *inode = dentry->d_inode;
4600         struct btrfs_root *root = BTRFS_I(inode)->root;
4601         int err;
4602
4603         if (btrfs_root_readonly(root))
4604                 return -EROFS;
4605
4606         err = inode_change_ok(inode, attr);
4607         if (err)
4608                 return err;
4609
4610         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4611                 err = btrfs_setsize(inode, attr);
4612                 if (err)
4613                         return err;
4614         }
4615
4616         if (attr->ia_valid) {
4617                 setattr_copy(inode, attr);
4618                 inode_inc_iversion(inode);
4619                 err = btrfs_dirty_inode(inode);
4620
4621                 if (!err && attr->ia_valid & ATTR_MODE)
4622                         err = btrfs_acl_chmod(inode);
4623         }
4624
4625         return err;
4626 }
4627
4628 void btrfs_evict_inode(struct inode *inode)
4629 {
4630         struct btrfs_trans_handle *trans;
4631         struct btrfs_root *root = BTRFS_I(inode)->root;
4632         struct btrfs_block_rsv *rsv, *global_rsv;
4633         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4634         int ret;
4635
4636         trace_btrfs_inode_evict(inode);
4637
4638         truncate_inode_pages(&inode->i_data, 0);
4639         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4640                                btrfs_is_free_space_inode(inode)))
4641                 goto no_delete;
4642
4643         if (is_bad_inode(inode)) {
4644                 btrfs_orphan_del(NULL, inode);
4645                 goto no_delete;
4646         }
4647         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4648         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4649
4650         if (root->fs_info->log_root_recovering) {
4651                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4652                                  &BTRFS_I(inode)->runtime_flags));
4653                 goto no_delete;
4654         }
4655
4656         if (inode->i_nlink > 0) {
4657                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4658                 goto no_delete;
4659         }
4660
4661         ret = btrfs_commit_inode_delayed_inode(inode);
4662         if (ret) {
4663                 btrfs_orphan_del(NULL, inode);
4664                 goto no_delete;
4665         }
4666
4667         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4668         if (!rsv) {
4669                 btrfs_orphan_del(NULL, inode);
4670                 goto no_delete;
4671         }
4672         rsv->size = min_size;
4673         rsv->failfast = 1;
4674         global_rsv = &root->fs_info->global_block_rsv;
4675
4676         btrfs_i_size_write(inode, 0);
4677
4678         /*
4679          * This is a bit simpler than btrfs_truncate since we've already
4680          * reserved our space for our orphan item in the unlink, so we just
4681          * need to reserve some slack space in case we add bytes and update
4682          * inode item when doing the truncate.
4683          */
4684         while (1) {
4685                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4686                                              BTRFS_RESERVE_FLUSH_LIMIT);
4687
4688                 /*
4689                  * Try and steal from the global reserve since we will
4690                  * likely not use this space anyway, we want to try as
4691                  * hard as possible to get this to work.
4692                  */
4693                 if (ret)
4694                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4695
4696                 if (ret) {
4697                         btrfs_warn(root->fs_info,
4698                                 "Could not get space for a delete, will truncate on mount %d",
4699                                 ret);
4700                         btrfs_orphan_del(NULL, inode);
4701                         btrfs_free_block_rsv(root, rsv);
4702                         goto no_delete;
4703                 }
4704
4705                 trans = btrfs_join_transaction(root);
4706                 if (IS_ERR(trans)) {
4707                         btrfs_orphan_del(NULL, inode);
4708                         btrfs_free_block_rsv(root, rsv);
4709                         goto no_delete;
4710                 }
4711
4712                 trans->block_rsv = rsv;
4713
4714                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4715                 if (ret != -ENOSPC)
4716                         break;
4717
4718                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4719                 btrfs_end_transaction(trans, root);
4720                 trans = NULL;
4721                 btrfs_btree_balance_dirty(root);
4722         }
4723
4724         btrfs_free_block_rsv(root, rsv);
4725
4726         if (ret == 0) {
4727                 trans->block_rsv = root->orphan_block_rsv;
4728                 ret = btrfs_orphan_del(trans, inode);
4729                 BUG_ON(ret);
4730         }
4731
4732         trans->block_rsv = &root->fs_info->trans_block_rsv;
4733         if (!(root == root->fs_info->tree_root ||
4734               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4735                 btrfs_return_ino(root, btrfs_ino(inode));
4736
4737         btrfs_end_transaction(trans, root);
4738         btrfs_btree_balance_dirty(root);
4739 no_delete:
4740         btrfs_remove_delayed_node(inode);
4741         clear_inode(inode);
4742         return;
4743 }
4744
4745 /*
4746  * this returns the key found in the dir entry in the location pointer.
4747  * If no dir entries were found, location->objectid is 0.
4748  */
4749 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4750                                struct btrfs_key *location)
4751 {
4752         const char *name = dentry->d_name.name;
4753         int namelen = dentry->d_name.len;
4754         struct btrfs_dir_item *di;
4755         struct btrfs_path *path;
4756         struct btrfs_root *root = BTRFS_I(dir)->root;
4757         int ret = 0;
4758
4759         path = btrfs_alloc_path();
4760         if (!path)
4761                 return -ENOMEM;
4762
4763         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4764                                     namelen, 0);
4765         if (IS_ERR(di))
4766                 ret = PTR_ERR(di);
4767
4768         if (IS_ERR_OR_NULL(di))
4769                 goto out_err;
4770
4771         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4772 out:
4773         btrfs_free_path(path);
4774         return ret;
4775 out_err:
4776         location->objectid = 0;
4777         goto out;
4778 }
4779
4780 /*
4781  * when we hit a tree root in a directory, the btrfs part of the inode
4782  * needs to be changed to reflect the root directory of the tree root.  This
4783  * is kind of like crossing a mount point.
4784  */
4785 static int fixup_tree_root_location(struct btrfs_root *root,
4786                                     struct inode *dir,
4787                                     struct dentry *dentry,
4788                                     struct btrfs_key *location,
4789                                     struct btrfs_root **sub_root)
4790 {
4791         struct btrfs_path *path;
4792         struct btrfs_root *new_root;
4793         struct btrfs_root_ref *ref;
4794         struct extent_buffer *leaf;
4795         int ret;
4796         int err = 0;
4797
4798         path = btrfs_alloc_path();
4799         if (!path) {
4800                 err = -ENOMEM;
4801                 goto out;
4802         }
4803
4804         err = -ENOENT;
4805         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4806                                   BTRFS_I(dir)->root->root_key.objectid,
4807                                   location->objectid);
4808         if (ret) {
4809                 if (ret < 0)
4810                         err = ret;
4811                 goto out;
4812         }
4813
4814         leaf = path->nodes[0];
4815         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4816         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4817             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4818                 goto out;
4819
4820         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4821                                    (unsigned long)(ref + 1),
4822                                    dentry->d_name.len);
4823         if (ret)
4824                 goto out;
4825
4826         btrfs_release_path(path);
4827
4828         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4829         if (IS_ERR(new_root)) {
4830                 err = PTR_ERR(new_root);
4831                 goto out;
4832         }
4833
4834         if (btrfs_root_refs(&new_root->root_item) == 0) {
4835                 err = -ENOENT;
4836                 goto out;
4837         }
4838
4839         *sub_root = new_root;
4840         location->objectid = btrfs_root_dirid(&new_root->root_item);
4841         location->type = BTRFS_INODE_ITEM_KEY;
4842         location->offset = 0;
4843         err = 0;
4844 out:
4845         btrfs_free_path(path);
4846         return err;
4847 }
4848
4849 static void inode_tree_add(struct inode *inode)
4850 {
4851         struct btrfs_root *root = BTRFS_I(inode)->root;
4852         struct btrfs_inode *entry;
4853         struct rb_node **p;
4854         struct rb_node *parent;
4855         u64 ino = btrfs_ino(inode);
4856
4857         if (inode_unhashed(inode))
4858                 return;
4859 again:
4860         parent = NULL;
4861         spin_lock(&root->inode_lock);
4862         p = &root->inode_tree.rb_node;
4863         while (*p) {
4864                 parent = *p;
4865                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4866
4867                 if (ino < btrfs_ino(&entry->vfs_inode))
4868                         p = &parent->rb_left;
4869                 else if (ino > btrfs_ino(&entry->vfs_inode))
4870                         p = &parent->rb_right;
4871                 else {
4872                         WARN_ON(!(entry->vfs_inode.i_state &
4873                                   (I_WILL_FREE | I_FREEING)));
4874                         rb_erase(parent, &root->inode_tree);
4875                         RB_CLEAR_NODE(parent);
4876                         spin_unlock(&root->inode_lock);
4877                         goto again;
4878                 }
4879         }
4880         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4881         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4882         spin_unlock(&root->inode_lock);
4883 }
4884
4885 static void inode_tree_del(struct inode *inode)
4886 {
4887         struct btrfs_root *root = BTRFS_I(inode)->root;
4888         int empty = 0;
4889
4890         spin_lock(&root->inode_lock);
4891         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4892                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4893                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4894                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4895         }
4896         spin_unlock(&root->inode_lock);
4897
4898         /*
4899          * Free space cache has inodes in the tree root, but the tree root has a
4900          * root_refs of 0, so this could end up dropping the tree root as a
4901          * snapshot, so we need the extra !root->fs_info->tree_root check to
4902          * make sure we don't drop it.
4903          */
4904         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4905             root != root->fs_info->tree_root) {
4906                 synchronize_srcu(&root->fs_info->subvol_srcu);
4907                 spin_lock(&root->inode_lock);
4908                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4909                 spin_unlock(&root->inode_lock);
4910                 if (empty)
4911                         btrfs_add_dead_root(root);
4912         }
4913 }
4914
4915 void btrfs_invalidate_inodes(struct btrfs_root *root)
4916 {
4917         struct rb_node *node;
4918         struct rb_node *prev;
4919         struct btrfs_inode *entry;
4920         struct inode *inode;
4921         u64 objectid = 0;
4922
4923         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4924
4925         spin_lock(&root->inode_lock);
4926 again:
4927         node = root->inode_tree.rb_node;
4928         prev = NULL;
4929         while (node) {
4930                 prev = node;
4931                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4932
4933                 if (objectid < btrfs_ino(&entry->vfs_inode))
4934                         node = node->rb_left;
4935                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4936                         node = node->rb_right;
4937                 else
4938                         break;
4939         }
4940         if (!node) {
4941                 while (prev) {
4942                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4943                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4944                                 node = prev;
4945                                 break;
4946                         }
4947                         prev = rb_next(prev);
4948                 }
4949         }
4950         while (node) {
4951                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4952                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4953                 inode = igrab(&entry->vfs_inode);
4954                 if (inode) {
4955                         spin_unlock(&root->inode_lock);
4956                         if (atomic_read(&inode->i_count) > 1)
4957                                 d_prune_aliases(inode);
4958                         /*
4959                          * btrfs_drop_inode will have it removed from
4960                          * the inode cache when its usage count
4961                          * hits zero.
4962                          */
4963                         iput(inode);
4964                         cond_resched();
4965                         spin_lock(&root->inode_lock);
4966                         goto again;
4967                 }
4968
4969                 if (cond_resched_lock(&root->inode_lock))
4970                         goto again;
4971
4972                 node = rb_next(node);
4973         }
4974         spin_unlock(&root->inode_lock);
4975 }
4976
4977 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4978 {
4979         struct btrfs_iget_args *args = p;
4980         inode->i_ino = args->ino;
4981         BTRFS_I(inode)->root = args->root;
4982         return 0;
4983 }
4984
4985 static int btrfs_find_actor(struct inode *inode, void *opaque)
4986 {
4987         struct btrfs_iget_args *args = opaque;
4988         return args->ino == btrfs_ino(inode) &&
4989                 args->root == BTRFS_I(inode)->root;
4990 }
4991
4992 static struct inode *btrfs_iget_locked(struct super_block *s,
4993                                        u64 objectid,
4994                                        struct btrfs_root *root)
4995 {
4996         struct inode *inode;
4997         struct btrfs_iget_args args;
4998         args.ino = objectid;
4999         args.root = root;
5000
5001         inode = iget5_locked(s, objectid, btrfs_find_actor,
5002                              btrfs_init_locked_inode,
5003                              (void *)&args);
5004         return inode;
5005 }
5006
5007 /* Get an inode object given its location and corresponding root.
5008  * Returns in *is_new if the inode was read from disk
5009  */
5010 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5011                          struct btrfs_root *root, int *new)
5012 {
5013         struct inode *inode;
5014
5015         inode = btrfs_iget_locked(s, location->objectid, root);
5016         if (!inode)
5017                 return ERR_PTR(-ENOMEM);
5018
5019         if (inode->i_state & I_NEW) {
5020                 BTRFS_I(inode)->root = root;
5021                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5022                 btrfs_read_locked_inode(inode);
5023                 if (!is_bad_inode(inode)) {
5024                         inode_tree_add(inode);
5025                         unlock_new_inode(inode);
5026                         if (new)
5027                                 *new = 1;
5028                 } else {
5029                         unlock_new_inode(inode);
5030                         iput(inode);
5031                         inode = ERR_PTR(-ESTALE);
5032                 }
5033         }
5034
5035         return inode;
5036 }
5037
5038 static struct inode *new_simple_dir(struct super_block *s,
5039                                     struct btrfs_key *key,
5040                                     struct btrfs_root *root)
5041 {
5042         struct inode *inode = new_inode(s);
5043
5044         if (!inode)
5045                 return ERR_PTR(-ENOMEM);
5046
5047         BTRFS_I(inode)->root = root;
5048         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5049         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5050
5051         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5052         inode->i_op = &btrfs_dir_ro_inode_operations;
5053         inode->i_fop = &simple_dir_operations;
5054         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5055         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5056
5057         return inode;
5058 }
5059
5060 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5061 {
5062         struct inode *inode;
5063         struct btrfs_root *root = BTRFS_I(dir)->root;
5064         struct btrfs_root *sub_root = root;
5065         struct btrfs_key location;
5066         int index;
5067         int ret = 0;
5068
5069         if (dentry->d_name.len > BTRFS_NAME_LEN)
5070                 return ERR_PTR(-ENAMETOOLONG);
5071
5072         ret = btrfs_inode_by_name(dir, dentry, &location);
5073         if (ret < 0)
5074                 return ERR_PTR(ret);
5075
5076         if (location.objectid == 0)
5077                 return NULL;
5078
5079         if (location.type == BTRFS_INODE_ITEM_KEY) {
5080                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5081                 return inode;
5082         }
5083
5084         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5085
5086         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5087         ret = fixup_tree_root_location(root, dir, dentry,
5088                                        &location, &sub_root);
5089         if (ret < 0) {
5090                 if (ret != -ENOENT)
5091                         inode = ERR_PTR(ret);
5092                 else
5093                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5094         } else {
5095                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5096         }
5097         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5098
5099         if (!IS_ERR(inode) && root != sub_root) {
5100                 down_read(&root->fs_info->cleanup_work_sem);
5101                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5102                         ret = btrfs_orphan_cleanup(sub_root);
5103                 up_read(&root->fs_info->cleanup_work_sem);
5104                 if (ret)
5105                         inode = ERR_PTR(ret);
5106         }
5107
5108         return inode;
5109 }
5110
5111 static int btrfs_dentry_delete(const struct dentry *dentry)
5112 {
5113         struct btrfs_root *root;
5114         struct inode *inode = dentry->d_inode;
5115
5116         if (!inode && !IS_ROOT(dentry))
5117                 inode = dentry->d_parent->d_inode;
5118
5119         if (inode) {
5120                 root = BTRFS_I(inode)->root;
5121                 if (btrfs_root_refs(&root->root_item) == 0)
5122                         return 1;
5123
5124                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5125                         return 1;
5126         }
5127         return 0;
5128 }
5129
5130 static void btrfs_dentry_release(struct dentry *dentry)
5131 {
5132         if (dentry->d_fsdata)
5133                 kfree(dentry->d_fsdata);
5134 }
5135
5136 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5137                                    unsigned int flags)
5138 {
5139         struct dentry *ret;
5140
5141         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5142         return ret;
5143 }
5144
5145 unsigned char btrfs_filetype_table[] = {
5146         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5147 };
5148
5149 static int btrfs_real_readdir(struct file *filp, void *dirent,
5150                               filldir_t filldir)
5151 {
5152         struct inode *inode = file_inode(filp);
5153         struct btrfs_root *root = BTRFS_I(inode)->root;
5154         struct btrfs_item *item;
5155         struct btrfs_dir_item *di;
5156         struct btrfs_key key;
5157         struct btrfs_key found_key;
5158         struct btrfs_path *path;
5159         struct list_head ins_list;
5160         struct list_head del_list;
5161         int ret;
5162         struct extent_buffer *leaf;
5163         int slot;
5164         unsigned char d_type;
5165         int over = 0;
5166         u32 di_cur;
5167         u32 di_total;
5168         u32 di_len;
5169         int key_type = BTRFS_DIR_INDEX_KEY;
5170         char tmp_name[32];
5171         char *name_ptr;
5172         int name_len;
5173         int is_curr = 0;        /* filp->f_pos points to the current index? */
5174
5175         /* FIXME, use a real flag for deciding about the key type */
5176         if (root->fs_info->tree_root == root)
5177                 key_type = BTRFS_DIR_ITEM_KEY;
5178
5179         /* special case for "." */
5180         if (filp->f_pos == 0) {
5181                 over = filldir(dirent, ".", 1,
5182                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5183                 if (over)
5184                         return 0;
5185                 filp->f_pos = 1;
5186         }
5187         /* special case for .., just use the back ref */
5188         if (filp->f_pos == 1) {
5189                 u64 pino = parent_ino(filp->f_path.dentry);
5190                 over = filldir(dirent, "..", 2,
5191                                filp->f_pos, pino, DT_DIR);
5192                 if (over)
5193                         return 0;
5194                 filp->f_pos = 2;
5195         }
5196         path = btrfs_alloc_path();
5197         if (!path)
5198                 return -ENOMEM;
5199
5200         path->reada = 1;
5201
5202         if (key_type == BTRFS_DIR_INDEX_KEY) {
5203                 INIT_LIST_HEAD(&ins_list);
5204                 INIT_LIST_HEAD(&del_list);
5205                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5206         }
5207
5208         btrfs_set_key_type(&key, key_type);
5209         key.offset = filp->f_pos;
5210         key.objectid = btrfs_ino(inode);
5211
5212         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5213         if (ret < 0)
5214                 goto err;
5215
5216         while (1) {
5217                 leaf = path->nodes[0];
5218                 slot = path->slots[0];
5219                 if (slot >= btrfs_header_nritems(leaf)) {
5220                         ret = btrfs_next_leaf(root, path);
5221                         if (ret < 0)
5222                                 goto err;
5223                         else if (ret > 0)
5224                                 break;
5225                         continue;
5226                 }
5227
5228                 item = btrfs_item_nr(leaf, slot);
5229                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5230
5231                 if (found_key.objectid != key.objectid)
5232                         break;
5233                 if (btrfs_key_type(&found_key) != key_type)
5234                         break;
5235                 if (found_key.offset < filp->f_pos)
5236                         goto next;
5237                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5238                     btrfs_should_delete_dir_index(&del_list,
5239                                                   found_key.offset))
5240                         goto next;
5241
5242                 filp->f_pos = found_key.offset;
5243                 is_curr = 1;
5244
5245                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5246                 di_cur = 0;
5247                 di_total = btrfs_item_size(leaf, item);
5248
5249                 while (di_cur < di_total) {
5250                         struct btrfs_key location;
5251
5252                         if (verify_dir_item(root, leaf, di))
5253                                 break;
5254
5255                         name_len = btrfs_dir_name_len(leaf, di);
5256                         if (name_len <= sizeof(tmp_name)) {
5257                                 name_ptr = tmp_name;
5258                         } else {
5259                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5260                                 if (!name_ptr) {
5261                                         ret = -ENOMEM;
5262                                         goto err;
5263                                 }
5264                         }
5265                         read_extent_buffer(leaf, name_ptr,
5266                                            (unsigned long)(di + 1), name_len);
5267
5268                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5269                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5270
5271
5272                         /* is this a reference to our own snapshot? If so
5273                          * skip it.
5274                          *
5275                          * In contrast to old kernels, we insert the snapshot's
5276                          * dir item and dir index after it has been created, so
5277                          * we won't find a reference to our own snapshot. We
5278                          * still keep the following code for backward
5279                          * compatibility.
5280                          */
5281                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5282                             location.objectid == root->root_key.objectid) {
5283                                 over = 0;
5284                                 goto skip;
5285                         }
5286                         over = filldir(dirent, name_ptr, name_len,
5287                                        found_key.offset, location.objectid,
5288                                        d_type);
5289
5290 skip:
5291                         if (name_ptr != tmp_name)
5292                                 kfree(name_ptr);
5293
5294                         if (over)
5295                                 goto nopos;
5296                         di_len = btrfs_dir_name_len(leaf, di) +
5297                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5298                         di_cur += di_len;
5299                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5300                 }
5301 next:
5302                 path->slots[0]++;
5303         }
5304
5305         if (key_type == BTRFS_DIR_INDEX_KEY) {
5306                 if (is_curr)
5307                         filp->f_pos++;
5308                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5309                                                       &ins_list);
5310                 if (ret)
5311                         goto nopos;
5312         }
5313
5314         /* Reached end of directory/root. Bump pos past the last item. */
5315         if (key_type == BTRFS_DIR_INDEX_KEY)
5316                 /*
5317                  * 32-bit glibc will use getdents64, but then strtol -
5318                  * so the last number we can serve is this.
5319                  */
5320                 filp->f_pos = 0x7fffffff;
5321         else
5322                 filp->f_pos++;
5323 nopos:
5324         ret = 0;
5325 err:
5326         if (key_type == BTRFS_DIR_INDEX_KEY)
5327                 btrfs_put_delayed_items(&ins_list, &del_list);
5328         btrfs_free_path(path);
5329         return ret;
5330 }
5331
5332 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5333 {
5334         struct btrfs_root *root = BTRFS_I(inode)->root;
5335         struct btrfs_trans_handle *trans;
5336         int ret = 0;
5337         bool nolock = false;
5338
5339         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5340                 return 0;
5341
5342         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5343                 nolock = true;
5344
5345         if (wbc->sync_mode == WB_SYNC_ALL) {
5346                 if (nolock)
5347                         trans = btrfs_join_transaction_nolock(root);
5348                 else
5349                         trans = btrfs_join_transaction(root);
5350                 if (IS_ERR(trans))
5351                         return PTR_ERR(trans);
5352                 ret = btrfs_commit_transaction(trans, root);
5353         }
5354         return ret;
5355 }
5356
5357 /*
5358  * This is somewhat expensive, updating the tree every time the
5359  * inode changes.  But, it is most likely to find the inode in cache.
5360  * FIXME, needs more benchmarking...there are no reasons other than performance
5361  * to keep or drop this code.
5362  */
5363 static int btrfs_dirty_inode(struct inode *inode)
5364 {
5365         struct btrfs_root *root = BTRFS_I(inode)->root;
5366         struct btrfs_trans_handle *trans;
5367         int ret;
5368
5369         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5370                 return 0;
5371
5372         trans = btrfs_join_transaction(root);
5373         if (IS_ERR(trans))
5374                 return PTR_ERR(trans);
5375
5376         ret = btrfs_update_inode(trans, root, inode);
5377         if (ret && ret == -ENOSPC) {
5378                 /* whoops, lets try again with the full transaction */
5379                 btrfs_end_transaction(trans, root);
5380                 trans = btrfs_start_transaction(root, 1);
5381                 if (IS_ERR(trans))
5382                         return PTR_ERR(trans);
5383
5384                 ret = btrfs_update_inode(trans, root, inode);
5385         }
5386         btrfs_end_transaction(trans, root);
5387         if (BTRFS_I(inode)->delayed_node)
5388                 btrfs_balance_delayed_items(root);
5389
5390         return ret;
5391 }
5392
5393 /*
5394  * This is a copy of file_update_time.  We need this so we can return error on
5395  * ENOSPC for updating the inode in the case of file write and mmap writes.
5396  */
5397 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5398                              int flags)
5399 {
5400         struct btrfs_root *root = BTRFS_I(inode)->root;
5401
5402         if (btrfs_root_readonly(root))
5403                 return -EROFS;
5404
5405         if (flags & S_VERSION)
5406                 inode_inc_iversion(inode);
5407         if (flags & S_CTIME)
5408                 inode->i_ctime = *now;
5409         if (flags & S_MTIME)
5410                 inode->i_mtime = *now;
5411         if (flags & S_ATIME)
5412                 inode->i_atime = *now;
5413         return btrfs_dirty_inode(inode);
5414 }
5415
5416 /*
5417  * find the highest existing sequence number in a directory
5418  * and then set the in-memory index_cnt variable to reflect
5419  * free sequence numbers
5420  */
5421 static int btrfs_set_inode_index_count(struct inode *inode)
5422 {
5423         struct btrfs_root *root = BTRFS_I(inode)->root;
5424         struct btrfs_key key, found_key;
5425         struct btrfs_path *path;
5426         struct extent_buffer *leaf;
5427         int ret;
5428
5429         key.objectid = btrfs_ino(inode);
5430         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5431         key.offset = (u64)-1;
5432
5433         path = btrfs_alloc_path();
5434         if (!path)
5435                 return -ENOMEM;
5436
5437         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5438         if (ret < 0)
5439                 goto out;
5440         /* FIXME: we should be able to handle this */
5441         if (ret == 0)
5442                 goto out;
5443         ret = 0;
5444
5445         /*
5446          * MAGIC NUMBER EXPLANATION:
5447          * since we search a directory based on f_pos we have to start at 2
5448          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5449          * else has to start at 2
5450          */
5451         if (path->slots[0] == 0) {
5452                 BTRFS_I(inode)->index_cnt = 2;
5453                 goto out;
5454         }
5455
5456         path->slots[0]--;
5457
5458         leaf = path->nodes[0];
5459         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5460
5461         if (found_key.objectid != btrfs_ino(inode) ||
5462             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5463                 BTRFS_I(inode)->index_cnt = 2;
5464                 goto out;
5465         }
5466
5467         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5468 out:
5469         btrfs_free_path(path);
5470         return ret;
5471 }
5472
5473 /*
5474  * helper to find a free sequence number in a given directory.  This current
5475  * code is very simple, later versions will do smarter things in the btree
5476  */
5477 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5478 {
5479         int ret = 0;
5480
5481         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5482                 ret = btrfs_inode_delayed_dir_index_count(dir);
5483                 if (ret) {
5484                         ret = btrfs_set_inode_index_count(dir);
5485                         if (ret)
5486                                 return ret;
5487                 }
5488         }
5489
5490         *index = BTRFS_I(dir)->index_cnt;
5491         BTRFS_I(dir)->index_cnt++;
5492
5493         return ret;
5494 }
5495
5496 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5497                                      struct btrfs_root *root,
5498                                      struct inode *dir,
5499                                      const char *name, int name_len,
5500                                      u64 ref_objectid, u64 objectid,
5501                                      umode_t mode, u64 *index)
5502 {
5503         struct inode *inode;
5504         struct btrfs_inode_item *inode_item;
5505         struct btrfs_key *location;
5506         struct btrfs_path *path;
5507         struct btrfs_inode_ref *ref;
5508         struct btrfs_key key[2];
5509         u32 sizes[2];
5510         unsigned long ptr;
5511         int ret;
5512         int owner;
5513
5514         path = btrfs_alloc_path();
5515         if (!path)
5516                 return ERR_PTR(-ENOMEM);
5517
5518         inode = new_inode(root->fs_info->sb);
5519         if (!inode) {
5520                 btrfs_free_path(path);
5521                 return ERR_PTR(-ENOMEM);
5522         }
5523
5524         /*
5525          * we have to initialize this early, so we can reclaim the inode
5526          * number if we fail afterwards in this function.
5527          */
5528         inode->i_ino = objectid;
5529
5530         if (dir) {
5531                 trace_btrfs_inode_request(dir);
5532
5533                 ret = btrfs_set_inode_index(dir, index);
5534                 if (ret) {
5535                         btrfs_free_path(path);
5536                         iput(inode);
5537                         return ERR_PTR(ret);
5538                 }
5539         }
5540         /*
5541          * index_cnt is ignored for everything but a dir,
5542          * btrfs_get_inode_index_count has an explanation for the magic
5543          * number
5544          */
5545         BTRFS_I(inode)->index_cnt = 2;
5546         BTRFS_I(inode)->root = root;
5547         BTRFS_I(inode)->generation = trans->transid;
5548         inode->i_generation = BTRFS_I(inode)->generation;
5549
5550         /*
5551          * We could have gotten an inode number from somebody who was fsynced
5552          * and then removed in this same transaction, so let's just set full
5553          * sync since it will be a full sync anyway and this will blow away the
5554          * old info in the log.
5555          */
5556         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5557
5558         if (S_ISDIR(mode))
5559                 owner = 0;
5560         else
5561                 owner = 1;
5562
5563         key[0].objectid = objectid;
5564         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5565         key[0].offset = 0;
5566
5567         /*
5568          * Start new inodes with an inode_ref. This is slightly more
5569          * efficient for small numbers of hard links since they will
5570          * be packed into one item. Extended refs will kick in if we
5571          * add more hard links than can fit in the ref item.
5572          */
5573         key[1].objectid = objectid;
5574         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5575         key[1].offset = ref_objectid;
5576
5577         sizes[0] = sizeof(struct btrfs_inode_item);
5578         sizes[1] = name_len + sizeof(*ref);
5579
5580         path->leave_spinning = 1;
5581         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5582         if (ret != 0)
5583                 goto fail;
5584
5585         inode_init_owner(inode, dir, mode);
5586         inode_set_bytes(inode, 0);
5587         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5588         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5589                                   struct btrfs_inode_item);
5590         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5591                              sizeof(*inode_item));
5592         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5593
5594         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5595                              struct btrfs_inode_ref);
5596         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5597         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5598         ptr = (unsigned long)(ref + 1);
5599         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5600
5601         btrfs_mark_buffer_dirty(path->nodes[0]);
5602         btrfs_free_path(path);
5603
5604         location = &BTRFS_I(inode)->location;
5605         location->objectid = objectid;
5606         location->offset = 0;
5607         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5608
5609         btrfs_inherit_iflags(inode, dir);
5610
5611         if (S_ISREG(mode)) {
5612                 if (btrfs_test_opt(root, NODATASUM))
5613                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5614                 if (btrfs_test_opt(root, NODATACOW))
5615                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5616                                 BTRFS_INODE_NODATASUM;
5617         }
5618
5619         insert_inode_hash(inode);
5620         inode_tree_add(inode);
5621
5622         trace_btrfs_inode_new(inode);
5623         btrfs_set_inode_last_trans(trans, inode);
5624
5625         btrfs_update_root_times(trans, root);
5626
5627         return inode;
5628 fail:
5629         if (dir)
5630                 BTRFS_I(dir)->index_cnt--;
5631         btrfs_free_path(path);
5632         iput(inode);
5633         return ERR_PTR(ret);
5634 }
5635
5636 static inline u8 btrfs_inode_type(struct inode *inode)
5637 {
5638         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5639 }
5640
5641 /*
5642  * utility function to add 'inode' into 'parent_inode' with
5643  * a give name and a given sequence number.
5644  * if 'add_backref' is true, also insert a backref from the
5645  * inode to the parent directory.
5646  */
5647 int btrfs_add_link(struct btrfs_trans_handle *trans,
5648                    struct inode *parent_inode, struct inode *inode,
5649                    const char *name, int name_len, int add_backref, u64 index)
5650 {
5651         int ret = 0;
5652         struct btrfs_key key;
5653         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5654         u64 ino = btrfs_ino(inode);
5655         u64 parent_ino = btrfs_ino(parent_inode);
5656
5657         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5658                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5659         } else {
5660                 key.objectid = ino;
5661                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5662                 key.offset = 0;
5663         }
5664
5665         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5666                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5667                                          key.objectid, root->root_key.objectid,
5668                                          parent_ino, index, name, name_len);
5669         } else if (add_backref) {
5670                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5671                                              parent_ino, index);
5672         }
5673
5674         /* Nothing to clean up yet */
5675         if (ret)
5676                 return ret;
5677
5678         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5679                                     parent_inode, &key,
5680                                     btrfs_inode_type(inode), index);
5681         if (ret == -EEXIST || ret == -EOVERFLOW)
5682                 goto fail_dir_item;
5683         else if (ret) {
5684                 btrfs_abort_transaction(trans, root, ret);
5685                 return ret;
5686         }
5687
5688         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5689                            name_len * 2);
5690         inode_inc_iversion(parent_inode);
5691         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5692         ret = btrfs_update_inode(trans, root, parent_inode);
5693         if (ret)
5694                 btrfs_abort_transaction(trans, root, ret);
5695         return ret;
5696
5697 fail_dir_item:
5698         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5699                 u64 local_index;
5700                 int err;
5701                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5702                                  key.objectid, root->root_key.objectid,
5703                                  parent_ino, &local_index, name, name_len);
5704
5705         } else if (add_backref) {
5706                 u64 local_index;
5707                 int err;
5708
5709                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5710                                           ino, parent_ino, &local_index);
5711         }
5712         return ret;
5713 }
5714
5715 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5716                             struct inode *dir, struct dentry *dentry,
5717                             struct inode *inode, int backref, u64 index)
5718 {
5719         int err = btrfs_add_link(trans, dir, inode,
5720                                  dentry->d_name.name, dentry->d_name.len,
5721                                  backref, index);
5722         if (err > 0)
5723                 err = -EEXIST;
5724         return err;
5725 }
5726
5727 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5728                         umode_t mode, dev_t rdev)
5729 {
5730         struct btrfs_trans_handle *trans;
5731         struct btrfs_root *root = BTRFS_I(dir)->root;
5732         struct inode *inode = NULL;
5733         int err;
5734         int drop_inode = 0;
5735         u64 objectid;
5736         u64 index = 0;
5737
5738         if (!new_valid_dev(rdev))
5739                 return -EINVAL;
5740
5741         /*
5742          * 2 for inode item and ref
5743          * 2 for dir items
5744          * 1 for xattr if selinux is on
5745          */
5746         trans = btrfs_start_transaction(root, 5);
5747         if (IS_ERR(trans))
5748                 return PTR_ERR(trans);
5749
5750         err = btrfs_find_free_ino(root, &objectid);
5751         if (err)
5752                 goto out_unlock;
5753
5754         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5755                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5756                                 mode, &index);
5757         if (IS_ERR(inode)) {
5758                 err = PTR_ERR(inode);
5759                 goto out_unlock;
5760         }
5761
5762         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5763         if (err) {
5764                 drop_inode = 1;
5765                 goto out_unlock;
5766         }
5767
5768         /*
5769         * If the active LSM wants to access the inode during
5770         * d_instantiate it needs these. Smack checks to see
5771         * if the filesystem supports xattrs by looking at the
5772         * ops vector.
5773         */
5774
5775         inode->i_op = &btrfs_special_inode_operations;
5776         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5777         if (err)
5778                 drop_inode = 1;
5779         else {
5780                 init_special_inode(inode, inode->i_mode, rdev);
5781                 btrfs_update_inode(trans, root, inode);
5782                 d_instantiate(dentry, inode);
5783         }
5784 out_unlock:
5785         btrfs_end_transaction(trans, root);
5786         btrfs_btree_balance_dirty(root);
5787         if (drop_inode) {
5788                 inode_dec_link_count(inode);
5789                 iput(inode);
5790         }
5791         return err;
5792 }
5793
5794 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5795                         umode_t mode, bool excl)
5796 {
5797         struct btrfs_trans_handle *trans;
5798         struct btrfs_root *root = BTRFS_I(dir)->root;
5799         struct inode *inode = NULL;
5800         int drop_inode_on_err = 0;
5801         int err;
5802         u64 objectid;
5803         u64 index = 0;
5804
5805         /*
5806          * 2 for inode item and ref
5807          * 2 for dir items
5808          * 1 for xattr if selinux is on
5809          */
5810         trans = btrfs_start_transaction(root, 5);
5811         if (IS_ERR(trans))
5812                 return PTR_ERR(trans);
5813
5814         err = btrfs_find_free_ino(root, &objectid);
5815         if (err)
5816                 goto out_unlock;
5817
5818         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5819                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5820                                 mode, &index);
5821         if (IS_ERR(inode)) {
5822                 err = PTR_ERR(inode);
5823                 goto out_unlock;
5824         }
5825         drop_inode_on_err = 1;
5826
5827         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5828         if (err)
5829                 goto out_unlock;
5830
5831         err = btrfs_update_inode(trans, root, inode);
5832         if (err)
5833                 goto out_unlock;
5834
5835         /*
5836         * If the active LSM wants to access the inode during
5837         * d_instantiate it needs these. Smack checks to see
5838         * if the filesystem supports xattrs by looking at the
5839         * ops vector.
5840         */
5841         inode->i_fop = &btrfs_file_operations;
5842         inode->i_op = &btrfs_file_inode_operations;
5843
5844         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5845         if (err)
5846                 goto out_unlock;
5847
5848         inode->i_mapping->a_ops = &btrfs_aops;
5849         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5850         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5851         d_instantiate(dentry, inode);
5852
5853 out_unlock:
5854         btrfs_end_transaction(trans, root);
5855         if (err && drop_inode_on_err) {
5856                 inode_dec_link_count(inode);
5857                 iput(inode);
5858         }
5859         btrfs_btree_balance_dirty(root);
5860         return err;
5861 }
5862
5863 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5864                       struct dentry *dentry)
5865 {
5866         struct btrfs_trans_handle *trans;
5867         struct btrfs_root *root = BTRFS_I(dir)->root;
5868         struct inode *inode = old_dentry->d_inode;
5869         u64 index;
5870         int err;
5871         int drop_inode = 0;
5872
5873         /* do not allow sys_link's with other subvols of the same device */
5874         if (root->objectid != BTRFS_I(inode)->root->objectid)
5875                 return -EXDEV;
5876
5877         if (inode->i_nlink >= BTRFS_LINK_MAX)
5878                 return -EMLINK;
5879
5880         err = btrfs_set_inode_index(dir, &index);
5881         if (err)
5882                 goto fail;
5883
5884         /*
5885          * 2 items for inode and inode ref
5886          * 2 items for dir items
5887          * 1 item for parent inode
5888          */
5889         trans = btrfs_start_transaction(root, 5);
5890         if (IS_ERR(trans)) {
5891                 err = PTR_ERR(trans);
5892                 goto fail;
5893         }
5894
5895         btrfs_inc_nlink(inode);
5896         inode_inc_iversion(inode);
5897         inode->i_ctime = CURRENT_TIME;
5898         ihold(inode);
5899         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5900
5901         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5902
5903         if (err) {
5904                 drop_inode = 1;
5905         } else {
5906                 struct dentry *parent = dentry->d_parent;
5907                 err = btrfs_update_inode(trans, root, inode);
5908                 if (err)
5909                         goto fail;
5910                 d_instantiate(dentry, inode);
5911                 btrfs_log_new_name(trans, inode, NULL, parent);
5912         }
5913
5914         btrfs_end_transaction(trans, root);
5915 fail:
5916         if (drop_inode) {
5917                 inode_dec_link_count(inode);
5918                 iput(inode);
5919         }
5920         btrfs_btree_balance_dirty(root);
5921         return err;
5922 }
5923
5924 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5925 {
5926         struct inode *inode = NULL;
5927         struct btrfs_trans_handle *trans;
5928         struct btrfs_root *root = BTRFS_I(dir)->root;
5929         int err = 0;
5930         int drop_on_err = 0;
5931         u64 objectid = 0;
5932         u64 index = 0;
5933
5934         /*
5935          * 2 items for inode and ref
5936          * 2 items for dir items
5937          * 1 for xattr if selinux is on
5938          */
5939         trans = btrfs_start_transaction(root, 5);
5940         if (IS_ERR(trans))
5941                 return PTR_ERR(trans);
5942
5943         err = btrfs_find_free_ino(root, &objectid);
5944         if (err)
5945                 goto out_fail;
5946
5947         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5948                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5949                                 S_IFDIR | mode, &index);
5950         if (IS_ERR(inode)) {
5951                 err = PTR_ERR(inode);
5952                 goto out_fail;
5953         }
5954
5955         drop_on_err = 1;
5956
5957         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5958         if (err)
5959                 goto out_fail;
5960
5961         inode->i_op = &btrfs_dir_inode_operations;
5962         inode->i_fop = &btrfs_dir_file_operations;
5963
5964         btrfs_i_size_write(inode, 0);
5965         err = btrfs_update_inode(trans, root, inode);
5966         if (err)
5967                 goto out_fail;
5968
5969         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5970                              dentry->d_name.len, 0, index);
5971         if (err)
5972                 goto out_fail;
5973
5974         d_instantiate(dentry, inode);
5975         drop_on_err = 0;
5976
5977 out_fail:
5978         btrfs_end_transaction(trans, root);
5979         if (drop_on_err)
5980                 iput(inode);
5981         btrfs_btree_balance_dirty(root);
5982         return err;
5983 }
5984
5985 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5986  * and an extent that you want to insert, deal with overlap and insert
5987  * the new extent into the tree.
5988  */
5989 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5990                                 struct extent_map *existing,
5991                                 struct extent_map *em,
5992                                 u64 map_start, u64 map_len)
5993 {
5994         u64 start_diff;
5995
5996         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5997         start_diff = map_start - em->start;
5998         em->start = map_start;
5999         em->len = map_len;
6000         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6001             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6002                 em->block_start += start_diff;
6003                 em->block_len -= start_diff;
6004         }
6005         return add_extent_mapping(em_tree, em, 0);
6006 }
6007
6008 static noinline int uncompress_inline(struct btrfs_path *path,
6009                                       struct inode *inode, struct page *page,
6010                                       size_t pg_offset, u64 extent_offset,
6011                                       struct btrfs_file_extent_item *item)
6012 {
6013         int ret;
6014         struct extent_buffer *leaf = path->nodes[0];
6015         char *tmp;
6016         size_t max_size;
6017         unsigned long inline_size;
6018         unsigned long ptr;
6019         int compress_type;
6020
6021         WARN_ON(pg_offset != 0);
6022         compress_type = btrfs_file_extent_compression(leaf, item);
6023         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6024         inline_size = btrfs_file_extent_inline_item_len(leaf,
6025                                         btrfs_item_nr(leaf, path->slots[0]));
6026         tmp = kmalloc(inline_size, GFP_NOFS);
6027         if (!tmp)
6028                 return -ENOMEM;
6029         ptr = btrfs_file_extent_inline_start(item);
6030
6031         read_extent_buffer(leaf, tmp, ptr, inline_size);
6032
6033         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6034         ret = btrfs_decompress(compress_type, tmp, page,
6035                                extent_offset, inline_size, max_size);
6036         if (ret) {
6037                 char *kaddr = kmap_atomic(page);
6038                 unsigned long copy_size = min_t(u64,
6039                                   PAGE_CACHE_SIZE - pg_offset,
6040                                   max_size - extent_offset);
6041                 memset(kaddr + pg_offset, 0, copy_size);
6042                 kunmap_atomic(kaddr);
6043         }
6044         kfree(tmp);
6045         return 0;
6046 }
6047
6048 /*
6049  * a bit scary, this does extent mapping from logical file offset to the disk.
6050  * the ugly parts come from merging extents from the disk with the in-ram
6051  * representation.  This gets more complex because of the data=ordered code,
6052  * where the in-ram extents might be locked pending data=ordered completion.
6053  *
6054  * This also copies inline extents directly into the page.
6055  */
6056
6057 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6058                                     size_t pg_offset, u64 start, u64 len,
6059                                     int create)
6060 {
6061         int ret;
6062         int err = 0;
6063         u64 bytenr;
6064         u64 extent_start = 0;
6065         u64 extent_end = 0;
6066         u64 objectid = btrfs_ino(inode);
6067         u32 found_type;
6068         struct btrfs_path *path = NULL;
6069         struct btrfs_root *root = BTRFS_I(inode)->root;
6070         struct btrfs_file_extent_item *item;
6071         struct extent_buffer *leaf;
6072         struct btrfs_key found_key;
6073         struct extent_map *em = NULL;
6074         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6075         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6076         struct btrfs_trans_handle *trans = NULL;
6077         int compress_type;
6078
6079 again:
6080         read_lock(&em_tree->lock);
6081         em = lookup_extent_mapping(em_tree, start, len);
6082         if (em)
6083                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6084         read_unlock(&em_tree->lock);
6085
6086         if (em) {
6087                 if (em->start > start || em->start + em->len <= start)
6088                         free_extent_map(em);
6089                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6090                         free_extent_map(em);
6091                 else
6092                         goto out;
6093         }
6094         em = alloc_extent_map();
6095         if (!em) {
6096                 err = -ENOMEM;
6097                 goto out;
6098         }
6099         em->bdev = root->fs_info->fs_devices->latest_bdev;
6100         em->start = EXTENT_MAP_HOLE;
6101         em->orig_start = EXTENT_MAP_HOLE;
6102         em->len = (u64)-1;
6103         em->block_len = (u64)-1;
6104
6105         if (!path) {
6106                 path = btrfs_alloc_path();
6107                 if (!path) {
6108                         err = -ENOMEM;
6109                         goto out;
6110                 }
6111                 /*
6112                  * Chances are we'll be called again, so go ahead and do
6113                  * readahead
6114                  */
6115                 path->reada = 1;
6116         }
6117
6118         ret = btrfs_lookup_file_extent(trans, root, path,
6119                                        objectid, start, trans != NULL);
6120         if (ret < 0) {
6121                 err = ret;
6122                 goto out;
6123         }
6124
6125         if (ret != 0) {
6126                 if (path->slots[0] == 0)
6127                         goto not_found;
6128                 path->slots[0]--;
6129         }
6130
6131         leaf = path->nodes[0];
6132         item = btrfs_item_ptr(leaf, path->slots[0],
6133                               struct btrfs_file_extent_item);
6134         /* are we inside the extent that was found? */
6135         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6136         found_type = btrfs_key_type(&found_key);
6137         if (found_key.objectid != objectid ||
6138             found_type != BTRFS_EXTENT_DATA_KEY) {
6139                 goto not_found;
6140         }
6141
6142         found_type = btrfs_file_extent_type(leaf, item);
6143         extent_start = found_key.offset;
6144         compress_type = btrfs_file_extent_compression(leaf, item);
6145         if (found_type == BTRFS_FILE_EXTENT_REG ||
6146             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6147                 extent_end = extent_start +
6148                        btrfs_file_extent_num_bytes(leaf, item);
6149         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6150                 size_t size;
6151                 size = btrfs_file_extent_inline_len(leaf, item);
6152                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6153         }
6154
6155         if (start >= extent_end) {
6156                 path->slots[0]++;
6157                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6158                         ret = btrfs_next_leaf(root, path);
6159                         if (ret < 0) {
6160                                 err = ret;
6161                                 goto out;
6162                         }
6163                         if (ret > 0)
6164                                 goto not_found;
6165                         leaf = path->nodes[0];
6166                 }
6167                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6168                 if (found_key.objectid != objectid ||
6169                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6170                         goto not_found;
6171                 if (start + len <= found_key.offset)
6172                         goto not_found;
6173                 em->start = start;
6174                 em->orig_start = start;
6175                 em->len = found_key.offset - start;
6176                 goto not_found_em;
6177         }
6178
6179         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6180         if (found_type == BTRFS_FILE_EXTENT_REG ||
6181             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6182                 em->start = extent_start;
6183                 em->len = extent_end - extent_start;
6184                 em->orig_start = extent_start -
6185                                  btrfs_file_extent_offset(leaf, item);
6186                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6187                                                                       item);
6188                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6189                 if (bytenr == 0) {
6190                         em->block_start = EXTENT_MAP_HOLE;
6191                         goto insert;
6192                 }
6193                 if (compress_type != BTRFS_COMPRESS_NONE) {
6194                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6195                         em->compress_type = compress_type;
6196                         em->block_start = bytenr;
6197                         em->block_len = em->orig_block_len;
6198                 } else {
6199                         bytenr += btrfs_file_extent_offset(leaf, item);
6200                         em->block_start = bytenr;
6201                         em->block_len = em->len;
6202                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6203                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6204                 }
6205                 goto insert;
6206         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6207                 unsigned long ptr;
6208                 char *map;
6209                 size_t size;
6210                 size_t extent_offset;
6211                 size_t copy_size;
6212
6213                 em->block_start = EXTENT_MAP_INLINE;
6214                 if (!page || create) {
6215                         em->start = extent_start;
6216                         em->len = extent_end - extent_start;
6217                         goto out;
6218                 }
6219
6220                 size = btrfs_file_extent_inline_len(leaf, item);
6221                 extent_offset = page_offset(page) + pg_offset - extent_start;
6222                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6223                                 size - extent_offset);
6224                 em->start = extent_start + extent_offset;
6225                 em->len = ALIGN(copy_size, root->sectorsize);
6226                 em->orig_block_len = em->len;
6227                 em->orig_start = em->start;
6228                 if (compress_type) {
6229                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6230                         em->compress_type = compress_type;
6231                 }
6232                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6233                 if (create == 0 && !PageUptodate(page)) {
6234                         if (btrfs_file_extent_compression(leaf, item) !=
6235                             BTRFS_COMPRESS_NONE) {
6236                                 ret = uncompress_inline(path, inode, page,
6237                                                         pg_offset,
6238                                                         extent_offset, item);
6239                                 BUG_ON(ret); /* -ENOMEM */
6240                         } else {
6241                                 map = kmap(page);
6242                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6243                                                    copy_size);
6244                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6245                                         memset(map + pg_offset + copy_size, 0,
6246                                                PAGE_CACHE_SIZE - pg_offset -
6247                                                copy_size);
6248                                 }
6249                                 kunmap(page);
6250                         }
6251                         flush_dcache_page(page);
6252                 } else if (create && PageUptodate(page)) {
6253                         BUG();
6254                         if (!trans) {
6255                                 kunmap(page);
6256                                 free_extent_map(em);
6257                                 em = NULL;
6258
6259                                 btrfs_release_path(path);
6260                                 trans = btrfs_join_transaction(root);
6261
6262                                 if (IS_ERR(trans))
6263                                         return ERR_CAST(trans);
6264                                 goto again;
6265                         }
6266                         map = kmap(page);
6267                         write_extent_buffer(leaf, map + pg_offset, ptr,
6268                                             copy_size);
6269                         kunmap(page);
6270                         btrfs_mark_buffer_dirty(leaf);
6271                 }
6272                 set_extent_uptodate(io_tree, em->start,
6273                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6274                 goto insert;
6275         } else {
6276                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6277         }
6278 not_found:
6279         em->start = start;
6280         em->orig_start = start;
6281         em->len = len;
6282 not_found_em:
6283         em->block_start = EXTENT_MAP_HOLE;
6284         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6285 insert:
6286         btrfs_release_path(path);
6287         if (em->start > start || extent_map_end(em) <= start) {
6288                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6289                         (unsigned long long)em->start,
6290                         (unsigned long long)em->len,
6291                         (unsigned long long)start,
6292                         (unsigned long long)len);
6293                 err = -EIO;
6294                 goto out;
6295         }
6296
6297         err = 0;
6298         write_lock(&em_tree->lock);
6299         ret = add_extent_mapping(em_tree, em, 0);
6300         /* it is possible that someone inserted the extent into the tree
6301          * while we had the lock dropped.  It is also possible that
6302          * an overlapping map exists in the tree
6303          */
6304         if (ret == -EEXIST) {
6305                 struct extent_map *existing;
6306
6307                 ret = 0;
6308
6309                 existing = lookup_extent_mapping(em_tree, start, len);
6310                 if (existing && (existing->start > start ||
6311                     existing->start + existing->len <= start)) {
6312                         free_extent_map(existing);
6313                         existing = NULL;
6314                 }
6315                 if (!existing) {
6316                         existing = lookup_extent_mapping(em_tree, em->start,
6317                                                          em->len);
6318                         if (existing) {
6319                                 err = merge_extent_mapping(em_tree, existing,
6320                                                            em, start,
6321                                                            root->sectorsize);
6322                                 free_extent_map(existing);
6323                                 if (err) {
6324                                         free_extent_map(em);
6325                                         em = NULL;
6326                                 }
6327                         } else {
6328                                 err = -EIO;
6329                                 free_extent_map(em);
6330                                 em = NULL;
6331                         }
6332                 } else {
6333                         free_extent_map(em);
6334                         em = existing;
6335                         err = 0;
6336                 }
6337         }
6338         write_unlock(&em_tree->lock);
6339 out:
6340
6341         if (em)
6342                 trace_btrfs_get_extent(root, em);
6343
6344         if (path)
6345                 btrfs_free_path(path);
6346         if (trans) {
6347                 ret = btrfs_end_transaction(trans, root);
6348                 if (!err)
6349                         err = ret;
6350         }
6351         if (err) {
6352                 free_extent_map(em);
6353                 return ERR_PTR(err);
6354         }
6355         BUG_ON(!em); /* Error is always set */
6356         return em;
6357 }
6358
6359 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6360                                            size_t pg_offset, u64 start, u64 len,
6361                                            int create)
6362 {
6363         struct extent_map *em;
6364         struct extent_map *hole_em = NULL;
6365         u64 range_start = start;
6366         u64 end;
6367         u64 found;
6368         u64 found_end;
6369         int err = 0;
6370
6371         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6372         if (IS_ERR(em))
6373                 return em;
6374         if (em) {
6375                 /*
6376                  * if our em maps to
6377                  * -  a hole or
6378                  * -  a pre-alloc extent,
6379                  * there might actually be delalloc bytes behind it.
6380                  */
6381                 if (em->block_start != EXTENT_MAP_HOLE &&
6382                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6383                         return em;
6384                 else
6385                         hole_em = em;
6386         }
6387
6388         /* check to see if we've wrapped (len == -1 or similar) */
6389         end = start + len;
6390         if (end < start)
6391                 end = (u64)-1;
6392         else
6393                 end -= 1;
6394
6395         em = NULL;
6396
6397         /* ok, we didn't find anything, lets look for delalloc */
6398         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6399                                  end, len, EXTENT_DELALLOC, 1);
6400         found_end = range_start + found;
6401         if (found_end < range_start)
6402                 found_end = (u64)-1;
6403
6404         /*
6405          * we didn't find anything useful, return
6406          * the original results from get_extent()
6407          */
6408         if (range_start > end || found_end <= start) {
6409                 em = hole_em;
6410                 hole_em = NULL;
6411                 goto out;
6412         }
6413
6414         /* adjust the range_start to make sure it doesn't
6415          * go backwards from the start they passed in
6416          */
6417         range_start = max(start,range_start);
6418         found = found_end - range_start;
6419
6420         if (found > 0) {
6421                 u64 hole_start = start;
6422                 u64 hole_len = len;
6423
6424                 em = alloc_extent_map();
6425                 if (!em) {
6426                         err = -ENOMEM;
6427                         goto out;
6428                 }
6429                 /*
6430                  * when btrfs_get_extent can't find anything it
6431                  * returns one huge hole
6432                  *
6433                  * make sure what it found really fits our range, and
6434                  * adjust to make sure it is based on the start from
6435                  * the caller
6436                  */
6437                 if (hole_em) {
6438                         u64 calc_end = extent_map_end(hole_em);
6439
6440                         if (calc_end <= start || (hole_em->start > end)) {
6441                                 free_extent_map(hole_em);
6442                                 hole_em = NULL;
6443                         } else {
6444                                 hole_start = max(hole_em->start, start);
6445                                 hole_len = calc_end - hole_start;
6446                         }
6447                 }
6448                 em->bdev = NULL;
6449                 if (hole_em && range_start > hole_start) {
6450                         /* our hole starts before our delalloc, so we
6451                          * have to return just the parts of the hole
6452                          * that go until  the delalloc starts
6453                          */
6454                         em->len = min(hole_len,
6455                                       range_start - hole_start);
6456                         em->start = hole_start;
6457                         em->orig_start = hole_start;
6458                         /*
6459                          * don't adjust block start at all,
6460                          * it is fixed at EXTENT_MAP_HOLE
6461                          */
6462                         em->block_start = hole_em->block_start;
6463                         em->block_len = hole_len;
6464                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6465                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6466                 } else {
6467                         em->start = range_start;
6468                         em->len = found;
6469                         em->orig_start = range_start;
6470                         em->block_start = EXTENT_MAP_DELALLOC;
6471                         em->block_len = found;
6472                 }
6473         } else if (hole_em) {
6474                 return hole_em;
6475         }
6476 out:
6477
6478         free_extent_map(hole_em);
6479         if (err) {
6480                 free_extent_map(em);
6481                 return ERR_PTR(err);
6482         }
6483         return em;
6484 }
6485
6486 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6487                                                   u64 start, u64 len)
6488 {
6489         struct btrfs_root *root = BTRFS_I(inode)->root;
6490         struct btrfs_trans_handle *trans;
6491         struct extent_map *em;
6492         struct btrfs_key ins;
6493         u64 alloc_hint;
6494         int ret;
6495
6496         trans = btrfs_join_transaction(root);
6497         if (IS_ERR(trans))
6498                 return ERR_CAST(trans);
6499
6500         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6501
6502         alloc_hint = get_extent_allocation_hint(inode, start, len);
6503         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6504                                    alloc_hint, &ins, 1);
6505         if (ret) {
6506                 em = ERR_PTR(ret);
6507                 goto out;
6508         }
6509
6510         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6511                               ins.offset, ins.offset, ins.offset, 0);
6512         if (IS_ERR(em))
6513                 goto out;
6514
6515         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6516                                            ins.offset, ins.offset, 0);
6517         if (ret) {
6518                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6519                 em = ERR_PTR(ret);
6520         }
6521 out:
6522         btrfs_end_transaction(trans, root);
6523         return em;
6524 }
6525
6526 /*
6527  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6528  * block must be cow'd
6529  */
6530 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6531                                       struct inode *inode, u64 offset, u64 *len,
6532                                       u64 *orig_start, u64 *orig_block_len,
6533                                       u64 *ram_bytes)
6534 {
6535         struct btrfs_path *path;
6536         int ret;
6537         struct extent_buffer *leaf;
6538         struct btrfs_root *root = BTRFS_I(inode)->root;
6539         struct btrfs_file_extent_item *fi;
6540         struct btrfs_key key;
6541         u64 disk_bytenr;
6542         u64 backref_offset;
6543         u64 extent_end;
6544         u64 num_bytes;
6545         int slot;
6546         int found_type;
6547
6548         path = btrfs_alloc_path();
6549         if (!path)
6550                 return -ENOMEM;
6551
6552         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6553                                        offset, 0);
6554         if (ret < 0)
6555                 goto out;
6556
6557         slot = path->slots[0];
6558         if (ret == 1) {
6559                 if (slot == 0) {
6560                         /* can't find the item, must cow */
6561                         ret = 0;
6562                         goto out;
6563                 }
6564                 slot--;
6565         }
6566         ret = 0;
6567         leaf = path->nodes[0];
6568         btrfs_item_key_to_cpu(leaf, &key, slot);
6569         if (key.objectid != btrfs_ino(inode) ||
6570             key.type != BTRFS_EXTENT_DATA_KEY) {
6571                 /* not our file or wrong item type, must cow */
6572                 goto out;
6573         }
6574
6575         if (key.offset > offset) {
6576                 /* Wrong offset, must cow */
6577                 goto out;
6578         }
6579
6580         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6581         found_type = btrfs_file_extent_type(leaf, fi);
6582         if (found_type != BTRFS_FILE_EXTENT_REG &&
6583             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6584                 /* not a regular extent, must cow */
6585                 goto out;
6586         }
6587         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6588         backref_offset = btrfs_file_extent_offset(leaf, fi);
6589
6590         *orig_start = key.offset - backref_offset;
6591         *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6592         *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6593
6594         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6595         if (extent_end < offset + *len) {
6596                 /* extent doesn't include our full range, must cow */
6597                 goto out;
6598         }
6599
6600         if (btrfs_extent_readonly(root, disk_bytenr))
6601                 goto out;
6602
6603         /*
6604          * look for other files referencing this extent, if we
6605          * find any we must cow
6606          */
6607         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6608                                   key.offset - backref_offset, disk_bytenr))
6609                 goto out;
6610
6611         /*
6612          * adjust disk_bytenr and num_bytes to cover just the bytes
6613          * in this extent we are about to write.  If there
6614          * are any csums in that range we have to cow in order
6615          * to keep the csums correct
6616          */
6617         disk_bytenr += backref_offset;
6618         disk_bytenr += offset - key.offset;
6619         num_bytes = min(offset + *len, extent_end) - offset;
6620         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6621                                 goto out;
6622         /*
6623          * all of the above have passed, it is safe to overwrite this extent
6624          * without cow
6625          */
6626         *len = num_bytes;
6627         ret = 1;
6628 out:
6629         btrfs_free_path(path);
6630         return ret;
6631 }
6632
6633 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6634                               struct extent_state **cached_state, int writing)
6635 {
6636         struct btrfs_ordered_extent *ordered;
6637         int ret = 0;
6638
6639         while (1) {
6640                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6641                                  0, cached_state);
6642                 /*
6643                  * We're concerned with the entire range that we're going to be
6644                  * doing DIO to, so we need to make sure theres no ordered
6645                  * extents in this range.
6646                  */
6647                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6648                                                      lockend - lockstart + 1);
6649
6650                 /*
6651                  * We need to make sure there are no buffered pages in this
6652                  * range either, we could have raced between the invalidate in
6653                  * generic_file_direct_write and locking the extent.  The
6654                  * invalidate needs to happen so that reads after a write do not
6655                  * get stale data.
6656                  */
6657                 if (!ordered && (!writing ||
6658                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6659                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6660                                     *cached_state)))
6661                         break;
6662
6663                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6664                                      cached_state, GFP_NOFS);
6665
6666                 if (ordered) {
6667                         btrfs_start_ordered_extent(inode, ordered, 1);
6668                         btrfs_put_ordered_extent(ordered);
6669                 } else {
6670                         /* Screw you mmap */
6671                         ret = filemap_write_and_wait_range(inode->i_mapping,
6672                                                            lockstart,
6673                                                            lockend);
6674                         if (ret)
6675                                 break;
6676
6677                         /*
6678                          * If we found a page that couldn't be invalidated just
6679                          * fall back to buffered.
6680                          */
6681                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6682                                         lockstart >> PAGE_CACHE_SHIFT,
6683                                         lockend >> PAGE_CACHE_SHIFT);
6684                         if (ret)
6685                                 break;
6686                 }
6687
6688                 cond_resched();
6689         }
6690
6691         return ret;
6692 }
6693
6694 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6695                                            u64 len, u64 orig_start,
6696                                            u64 block_start, u64 block_len,
6697                                            u64 orig_block_len, u64 ram_bytes,
6698                                            int type)
6699 {
6700         struct extent_map_tree *em_tree;
6701         struct extent_map *em;
6702         struct btrfs_root *root = BTRFS_I(inode)->root;
6703         int ret;
6704
6705         em_tree = &BTRFS_I(inode)->extent_tree;
6706         em = alloc_extent_map();
6707         if (!em)
6708                 return ERR_PTR(-ENOMEM);
6709
6710         em->start = start;
6711         em->orig_start = orig_start;
6712         em->mod_start = start;
6713         em->mod_len = len;
6714         em->len = len;
6715         em->block_len = block_len;
6716         em->block_start = block_start;
6717         em->bdev = root->fs_info->fs_devices->latest_bdev;
6718         em->orig_block_len = orig_block_len;
6719         em->ram_bytes = ram_bytes;
6720         em->generation = -1;
6721         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6722         if (type == BTRFS_ORDERED_PREALLOC)
6723                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6724
6725         do {
6726                 btrfs_drop_extent_cache(inode, em->start,
6727                                 em->start + em->len - 1, 0);
6728                 write_lock(&em_tree->lock);
6729                 ret = add_extent_mapping(em_tree, em, 1);
6730                 write_unlock(&em_tree->lock);
6731         } while (ret == -EEXIST);
6732
6733         if (ret) {
6734                 free_extent_map(em);
6735                 return ERR_PTR(ret);
6736         }
6737
6738         return em;
6739 }
6740
6741
6742 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6743                                    struct buffer_head *bh_result, int create)
6744 {
6745         struct extent_map *em;
6746         struct btrfs_root *root = BTRFS_I(inode)->root;
6747         struct extent_state *cached_state = NULL;
6748         u64 start = iblock << inode->i_blkbits;
6749         u64 lockstart, lockend;
6750         u64 len = bh_result->b_size;
6751         struct btrfs_trans_handle *trans;
6752         int unlock_bits = EXTENT_LOCKED;
6753         int ret = 0;
6754
6755         if (create)
6756                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6757         else
6758                 len = min_t(u64, len, root->sectorsize);
6759
6760         lockstart = start;
6761         lockend = start + len - 1;
6762
6763         /*
6764          * If this errors out it's because we couldn't invalidate pagecache for
6765          * this range and we need to fallback to buffered.
6766          */
6767         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6768                 return -ENOTBLK;
6769
6770         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6771         if (IS_ERR(em)) {
6772                 ret = PTR_ERR(em);
6773                 goto unlock_err;
6774         }
6775
6776         /*
6777          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6778          * io.  INLINE is special, and we could probably kludge it in here, but
6779          * it's still buffered so for safety lets just fall back to the generic
6780          * buffered path.
6781          *
6782          * For COMPRESSED we _have_ to read the entire extent in so we can
6783          * decompress it, so there will be buffering required no matter what we
6784          * do, so go ahead and fallback to buffered.
6785          *
6786          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6787          * to buffered IO.  Don't blame me, this is the price we pay for using
6788          * the generic code.
6789          */
6790         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6791             em->block_start == EXTENT_MAP_INLINE) {
6792                 free_extent_map(em);
6793                 ret = -ENOTBLK;
6794                 goto unlock_err;
6795         }
6796
6797         /* Just a good old fashioned hole, return */
6798         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6799                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6800                 free_extent_map(em);
6801                 goto unlock_err;
6802         }
6803
6804         /*
6805          * We don't allocate a new extent in the following cases
6806          *
6807          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6808          * existing extent.
6809          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6810          * just use the extent.
6811          *
6812          */
6813         if (!create) {
6814                 len = min(len, em->len - (start - em->start));
6815                 lockstart = start + len;
6816                 goto unlock;
6817         }
6818
6819         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6820             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6821              em->block_start != EXTENT_MAP_HOLE)) {
6822                 int type;
6823                 int ret;
6824                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6825
6826                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6827                         type = BTRFS_ORDERED_PREALLOC;
6828                 else
6829                         type = BTRFS_ORDERED_NOCOW;
6830                 len = min(len, em->len - (start - em->start));
6831                 block_start = em->block_start + (start - em->start);
6832
6833                 /*
6834                  * we're not going to log anything, but we do need
6835                  * to make sure the current transaction stays open
6836                  * while we look for nocow cross refs
6837                  */
6838                 trans = btrfs_join_transaction(root);
6839                 if (IS_ERR(trans))
6840                         goto must_cow;
6841
6842                 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6843                                       &orig_block_len, &ram_bytes) == 1) {
6844                         if (type == BTRFS_ORDERED_PREALLOC) {
6845                                 free_extent_map(em);
6846                                 em = create_pinned_em(inode, start, len,
6847                                                        orig_start,
6848                                                        block_start, len,
6849                                                        orig_block_len,
6850                                                        ram_bytes, type);
6851                                 if (IS_ERR(em)) {
6852                                         btrfs_end_transaction(trans, root);
6853                                         goto unlock_err;
6854                                 }
6855                         }
6856
6857                         ret = btrfs_add_ordered_extent_dio(inode, start,
6858                                            block_start, len, len, type);
6859                         btrfs_end_transaction(trans, root);
6860                         if (ret) {
6861                                 free_extent_map(em);
6862                                 goto unlock_err;
6863                         }
6864                         goto unlock;
6865                 }
6866                 btrfs_end_transaction(trans, root);
6867         }
6868 must_cow:
6869         /*
6870          * this will cow the extent, reset the len in case we changed
6871          * it above
6872          */
6873         len = bh_result->b_size;
6874         free_extent_map(em);
6875         em = btrfs_new_extent_direct(inode, start, len);
6876         if (IS_ERR(em)) {
6877                 ret = PTR_ERR(em);
6878                 goto unlock_err;
6879         }
6880         len = min(len, em->len - (start - em->start));
6881 unlock:
6882         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6883                 inode->i_blkbits;
6884         bh_result->b_size = len;
6885         bh_result->b_bdev = em->bdev;
6886         set_buffer_mapped(bh_result);
6887         if (create) {
6888                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6889                         set_buffer_new(bh_result);
6890
6891                 /*
6892                  * Need to update the i_size under the extent lock so buffered
6893                  * readers will get the updated i_size when we unlock.
6894                  */
6895                 if (start + len > i_size_read(inode))
6896                         i_size_write(inode, start + len);
6897
6898                 spin_lock(&BTRFS_I(inode)->lock);
6899                 BTRFS_I(inode)->outstanding_extents++;
6900                 spin_unlock(&BTRFS_I(inode)->lock);
6901
6902                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6903                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6904                                      &cached_state, GFP_NOFS);
6905                 BUG_ON(ret);
6906         }
6907
6908         /*
6909          * In the case of write we need to clear and unlock the entire range,
6910          * in the case of read we need to unlock only the end area that we
6911          * aren't using if there is any left over space.
6912          */
6913         if (lockstart < lockend) {
6914                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6915                                  lockend, unlock_bits, 1, 0,
6916                                  &cached_state, GFP_NOFS);
6917         } else {
6918                 free_extent_state(cached_state);
6919         }
6920
6921         free_extent_map(em);
6922
6923         return 0;
6924
6925 unlock_err:
6926         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6927                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6928         return ret;
6929 }
6930
6931 struct btrfs_dio_private {
6932         struct inode *inode;
6933         u64 logical_offset;
6934         u64 disk_bytenr;
6935         u64 bytes;
6936         void *private;
6937
6938         /* number of bios pending for this dio */
6939         atomic_t pending_bios;
6940
6941         /* IO errors */
6942         int errors;
6943
6944         /* orig_bio is our btrfs_io_bio */
6945         struct bio *orig_bio;
6946
6947         /* dio_bio came from fs/direct-io.c */
6948         struct bio *dio_bio;
6949 };
6950
6951 static void btrfs_endio_direct_read(struct bio *bio, int err)
6952 {
6953         struct btrfs_dio_private *dip = bio->bi_private;
6954         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6955         struct bio_vec *bvec = bio->bi_io_vec;
6956         struct inode *inode = dip->inode;
6957         struct btrfs_root *root = BTRFS_I(inode)->root;
6958         struct bio *dio_bio;
6959         u64 start;
6960
6961         start = dip->logical_offset;
6962         do {
6963                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6964                         struct page *page = bvec->bv_page;
6965                         char *kaddr;
6966                         u32 csum = ~(u32)0;
6967                         u64 private = ~(u32)0;
6968                         unsigned long flags;
6969
6970                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6971                                               start, &private))
6972                                 goto failed;
6973                         local_irq_save(flags);
6974                         kaddr = kmap_atomic(page);
6975                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6976                                                csum, bvec->bv_len);
6977                         btrfs_csum_final(csum, (char *)&csum);
6978                         kunmap_atomic(kaddr);
6979                         local_irq_restore(flags);
6980
6981                         flush_dcache_page(bvec->bv_page);
6982                         if (csum != private) {
6983 failed:
6984                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6985                                         (unsigned long long)btrfs_ino(inode),
6986                                         (unsigned long long)start,
6987                                         csum, (unsigned)private);
6988                                 err = -EIO;
6989                         }
6990                 }
6991
6992                 start += bvec->bv_len;
6993                 bvec++;
6994         } while (bvec <= bvec_end);
6995
6996         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6997                       dip->logical_offset + dip->bytes - 1);
6998         dio_bio = dip->dio_bio;
6999
7000         kfree(dip);
7001
7002         /* If we had a csum failure make sure to clear the uptodate flag */
7003         if (err)
7004                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7005         dio_end_io(dio_bio, err);
7006         bio_put(bio);
7007 }
7008
7009 static void btrfs_endio_direct_write(struct bio *bio, int err)
7010 {
7011         struct btrfs_dio_private *dip = bio->bi_private;
7012         struct inode *inode = dip->inode;
7013         struct btrfs_root *root = BTRFS_I(inode)->root;
7014         struct btrfs_ordered_extent *ordered = NULL;
7015         u64 ordered_offset = dip->logical_offset;
7016         u64 ordered_bytes = dip->bytes;
7017         struct bio *dio_bio;
7018         int ret;
7019
7020         if (err)
7021                 goto out_done;
7022 again:
7023         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7024                                                    &ordered_offset,
7025                                                    ordered_bytes, !err);
7026         if (!ret)
7027                 goto out_test;
7028
7029         ordered->work.func = finish_ordered_fn;
7030         ordered->work.flags = 0;
7031         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7032                            &ordered->work);
7033 out_test:
7034         /*
7035          * our bio might span multiple ordered extents.  If we haven't
7036          * completed the accounting for the whole dio, go back and try again
7037          */
7038         if (ordered_offset < dip->logical_offset + dip->bytes) {
7039                 ordered_bytes = dip->logical_offset + dip->bytes -
7040                         ordered_offset;
7041                 ordered = NULL;
7042                 goto again;
7043         }
7044 out_done:
7045         dio_bio = dip->dio_bio;
7046
7047         kfree(dip);
7048
7049         /* If we had an error make sure to clear the uptodate flag */
7050         if (err)
7051                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7052         dio_end_io(dio_bio, err);
7053         bio_put(bio);
7054 }
7055
7056 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7057                                     struct bio *bio, int mirror_num,
7058                                     unsigned long bio_flags, u64 offset)
7059 {
7060         int ret;
7061         struct btrfs_root *root = BTRFS_I(inode)->root;
7062         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7063         BUG_ON(ret); /* -ENOMEM */
7064         return 0;
7065 }
7066
7067 static void btrfs_end_dio_bio(struct bio *bio, int err)
7068 {
7069         struct btrfs_dio_private *dip = bio->bi_private;
7070
7071         if (err) {
7072                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7073                       "sector %#Lx len %u err no %d\n",
7074                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7075                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7076                 dip->errors = 1;
7077
7078                 /*
7079                  * before atomic variable goto zero, we must make sure
7080                  * dip->errors is perceived to be set.
7081                  */
7082                 smp_mb__before_atomic_dec();
7083         }
7084
7085         /* if there are more bios still pending for this dio, just exit */
7086         if (!atomic_dec_and_test(&dip->pending_bios))
7087                 goto out;
7088
7089         if (dip->errors) {
7090                 bio_io_error(dip->orig_bio);
7091         } else {
7092                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7093                 bio_endio(dip->orig_bio, 0);
7094         }
7095 out:
7096         bio_put(bio);
7097 }
7098
7099 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7100                                        u64 first_sector, gfp_t gfp_flags)
7101 {
7102         int nr_vecs = bio_get_nr_vecs(bdev);
7103         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7104 }
7105
7106 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7107                                          int rw, u64 file_offset, int skip_sum,
7108                                          int async_submit)
7109 {
7110         int write = rw & REQ_WRITE;
7111         struct btrfs_root *root = BTRFS_I(inode)->root;
7112         int ret;
7113
7114         if (async_submit)
7115                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7116
7117         bio_get(bio);
7118
7119         if (!write) {
7120                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7121                 if (ret)
7122                         goto err;
7123         }
7124
7125         if (skip_sum)
7126                 goto map;
7127
7128         if (write && async_submit) {
7129                 ret = btrfs_wq_submit_bio(root->fs_info,
7130                                    inode, rw, bio, 0, 0,
7131                                    file_offset,
7132                                    __btrfs_submit_bio_start_direct_io,
7133                                    __btrfs_submit_bio_done);
7134                 goto err;
7135         } else if (write) {
7136                 /*
7137                  * If we aren't doing async submit, calculate the csum of the
7138                  * bio now.
7139                  */
7140                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7141                 if (ret)
7142                         goto err;
7143         } else if (!skip_sum) {
7144                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7145                 if (ret)
7146                         goto err;
7147         }
7148
7149 map:
7150         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7151 err:
7152         bio_put(bio);
7153         return ret;
7154 }
7155
7156 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7157                                     int skip_sum)
7158 {
7159         struct inode *inode = dip->inode;
7160         struct btrfs_root *root = BTRFS_I(inode)->root;
7161         struct bio *bio;
7162         struct bio *orig_bio = dip->orig_bio;
7163         struct bio_vec *bvec = orig_bio->bi_io_vec;
7164         u64 start_sector = orig_bio->bi_sector;
7165         u64 file_offset = dip->logical_offset;
7166         u64 submit_len = 0;
7167         u64 map_length;
7168         int nr_pages = 0;
7169         int ret = 0;
7170         int async_submit = 0;
7171
7172         map_length = orig_bio->bi_size;
7173         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7174                               &map_length, NULL, 0);
7175         if (ret) {
7176                 bio_put(orig_bio);
7177                 return -EIO;
7178         }
7179         if (map_length >= orig_bio->bi_size) {
7180                 bio = orig_bio;
7181                 goto submit;
7182         }
7183
7184         /* async crcs make it difficult to collect full stripe writes. */
7185         if (btrfs_get_alloc_profile(root, 1) &
7186             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7187                 async_submit = 0;
7188         else
7189                 async_submit = 1;
7190
7191         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7192         if (!bio)
7193                 return -ENOMEM;
7194         bio->bi_private = dip;
7195         bio->bi_end_io = btrfs_end_dio_bio;
7196         atomic_inc(&dip->pending_bios);
7197
7198         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7199                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7200                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7201                                  bvec->bv_offset) < bvec->bv_len)) {
7202                         /*
7203                          * inc the count before we submit the bio so
7204                          * we know the end IO handler won't happen before
7205                          * we inc the count. Otherwise, the dip might get freed
7206                          * before we're done setting it up
7207                          */
7208                         atomic_inc(&dip->pending_bios);
7209                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7210                                                      file_offset, skip_sum,
7211                                                      async_submit);
7212                         if (ret) {
7213                                 bio_put(bio);
7214                                 atomic_dec(&dip->pending_bios);
7215                                 goto out_err;
7216                         }
7217
7218                         start_sector += submit_len >> 9;
7219                         file_offset += submit_len;
7220
7221                         submit_len = 0;
7222                         nr_pages = 0;
7223
7224                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7225                                                   start_sector, GFP_NOFS);
7226                         if (!bio)
7227                                 goto out_err;
7228                         bio->bi_private = dip;
7229                         bio->bi_end_io = btrfs_end_dio_bio;
7230
7231                         map_length = orig_bio->bi_size;
7232                         ret = btrfs_map_block(root->fs_info, rw,
7233                                               start_sector << 9,
7234                                               &map_length, NULL, 0);
7235                         if (ret) {
7236                                 bio_put(bio);
7237                                 goto out_err;
7238                         }
7239                 } else {
7240                         submit_len += bvec->bv_len;
7241                         nr_pages ++;
7242                         bvec++;
7243                 }
7244         }
7245
7246 submit:
7247         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7248                                      async_submit);
7249         if (!ret)
7250                 return 0;
7251
7252         bio_put(bio);
7253 out_err:
7254         dip->errors = 1;
7255         /*
7256          * before atomic variable goto zero, we must
7257          * make sure dip->errors is perceived to be set.
7258          */
7259         smp_mb__before_atomic_dec();
7260         if (atomic_dec_and_test(&dip->pending_bios))
7261                 bio_io_error(dip->orig_bio);
7262
7263         /* bio_end_io() will handle error, so we needn't return it */
7264         return 0;
7265 }
7266
7267 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7268                                 struct inode *inode, loff_t file_offset)
7269 {
7270         struct btrfs_root *root = BTRFS_I(inode)->root;
7271         struct btrfs_dio_private *dip;
7272         struct bio_vec *bvec = dio_bio->bi_io_vec;
7273         struct bio *io_bio;
7274         int skip_sum;
7275         int write = rw & REQ_WRITE;
7276         int ret = 0;
7277
7278         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7279
7280         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7281
7282         if (!io_bio) {
7283                 ret = -ENOMEM;
7284                 goto free_ordered;
7285         }
7286
7287         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7288         if (!dip) {
7289                 ret = -ENOMEM;
7290                 goto free_io_bio;
7291         }
7292
7293         dip->private = dio_bio->bi_private;
7294         io_bio->bi_private = dio_bio->bi_private;
7295         dip->inode = inode;
7296         dip->logical_offset = file_offset;
7297
7298         dip->bytes = 0;
7299         do {
7300                 dip->bytes += bvec->bv_len;
7301                 bvec++;
7302         } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7303
7304         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7305         io_bio->bi_private = dip;
7306         dip->errors = 0;
7307         dip->orig_bio = io_bio;
7308         dip->dio_bio = dio_bio;
7309         atomic_set(&dip->pending_bios, 0);
7310
7311         if (write)
7312                 io_bio->bi_end_io = btrfs_endio_direct_write;
7313         else
7314                 io_bio->bi_end_io = btrfs_endio_direct_read;
7315
7316         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7317         if (!ret)
7318                 return;
7319
7320 free_io_bio:
7321         bio_put(io_bio);
7322
7323 free_ordered:
7324         /*
7325          * If this is a write, we need to clean up the reserved space and kill
7326          * the ordered extent.
7327          */
7328         if (write) {
7329                 struct btrfs_ordered_extent *ordered;
7330                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7331                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7332                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7333                         btrfs_free_reserved_extent(root, ordered->start,
7334                                                    ordered->disk_len);
7335                 btrfs_put_ordered_extent(ordered);
7336                 btrfs_put_ordered_extent(ordered);
7337         }
7338         bio_endio(dio_bio, ret);
7339 }
7340
7341 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7342                         const struct iovec *iov, loff_t offset,
7343                         unsigned long nr_segs)
7344 {
7345         int seg;
7346         int i;
7347         size_t size;
7348         unsigned long addr;
7349         unsigned blocksize_mask = root->sectorsize - 1;
7350         ssize_t retval = -EINVAL;
7351         loff_t end = offset;
7352
7353         if (offset & blocksize_mask)
7354                 goto out;
7355
7356         /* Check the memory alignment.  Blocks cannot straddle pages */
7357         for (seg = 0; seg < nr_segs; seg++) {
7358                 addr = (unsigned long)iov[seg].iov_base;
7359                 size = iov[seg].iov_len;
7360                 end += size;
7361                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7362                         goto out;
7363
7364                 /* If this is a write we don't need to check anymore */
7365                 if (rw & WRITE)
7366                         continue;
7367
7368                 /*
7369                  * Check to make sure we don't have duplicate iov_base's in this
7370                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7371                  * when reading back.
7372                  */
7373                 for (i = seg + 1; i < nr_segs; i++) {
7374                         if (iov[seg].iov_base == iov[i].iov_base)
7375                                 goto out;
7376                 }
7377         }
7378         retval = 0;
7379 out:
7380         return retval;
7381 }
7382
7383 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7384                         const struct iovec *iov, loff_t offset,
7385                         unsigned long nr_segs)
7386 {
7387         struct file *file = iocb->ki_filp;
7388         struct inode *inode = file->f_mapping->host;
7389         size_t count = 0;
7390         int flags = 0;
7391         bool wakeup = true;
7392         bool relock = false;
7393         ssize_t ret;
7394
7395         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7396                             offset, nr_segs))
7397                 return 0;
7398
7399         atomic_inc(&inode->i_dio_count);
7400         smp_mb__after_atomic_inc();
7401
7402         if (rw & WRITE) {
7403                 count = iov_length(iov, nr_segs);
7404                 /*
7405                  * If the write DIO is beyond the EOF, we need update
7406                  * the isize, but it is protected by i_mutex. So we can
7407                  * not unlock the i_mutex at this case.
7408                  */
7409                 if (offset + count <= inode->i_size) {
7410                         mutex_unlock(&inode->i_mutex);
7411                         relock = true;
7412                 }
7413                 ret = btrfs_delalloc_reserve_space(inode, count);
7414                 if (ret)
7415                         goto out;
7416         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7417                                      &BTRFS_I(inode)->runtime_flags))) {
7418                 inode_dio_done(inode);
7419                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7420                 wakeup = false;
7421         }
7422
7423         ret = __blockdev_direct_IO(rw, iocb, inode,
7424                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7425                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7426                         btrfs_submit_direct, flags);
7427         if (rw & WRITE) {
7428                 if (ret < 0 && ret != -EIOCBQUEUED)
7429                         btrfs_delalloc_release_space(inode, count);
7430                 else if (ret >= 0 && (size_t)ret < count)
7431                         btrfs_delalloc_release_space(inode,
7432                                                      count - (size_t)ret);
7433                 else
7434                         btrfs_delalloc_release_metadata(inode, 0);
7435         }
7436 out:
7437         if (wakeup)
7438                 inode_dio_done(inode);
7439         if (relock)
7440                 mutex_lock(&inode->i_mutex);
7441
7442         return ret;
7443 }
7444
7445 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7446
7447 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7448                 __u64 start, __u64 len)
7449 {
7450         int     ret;
7451
7452         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7453         if (ret)
7454                 return ret;
7455
7456         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7457 }
7458
7459 int btrfs_readpage(struct file *file, struct page *page)
7460 {
7461         struct extent_io_tree *tree;
7462         tree = &BTRFS_I(page->mapping->host)->io_tree;
7463         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7464 }
7465
7466 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7467 {
7468         struct extent_io_tree *tree;
7469
7470
7471         if (current->flags & PF_MEMALLOC) {
7472                 redirty_page_for_writepage(wbc, page);
7473                 unlock_page(page);
7474                 return 0;
7475         }
7476         tree = &BTRFS_I(page->mapping->host)->io_tree;
7477         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7478 }
7479
7480 static int btrfs_writepages(struct address_space *mapping,
7481                             struct writeback_control *wbc)
7482 {
7483         struct extent_io_tree *tree;
7484
7485         tree = &BTRFS_I(mapping->host)->io_tree;
7486         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7487 }
7488
7489 static int
7490 btrfs_readpages(struct file *file, struct address_space *mapping,
7491                 struct list_head *pages, unsigned nr_pages)
7492 {
7493         struct extent_io_tree *tree;
7494         tree = &BTRFS_I(mapping->host)->io_tree;
7495         return extent_readpages(tree, mapping, pages, nr_pages,
7496                                 btrfs_get_extent);
7497 }
7498 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7499 {
7500         struct extent_io_tree *tree;
7501         struct extent_map_tree *map;
7502         int ret;
7503
7504         tree = &BTRFS_I(page->mapping->host)->io_tree;
7505         map = &BTRFS_I(page->mapping->host)->extent_tree;
7506         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7507         if (ret == 1) {
7508                 ClearPagePrivate(page);
7509                 set_page_private(page, 0);
7510                 page_cache_release(page);
7511         }
7512         return ret;
7513 }
7514
7515 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7516 {
7517         if (PageWriteback(page) || PageDirty(page))
7518                 return 0;
7519         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7520 }
7521
7522 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7523 {
7524         struct inode *inode = page->mapping->host;
7525         struct extent_io_tree *tree;
7526         struct btrfs_ordered_extent *ordered;
7527         struct extent_state *cached_state = NULL;
7528         u64 page_start = page_offset(page);
7529         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7530
7531         /*
7532          * we have the page locked, so new writeback can't start,
7533          * and the dirty bit won't be cleared while we are here.
7534          *
7535          * Wait for IO on this page so that we can safely clear
7536          * the PagePrivate2 bit and do ordered accounting
7537          */
7538         wait_on_page_writeback(page);
7539
7540         tree = &BTRFS_I(inode)->io_tree;
7541         if (offset) {
7542                 btrfs_releasepage(page, GFP_NOFS);
7543                 return;
7544         }
7545         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7546         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7547         if (ordered) {
7548                 /*
7549                  * IO on this page will never be started, so we need
7550                  * to account for any ordered extents now
7551                  */
7552                 clear_extent_bit(tree, page_start, page_end,
7553                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7554                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7555                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7556                 /*
7557                  * whoever cleared the private bit is responsible
7558                  * for the finish_ordered_io
7559                  */
7560                 if (TestClearPagePrivate2(page) &&
7561                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7562                                                    PAGE_CACHE_SIZE, 1)) {
7563                         btrfs_finish_ordered_io(ordered);
7564                 }
7565                 btrfs_put_ordered_extent(ordered);
7566                 cached_state = NULL;
7567                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7568         }
7569         clear_extent_bit(tree, page_start, page_end,
7570                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7571                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7572                  &cached_state, GFP_NOFS);
7573         __btrfs_releasepage(page, GFP_NOFS);
7574
7575         ClearPageChecked(page);
7576         if (PagePrivate(page)) {
7577                 ClearPagePrivate(page);
7578                 set_page_private(page, 0);
7579                 page_cache_release(page);
7580         }
7581 }
7582
7583 /*
7584  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7585  * called from a page fault handler when a page is first dirtied. Hence we must
7586  * be careful to check for EOF conditions here. We set the page up correctly
7587  * for a written page which means we get ENOSPC checking when writing into
7588  * holes and correct delalloc and unwritten extent mapping on filesystems that
7589  * support these features.
7590  *
7591  * We are not allowed to take the i_mutex here so we have to play games to
7592  * protect against truncate races as the page could now be beyond EOF.  Because
7593  * vmtruncate() writes the inode size before removing pages, once we have the
7594  * page lock we can determine safely if the page is beyond EOF. If it is not
7595  * beyond EOF, then the page is guaranteed safe against truncation until we
7596  * unlock the page.
7597  */
7598 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7599 {
7600         struct page *page = vmf->page;
7601         struct inode *inode = file_inode(vma->vm_file);
7602         struct btrfs_root *root = BTRFS_I(inode)->root;
7603         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7604         struct btrfs_ordered_extent *ordered;
7605         struct extent_state *cached_state = NULL;
7606         char *kaddr;
7607         unsigned long zero_start;
7608         loff_t size;
7609         int ret;
7610         int reserved = 0;
7611         u64 page_start;
7612         u64 page_end;
7613
7614         sb_start_pagefault(inode->i_sb);
7615         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7616         if (!ret) {
7617                 ret = file_update_time(vma->vm_file);
7618                 reserved = 1;
7619         }
7620         if (ret) {
7621                 if (ret == -ENOMEM)
7622                         ret = VM_FAULT_OOM;
7623                 else /* -ENOSPC, -EIO, etc */
7624                         ret = VM_FAULT_SIGBUS;
7625                 if (reserved)
7626                         goto out;
7627                 goto out_noreserve;
7628         }
7629
7630         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7631 again:
7632         lock_page(page);
7633         size = i_size_read(inode);
7634         page_start = page_offset(page);
7635         page_end = page_start + PAGE_CACHE_SIZE - 1;
7636
7637         if ((page->mapping != inode->i_mapping) ||
7638             (page_start >= size)) {
7639                 /* page got truncated out from underneath us */
7640                 goto out_unlock;
7641         }
7642         wait_on_page_writeback(page);
7643
7644         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7645         set_page_extent_mapped(page);
7646
7647         /*
7648          * we can't set the delalloc bits if there are pending ordered
7649          * extents.  Drop our locks and wait for them to finish
7650          */
7651         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7652         if (ordered) {
7653                 unlock_extent_cached(io_tree, page_start, page_end,
7654                                      &cached_state, GFP_NOFS);
7655                 unlock_page(page);
7656                 btrfs_start_ordered_extent(inode, ordered, 1);
7657                 btrfs_put_ordered_extent(ordered);
7658                 goto again;
7659         }
7660
7661         /*
7662          * XXX - page_mkwrite gets called every time the page is dirtied, even
7663          * if it was already dirty, so for space accounting reasons we need to
7664          * clear any delalloc bits for the range we are fixing to save.  There
7665          * is probably a better way to do this, but for now keep consistent with
7666          * prepare_pages in the normal write path.
7667          */
7668         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7669                           EXTENT_DIRTY | EXTENT_DELALLOC |
7670                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7671                           0, 0, &cached_state, GFP_NOFS);
7672
7673         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7674                                         &cached_state);
7675         if (ret) {
7676                 unlock_extent_cached(io_tree, page_start, page_end,
7677                                      &cached_state, GFP_NOFS);
7678                 ret = VM_FAULT_SIGBUS;
7679                 goto out_unlock;
7680         }
7681         ret = 0;
7682
7683         /* page is wholly or partially inside EOF */
7684         if (page_start + PAGE_CACHE_SIZE > size)
7685                 zero_start = size & ~PAGE_CACHE_MASK;
7686         else
7687                 zero_start = PAGE_CACHE_SIZE;
7688
7689         if (zero_start != PAGE_CACHE_SIZE) {
7690                 kaddr = kmap(page);
7691                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7692                 flush_dcache_page(page);
7693                 kunmap(page);
7694         }
7695         ClearPageChecked(page);
7696         set_page_dirty(page);
7697         SetPageUptodate(page);
7698
7699         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7700         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7701         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7702
7703         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7704
7705 out_unlock:
7706         if (!ret) {
7707                 sb_end_pagefault(inode->i_sb);
7708                 return VM_FAULT_LOCKED;
7709         }
7710         unlock_page(page);
7711 out:
7712         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7713 out_noreserve:
7714         sb_end_pagefault(inode->i_sb);
7715         return ret;
7716 }
7717
7718 static int btrfs_truncate(struct inode *inode)
7719 {
7720         struct btrfs_root *root = BTRFS_I(inode)->root;
7721         struct btrfs_block_rsv *rsv;
7722         int ret;
7723         int err = 0;
7724         struct btrfs_trans_handle *trans;
7725         u64 mask = root->sectorsize - 1;
7726         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7727
7728         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7729         if (ret)
7730                 return ret;
7731
7732         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7733         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7734
7735         /*
7736          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7737          * 3 things going on here
7738          *
7739          * 1) We need to reserve space for our orphan item and the space to
7740          * delete our orphan item.  Lord knows we don't want to have a dangling
7741          * orphan item because we didn't reserve space to remove it.
7742          *
7743          * 2) We need to reserve space to update our inode.
7744          *
7745          * 3) We need to have something to cache all the space that is going to
7746          * be free'd up by the truncate operation, but also have some slack
7747          * space reserved in case it uses space during the truncate (thank you
7748          * very much snapshotting).
7749          *
7750          * And we need these to all be seperate.  The fact is we can use alot of
7751          * space doing the truncate, and we have no earthly idea how much space
7752          * we will use, so we need the truncate reservation to be seperate so it
7753          * doesn't end up using space reserved for updating the inode or
7754          * removing the orphan item.  We also need to be able to stop the
7755          * transaction and start a new one, which means we need to be able to
7756          * update the inode several times, and we have no idea of knowing how
7757          * many times that will be, so we can't just reserve 1 item for the
7758          * entirety of the opration, so that has to be done seperately as well.
7759          * Then there is the orphan item, which does indeed need to be held on
7760          * to for the whole operation, and we need nobody to touch this reserved
7761          * space except the orphan code.
7762          *
7763          * So that leaves us with
7764          *
7765          * 1) root->orphan_block_rsv - for the orphan deletion.
7766          * 2) rsv - for the truncate reservation, which we will steal from the
7767          * transaction reservation.
7768          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7769          * updating the inode.
7770          */
7771         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7772         if (!rsv)
7773                 return -ENOMEM;
7774         rsv->size = min_size;
7775         rsv->failfast = 1;
7776
7777         /*
7778          * 1 for the truncate slack space
7779          * 1 for updating the inode.
7780          */
7781         trans = btrfs_start_transaction(root, 2);
7782         if (IS_ERR(trans)) {
7783                 err = PTR_ERR(trans);
7784                 goto out;
7785         }
7786
7787         /* Migrate the slack space for the truncate to our reserve */
7788         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7789                                       min_size);
7790         BUG_ON(ret);
7791
7792         /*
7793          * setattr is responsible for setting the ordered_data_close flag,
7794          * but that is only tested during the last file release.  That
7795          * could happen well after the next commit, leaving a great big
7796          * window where new writes may get lost if someone chooses to write
7797          * to this file after truncating to zero
7798          *
7799          * The inode doesn't have any dirty data here, and so if we commit
7800          * this is a noop.  If someone immediately starts writing to the inode
7801          * it is very likely we'll catch some of their writes in this
7802          * transaction, and the commit will find this file on the ordered
7803          * data list with good things to send down.
7804          *
7805          * This is a best effort solution, there is still a window where
7806          * using truncate to replace the contents of the file will
7807          * end up with a zero length file after a crash.
7808          */
7809         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7810                                            &BTRFS_I(inode)->runtime_flags))
7811                 btrfs_add_ordered_operation(trans, root, inode);
7812
7813         /*
7814          * So if we truncate and then write and fsync we normally would just
7815          * write the extents that changed, which is a problem if we need to
7816          * first truncate that entire inode.  So set this flag so we write out
7817          * all of the extents in the inode to the sync log so we're completely
7818          * safe.
7819          */
7820         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7821         trans->block_rsv = rsv;
7822
7823         while (1) {
7824                 ret = btrfs_truncate_inode_items(trans, root, inode,
7825                                                  inode->i_size,
7826                                                  BTRFS_EXTENT_DATA_KEY);
7827                 if (ret != -ENOSPC) {
7828                         err = ret;
7829                         break;
7830                 }
7831
7832                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7833                 ret = btrfs_update_inode(trans, root, inode);
7834                 if (ret) {
7835                         err = ret;
7836                         break;
7837                 }
7838
7839                 btrfs_end_transaction(trans, root);
7840                 btrfs_btree_balance_dirty(root);
7841
7842                 trans = btrfs_start_transaction(root, 2);
7843                 if (IS_ERR(trans)) {
7844                         ret = err = PTR_ERR(trans);
7845                         trans = NULL;
7846                         break;
7847                 }
7848
7849                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7850                                               rsv, min_size);
7851                 BUG_ON(ret);    /* shouldn't happen */
7852                 trans->block_rsv = rsv;
7853         }
7854
7855         if (ret == 0 && inode->i_nlink > 0) {
7856                 trans->block_rsv = root->orphan_block_rsv;
7857                 ret = btrfs_orphan_del(trans, inode);
7858                 if (ret)
7859                         err = ret;
7860         }
7861
7862         if (trans) {
7863                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7864                 ret = btrfs_update_inode(trans, root, inode);
7865                 if (ret && !err)
7866                         err = ret;
7867
7868                 ret = btrfs_end_transaction(trans, root);
7869                 btrfs_btree_balance_dirty(root);
7870         }
7871
7872 out:
7873         btrfs_free_block_rsv(root, rsv);
7874
7875         if (ret && !err)
7876                 err = ret;
7877
7878         return err;
7879 }
7880
7881 /*
7882  * create a new subvolume directory/inode (helper for the ioctl).
7883  */
7884 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7885                              struct btrfs_root *new_root, u64 new_dirid)
7886 {
7887         struct inode *inode;
7888         int err;
7889         u64 index = 0;
7890
7891         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7892                                 new_dirid, new_dirid,
7893                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7894                                 &index);
7895         if (IS_ERR(inode))
7896                 return PTR_ERR(inode);
7897         inode->i_op = &btrfs_dir_inode_operations;
7898         inode->i_fop = &btrfs_dir_file_operations;
7899
7900         set_nlink(inode, 1);
7901         btrfs_i_size_write(inode, 0);
7902
7903         err = btrfs_update_inode(trans, new_root, inode);
7904
7905         iput(inode);
7906         return err;
7907 }
7908
7909 struct inode *btrfs_alloc_inode(struct super_block *sb)
7910 {
7911         struct btrfs_inode *ei;
7912         struct inode *inode;
7913
7914         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7915         if (!ei)
7916                 return NULL;
7917
7918         ei->root = NULL;
7919         ei->generation = 0;
7920         ei->last_trans = 0;
7921         ei->last_sub_trans = 0;
7922         ei->logged_trans = 0;
7923         ei->delalloc_bytes = 0;
7924         ei->disk_i_size = 0;
7925         ei->flags = 0;
7926         ei->csum_bytes = 0;
7927         ei->index_cnt = (u64)-1;
7928         ei->last_unlink_trans = 0;
7929         ei->last_log_commit = 0;
7930
7931         spin_lock_init(&ei->lock);
7932         ei->outstanding_extents = 0;
7933         ei->reserved_extents = 0;
7934
7935         ei->runtime_flags = 0;
7936         ei->force_compress = BTRFS_COMPRESS_NONE;
7937
7938         ei->delayed_node = NULL;
7939
7940         inode = &ei->vfs_inode;
7941         extent_map_tree_init(&ei->extent_tree);
7942         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7943         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7944         ei->io_tree.track_uptodate = 1;
7945         ei->io_failure_tree.track_uptodate = 1;
7946         atomic_set(&ei->sync_writers, 0);
7947         mutex_init(&ei->log_mutex);
7948         mutex_init(&ei->delalloc_mutex);
7949         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7950         INIT_LIST_HEAD(&ei->delalloc_inodes);
7951         INIT_LIST_HEAD(&ei->ordered_operations);
7952         RB_CLEAR_NODE(&ei->rb_node);
7953
7954         return inode;
7955 }
7956
7957 static void btrfs_i_callback(struct rcu_head *head)
7958 {
7959         struct inode *inode = container_of(head, struct inode, i_rcu);
7960         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7961 }
7962
7963 void btrfs_destroy_inode(struct inode *inode)
7964 {
7965         struct btrfs_ordered_extent *ordered;
7966         struct btrfs_root *root = BTRFS_I(inode)->root;
7967
7968         WARN_ON(!hlist_empty(&inode->i_dentry));
7969         WARN_ON(inode->i_data.nrpages);
7970         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7971         WARN_ON(BTRFS_I(inode)->reserved_extents);
7972         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7973         WARN_ON(BTRFS_I(inode)->csum_bytes);
7974
7975         /*
7976          * This can happen where we create an inode, but somebody else also
7977          * created the same inode and we need to destroy the one we already
7978          * created.
7979          */
7980         if (!root)
7981                 goto free;
7982
7983         /*
7984          * Make sure we're properly removed from the ordered operation
7985          * lists.
7986          */
7987         smp_mb();
7988         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7989                 spin_lock(&root->fs_info->ordered_extent_lock);
7990                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7991                 spin_unlock(&root->fs_info->ordered_extent_lock);
7992         }
7993
7994         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7995                      &BTRFS_I(inode)->runtime_flags)) {
7996                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7997                         (unsigned long long)btrfs_ino(inode));
7998                 atomic_dec(&root->orphan_inodes);
7999         }
8000
8001         while (1) {
8002                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8003                 if (!ordered)
8004                         break;
8005                 else {
8006                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8007                                 (unsigned long long)ordered->file_offset,
8008                                 (unsigned long long)ordered->len);
8009                         btrfs_remove_ordered_extent(inode, ordered);
8010                         btrfs_put_ordered_extent(ordered);
8011                         btrfs_put_ordered_extent(ordered);
8012                 }
8013         }
8014         inode_tree_del(inode);
8015         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8016 free:
8017         call_rcu(&inode->i_rcu, btrfs_i_callback);
8018 }
8019
8020 int btrfs_drop_inode(struct inode *inode)
8021 {
8022         struct btrfs_root *root = BTRFS_I(inode)->root;
8023
8024         if (root == NULL)
8025                 return 1;
8026
8027         /* the snap/subvol tree is on deleting */
8028         if (btrfs_root_refs(&root->root_item) == 0 &&
8029             root != root->fs_info->tree_root)
8030                 return 1;
8031         else
8032                 return generic_drop_inode(inode);
8033 }
8034
8035 static void init_once(void *foo)
8036 {
8037         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8038
8039         inode_init_once(&ei->vfs_inode);
8040 }
8041
8042 void btrfs_destroy_cachep(void)
8043 {
8044         /*
8045          * Make sure all delayed rcu free inodes are flushed before we
8046          * destroy cache.
8047          */
8048         rcu_barrier();
8049         if (btrfs_inode_cachep)
8050                 kmem_cache_destroy(btrfs_inode_cachep);
8051         if (btrfs_trans_handle_cachep)
8052                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8053         if (btrfs_transaction_cachep)
8054                 kmem_cache_destroy(btrfs_transaction_cachep);
8055         if (btrfs_path_cachep)
8056                 kmem_cache_destroy(btrfs_path_cachep);
8057         if (btrfs_free_space_cachep)
8058                 kmem_cache_destroy(btrfs_free_space_cachep);
8059         if (btrfs_delalloc_work_cachep)
8060                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8061 }
8062
8063 int btrfs_init_cachep(void)
8064 {
8065         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8066                         sizeof(struct btrfs_inode), 0,
8067                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8068         if (!btrfs_inode_cachep)
8069                 goto fail;
8070
8071         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8072                         sizeof(struct btrfs_trans_handle), 0,
8073                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8074         if (!btrfs_trans_handle_cachep)
8075                 goto fail;
8076
8077         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8078                         sizeof(struct btrfs_transaction), 0,
8079                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8080         if (!btrfs_transaction_cachep)
8081                 goto fail;
8082
8083         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8084                         sizeof(struct btrfs_path), 0,
8085                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8086         if (!btrfs_path_cachep)
8087                 goto fail;
8088
8089         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8090                         sizeof(struct btrfs_free_space), 0,
8091                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8092         if (!btrfs_free_space_cachep)
8093                 goto fail;
8094
8095         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8096                         sizeof(struct btrfs_delalloc_work), 0,
8097                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8098                         NULL);
8099         if (!btrfs_delalloc_work_cachep)
8100                 goto fail;
8101
8102         return 0;
8103 fail:
8104         btrfs_destroy_cachep();
8105         return -ENOMEM;
8106 }
8107
8108 static int btrfs_getattr(struct vfsmount *mnt,
8109                          struct dentry *dentry, struct kstat *stat)
8110 {
8111         u64 delalloc_bytes;
8112         struct inode *inode = dentry->d_inode;
8113         u32 blocksize = inode->i_sb->s_blocksize;
8114
8115         generic_fillattr(inode, stat);
8116         stat->dev = BTRFS_I(inode)->root->anon_dev;
8117         stat->blksize = PAGE_CACHE_SIZE;
8118
8119         spin_lock(&BTRFS_I(inode)->lock);
8120         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8121         spin_unlock(&BTRFS_I(inode)->lock);
8122         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8123                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8124         return 0;
8125 }
8126
8127 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8128                            struct inode *new_dir, struct dentry *new_dentry)
8129 {
8130         struct btrfs_trans_handle *trans;
8131         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8132         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8133         struct inode *new_inode = new_dentry->d_inode;
8134         struct inode *old_inode = old_dentry->d_inode;
8135         struct timespec ctime = CURRENT_TIME;
8136         u64 index = 0;
8137         u64 root_objectid;
8138         int ret;
8139         u64 old_ino = btrfs_ino(old_inode);
8140
8141         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8142                 return -EPERM;
8143
8144         /* we only allow rename subvolume link between subvolumes */
8145         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8146                 return -EXDEV;
8147
8148         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8149             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8150                 return -ENOTEMPTY;
8151
8152         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8153             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8154                 return -ENOTEMPTY;
8155
8156
8157         /* check for collisions, even if the  name isn't there */
8158         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8159                              new_dentry->d_name.name,
8160                              new_dentry->d_name.len);
8161
8162         if (ret) {
8163                 if (ret == -EEXIST) {
8164                         /* we shouldn't get
8165                          * eexist without a new_inode */
8166                         if (!new_inode) {
8167                                 WARN_ON(1);
8168                                 return ret;
8169                         }
8170                 } else {
8171                         /* maybe -EOVERFLOW */
8172                         return ret;
8173                 }
8174         }
8175         ret = 0;
8176
8177         /*
8178          * we're using rename to replace one file with another.
8179          * and the replacement file is large.  Start IO on it now so
8180          * we don't add too much work to the end of the transaction
8181          */
8182         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8183             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8184                 filemap_flush(old_inode->i_mapping);
8185
8186         /* close the racy window with snapshot create/destroy ioctl */
8187         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8188                 down_read(&root->fs_info->subvol_sem);
8189         /*
8190          * We want to reserve the absolute worst case amount of items.  So if
8191          * both inodes are subvols and we need to unlink them then that would
8192          * require 4 item modifications, but if they are both normal inodes it
8193          * would require 5 item modifications, so we'll assume their normal
8194          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8195          * should cover the worst case number of items we'll modify.
8196          */
8197         trans = btrfs_start_transaction(root, 11);
8198         if (IS_ERR(trans)) {
8199                 ret = PTR_ERR(trans);
8200                 goto out_notrans;
8201         }
8202
8203         if (dest != root)
8204                 btrfs_record_root_in_trans(trans, dest);
8205
8206         ret = btrfs_set_inode_index(new_dir, &index);
8207         if (ret)
8208                 goto out_fail;
8209
8210         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8211                 /* force full log commit if subvolume involved. */
8212                 root->fs_info->last_trans_log_full_commit = trans->transid;
8213         } else {
8214                 ret = btrfs_insert_inode_ref(trans, dest,
8215                                              new_dentry->d_name.name,
8216                                              new_dentry->d_name.len,
8217                                              old_ino,
8218                                              btrfs_ino(new_dir), index);
8219                 if (ret)
8220                         goto out_fail;
8221                 /*
8222                  * this is an ugly little race, but the rename is required
8223                  * to make sure that if we crash, the inode is either at the
8224                  * old name or the new one.  pinning the log transaction lets
8225                  * us make sure we don't allow a log commit to come in after
8226                  * we unlink the name but before we add the new name back in.
8227                  */
8228                 btrfs_pin_log_trans(root);
8229         }
8230         /*
8231          * make sure the inode gets flushed if it is replacing
8232          * something.
8233          */
8234         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8235                 btrfs_add_ordered_operation(trans, root, old_inode);
8236
8237         inode_inc_iversion(old_dir);
8238         inode_inc_iversion(new_dir);
8239         inode_inc_iversion(old_inode);
8240         old_dir->i_ctime = old_dir->i_mtime = ctime;
8241         new_dir->i_ctime = new_dir->i_mtime = ctime;
8242         old_inode->i_ctime = ctime;
8243
8244         if (old_dentry->d_parent != new_dentry->d_parent)
8245                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8246
8247         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8248                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8249                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8250                                         old_dentry->d_name.name,
8251                                         old_dentry->d_name.len);
8252         } else {
8253                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8254                                         old_dentry->d_inode,
8255                                         old_dentry->d_name.name,
8256                                         old_dentry->d_name.len);
8257                 if (!ret)
8258                         ret = btrfs_update_inode(trans, root, old_inode);
8259         }
8260         if (ret) {
8261                 btrfs_abort_transaction(trans, root, ret);
8262                 goto out_fail;
8263         }
8264
8265         if (new_inode) {
8266                 inode_inc_iversion(new_inode);
8267                 new_inode->i_ctime = CURRENT_TIME;
8268                 if (unlikely(btrfs_ino(new_inode) ==
8269                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8270                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8271                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8272                                                 root_objectid,
8273                                                 new_dentry->d_name.name,
8274                                                 new_dentry->d_name.len);
8275                         BUG_ON(new_inode->i_nlink == 0);
8276                 } else {
8277                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8278                                                  new_dentry->d_inode,
8279                                                  new_dentry->d_name.name,
8280                                                  new_dentry->d_name.len);
8281                 }
8282                 if (!ret && new_inode->i_nlink == 0) {
8283                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8284                         BUG_ON(ret);
8285                 }
8286                 if (ret) {
8287                         btrfs_abort_transaction(trans, root, ret);
8288                         goto out_fail;
8289                 }
8290         }
8291
8292         ret = btrfs_add_link(trans, new_dir, old_inode,
8293                              new_dentry->d_name.name,
8294                              new_dentry->d_name.len, 0, index);
8295         if (ret) {
8296                 btrfs_abort_transaction(trans, root, ret);
8297                 goto out_fail;
8298         }
8299
8300         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8301                 struct dentry *parent = new_dentry->d_parent;
8302                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8303                 btrfs_end_log_trans(root);
8304         }
8305 out_fail:
8306         btrfs_end_transaction(trans, root);
8307 out_notrans:
8308         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8309                 up_read(&root->fs_info->subvol_sem);
8310
8311         return ret;
8312 }
8313
8314 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8315 {
8316         struct btrfs_delalloc_work *delalloc_work;
8317
8318         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8319                                      work);
8320         if (delalloc_work->wait)
8321                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8322         else
8323                 filemap_flush(delalloc_work->inode->i_mapping);
8324
8325         if (delalloc_work->delay_iput)
8326                 btrfs_add_delayed_iput(delalloc_work->inode);
8327         else
8328                 iput(delalloc_work->inode);
8329         complete(&delalloc_work->completion);
8330 }
8331
8332 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8333                                                     int wait, int delay_iput)
8334 {
8335         struct btrfs_delalloc_work *work;
8336
8337         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8338         if (!work)
8339                 return NULL;
8340
8341         init_completion(&work->completion);
8342         INIT_LIST_HEAD(&work->list);
8343         work->inode = inode;
8344         work->wait = wait;
8345         work->delay_iput = delay_iput;
8346         work->work.func = btrfs_run_delalloc_work;
8347
8348         return work;
8349 }
8350
8351 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8352 {
8353         wait_for_completion(&work->completion);
8354         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8355 }
8356
8357 /*
8358  * some fairly slow code that needs optimization. This walks the list
8359  * of all the inodes with pending delalloc and forces them to disk.
8360  */
8361 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8362 {
8363         struct btrfs_inode *binode;
8364         struct inode *inode;
8365         struct btrfs_delalloc_work *work, *next;
8366         struct list_head works;
8367         struct list_head splice;
8368         int ret = 0;
8369
8370         if (root->fs_info->sb->s_flags & MS_RDONLY)
8371                 return -EROFS;
8372
8373         INIT_LIST_HEAD(&works);
8374         INIT_LIST_HEAD(&splice);
8375
8376         spin_lock(&root->fs_info->delalloc_lock);
8377         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8378         while (!list_empty(&splice)) {
8379                 binode = list_entry(splice.next, struct btrfs_inode,
8380                                     delalloc_inodes);
8381
8382                 list_del_init(&binode->delalloc_inodes);
8383
8384                 inode = igrab(&binode->vfs_inode);
8385                 if (!inode) {
8386                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8387                                   &binode->runtime_flags);
8388                         continue;
8389                 }
8390
8391                 list_add_tail(&binode->delalloc_inodes,
8392                               &root->fs_info->delalloc_inodes);
8393                 spin_unlock(&root->fs_info->delalloc_lock);
8394
8395                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8396                 if (unlikely(!work)) {
8397                         ret = -ENOMEM;
8398                         goto out;
8399                 }
8400                 list_add_tail(&work->list, &works);
8401                 btrfs_queue_worker(&root->fs_info->flush_workers,
8402                                    &work->work);
8403
8404                 cond_resched();
8405                 spin_lock(&root->fs_info->delalloc_lock);
8406         }
8407         spin_unlock(&root->fs_info->delalloc_lock);
8408
8409         list_for_each_entry_safe(work, next, &works, list) {
8410                 list_del_init(&work->list);
8411                 btrfs_wait_and_free_delalloc_work(work);
8412         }
8413
8414         /* the filemap_flush will queue IO into the worker threads, but
8415          * we have to make sure the IO is actually started and that
8416          * ordered extents get created before we return
8417          */
8418         atomic_inc(&root->fs_info->async_submit_draining);
8419         while (atomic_read(&root->fs_info->nr_async_submits) ||
8420               atomic_read(&root->fs_info->async_delalloc_pages)) {
8421                 wait_event(root->fs_info->async_submit_wait,
8422                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8423                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8424         }
8425         atomic_dec(&root->fs_info->async_submit_draining);
8426         return 0;
8427 out:
8428         list_for_each_entry_safe(work, next, &works, list) {
8429                 list_del_init(&work->list);
8430                 btrfs_wait_and_free_delalloc_work(work);
8431         }
8432
8433         if (!list_empty_careful(&splice)) {
8434                 spin_lock(&root->fs_info->delalloc_lock);
8435                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8436                 spin_unlock(&root->fs_info->delalloc_lock);
8437         }
8438         return ret;
8439 }
8440
8441 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8442                          const char *symname)
8443 {
8444         struct btrfs_trans_handle *trans;
8445         struct btrfs_root *root = BTRFS_I(dir)->root;
8446         struct btrfs_path *path;
8447         struct btrfs_key key;
8448         struct inode *inode = NULL;
8449         int err;
8450         int drop_inode = 0;
8451         u64 objectid;
8452         u64 index = 0 ;
8453         int name_len;
8454         int datasize;
8455         unsigned long ptr;
8456         struct btrfs_file_extent_item *ei;
8457         struct extent_buffer *leaf;
8458
8459         name_len = strlen(symname) + 1;
8460         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8461                 return -ENAMETOOLONG;
8462
8463         /*
8464          * 2 items for inode item and ref
8465          * 2 items for dir items
8466          * 1 item for xattr if selinux is on
8467          */
8468         trans = btrfs_start_transaction(root, 5);
8469         if (IS_ERR(trans))
8470                 return PTR_ERR(trans);
8471
8472         err = btrfs_find_free_ino(root, &objectid);
8473         if (err)
8474                 goto out_unlock;
8475
8476         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8477                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8478                                 S_IFLNK|S_IRWXUGO, &index);
8479         if (IS_ERR(inode)) {
8480                 err = PTR_ERR(inode);
8481                 goto out_unlock;
8482         }
8483
8484         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8485         if (err) {
8486                 drop_inode = 1;
8487                 goto out_unlock;
8488         }
8489
8490         /*
8491         * If the active LSM wants to access the inode during
8492         * d_instantiate it needs these. Smack checks to see
8493         * if the filesystem supports xattrs by looking at the
8494         * ops vector.
8495         */
8496         inode->i_fop = &btrfs_file_operations;
8497         inode->i_op = &btrfs_file_inode_operations;
8498
8499         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8500         if (err)
8501                 drop_inode = 1;
8502         else {
8503                 inode->i_mapping->a_ops = &btrfs_aops;
8504                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8505                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8506         }
8507         if (drop_inode)
8508                 goto out_unlock;
8509
8510         path = btrfs_alloc_path();
8511         if (!path) {
8512                 err = -ENOMEM;
8513                 drop_inode = 1;
8514                 goto out_unlock;
8515         }
8516         key.objectid = btrfs_ino(inode);
8517         key.offset = 0;
8518         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8519         datasize = btrfs_file_extent_calc_inline_size(name_len);
8520         err = btrfs_insert_empty_item(trans, root, path, &key,
8521                                       datasize);
8522         if (err) {
8523                 drop_inode = 1;
8524                 btrfs_free_path(path);
8525                 goto out_unlock;
8526         }
8527         leaf = path->nodes[0];
8528         ei = btrfs_item_ptr(leaf, path->slots[0],
8529                             struct btrfs_file_extent_item);
8530         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8531         btrfs_set_file_extent_type(leaf, ei,
8532                                    BTRFS_FILE_EXTENT_INLINE);
8533         btrfs_set_file_extent_encryption(leaf, ei, 0);
8534         btrfs_set_file_extent_compression(leaf, ei, 0);
8535         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8536         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8537
8538         ptr = btrfs_file_extent_inline_start(ei);
8539         write_extent_buffer(leaf, symname, ptr, name_len);
8540         btrfs_mark_buffer_dirty(leaf);
8541         btrfs_free_path(path);
8542
8543         inode->i_op = &btrfs_symlink_inode_operations;
8544         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8545         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8546         inode_set_bytes(inode, name_len);
8547         btrfs_i_size_write(inode, name_len - 1);
8548         err = btrfs_update_inode(trans, root, inode);
8549         if (err)
8550                 drop_inode = 1;
8551
8552 out_unlock:
8553         if (!err)
8554                 d_instantiate(dentry, inode);
8555         btrfs_end_transaction(trans, root);
8556         if (drop_inode) {
8557                 inode_dec_link_count(inode);
8558                 iput(inode);
8559         }
8560         btrfs_btree_balance_dirty(root);
8561         return err;
8562 }
8563
8564 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8565                                        u64 start, u64 num_bytes, u64 min_size,
8566                                        loff_t actual_len, u64 *alloc_hint,
8567                                        struct btrfs_trans_handle *trans)
8568 {
8569         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8570         struct extent_map *em;
8571         struct btrfs_root *root = BTRFS_I(inode)->root;
8572         struct btrfs_key ins;
8573         u64 cur_offset = start;
8574         u64 i_size;
8575         u64 cur_bytes;
8576         int ret = 0;
8577         bool own_trans = true;
8578
8579         if (trans)
8580                 own_trans = false;
8581         while (num_bytes > 0) {
8582                 if (own_trans) {
8583                         trans = btrfs_start_transaction(root, 3);
8584                         if (IS_ERR(trans)) {
8585                                 ret = PTR_ERR(trans);
8586                                 break;
8587                         }
8588                 }
8589
8590                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8591                 cur_bytes = max(cur_bytes, min_size);
8592                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8593                                            min_size, 0, *alloc_hint, &ins, 1);
8594                 if (ret) {
8595                         if (own_trans)
8596                                 btrfs_end_transaction(trans, root);
8597                         break;
8598                 }
8599
8600                 ret = insert_reserved_file_extent(trans, inode,
8601                                                   cur_offset, ins.objectid,
8602                                                   ins.offset, ins.offset,
8603                                                   ins.offset, 0, 0, 0,
8604                                                   BTRFS_FILE_EXTENT_PREALLOC);
8605                 if (ret) {
8606                         btrfs_abort_transaction(trans, root, ret);
8607                         if (own_trans)
8608                                 btrfs_end_transaction(trans, root);
8609                         break;
8610                 }
8611                 btrfs_drop_extent_cache(inode, cur_offset,
8612                                         cur_offset + ins.offset -1, 0);
8613
8614                 em = alloc_extent_map();
8615                 if (!em) {
8616                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8617                                 &BTRFS_I(inode)->runtime_flags);
8618                         goto next;
8619                 }
8620
8621                 em->start = cur_offset;
8622                 em->orig_start = cur_offset;
8623                 em->len = ins.offset;
8624                 em->block_start = ins.objectid;
8625                 em->block_len = ins.offset;
8626                 em->orig_block_len = ins.offset;
8627                 em->ram_bytes = ins.offset;
8628                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8629                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8630                 em->generation = trans->transid;
8631
8632                 while (1) {
8633                         write_lock(&em_tree->lock);
8634                         ret = add_extent_mapping(em_tree, em, 1);
8635                         write_unlock(&em_tree->lock);
8636                         if (ret != -EEXIST)
8637                                 break;
8638                         btrfs_drop_extent_cache(inode, cur_offset,
8639                                                 cur_offset + ins.offset - 1,
8640                                                 0);
8641                 }
8642                 free_extent_map(em);
8643 next:
8644                 num_bytes -= ins.offset;
8645                 cur_offset += ins.offset;
8646                 *alloc_hint = ins.objectid + ins.offset;
8647
8648                 inode_inc_iversion(inode);
8649                 inode->i_ctime = CURRENT_TIME;
8650                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8651                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8652                     (actual_len > inode->i_size) &&
8653                     (cur_offset > inode->i_size)) {
8654                         if (cur_offset > actual_len)
8655                                 i_size = actual_len;
8656                         else
8657                                 i_size = cur_offset;
8658                         i_size_write(inode, i_size);
8659                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8660                 }
8661
8662                 ret = btrfs_update_inode(trans, root, inode);
8663
8664                 if (ret) {
8665                         btrfs_abort_transaction(trans, root, ret);
8666                         if (own_trans)
8667                                 btrfs_end_transaction(trans, root);
8668                         break;
8669                 }
8670
8671                 if (own_trans)
8672                         btrfs_end_transaction(trans, root);
8673         }
8674         return ret;
8675 }
8676
8677 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8678                               u64 start, u64 num_bytes, u64 min_size,
8679                               loff_t actual_len, u64 *alloc_hint)
8680 {
8681         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8682                                            min_size, actual_len, alloc_hint,
8683                                            NULL);
8684 }
8685
8686 int btrfs_prealloc_file_range_trans(struct inode *inode,
8687                                     struct btrfs_trans_handle *trans, int mode,
8688                                     u64 start, u64 num_bytes, u64 min_size,
8689                                     loff_t actual_len, u64 *alloc_hint)
8690 {
8691         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8692                                            min_size, actual_len, alloc_hint, trans);
8693 }
8694
8695 static int btrfs_set_page_dirty(struct page *page)
8696 {
8697         return __set_page_dirty_nobuffers(page);
8698 }
8699
8700 static int btrfs_permission(struct inode *inode, int mask)
8701 {
8702         struct btrfs_root *root = BTRFS_I(inode)->root;
8703         umode_t mode = inode->i_mode;
8704
8705         if (mask & MAY_WRITE &&
8706             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8707                 if (btrfs_root_readonly(root))
8708                         return -EROFS;
8709                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8710                         return -EACCES;
8711         }
8712         return generic_permission(inode, mask);
8713 }
8714
8715 static const struct inode_operations btrfs_dir_inode_operations = {
8716         .getattr        = btrfs_getattr,
8717         .lookup         = btrfs_lookup,
8718         .create         = btrfs_create,
8719         .unlink         = btrfs_unlink,
8720         .link           = btrfs_link,
8721         .mkdir          = btrfs_mkdir,
8722         .rmdir          = btrfs_rmdir,
8723         .rename         = btrfs_rename,
8724         .symlink        = btrfs_symlink,
8725         .setattr        = btrfs_setattr,
8726         .mknod          = btrfs_mknod,
8727         .setxattr       = btrfs_setxattr,
8728         .getxattr       = btrfs_getxattr,
8729         .listxattr      = btrfs_listxattr,
8730         .removexattr    = btrfs_removexattr,
8731         .permission     = btrfs_permission,
8732         .get_acl        = btrfs_get_acl,
8733 };
8734 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8735         .lookup         = btrfs_lookup,
8736         .permission     = btrfs_permission,
8737         .get_acl        = btrfs_get_acl,
8738 };
8739
8740 static const struct file_operations btrfs_dir_file_operations = {
8741         .llseek         = generic_file_llseek,
8742         .read           = generic_read_dir,
8743         .readdir        = btrfs_real_readdir,
8744         .unlocked_ioctl = btrfs_ioctl,
8745 #ifdef CONFIG_COMPAT
8746         .compat_ioctl   = btrfs_ioctl,
8747 #endif
8748         .release        = btrfs_release_file,
8749         .fsync          = btrfs_sync_file,
8750 };
8751
8752 static struct extent_io_ops btrfs_extent_io_ops = {
8753         .fill_delalloc = run_delalloc_range,
8754         .submit_bio_hook = btrfs_submit_bio_hook,
8755         .merge_bio_hook = btrfs_merge_bio_hook,
8756         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8757         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8758         .writepage_start_hook = btrfs_writepage_start_hook,
8759         .set_bit_hook = btrfs_set_bit_hook,
8760         .clear_bit_hook = btrfs_clear_bit_hook,
8761         .merge_extent_hook = btrfs_merge_extent_hook,
8762         .split_extent_hook = btrfs_split_extent_hook,
8763 };
8764
8765 /*
8766  * btrfs doesn't support the bmap operation because swapfiles
8767  * use bmap to make a mapping of extents in the file.  They assume
8768  * these extents won't change over the life of the file and they
8769  * use the bmap result to do IO directly to the drive.
8770  *
8771  * the btrfs bmap call would return logical addresses that aren't
8772  * suitable for IO and they also will change frequently as COW
8773  * operations happen.  So, swapfile + btrfs == corruption.
8774  *
8775  * For now we're avoiding this by dropping bmap.
8776  */
8777 static const struct address_space_operations btrfs_aops = {
8778         .readpage       = btrfs_readpage,
8779         .writepage      = btrfs_writepage,
8780         .writepages     = btrfs_writepages,
8781         .readpages      = btrfs_readpages,
8782         .direct_IO      = btrfs_direct_IO,
8783         .invalidatepage = btrfs_invalidatepage,
8784         .releasepage    = btrfs_releasepage,
8785         .set_page_dirty = btrfs_set_page_dirty,
8786         .error_remove_page = generic_error_remove_page,
8787 };
8788
8789 static const struct address_space_operations btrfs_symlink_aops = {
8790         .readpage       = btrfs_readpage,
8791         .writepage      = btrfs_writepage,
8792         .invalidatepage = btrfs_invalidatepage,
8793         .releasepage    = btrfs_releasepage,
8794 };
8795
8796 static const struct inode_operations btrfs_file_inode_operations = {
8797         .getattr        = btrfs_getattr,
8798         .setattr        = btrfs_setattr,
8799         .setxattr       = btrfs_setxattr,
8800         .getxattr       = btrfs_getxattr,
8801         .listxattr      = btrfs_listxattr,
8802         .removexattr    = btrfs_removexattr,
8803         .permission     = btrfs_permission,
8804         .fiemap         = btrfs_fiemap,
8805         .get_acl        = btrfs_get_acl,
8806         .update_time    = btrfs_update_time,
8807 };
8808 static const struct inode_operations btrfs_special_inode_operations = {
8809         .getattr        = btrfs_getattr,
8810         .setattr        = btrfs_setattr,
8811         .permission     = btrfs_permission,
8812         .setxattr       = btrfs_setxattr,
8813         .getxattr       = btrfs_getxattr,
8814         .listxattr      = btrfs_listxattr,
8815         .removexattr    = btrfs_removexattr,
8816         .get_acl        = btrfs_get_acl,
8817         .update_time    = btrfs_update_time,
8818 };
8819 static const struct inode_operations btrfs_symlink_inode_operations = {
8820         .readlink       = generic_readlink,
8821         .follow_link    = page_follow_link_light,
8822         .put_link       = page_put_link,
8823         .getattr        = btrfs_getattr,
8824         .setattr        = btrfs_setattr,
8825         .permission     = btrfs_permission,
8826         .setxattr       = btrfs_setxattr,
8827         .getxattr       = btrfs_getxattr,
8828         .listxattr      = btrfs_listxattr,
8829         .removexattr    = btrfs_removexattr,
8830         .get_acl        = btrfs_get_acl,
8831         .update_time    = btrfs_update_time,
8832 };
8833
8834 const struct dentry_operations btrfs_dentry_operations = {
8835         .d_delete       = btrfs_dentry_delete,
8836         .d_release      = btrfs_dentry_release,
8837 };