fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54 #include "inode-map.h"
  55
  56 struct btrfs_iget_args {
  57         u64 ino;
  58         struct btrfs_root *root;
  59 };
  60
  61 static const struct inode_operations btrfs_dir_inode_operations;
  62 static const struct inode_operations btrfs_symlink_inode_operations;
  63 static const struct inode_operations btrfs_dir_ro_inode_operations;
  64 static const struct inode_operations btrfs_special_inode_operations;
  65 static const struct inode_operations btrfs_file_inode_operations;
  66 static const struct address_space_operations btrfs_aops;
  67 static const struct address_space_operations btrfs_symlink_aops;
  68 static const struct file_operations btrfs_dir_file_operations;
  69 static struct extent_io_ops btrfs_extent_io_ops;
  70
  71 static struct kmem_cache *btrfs_inode_cachep;
  72 struct kmem_cache *btrfs_trans_handle_cachep;
  73 struct kmem_cache *btrfs_transaction_cachep;
  74 struct kmem_cache *btrfs_path_cachep;
  75 struct kmem_cache *btrfs_free_space_cachep;
  76
  77 #define S_SHIFT 12
  78 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  79         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  80         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  81         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  82         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  83         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  84         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  85         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  86 };
  87
  88 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  89 static int btrfs_truncate(struct inode *inode);
  90 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  91 static noinline int cow_file_range(struct inode *inode,
  92                                    struct page *locked_page,
  93                                    u64 start, u64 end, int *page_started,
  94                                    unsigned long *nr_written, int unlock);
  95
  96 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  97                                      struct inode *inode,  struct inode *dir)
  98 {
  99         int err;
 100
 101         err = btrfs_init_acl(trans, inode, dir);
 102         if (!err)
 103                 err = btrfs_xattr_security_init(trans, inode, dir);
 104         return err;
 105 }
 106
 107 /*
 108  * this does all the hard work for inserting an inline extent into
 109  * the btree.  The caller should have done a btrfs_drop_extents so that
 110  * no overlapping inline items exist in the btree
 111  */
 112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 113                                 struct btrfs_root *root, struct inode *inode,
 114                                 u64 start, size_t size, size_t compressed_size,
 115                                 int compress_type,
 116                                 struct page **compressed_pages)
 117 {
 118         struct btrfs_key key;
 119         struct btrfs_path *path;
 120         struct extent_buffer *leaf;
 121         struct page *page = NULL;
 122         char *kaddr;
 123         unsigned long ptr;
 124         struct btrfs_file_extent_item *ei;
 125         int err = 0;
 126         int ret;
 127         size_t cur_size = size;
 128         size_t datasize;
 129         unsigned long offset;
 130
 131         if (compressed_size && compressed_pages)
 132                 cur_size = compressed_size;
 133
 134         path = btrfs_alloc_path();
 135         if (!path)
 136                 return -ENOMEM;
 137
 138         path->leave_spinning = 1;
 139         btrfs_set_trans_block_group(trans, inode);
 140
 141         key.objectid = btrfs_ino(inode);
 142         key.offset = start;
 143         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 144         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 145
 146         inode_add_bytes(inode, size);
 147         ret = btrfs_insert_empty_item(trans, root, path, &key,
 148                                       datasize);
 149         BUG_ON(ret);
 150         if (ret) {
 151                 err = ret;
 152                 goto fail;
 153         }
 154         leaf = path->nodes[0];
 155         ei = btrfs_item_ptr(leaf, path->slots[0],
 156                             struct btrfs_file_extent_item);
 157         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 158         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 159         btrfs_set_file_extent_encryption(leaf, ei, 0);
 160         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 161         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 162         ptr = btrfs_file_extent_inline_start(ei);
 163
 164         if (compress_type != BTRFS_COMPRESS_NONE) {
 165                 struct page *cpage;
 166                 int i = 0;
 167                 while (compressed_size > 0) {
 168                         cpage = compressed_pages[i];
 169                         cur_size = min_t(unsigned long, compressed_size,
 170                                        PAGE_CACHE_SIZE);
 171
 172                         kaddr = kmap_atomic(cpage, KM_USER0);
 173                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 174                         kunmap_atomic(kaddr, KM_USER0);
 175
 176                         i++;
 177                         ptr += cur_size;
 178                         compressed_size -= cur_size;
 179                 }
 180                 btrfs_set_file_extent_compression(leaf, ei,
 181                                                   compress_type);
 182         } else {
 183                 page = find_get_page(inode->i_mapping,
 184                                      start >> PAGE_CACHE_SHIFT);
 185                 btrfs_set_file_extent_compression(leaf, ei, 0);
 186                 kaddr = kmap_atomic(page, KM_USER0);
 187                 offset = start & (PAGE_CACHE_SIZE - 1);
 188                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 189                 kunmap_atomic(kaddr, KM_USER0);
 190                 page_cache_release(page);
 191         }
 192         btrfs_mark_buffer_dirty(leaf);
 193         btrfs_free_path(path);
 194
 195         /*
 196          * we're an inline extent, so nobody can
 197          * extend the file past i_size without locking
 198          * a page we already have locked.
 199          *
 200          * We must do any isize and inode updates
 201          * before we unlock the pages.  Otherwise we
 202          * could end up racing with unlink.
 203          */
 204         BTRFS_I(inode)->disk_i_size = inode->i_size;
 205         btrfs_update_inode(trans, root, inode);
 206
 207         return 0;
 208 fail:
 209         btrfs_free_path(path);
 210         return err;
 211 }
 212
 213
 214 /*
 215  * conditionally insert an inline extent into the file.  This
 216  * does the checks required to make sure the data is small enough
 217  * to fit as an inline extent.
 218  */
 219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 220                                  struct btrfs_root *root,
 221                                  struct inode *inode, u64 start, u64 end,
 222                                  size_t compressed_size, int compress_type,
 223                                  struct page **compressed_pages)
 224 {
 225         u64 isize = i_size_read(inode);
 226         u64 actual_end = min(end + 1, isize);
 227         u64 inline_len = actual_end - start;
 228         u64 aligned_end = (end + root->sectorsize - 1) &
 229                         ~((u64)root->sectorsize - 1);
 230         u64 hint_byte;
 231         u64 data_len = inline_len;
 232         int ret;
 233
 234         if (compressed_size)
 235                 data_len = compressed_size;
 236
 237         if (start > 0 ||
 238             actual_end >= PAGE_CACHE_SIZE ||
 239             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 240             (!compressed_size &&
 241             (actual_end & (root->sectorsize - 1)) == 0) ||
 242             end + 1 < isize ||
 243             data_len > root->fs_info->max_inline) {
 244                 return 1;
 245         }
 246
 247         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 248                                  &hint_byte, 1);
 249         BUG_ON(ret);
 250
 251         if (isize > actual_end)
 252                 inline_len = min_t(u64, isize, actual_end);
 253         ret = insert_inline_extent(trans, root, inode, start,
 254                                    inline_len, compressed_size,
 255                                    compress_type, compressed_pages);
 256         BUG_ON(ret);
 257         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 258         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 259         return 0;
 260 }
 261
 262 struct async_extent {
 263         u64 start;
 264         u64 ram_size;
 265         u64 compressed_size;
 266         struct page **pages;
 267         unsigned long nr_pages;
 268         int compress_type;
 269         struct list_head list;
 270 };
 271
 272 struct async_cow {
 273         struct inode *inode;
 274         struct btrfs_root *root;
 275         struct page *locked_page;
 276         u64 start;
 277         u64 end;
 278         struct list_head extents;
 279         struct btrfs_work work;
 280 };
 281
 282 static noinline int add_async_extent(struct async_cow *cow,
 283                                      u64 start, u64 ram_size,
 284                                      u64 compressed_size,
 285                                      struct page **pages,
 286                                      unsigned long nr_pages,
 287                                      int compress_type)
 288 {
 289         struct async_extent *async_extent;
 290
 291         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 292         BUG_ON(!async_extent);
 293         async_extent->start = start;
 294         async_extent->ram_size = ram_size;
 295         async_extent->compressed_size = compressed_size;
 296         async_extent->pages = pages;
 297         async_extent->nr_pages = nr_pages;
 298         async_extent->compress_type = compress_type;
 299         list_add_tail(&async_extent->list, &cow->extents);
 300         return 0;
 301 }
 302
 303 /*
 304  * we create compressed extents in two phases.  The first
 305  * phase compresses a range of pages that have already been
 306  * locked (both pages and state bits are locked).
 307  *
 308  * This is done inside an ordered work queue, and the compression
 309  * is spread across many cpus.  The actual IO submission is step
 310  * two, and the ordered work queue takes care of making sure that
 311  * happens in the same order things were put onto the queue by
 312  * writepages and friends.
 313  *
 314  * If this code finds it can't get good compression, it puts an
 315  * entry onto the work queue to write the uncompressed bytes.  This
 316  * makes sure that both compressed inodes and uncompressed inodes
 317  * are written in the same order that pdflush sent them down.
 318  */
 319 static noinline int compress_file_range(struct inode *inode,
 320                                         struct page *locked_page,
 321                                         u64 start, u64 end,
 322                                         struct async_cow *async_cow,
 323                                         int *num_added)
 324 {
 325         struct btrfs_root *root = BTRFS_I(inode)->root;
 326         struct btrfs_trans_handle *trans;
 327         u64 num_bytes;
 328         u64 blocksize = root->sectorsize;
 329         u64 actual_end;
 330         u64 isize = i_size_read(inode);
 331         int ret = 0;
 332         struct page **pages = NULL;
 333         unsigned long nr_pages;
 334         unsigned long nr_pages_ret = 0;
 335         unsigned long total_compressed = 0;
 336         unsigned long total_in = 0;
 337         unsigned long max_compressed = 128 * 1024;
 338         unsigned long max_uncompressed = 128 * 1024;
 339         int i;
 340         int will_compress;
 341         int compress_type = root->fs_info->compress_type;
 342
 343         actual_end = min_t(u64, isize, end + 1);
 344 again:
 345         will_compress = 0;
 346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 348
 349         /*
 350          * we don't want to send crud past the end of i_size through
 351          * compression, that's just a waste of CPU time.  So, if the
 352          * end of the file is before the start of our current
 353          * requested range of bytes, we bail out to the uncompressed
 354          * cleanup code that can deal with all of this.
 355          *
 356          * It isn't really the fastest way to fix things, but this is a
 357          * very uncommon corner.
 358          */
 359         if (actual_end <= start)
 360                 goto cleanup_and_bail_uncompressed;
 361
 362         total_compressed = actual_end - start;
 363
 364         /* we want to make sure that amount of ram required to uncompress
 365          * an extent is reasonable, so we limit the total size in ram
 366          * of a compressed extent to 128k.  This is a crucial number
 367          * because it also controls how easily we can spread reads across
 368          * cpus for decompression.
 369          *
 370          * We also want to make sure the amount of IO required to do
 371          * a random read is reasonably small, so we limit the size of
 372          * a compressed extent to 128k.
 373          */
 374         total_compressed = min(total_compressed, max_uncompressed);
 375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 376         num_bytes = max(blocksize,  num_bytes);
 377         total_in = 0;
 378         ret = 0;
 379
 380         /*
 381          * we do compression for mount -o compress and when the
 382          * inode has not been flagged as nocompress.  This flag can
 383          * change at any time if we discover bad compression ratios.
 384          */
 385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 386             (btrfs_test_opt(root, COMPRESS) ||
 387              (BTRFS_I(inode)->force_compress) ||
 388              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 389                 WARN_ON(pages);
 390                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 391                 BUG_ON(!pages);
 392
 393                 if (BTRFS_I(inode)->force_compress)
 394                         compress_type = BTRFS_I(inode)->force_compress;
 395
 396                 ret = btrfs_compress_pages(compress_type,
 397                                            inode->i_mapping, start,
 398                                            total_compressed, pages,
 399                                            nr_pages, &nr_pages_ret,
 400                                            &total_in,
 401                                            &total_compressed,
 402                                            max_compressed);
 403
 404                 if (!ret) {
 405                         unsigned long offset = total_compressed &
 406                                 (PAGE_CACHE_SIZE - 1);
 407                         struct page *page = pages[nr_pages_ret - 1];
 408                         char *kaddr;
 409
 410                         /* zero the tail end of the last page, we might be
 411                          * sending it down to disk
 412                          */
 413                         if (offset) {
 414                                 kaddr = kmap_atomic(page, KM_USER0);
 415                                 memset(kaddr + offset, 0,
 416                                        PAGE_CACHE_SIZE - offset);
 417                                 kunmap_atomic(kaddr, KM_USER0);
 418                         }
 419                         will_compress = 1;
 420                 }
 421         }
 422         if (start == 0) {
 423                 trans = btrfs_join_transaction(root, 1);
 424                 BUG_ON(IS_ERR(trans));
 425                 btrfs_set_trans_block_group(trans, inode);
 426                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 427
 428                 /* lets try to make an inline extent */
 429                 if (ret || total_in < (actual_end - start)) {
 430                         /* we didn't compress the entire range, try
 431                          * to make an uncompressed inline extent.
 432                          */
 433                         ret = cow_file_range_inline(trans, root, inode,
 434                                                     start, end, 0, 0, NULL);
 435                 } else {
 436                         /* try making a compressed inline extent */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end,
 439                                                     total_compressed,
 440                                                     compress_type, pages);
 441                 }
 442                 if (ret == 0) {
 443                         /*
 444                          * inline extent creation worked, we don't need
 445                          * to create any more async work items.  Unlock
 446                          * and free up our temp pages.
 447                          */
 448                         extent_clear_unlock_delalloc(inode,
 449                              &BTRFS_I(inode)->io_tree,
 450                              start, end, NULL,
 451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 452                              EXTENT_CLEAR_DELALLOC |
 453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 454
 455                         btrfs_end_transaction(trans, root);
 456                         goto free_pages_out;
 457                 }
 458                 btrfs_end_transaction(trans, root);
 459         }
 460
 461         if (will_compress) {
 462                 /*
 463                  * we aren't doing an inline extent round the compressed size
 464                  * up to a block size boundary so the allocator does sane
 465                  * things
 466                  */
 467                 total_compressed = (total_compressed + blocksize - 1) &
 468                         ~(blocksize - 1);
 469
 470                 /*
 471                  * one last check to make sure the compression is really a
 472                  * win, compare the page count read with the blocks on disk
 473                  */
 474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 475                         ~(PAGE_CACHE_SIZE - 1);
 476                 if (total_compressed >= total_in) {
 477                         will_compress = 0;
 478                 } else {
 479                         num_bytes = total_in;
 480                 }
 481         }
 482         if (!will_compress && pages) {
 483                 /*
 484                  * the compression code ran but failed to make things smaller,
 485                  * free any pages it allocated and our page pointer array
 486                  */
 487                 for (i = 0; i < nr_pages_ret; i++) {
 488                         WARN_ON(pages[i]->mapping);
 489                         page_cache_release(pages[i]);
 490                 }
 491                 kfree(pages);
 492                 pages = NULL;
 493                 total_compressed = 0;
 494                 nr_pages_ret = 0;
 495
 496                 /* flag the file so we don't compress in the future */
 497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 498                     !(BTRFS_I(inode)->force_compress)) {
 499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 500                 }
 501         }
 502         if (will_compress) {
 503                 *num_added += 1;
 504
 505                 /* the async work queues will take care of doing actual
 506                  * allocation on disk for these compressed pages,
 507                  * and will submit them to the elevator.
 508                  */
 509                 add_async_extent(async_cow, start, num_bytes,
 510                                  total_compressed, pages, nr_pages_ret,
 511                                  compress_type);
 512
 513                 if (start + num_bytes < end) {
 514                         start += num_bytes;
 515                         pages = NULL;
 516                         cond_resched();
 517                         goto again;
 518                 }
 519         } else {
 520 cleanup_and_bail_uncompressed:
 521                 /*
 522                  * No compression, but we still need to write the pages in
 523                  * the file we've been given so far.  redirty the locked
 524                  * page if it corresponds to our extent and set things up
 525                  * for the async work queue to run cow_file_range to do
 526                  * the normal delalloc dance
 527                  */
 528                 if (page_offset(locked_page) >= start &&
 529                     page_offset(locked_page) <= end) {
 530                         __set_page_dirty_nobuffers(locked_page);
 531                         /* unlocked later on in the async handlers */
 532                 }
 533                 add_async_extent(async_cow, start, end - start + 1,
 534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 535                 *num_added += 1;
 536         }
 537
 538 out:
 539         return 0;
 540
 541 free_pages_out:
 542         for (i = 0; i < nr_pages_ret; i++) {
 543                 WARN_ON(pages[i]->mapping);
 544                 page_cache_release(pages[i]);
 545         }
 546         kfree(pages);
 547
 548         goto out;
 549 }
 550
 551 /*
 552  * phase two of compressed writeback.  This is the ordered portion
 553  * of the code, which only gets called in the order the work was
 554  * queued.  We walk all the async extents created by compress_file_range
 555  * and send them down to the disk.
 556  */
 557 static noinline int submit_compressed_extents(struct inode *inode,
 558                                               struct async_cow *async_cow)
 559 {
 560         struct async_extent *async_extent;
 561         u64 alloc_hint = 0;
 562         struct btrfs_trans_handle *trans;
 563         struct btrfs_key ins;
 564         struct extent_map *em;
 565         struct btrfs_root *root = BTRFS_I(inode)->root;
 566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 567         struct extent_io_tree *io_tree;
 568         int ret = 0;
 569
 570         if (list_empty(&async_cow->extents))
 571                 return 0;
 572
 573
 574         while (!list_empty(&async_cow->extents)) {
 575                 async_extent = list_entry(async_cow->extents.next,
 576                                           struct async_extent, list);
 577                 list_del(&async_extent->list);
 578
 579                 io_tree = &BTRFS_I(inode)->io_tree;
 580
 581 retry:
 582                 /* did the compression code fall back to uncompressed IO? */
 583                 if (!async_extent->pages) {
 584                         int page_started = 0;
 585                         unsigned long nr_written = 0;
 586
 587                         lock_extent(io_tree, async_extent->start,
 588                                          async_extent->start +
 589                                          async_extent->ram_size - 1, GFP_NOFS);
 590
 591                         /* allocate blocks */
 592                         ret = cow_file_range(inode, async_cow->locked_page,
 593                                              async_extent->start,
 594                                              async_extent->start +
 595                                              async_extent->ram_size - 1,
 596                                              &page_started, &nr_written, 0);
 597
 598                         /*
 599                          * if page_started, cow_file_range inserted an
 600                          * inline extent and took care of all the unlocking
 601                          * and IO for us.  Otherwise, we need to submit
 602                          * all those pages down to the drive.
 603                          */
 604                         if (!page_started && !ret)
 605                                 extent_write_locked_range(io_tree,
 606                                                   inode, async_extent->start,
 607                                                   async_extent->start +
 608                                                   async_extent->ram_size - 1,
 609                                                   btrfs_get_extent,
 610                                                   WB_SYNC_ALL);
 611                         kfree(async_extent);
 612                         cond_resched();
 613                         continue;
 614                 }
 615
 616                 lock_extent(io_tree, async_extent->start,
 617                             async_extent->start + async_extent->ram_size - 1,
 618                             GFP_NOFS);
 619
 620                 trans = btrfs_join_transaction(root, 1);
 621                 BUG_ON(IS_ERR(trans));
 622                 ret = btrfs_reserve_extent(trans, root,
 623                                            async_extent->compressed_size,
 624                                            async_extent->compressed_size,
 625                                            0, alloc_hint,
 626                                            (u64)-1, &ins, 1);
 627                 btrfs_end_transaction(trans, root);
 628
 629                 if (ret) {
 630                         int i;
 631                         for (i = 0; i < async_extent->nr_pages; i++) {
 632                                 WARN_ON(async_extent->pages[i]->mapping);
 633                                 page_cache_release(async_extent->pages[i]);
 634                         }
 635                         kfree(async_extent->pages);
 636                         async_extent->nr_pages = 0;
 637                         async_extent->pages = NULL;
 638                         unlock_extent(io_tree, async_extent->start,
 639                                       async_extent->start +
 640                                       async_extent->ram_size - 1, GFP_NOFS);
 641                         goto retry;
 642                 }
 643
 644                 /*
 645                  * here we're doing allocation and writeback of the
 646                  * compressed pages
 647                  */
 648                 btrfs_drop_extent_cache(inode, async_extent->start,
 649                                         async_extent->start +
 650                                         async_extent->ram_size - 1, 0);
 651
 652                 em = alloc_extent_map(GFP_NOFS);
 653                 BUG_ON(!em);
 654                 em->start = async_extent->start;
 655                 em->len = async_extent->ram_size;
 656                 em->orig_start = em->start;
 657
 658                 em->block_start = ins.objectid;
 659                 em->block_len = ins.offset;
 660                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 661                 em->compress_type = async_extent->compress_type;
 662                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 663                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 664
 665                 while (1) {
 666                         write_lock(&em_tree->lock);
 667                         ret = add_extent_mapping(em_tree, em);
 668                         write_unlock(&em_tree->lock);
 669                         if (ret != -EEXIST) {
 670                                 free_extent_map(em);
 671                                 break;
 672                         }
 673                         btrfs_drop_extent_cache(inode, async_extent->start,
 674                                                 async_extent->start +
 675                                                 async_extent->ram_size - 1, 0);
 676                 }
 677
 678                 ret = btrfs_add_ordered_extent_compress(inode,
 679                                                 async_extent->start,
 680                                                 ins.objectid,
 681                                                 async_extent->ram_size,
 682                                                 ins.offset,
 683                                                 BTRFS_ORDERED_COMPRESSED,
 684                                                 async_extent->compress_type);
 685                 BUG_ON(ret);
 686
 687                 /*
 688                  * clear dirty, set writeback and unlock the pages.
 689                  */
 690                 extent_clear_unlock_delalloc(inode,
 691                                 &BTRFS_I(inode)->io_tree,
 692                                 async_extent->start,
 693                                 async_extent->start +
 694                                 async_extent->ram_size - 1,
 695                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 696                                 EXTENT_CLEAR_UNLOCK |
 697                                 EXTENT_CLEAR_DELALLOC |
 698                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 699
 700                 ret = btrfs_submit_compressed_write(inode,
 701                                     async_extent->start,
 702                                     async_extent->ram_size,
 703                                     ins.objectid,
 704                                     ins.offset, async_extent->pages,
 705                                     async_extent->nr_pages);
 706
 707                 BUG_ON(ret);
 708                 alloc_hint = ins.objectid + ins.offset;
 709                 kfree(async_extent);
 710                 cond_resched();
 711         }
 712
 713         return 0;
 714 }
 715
 716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 717                                       u64 num_bytes)
 718 {
 719         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 720         struct extent_map *em;
 721         u64 alloc_hint = 0;
 722
 723         read_lock(&em_tree->lock);
 724         em = search_extent_mapping(em_tree, start, num_bytes);
 725         if (em) {
 726                 /*
 727                  * if block start isn't an actual block number then find the
 728                  * first block in this inode and use that as a hint.  If that
 729                  * block is also bogus then just don't worry about it.
 730                  */
 731                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 732                         free_extent_map(em);
 733                         em = search_extent_mapping(em_tree, 0, 0);
 734                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 735                                 alloc_hint = em->block_start;
 736                         if (em)
 737                                 free_extent_map(em);
 738                 } else {
 739                         alloc_hint = em->block_start;
 740                         free_extent_map(em);
 741                 }
 742         }
 743         read_unlock(&em_tree->lock);
 744
 745         return alloc_hint;
 746 }
 747
 748 /*
 749  * when extent_io.c finds a delayed allocation range in the file,
 750  * the call backs end up in this code.  The basic idea is to
 751  * allocate extents on disk for the range, and create ordered data structs
 752  * in ram to track those extents.
 753  *
 754  * locked_page is the page that writepage had locked already.  We use
 755  * it to make sure we don't do extra locks or unlocks.
 756  *
 757  * *page_started is set to one if we unlock locked_page and do everything
 758  * required to start IO on it.  It may be clean and already done with
 759  * IO when we return.
 760  */
 761 static noinline int cow_file_range(struct inode *inode,
 762                                    struct page *locked_page,
 763                                    u64 start, u64 end, int *page_started,
 764                                    unsigned long *nr_written,
 765                                    int unlock)
 766 {
 767         struct btrfs_root *root = BTRFS_I(inode)->root;
 768         struct btrfs_trans_handle *trans;
 769         u64 alloc_hint = 0;
 770         u64 num_bytes;
 771         unsigned long ram_size;
 772         u64 disk_num_bytes;
 773         u64 cur_alloc_size;
 774         u64 blocksize = root->sectorsize;
 775         struct btrfs_key ins;
 776         struct extent_map *em;
 777         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 778         int ret = 0;
 779
 780         BUG_ON(root == root->fs_info->tree_root);
 781         trans = btrfs_join_transaction(root, 1);
 782         BUG_ON(IS_ERR(trans));
 783         btrfs_set_trans_block_group(trans, inode);
 784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 785
 786         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 787         num_bytes = max(blocksize,  num_bytes);
 788         disk_num_bytes = num_bytes;
 789         ret = 0;
 790
 791         if (start == 0) {
 792                 /* lets try to make an inline extent */
 793                 ret = cow_file_range_inline(trans, root, inode,
 794                                             start, end, 0, 0, NULL);
 795                 if (ret == 0) {
 796                         extent_clear_unlock_delalloc(inode,
 797                                      &BTRFS_I(inode)->io_tree,
 798                                      start, end, NULL,
 799                                      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
 806                         *nr_written = *nr_written +
 807                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 808                         *page_started = 1;
 809                         ret = 0;
 810                         goto out;
 811                 }
 812         }
 813
 814         BUG_ON(disk_num_bytes >
 815                btrfs_super_total_bytes(&root->fs_info->super_copy));
 816
 817         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 818         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 819
 820         while (disk_num_bytes > 0) {
 821                 unsigned long op;
 822
 823                 cur_alloc_size = disk_num_bytes;
 824                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 825                                            root->sectorsize, 0, alloc_hint,
 826                                            (u64)-1, &ins, 1);
 827                 BUG_ON(ret);
 828
 829                 em = alloc_extent_map(GFP_NOFS);
 830                 BUG_ON(!em);
 831                 em->start = start;
 832                 em->orig_start = em->start;
 833                 ram_size = ins.offset;
 834                 em->len = ins.offset;
 835
 836                 em->block_start = ins.objectid;
 837                 em->block_len = ins.offset;
 838                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 839                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 840
 841                 while (1) {
 842                         write_lock(&em_tree->lock);
 843                         ret = add_extent_mapping(em_tree, em);
 844                         write_unlock(&em_tree->lock);
 845                         if (ret != -EEXIST) {
 846                                 free_extent_map(em);
 847                                 break;
 848                         }
 849                         btrfs_drop_extent_cache(inode, start,
 850                                                 start + ram_size - 1, 0);
 851                 }
 852
 853                 cur_alloc_size = ins.offset;
 854                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 855                                                ram_size, cur_alloc_size, 0);
 856                 BUG_ON(ret);
 857
 858                 if (root->root_key.objectid ==
 859                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 860                         ret = btrfs_reloc_clone_csums(inode, start,
 861                                                       cur_alloc_size);
 862                         BUG_ON(ret);
 863                 }
 864
 865                 if (disk_num_bytes < cur_alloc_size)
 866                         break;
 867
 868                 /* we're not doing compressed IO, don't unlock the first
 869                  * page (which the caller expects to stay locked), don't
 870                  * clear any dirty bits and don't set any writeback bits
 871                  *
 872                  * Do set the Private2 bit so we know this page was properly
 873                  * setup for writepage
 874                  */
 875                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 876                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 877                         EXTENT_SET_PRIVATE2;
 878
 879                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 880                                              start, start + ram_size - 1,
 881                                              locked_page, op);
 882                 disk_num_bytes -= cur_alloc_size;
 883                 num_bytes -= cur_alloc_size;
 884                 alloc_hint = ins.objectid + ins.offset;
 885                 start += cur_alloc_size;
 886         }
 887 out:
 888         ret = 0;
 889         btrfs_end_transaction(trans, root);
 890
 891         return ret;
 892 }
 893
 894 /*
 895  * work queue call back to started compression on a file and pages
 896  */
 897 static noinline void async_cow_start(struct btrfs_work *work)
 898 {
 899         struct async_cow *async_cow;
 900         int num_added = 0;
 901         async_cow = container_of(work, struct async_cow, work);
 902
 903         compress_file_range(async_cow->inode, async_cow->locked_page,
 904                             async_cow->start, async_cow->end, async_cow,
 905                             &num_added);
 906         if (num_added == 0)
 907                 async_cow->inode = NULL;
 908 }
 909
 910 /*
 911  * work queue call back to submit previously compressed pages
 912  */
 913 static noinline void async_cow_submit(struct btrfs_work *work)
 914 {
 915         struct async_cow *async_cow;
 916         struct btrfs_root *root;
 917         unsigned long nr_pages;
 918
 919         async_cow = container_of(work, struct async_cow, work);
 920
 921         root = async_cow->root;
 922         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 923                 PAGE_CACHE_SHIFT;
 924
 925         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 926
 927         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 928             5 * 1042 * 1024 &&
 929             waitqueue_active(&root->fs_info->async_submit_wait))
 930                 wake_up(&root->fs_info->async_submit_wait);
 931
 932         if (async_cow->inode)
 933                 submit_compressed_extents(async_cow->inode, async_cow);
 934 }
 935
 936 static noinline void async_cow_free(struct btrfs_work *work)
 937 {
 938         struct async_cow *async_cow;
 939         async_cow = container_of(work, struct async_cow, work);
 940         kfree(async_cow);
 941 }
 942
 943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 944                                 u64 start, u64 end, int *page_started,
 945                                 unsigned long *nr_written)
 946 {
 947         struct async_cow *async_cow;
 948         struct btrfs_root *root = BTRFS_I(inode)->root;
 949         unsigned long nr_pages;
 950         u64 cur_end;
 951         int limit = 10 * 1024 * 1042;
 952
 953         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 954                          1, 0, NULL, GFP_NOFS);
 955         while (start < end) {
 956                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 957                 async_cow->inode = inode;
 958                 async_cow->root = root;
 959                 async_cow->locked_page = locked_page;
 960                 async_cow->start = start;
 961
 962                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 963                         cur_end = end;
 964                 else
 965                         cur_end = min(end, start + 512 * 1024 - 1);
 966
 967                 async_cow->end = cur_end;
 968                 INIT_LIST_HEAD(&async_cow->extents);
 969
 970                 async_cow->work.func = async_cow_start;
 971                 async_cow->work.ordered_func = async_cow_submit;
 972                 async_cow->work.ordered_free = async_cow_free;
 973                 async_cow->work.flags = 0;
 974
 975                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 976                         PAGE_CACHE_SHIFT;
 977                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 978
 979                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 980                                    &async_cow->work);
 981
 982                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 983                         wait_event(root->fs_info->async_submit_wait,
 984                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 985                             limit));
 986                 }
 987
 988                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 989                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 990                         wait_event(root->fs_info->async_submit_wait,
 991                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 992                            0));
 993                 }
 994
 995                 *nr_written += nr_pages;
 996                 start = cur_end + 1;
 997         }
 998         *page_started = 1;
 999         return 0;
1000 }
1001
1002 static noinline int csum_exist_in_range(struct btrfs_root *root,
1003                                         u64 bytenr, u64 num_bytes)
1004 {
1005         int ret;
1006         struct btrfs_ordered_sum *sums;
1007         LIST_HEAD(list);
1008
1009         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1010                                        bytenr + num_bytes - 1, &list);
1011         if (ret == 0 && list_empty(&list))
1012                 return 0;
1013
1014         while (!list_empty(&list)) {
1015                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1016                 list_del(&sums->list);
1017                 kfree(sums);
1018         }
1019         return 1;
1020 }
1021
1022 /*
1023  * when nowcow writeback call back.  This checks for snapshots or COW copies
1024  * of the extents that exist in the file, and COWs the file as required.
1025  *
1026  * If no cow copies or snapshots exist, we write directly to the existing
1027  * blocks on disk
1028  */
1029 static noinline int run_delalloc_nocow(struct inode *inode,
1030                                        struct page *locked_page,
1031                               u64 start, u64 end, int *page_started, int force,
1032                               unsigned long *nr_written)
1033 {
1034         struct btrfs_root *root = BTRFS_I(inode)->root;
1035         struct btrfs_trans_handle *trans;
1036         struct extent_buffer *leaf;
1037         struct btrfs_path *path;
1038         struct btrfs_file_extent_item *fi;
1039         struct btrfs_key found_key;
1040         u64 cow_start;
1041         u64 cur_offset;
1042         u64 extent_end;
1043         u64 extent_offset;
1044         u64 disk_bytenr;
1045         u64 num_bytes;
1046         int extent_type;
1047         int ret;
1048         int type;
1049         int nocow;
1050         int check_prev = 1;
1051         bool nolock = false;
1052         u64 ino = btrfs_ino(inode);
1053
1054         path = btrfs_alloc_path();
1055         BUG_ON(!path);
1056         if (root == root->fs_info->tree_root) {
1057                 nolock = true;
1058                 trans = btrfs_join_transaction_nolock(root, 1);
1059         } else {
1060                 trans = btrfs_join_transaction(root, 1);
1061         }
1062         BUG_ON(IS_ERR(trans));
1063
1064         cow_start = (u64)-1;
1065         cur_offset = start;
1066         while (1) {
1067                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1068                                                cur_offset, 0);
1069                 BUG_ON(ret < 0);
1070                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071                         leaf = path->nodes[0];
1072                         btrfs_item_key_to_cpu(leaf, &found_key,
1073                                               path->slots[0] - 1);
1074                         if (found_key.objectid == ino &&
1075                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1076                                 path->slots[0]--;
1077                 }
1078                 check_prev = 0;
1079 next_slot:
1080                 leaf = path->nodes[0];
1081                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082                         ret = btrfs_next_leaf(root, path);
1083                         if (ret < 0)
1084                                 BUG_ON(1);
1085                         if (ret > 0)
1086                                 break;
1087                         leaf = path->nodes[0];
1088                 }
1089
1090                 nocow = 0;
1091                 disk_bytenr = 0;
1092                 num_bytes = 0;
1093                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094
1095                 if (found_key.objectid > ino ||
1096                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097                     found_key.offset > end)
1098                         break;
1099
1100                 if (found_key.offset > cur_offset) {
1101                         extent_end = found_key.offset;
1102                         extent_type = 0;
1103                         goto out_check;
1104                 }
1105
1106                 fi = btrfs_item_ptr(leaf, path->slots[0],
1107                                     struct btrfs_file_extent_item);
1108                 extent_type = btrfs_file_extent_type(leaf, fi);
1109
1110                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1114                         extent_end = found_key.offset +
1115                                 btrfs_file_extent_num_bytes(leaf, fi);
1116                         if (extent_end <= start) {
1117                                 path->slots[0]++;
1118                                 goto next_slot;
1119                         }
1120                         if (disk_bytenr == 0)
1121                                 goto out_check;
1122                         if (btrfs_file_extent_compression(leaf, fi) ||
1123                             btrfs_file_extent_encryption(leaf, fi) ||
1124                             btrfs_file_extent_other_encoding(leaf, fi))
1125                                 goto out_check;
1126                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127                                 goto out_check;
1128                         if (btrfs_extent_readonly(root, disk_bytenr))
1129                                 goto out_check;
1130                         if (btrfs_cross_ref_exist(trans, root, ino,
1131                                                   found_key.offset -
1132                                                   extent_offset, disk_bytenr))
1133                                 goto out_check;
1134                         disk_bytenr += extent_offset;
1135                         disk_bytenr += cur_offset - found_key.offset;
1136                         num_bytes = min(end + 1, extent_end) - cur_offset;
1137                         /*
1138                          * force cow if csum exists in the range.
1139                          * this ensure that csum for a given extent are
1140                          * either valid or do not exist.
1141                          */
1142                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143                                 goto out_check;
1144                         nocow = 1;
1145                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146                         extent_end = found_key.offset +
1147                                 btrfs_file_extent_inline_len(leaf, fi);
1148                         extent_end = ALIGN(extent_end, root->sectorsize);
1149                 } else {
1150                         BUG_ON(1);
1151                 }
1152 out_check:
1153                 if (extent_end <= start) {
1154                         path->slots[0]++;
1155                         goto next_slot;
1156                 }
1157                 if (!nocow) {
1158                         if (cow_start == (u64)-1)
1159                                 cow_start = cur_offset;
1160                         cur_offset = extent_end;
1161                         if (cur_offset > end)
1162                                 break;
1163                         path->slots[0]++;
1164                         goto next_slot;
1165                 }
1166
1167                 btrfs_release_path(root, path);
1168                 if (cow_start != (u64)-1) {
1169                         ret = cow_file_range(inode, locked_page, cow_start,
1170                                         found_key.offset - 1, page_started,
1171                                         nr_written, 1);
1172                         BUG_ON(ret);
1173                         cow_start = (u64)-1;
1174                 }
1175
1176                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177                         struct extent_map *em;
1178                         struct extent_map_tree *em_tree;
1179                         em_tree = &BTRFS_I(inode)->extent_tree;
1180                         em = alloc_extent_map(GFP_NOFS);
1181                         BUG_ON(!em);
1182                         em->start = cur_offset;
1183                         em->orig_start = em->start;
1184                         em->len = num_bytes;
1185                         em->block_len = num_bytes;
1186                         em->block_start = disk_bytenr;
1187                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1188                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189                         while (1) {
1190                                 write_lock(&em_tree->lock);
1191                                 ret = add_extent_mapping(em_tree, em);
1192                                 write_unlock(&em_tree->lock);
1193                                 if (ret != -EEXIST) {
1194                                         free_extent_map(em);
1195                                         break;
1196                                 }
1197                                 btrfs_drop_extent_cache(inode, em->start,
1198                                                 em->start + em->len - 1, 0);
1199                         }
1200                         type = BTRFS_ORDERED_PREALLOC;
1201                 } else {
1202                         type = BTRFS_ORDERED_NOCOW;
1203                 }
1204
1205                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206                                                num_bytes, num_bytes, type);
1207                 BUG_ON(ret);
1208
1209                 if (root->root_key.objectid ==
1210                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1212                                                       num_bytes);
1213                         BUG_ON(ret);
1214                 }
1215
1216                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217                                 cur_offset, cur_offset + num_bytes - 1,
1218                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220                                 EXTENT_SET_PRIVATE2);
1221                 cur_offset = extent_end;
1222                 if (cur_offset > end)
1223                         break;
1224         }
1225         btrfs_release_path(root, path);
1226
1227         if (cur_offset <= end && cow_start == (u64)-1)
1228                 cow_start = cur_offset;
1229         if (cow_start != (u64)-1) {
1230                 ret = cow_file_range(inode, locked_page, cow_start, end,
1231                                      page_started, nr_written, 1);
1232                 BUG_ON(ret);
1233         }
1234
1235         if (nolock) {
1236                 ret = btrfs_end_transaction_nolock(trans, root);
1237                 BUG_ON(ret);
1238         } else {
1239                 ret = btrfs_end_transaction(trans, root);
1240                 BUG_ON(ret);
1241         }
1242         btrfs_free_path(path);
1243         return 0;
1244 }
1245
1246 /*
1247  * extent_io.c call back to do delayed allocation processing
1248  */
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250                               u64 start, u64 end, int *page_started,
1251                               unsigned long *nr_written)
1252 {
1253         int ret;
1254         struct btrfs_root *root = BTRFS_I(inode)->root;
1255
1256         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258                                          page_started, 1, nr_written);
1259         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261                                          page_started, 0, nr_written);
1262         else if (!btrfs_test_opt(root, COMPRESS) &&
1263                  !(BTRFS_I(inode)->force_compress) &&
1264                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265                 ret = cow_file_range(inode, locked_page, start, end,
1266                                       page_started, nr_written, 1);
1267         else
1268                 ret = cow_file_range_async(inode, locked_page, start, end,
1269                                            page_started, nr_written);
1270         return ret;
1271 }
1272
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274                                    struct extent_state *orig, u64 split)
1275 {
1276         /* not delalloc, ignore it */
1277         if (!(orig->state & EXTENT_DELALLOC))
1278                 return 0;
1279
1280         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1281         return 0;
1282 }
1283
1284 /*
1285  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286  * extents so we can keep track of new extents that are just merged onto old
1287  * extents, such as when we are doing sequential writes, so we can properly
1288  * account for the metadata space we'll need.
1289  */
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291                                    struct extent_state *new,
1292                                    struct extent_state *other)
1293 {
1294         /* not delalloc, ignore it */
1295         if (!(other->state & EXTENT_DELALLOC))
1296                 return 0;
1297
1298         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1299         return 0;
1300 }
1301
1302 /*
1303  * extent_io.c set_bit_hook, used to track delayed allocation
1304  * bytes in this file, and to maintain the list of inodes that
1305  * have pending delalloc work to be done.
1306  */
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308                               struct extent_state *state, int *bits)
1309 {
1310
1311         /*
1312          * set_bit and clear bit hooks normally require _irqsave/restore
1313          * but in this case, we are only testeing for the DELALLOC
1314          * bit, which is only set or cleared with irqs on
1315          */
1316         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317                 struct btrfs_root *root = BTRFS_I(inode)->root;
1318                 u64 len = state->end + 1 - state->start;
1319                 int do_list = (root->root_key.objectid !=
1320                                BTRFS_ROOT_TREE_OBJECTID);
1321
1322                 if (*bits & EXTENT_FIRST_DELALLOC)
1323                         *bits &= ~EXTENT_FIRST_DELALLOC;
1324                 else
1325                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326
1327                 spin_lock(&root->fs_info->delalloc_lock);
1328                 BTRFS_I(inode)->delalloc_bytes += len;
1329                 root->fs_info->delalloc_bytes += len;
1330                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332                                       &root->fs_info->delalloc_inodes);
1333                 }
1334                 spin_unlock(&root->fs_info->delalloc_lock);
1335         }
1336         return 0;
1337 }
1338
1339 /*
1340  * extent_io.c clear_bit_hook, see set_bit_hook for why
1341  */
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343                                 struct extent_state *state, int *bits)
1344 {
1345         /*
1346          * set_bit and clear bit hooks normally require _irqsave/restore
1347          * but in this case, we are only testeing for the DELALLOC
1348          * bit, which is only set or cleared with irqs on
1349          */
1350         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351                 struct btrfs_root *root = BTRFS_I(inode)->root;
1352                 u64 len = state->end + 1 - state->start;
1353                 int do_list = (root->root_key.objectid !=
1354                                BTRFS_ROOT_TREE_OBJECTID);
1355
1356                 if (*bits & EXTENT_FIRST_DELALLOC)
1357                         *bits &= ~EXTENT_FIRST_DELALLOC;
1358                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360
1361                 if (*bits & EXTENT_DO_ACCOUNTING)
1362                         btrfs_delalloc_release_metadata(inode, len);
1363
1364                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365                     && do_list)
1366                         btrfs_free_reserved_data_space(inode, len);
1367
1368                 spin_lock(&root->fs_info->delalloc_lock);
1369                 root->fs_info->delalloc_bytes -= len;
1370                 BTRFS_I(inode)->delalloc_bytes -= len;
1371
1372                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375                 }
1376                 spin_unlock(&root->fs_info->delalloc_lock);
1377         }
1378         return 0;
1379 }
1380
1381 /*
1382  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383  * we don't create bios that span stripes or chunks
1384  */
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386                          size_t size, struct bio *bio,
1387                          unsigned long bio_flags)
1388 {
1389         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390         struct btrfs_mapping_tree *map_tree;
1391         u64 logical = (u64)bio->bi_sector << 9;
1392         u64 length = 0;
1393         u64 map_length;
1394         int ret;
1395
1396         if (bio_flags & EXTENT_BIO_COMPRESSED)
1397                 return 0;
1398
1399         length = bio->bi_size;
1400         map_tree = &root->fs_info->mapping_tree;
1401         map_length = length;
1402         ret = btrfs_map_block(map_tree, READ, logical,
1403                               &map_length, NULL, 0);
1404
1405         if (map_length < length + size)
1406                 return 1;
1407         return ret;
1408 }
1409
1410 /*
1411  * in order to insert checksums into the metadata in large chunks,
1412  * we wait until bio submission time.   All the pages in the bio are
1413  * checksummed and sums are attached onto the ordered extent record.
1414  *
1415  * At IO completion time the cums attached on the ordered extent record
1416  * are inserted into the btree
1417  */
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419                                     struct bio *bio, int mirror_num,
1420                                     unsigned long bio_flags,
1421                                     u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425
1426         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1427         BUG_ON(ret);
1428         return 0;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440                           int mirror_num, unsigned long bio_flags,
1441                           u64 bio_offset)
1442 {
1443         struct btrfs_root *root = BTRFS_I(inode)->root;
1444         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1445 }
1446
1447 /*
1448  * extent_io.c submission hook. This does the right thing for csum calculation
1449  * on write, or reading the csums from the tree before a read
1450  */
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452                           int mirror_num, unsigned long bio_flags,
1453                           u64 bio_offset)
1454 {
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456         int ret = 0;
1457         int skip_sum;
1458
1459         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460
1461         if (root == root->fs_info->tree_root)
1462                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463         else
1464                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465         BUG_ON(ret);
1466
1467         if (!(rw & REQ_WRITE)) {
1468                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469                         return btrfs_submit_compressed_read(inode, bio,
1470                                                     mirror_num, bio_flags);
1471                 } else if (!skip_sum) {
1472                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1473                         if (ret)
1474                                 return ret;
1475                 }
1476                 goto mapit;
1477         } else if (!skip_sum) {
1478                 /* csum items have already been cloned */
1479                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480                         goto mapit;
1481                 /* we're doing a write, do the async checksumming */
1482                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483                                    inode, rw, bio, mirror_num,
1484                                    bio_flags, bio_offset,
1485                                    __btrfs_submit_bio_start,
1486                                    __btrfs_submit_bio_done);
1487         }
1488
1489 mapit:
1490         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1491 }
1492
1493 /*
1494  * given a list of ordered sums record them in the inode.  This happens
1495  * at IO completion time based on sums calculated at bio submission time.
1496  */
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498                              struct inode *inode, u64 file_offset,
1499                              struct list_head *list)
1500 {
1501         struct btrfs_ordered_sum *sum;
1502
1503         btrfs_set_trans_block_group(trans, inode);
1504
1505         list_for_each_entry(sum, list, list) {
1506                 btrfs_csum_file_blocks(trans,
1507                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1508         }
1509         return 0;
1510 }
1511
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513                               struct extent_state **cached_state)
1514 {
1515         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1516                 WARN_ON(1);
1517         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518                                    cached_state, GFP_NOFS);
1519 }
1520
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1523         struct page *page;
1524         struct btrfs_work work;
1525 };
1526
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1528 {
1529         struct btrfs_writepage_fixup *fixup;
1530         struct btrfs_ordered_extent *ordered;
1531         struct extent_state *cached_state = NULL;
1532         struct page *page;
1533         struct inode *inode;
1534         u64 page_start;
1535         u64 page_end;
1536
1537         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538         page = fixup->page;
1539 again:
1540         lock_page(page);
1541         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542                 ClearPageChecked(page);
1543                 goto out_page;
1544         }
1545
1546         inode = page->mapping->host;
1547         page_start = page_offset(page);
1548         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1549
1550         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551                          &cached_state, GFP_NOFS);
1552
1553         /* already ordered? We're done */
1554         if (PagePrivate2(page))
1555                 goto out;
1556
1557         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1558         if (ordered) {
1559                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560                                      page_end, &cached_state, GFP_NOFS);
1561                 unlock_page(page);
1562                 btrfs_start_ordered_extent(inode, ordered, 1);
1563                 goto again;
1564         }
1565
1566         BUG();
1567         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568         ClearPageChecked(page);
1569 out:
1570         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571                              &cached_state, GFP_NOFS);
1572 out_page:
1573         unlock_page(page);
1574         page_cache_release(page);
1575         kfree(fixup);
1576 }
1577
1578 /*
1579  * There are a few paths in the higher layers of the kernel that directly
1580  * set the page dirty bit without asking the filesystem if it is a
1581  * good idea.  This causes problems because we want to make sure COW
1582  * properly happens and the data=ordered rules are followed.
1583  *
1584  * In our case any range that doesn't have the ORDERED bit set
1585  * hasn't been properly setup for IO.  We kick off an async process
1586  * to fix it up.  The async helper will wait for ordered extents, set
1587  * the delalloc bit and make it safe to write the page.
1588  */
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1590 {
1591         struct inode *inode = page->mapping->host;
1592         struct btrfs_writepage_fixup *fixup;
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594
1595         /* this page is properly in the ordered list */
1596         if (TestClearPagePrivate2(page))
1597                 return 0;
1598
1599         if (PageChecked(page))
1600                 return -EAGAIN;
1601
1602         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1603         if (!fixup)
1604                 return -EAGAIN;
1605
1606         SetPageChecked(page);
1607         page_cache_get(page);
1608         fixup->work.func = btrfs_writepage_fixup_worker;
1609         fixup->page = page;
1610         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1611         return -EAGAIN;
1612 }
1613
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615                                        struct inode *inode, u64 file_pos,
1616                                        u64 disk_bytenr, u64 disk_num_bytes,
1617                                        u64 num_bytes, u64 ram_bytes,
1618                                        u8 compression, u8 encryption,
1619                                        u16 other_encoding, int extent_type)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         struct btrfs_file_extent_item *fi;
1623         struct btrfs_path *path;
1624         struct extent_buffer *leaf;
1625         struct btrfs_key ins;
1626         u64 hint;
1627         int ret;
1628
1629         path = btrfs_alloc_path();
1630         BUG_ON(!path);
1631
1632         path->leave_spinning = 1;
1633
1634         /*
1635          * we may be replacing one extent in the tree with another.
1636          * The new extent is pinned in the extent map, and we don't want
1637          * to drop it from the cache until it is completely in the btree.
1638          *
1639          * So, tell btrfs_drop_extents to leave this extent in the cache.
1640          * the caller is expected to unpin it and allow it to be merged
1641          * with the others.
1642          */
1643         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1644                                  &hint, 0);
1645         BUG_ON(ret);
1646
1647         ins.objectid = btrfs_ino(inode);
1648         ins.offset = file_pos;
1649         ins.type = BTRFS_EXTENT_DATA_KEY;
1650         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1651         BUG_ON(ret);
1652         leaf = path->nodes[0];
1653         fi = btrfs_item_ptr(leaf, path->slots[0],
1654                             struct btrfs_file_extent_item);
1655         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656         btrfs_set_file_extent_type(leaf, fi, extent_type);
1657         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659         btrfs_set_file_extent_offset(leaf, fi, 0);
1660         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662         btrfs_set_file_extent_compression(leaf, fi, compression);
1663         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1665
1666         btrfs_unlock_up_safe(path, 1);
1667         btrfs_set_lock_blocking(leaf);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670
1671         inode_add_bytes(inode, num_bytes);
1672
1673         ins.objectid = disk_bytenr;
1674         ins.offset = disk_num_bytes;
1675         ins.type = BTRFS_EXTENT_ITEM_KEY;
1676         ret = btrfs_alloc_reserved_file_extent(trans, root,
1677                                         root->root_key.objectid,
1678                                         btrfs_ino(inode), file_pos, &ins);
1679         BUG_ON(ret);
1680         btrfs_free_path(path);
1681
1682         return 0;
1683 }
1684
1685 /*
1686  * helper function for btrfs_finish_ordered_io, this
1687  * just reads in some of the csum leaves to prime them into ram
1688  * before we start the transaction.  It limits the amount of btree
1689  * reads required while inside the transaction.
1690  */
1691 /* as ordered data IO finishes, this gets called so we can finish
1692  * an ordered extent if the range of bytes in the file it covers are
1693  * fully written.
1694  */
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct btrfs_trans_handle *trans = NULL;
1699         struct btrfs_ordered_extent *ordered_extent = NULL;
1700         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701         struct extent_state *cached_state = NULL;
1702         int compress_type = 0;
1703         int ret;
1704         bool nolock = false;
1705
1706         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1707                                              end - start + 1);
1708         if (!ret)
1709                 return 0;
1710         BUG_ON(!ordered_extent);
1711
1712         nolock = (root == root->fs_info->tree_root);
1713
1714         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715                 BUG_ON(!list_empty(&ordered_extent->list));
1716                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717                 if (!ret) {
1718                         if (nolock)
1719                                 trans = btrfs_join_transaction_nolock(root, 1);
1720                         else
1721                                 trans = btrfs_join_transaction(root, 1);
1722                         BUG_ON(IS_ERR(trans));
1723                         btrfs_set_trans_block_group(trans, inode);
1724                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725                         ret = btrfs_update_inode(trans, root, inode);
1726                         BUG_ON(ret);
1727                 }
1728                 goto out;
1729         }
1730
1731         lock_extent_bits(io_tree, ordered_extent->file_offset,
1732                          ordered_extent->file_offset + ordered_extent->len - 1,
1733                          0, &cached_state, GFP_NOFS);
1734
1735         if (nolock)
1736                 trans = btrfs_join_transaction_nolock(root, 1);
1737         else
1738                 trans = btrfs_join_transaction(root, 1);
1739         BUG_ON(IS_ERR(trans));
1740         btrfs_set_trans_block_group(trans, inode);
1741         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1742
1743         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744                 compress_type = ordered_extent->compress_type;
1745         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746                 BUG_ON(compress_type);
1747                 ret = btrfs_mark_extent_written(trans, inode,
1748                                                 ordered_extent->file_offset,
1749                                                 ordered_extent->file_offset +
1750                                                 ordered_extent->len);
1751                 BUG_ON(ret);
1752         } else {
1753                 BUG_ON(root == root->fs_info->tree_root);
1754                 ret = insert_reserved_file_extent(trans, inode,
1755                                                 ordered_extent->file_offset,
1756                                                 ordered_extent->start,
1757                                                 ordered_extent->disk_len,
1758                                                 ordered_extent->len,
1759                                                 ordered_extent->len,
1760                                                 compress_type, 0, 0,
1761                                                 BTRFS_FILE_EXTENT_REG);
1762                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763                                    ordered_extent->file_offset,
1764                                    ordered_extent->len);
1765                 BUG_ON(ret);
1766         }
1767         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768                              ordered_extent->file_offset +
1769                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1770
1771         add_pending_csums(trans, inode, ordered_extent->file_offset,
1772                           &ordered_extent->list);
1773
1774         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775         if (!ret) {
1776                 ret = btrfs_update_inode(trans, root, inode);
1777                 BUG_ON(ret);
1778         }
1779         ret = 0;
1780 out:
1781         if (nolock) {
1782                 if (trans)
1783                         btrfs_end_transaction_nolock(trans, root);
1784         } else {
1785                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1786                 if (trans)
1787                         btrfs_end_transaction(trans, root);
1788         }
1789
1790         /* once for us */
1791         btrfs_put_ordered_extent(ordered_extent);
1792         /* once for the tree */
1793         btrfs_put_ordered_extent(ordered_extent);
1794
1795         return 0;
1796 }
1797
1798 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1799                                 struct extent_state *state, int uptodate)
1800 {
1801         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1802
1803         ClearPagePrivate2(page);
1804         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1805 }
1806
1807 /*
1808  * When IO fails, either with EIO or csum verification fails, we
1809  * try other mirrors that might have a good copy of the data.  This
1810  * io_failure_record is used to record state as we go through all the
1811  * mirrors.  If another mirror has good data, the page is set up to date
1812  * and things continue.  If a good mirror can't be found, the original
1813  * bio end_io callback is called to indicate things have failed.
1814  */
1815 struct io_failure_record {
1816         struct page *page;
1817         u64 start;
1818         u64 len;
1819         u64 logical;
1820         unsigned long bio_flags;
1821         int last_mirror;
1822 };
1823
1824 static int btrfs_io_failed_hook(struct bio *failed_bio,
1825                          struct page *page, u64 start, u64 end,
1826                          struct extent_state *state)
1827 {
1828         struct io_failure_record *failrec = NULL;
1829         u64 private;
1830         struct extent_map *em;
1831         struct inode *inode = page->mapping->host;
1832         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1833         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1834         struct bio *bio;
1835         int num_copies;
1836         int ret;
1837         int rw;
1838         u64 logical;
1839
1840         ret = get_state_private(failure_tree, start, &private);
1841         if (ret) {
1842                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1843                 if (!failrec)
1844                         return -ENOMEM;
1845                 failrec->start = start;
1846                 failrec->len = end - start + 1;
1847                 failrec->last_mirror = 0;
1848                 failrec->bio_flags = 0;
1849
1850                 read_lock(&em_tree->lock);
1851                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1852                 if (em->start > start || em->start + em->len < start) {
1853                         free_extent_map(em);
1854                         em = NULL;
1855                 }
1856                 read_unlock(&em_tree->lock);
1857
1858                 if (!em || IS_ERR(em)) {
1859                         kfree(failrec);
1860                         return -EIO;
1861                 }
1862                 logical = start - em->start;
1863                 logical = em->block_start + logical;
1864                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1865                         logical = em->block_start;
1866                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1867                         extent_set_compress_type(&failrec->bio_flags,
1868                                                  em->compress_type);
1869                 }
1870                 failrec->logical = logical;
1871                 free_extent_map(em);
1872                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1873                                 EXTENT_DIRTY, GFP_NOFS);
1874                 set_state_private(failure_tree, start,
1875                                  (u64)(unsigned long)failrec);
1876         } else {
1877                 failrec = (struct io_failure_record *)(unsigned long)private;
1878         }
1879         num_copies = btrfs_num_copies(
1880                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1881                               failrec->logical, failrec->len);
1882         failrec->last_mirror++;
1883         if (!state) {
1884                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1885                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1886                                                     failrec->start,
1887                                                     EXTENT_LOCKED);
1888                 if (state && state->start != failrec->start)
1889                         state = NULL;
1890                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1891         }
1892         if (!state || failrec->last_mirror > num_copies) {
1893                 set_state_private(failure_tree, failrec->start, 0);
1894                 clear_extent_bits(failure_tree, failrec->start,
1895                                   failrec->start + failrec->len - 1,
1896                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1897                 kfree(failrec);
1898                 return -EIO;
1899         }
1900         bio = bio_alloc(GFP_NOFS, 1);
1901         bio->bi_private = state;
1902         bio->bi_end_io = failed_bio->bi_end_io;
1903         bio->bi_sector = failrec->logical >> 9;
1904         bio->bi_bdev = failed_bio->bi_bdev;
1905         bio->bi_size = 0;
1906
1907         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1908         if (failed_bio->bi_rw & REQ_WRITE)
1909                 rw = WRITE;
1910         else
1911                 rw = READ;
1912
1913         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1914                                                       failrec->last_mirror,
1915                                                       failrec->bio_flags, 0);
1916         return ret;
1917 }
1918
1919 /*
1920  * each time an IO finishes, we do a fast check in the IO failure tree
1921  * to see if we need to process or clean up an io_failure_record
1922  */
1923 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1924 {
1925         u64 private;
1926         u64 private_failure;
1927         struct io_failure_record *failure;
1928         int ret;
1929
1930         private = 0;
1931         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1932                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1933                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1934                                         start, &private_failure);
1935                 if (ret == 0) {
1936                         failure = (struct io_failure_record *)(unsigned long)
1937                                    private_failure;
1938                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1939                                           failure->start, 0);
1940                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1941                                           failure->start,
1942                                           failure->start + failure->len - 1,
1943                                           EXTENT_DIRTY | EXTENT_LOCKED,
1944                                           GFP_NOFS);
1945                         kfree(failure);
1946                 }
1947         }
1948         return 0;
1949 }
1950
1951 /*
1952  * when reads are done, we need to check csums to verify the data is correct
1953  * if there's a match, we allow the bio to finish.  If not, we go through
1954  * the io_failure_record routines to find good copies
1955  */
1956 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1957                                struct extent_state *state)
1958 {
1959         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1960         struct inode *inode = page->mapping->host;
1961         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1962         char *kaddr;
1963         u64 private = ~(u32)0;
1964         int ret;
1965         struct btrfs_root *root = BTRFS_I(inode)->root;
1966         u32 csum = ~(u32)0;
1967
1968         if (PageChecked(page)) {
1969                 ClearPageChecked(page);
1970                 goto good;
1971         }
1972
1973         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1974                 return 0;
1975
1976         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1977             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1978                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1979                                   GFP_NOFS);
1980                 return 0;
1981         }
1982
1983         if (state && state->start == start) {
1984                 private = state->private;
1985                 ret = 0;
1986         } else {
1987                 ret = get_state_private(io_tree, start, &private);
1988         }
1989         kaddr = kmap_atomic(page, KM_USER0);
1990         if (ret)
1991                 goto zeroit;
1992
1993         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1994         btrfs_csum_final(csum, (char *)&csum);
1995         if (csum != private)
1996                 goto zeroit;
1997
1998         kunmap_atomic(kaddr, KM_USER0);
1999 good:
2000         /* if the io failure tree for this inode is non-empty,
2001          * check to see if we've recovered from a failed IO
2002          */
2003         btrfs_clean_io_failures(inode, start);
2004         return 0;
2005
2006 zeroit:
2007         if (printk_ratelimit()) {
2008                 printk(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2009                        "private %llu\n",
2010                        (unsigned long long)btrfs_ino(page->mapping->host),
2011                        (unsigned long long)start, csum,
2012                        (unsigned long long)private);
2013         }
2014         memset(kaddr + offset, 1, end - start + 1);
2015         flush_dcache_page(page);
2016         kunmap_atomic(kaddr, KM_USER0);
2017         if (private == 0)
2018                 return 0;
2019         return -EIO;
2020 }
2021
2022 struct delayed_iput {
2023         struct list_head list;
2024         struct inode *inode;
2025 };
2026
2027 void btrfs_add_delayed_iput(struct inode *inode)
2028 {
2029         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2030         struct delayed_iput *delayed;
2031
2032         if (atomic_add_unless(&inode->i_count, -1, 1))
2033                 return;
2034
2035         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2036         delayed->inode = inode;
2037
2038         spin_lock(&fs_info->delayed_iput_lock);
2039         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2040         spin_unlock(&fs_info->delayed_iput_lock);
2041 }
2042
2043 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2044 {
2045         LIST_HEAD(list);
2046         struct btrfs_fs_info *fs_info = root->fs_info;
2047         struct delayed_iput *delayed;
2048         int empty;
2049
2050         spin_lock(&fs_info->delayed_iput_lock);
2051         empty = list_empty(&fs_info->delayed_iputs);
2052         spin_unlock(&fs_info->delayed_iput_lock);
2053         if (empty)
2054                 return;
2055
2056         down_read(&root->fs_info->cleanup_work_sem);
2057         spin_lock(&fs_info->delayed_iput_lock);
2058         list_splice_init(&fs_info->delayed_iputs, &list);
2059         spin_unlock(&fs_info->delayed_iput_lock);
2060
2061         while (!list_empty(&list)) {
2062                 delayed = list_entry(list.next, struct delayed_iput, list);
2063                 list_del(&delayed->list);
2064                 iput(delayed->inode);
2065                 kfree(delayed);
2066         }
2067         up_read(&root->fs_info->cleanup_work_sem);
2068 }
2069
2070 /*
2071  * calculate extra metadata reservation when snapshotting a subvolume
2072  * contains orphan files.
2073  */
2074 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2075                                 struct btrfs_pending_snapshot *pending,
2076                                 u64 *bytes_to_reserve)
2077 {
2078         struct btrfs_root *root;
2079         struct btrfs_block_rsv *block_rsv;
2080         u64 num_bytes;
2081         int index;
2082
2083         root = pending->root;
2084         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2085                 return;
2086
2087         block_rsv = root->orphan_block_rsv;
2088
2089         /* orphan block reservation for the snapshot */
2090         num_bytes = block_rsv->size;
2091
2092         /*
2093          * after the snapshot is created, COWing tree blocks may use more
2094          * space than it frees. So we should make sure there is enough
2095          * reserved space.
2096          */
2097         index = trans->transid & 0x1;
2098         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2099                 num_bytes += block_rsv->size -
2100                              (block_rsv->reserved + block_rsv->freed[index]);
2101         }
2102
2103         *bytes_to_reserve += num_bytes;
2104 }
2105
2106 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2107                                 struct btrfs_pending_snapshot *pending)
2108 {
2109         struct btrfs_root *root = pending->root;
2110         struct btrfs_root *snap = pending->snap;
2111         struct btrfs_block_rsv *block_rsv;
2112         u64 num_bytes;
2113         int index;
2114         int ret;
2115
2116         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2117                 return;
2118
2119         /* refill source subvolume's orphan block reservation */
2120         block_rsv = root->orphan_block_rsv;
2121         index = trans->transid & 0x1;
2122         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2123                 num_bytes = block_rsv->size -
2124                             (block_rsv->reserved + block_rsv->freed[index]);
2125                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2126                                               root->orphan_block_rsv,
2127                                               num_bytes);
2128                 BUG_ON(ret);
2129         }
2130
2131         /* setup orphan block reservation for the snapshot */
2132         block_rsv = btrfs_alloc_block_rsv(snap);
2133         BUG_ON(!block_rsv);
2134
2135         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2136         snap->orphan_block_rsv = block_rsv;
2137
2138         num_bytes = root->orphan_block_rsv->size;
2139         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2140                                       block_rsv, num_bytes);
2141         BUG_ON(ret);
2142
2143 #if 0
2144         /* insert orphan item for the snapshot */
2145         WARN_ON(!root->orphan_item_inserted);
2146         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2147                                        snap->root_key.objectid);
2148         BUG_ON(ret);
2149         snap->orphan_item_inserted = 1;
2150 #endif
2151 }
2152
2153 enum btrfs_orphan_cleanup_state {
2154         ORPHAN_CLEANUP_STARTED  = 1,
2155         ORPHAN_CLEANUP_DONE     = 2,
2156 };
2157
2158 /*
2159  * This is called in transaction commmit time. If there are no orphan
2160  * files in the subvolume, it removes orphan item and frees block_rsv
2161  * structure.
2162  */
2163 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2164                               struct btrfs_root *root)
2165 {
2166         int ret;
2167
2168         if (!list_empty(&root->orphan_list) ||
2169             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2170                 return;
2171
2172         if (root->orphan_item_inserted &&
2173             btrfs_root_refs(&root->root_item) > 0) {
2174                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2175                                             root->root_key.objectid);
2176                 BUG_ON(ret);
2177                 root->orphan_item_inserted = 0;
2178         }
2179
2180         if (root->orphan_block_rsv) {
2181                 WARN_ON(root->orphan_block_rsv->size > 0);
2182                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2183                 root->orphan_block_rsv = NULL;
2184         }
2185 }
2186
2187 /*
2188  * This creates an orphan entry for the given inode in case something goes
2189  * wrong in the middle of an unlink/truncate.
2190  *
2191  * NOTE: caller of this function should reserve 5 units of metadata for
2192  *       this function.
2193  */
2194 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2195 {
2196         struct btrfs_root *root = BTRFS_I(inode)->root;
2197         struct btrfs_block_rsv *block_rsv = NULL;
2198         int reserve = 0;
2199         int insert = 0;
2200         int ret;
2201
2202         if (!root->orphan_block_rsv) {
2203                 block_rsv = btrfs_alloc_block_rsv(root);
2204                 BUG_ON(!block_rsv);
2205         }
2206
2207         spin_lock(&root->orphan_lock);
2208         if (!root->orphan_block_rsv) {
2209                 root->orphan_block_rsv = block_rsv;
2210         } else if (block_rsv) {
2211                 btrfs_free_block_rsv(root, block_rsv);
2212                 block_rsv = NULL;
2213         }
2214
2215         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2216                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2217 #if 0
2218                 /*
2219                  * For proper ENOSPC handling, we should do orphan
2220                  * cleanup when mounting. But this introduces backward
2221                  * compatibility issue.
2222                  */
2223                 if (!xchg(&root->orphan_item_inserted, 1))
2224                         insert = 2;
2225                 else
2226                         insert = 1;
2227 #endif
2228                 insert = 1;
2229         }
2230
2231         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2232                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2233                 reserve = 1;
2234         }
2235         spin_unlock(&root->orphan_lock);
2236
2237         if (block_rsv)
2238                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2239
2240         /* grab metadata reservation from transaction handle */
2241         if (reserve) {
2242                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2243                 BUG_ON(ret);
2244         }
2245
2246         /* insert an orphan item to track this unlinked/truncated file */
2247         if (insert >= 1) {
2248                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2249                 BUG_ON(ret);
2250         }
2251
2252         /* insert an orphan item to track subvolume contains orphan files */
2253         if (insert >= 2) {
2254                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2255                                                root->root_key.objectid);
2256                 BUG_ON(ret);
2257         }
2258         return 0;
2259 }
2260
2261 /*
2262  * We have done the truncate/delete so we can go ahead and remove the orphan
2263  * item for this particular inode.
2264  */
2265 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2266 {
2267         struct btrfs_root *root = BTRFS_I(inode)->root;
2268         int delete_item = 0;
2269         int release_rsv = 0;
2270         int ret = 0;
2271
2272         spin_lock(&root->orphan_lock);
2273         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2274                 list_del_init(&BTRFS_I(inode)->i_orphan);
2275                 delete_item = 1;
2276         }
2277
2278         if (BTRFS_I(inode)->orphan_meta_reserved) {
2279                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2280                 release_rsv = 1;
2281         }
2282         spin_unlock(&root->orphan_lock);
2283
2284         if (trans && delete_item) {
2285                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2286                 BUG_ON(ret);
2287         }
2288
2289         if (release_rsv)
2290                 btrfs_orphan_release_metadata(inode);
2291
2292         return 0;
2293 }
2294
2295 /*
2296  * this cleans up any orphans that may be left on the list from the last use
2297  * of this root.
2298  */
2299 int btrfs_orphan_cleanup(struct btrfs_root *root)
2300 {
2301         struct btrfs_path *path;
2302         struct extent_buffer *leaf;
2303         struct btrfs_key key, found_key;
2304         struct btrfs_trans_handle *trans;
2305         struct inode *inode;
2306         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2307
2308         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2309                 return 0;
2310
2311         path = btrfs_alloc_path();
2312         if (!path) {
2313                 ret = -ENOMEM;
2314                 goto out;
2315         }
2316         path->reada = -1;
2317
2318         key.objectid = BTRFS_ORPHAN_OBJECTID;
2319         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2320         key.offset = (u64)-1;
2321
2322         while (1) {
2323                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2324                 if (ret < 0)
2325                         goto out;
2326
2327                 /*
2328                  * if ret == 0 means we found what we were searching for, which
2329                  * is weird, but possible, so only screw with path if we didnt
2330                  * find the key and see if we have stuff that matches
2331                  */
2332                 if (ret > 0) {
2333                         ret = 0;
2334                         if (path->slots[0] == 0)
2335                                 break;
2336                         path->slots[0]--;
2337                 }
2338
2339                 /* pull out the item */
2340                 leaf = path->nodes[0];
2341                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2342
2343                 /* make sure the item matches what we want */
2344                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2345                         break;
2346                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2347                         break;
2348
2349                 /* release the path since we're done with it */
2350                 btrfs_release_path(root, path);
2351
2352                 /*
2353                  * this is where we are basically btrfs_lookup, without the
2354                  * crossing root thing.  we store the inode number in the
2355                  * offset of the orphan item.
2356                  */
2357                 found_key.objectid = found_key.offset;
2358                 found_key.type = BTRFS_INODE_ITEM_KEY;
2359                 found_key.offset = 0;
2360                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2361                 if (IS_ERR(inode)) {
2362                         ret = PTR_ERR(inode);
2363                         goto out;
2364                 }
2365
2366                 /*
2367                  * add this inode to the orphan list so btrfs_orphan_del does
2368                  * the proper thing when we hit it
2369                  */
2370                 spin_lock(&root->orphan_lock);
2371                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2372                 spin_unlock(&root->orphan_lock);
2373
2374                 /*
2375                  * if this is a bad inode, means we actually succeeded in
2376                  * removing the inode, but not the orphan record, which means
2377                  * we need to manually delete the orphan since iput will just
2378                  * do a destroy_inode
2379                  */
2380                 if (is_bad_inode(inode)) {
2381                         trans = btrfs_start_transaction(root, 0);
2382                         if (IS_ERR(trans)) {
2383                                 ret = PTR_ERR(trans);
2384                                 goto out;
2385                         }
2386                         btrfs_orphan_del(trans, inode);
2387                         btrfs_end_transaction(trans, root);
2388                         iput(inode);
2389                         continue;
2390                 }
2391
2392                 /* if we have links, this was a truncate, lets do that */
2393                 if (inode->i_nlink) {
2394                         if (!S_ISREG(inode->i_mode)) {
2395                                 WARN_ON(1);
2396                                 iput(inode);
2397                                 continue;
2398                         }
2399                         nr_truncate++;
2400                         ret = btrfs_truncate(inode);
2401                 } else {
2402                         nr_unlink++;
2403                 }
2404
2405                 /* this will do delete_inode and everything for us */
2406                 iput(inode);
2407                 if (ret)
2408                         goto out;
2409         }
2410         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2411
2412         if (root->orphan_block_rsv)
2413                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2414                                         (u64)-1);
2415
2416         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2417                 trans = btrfs_join_transaction(root, 1);
2418                 if (!IS_ERR(trans))
2419                         btrfs_end_transaction(trans, root);
2420         }
2421
2422         if (nr_unlink)
2423                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2424         if (nr_truncate)
2425                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2426
2427 out:
2428         if (ret)
2429                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2430         btrfs_free_path(path);
2431         return ret;
2432 }
2433
2434 /*
2435  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2436  * don't find any xattrs, we know there can't be any acls.
2437  *
2438  * slot is the slot the inode is in, objectid is the objectid of the inode
2439  */
2440 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2441                                           int slot, u64 objectid)
2442 {
2443         u32 nritems = btrfs_header_nritems(leaf);
2444         struct btrfs_key found_key;
2445         int scanned = 0;
2446
2447         slot++;
2448         while (slot < nritems) {
2449                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2450
2451                 /* we found a different objectid, there must not be acls */
2452                 if (found_key.objectid != objectid)
2453                         return 0;
2454
2455                 /* we found an xattr, assume we've got an acl */
2456                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2457                         return 1;
2458
2459                 /*
2460                  * we found a key greater than an xattr key, there can't
2461                  * be any acls later on
2462                  */
2463                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2464                         return 0;
2465
2466                 slot++;
2467                 scanned++;
2468
2469                 /*
2470                  * it goes inode, inode backrefs, xattrs, extents,
2471                  * so if there are a ton of hard links to an inode there can
2472                  * be a lot of backrefs.  Don't waste time searching too hard,
2473                  * this is just an optimization
2474                  */
2475                 if (scanned >= 8)
2476                         break;
2477         }
2478         /* we hit the end of the leaf before we found an xattr or
2479          * something larger than an xattr.  We have to assume the inode
2480          * has acls
2481          */
2482         return 1;
2483 }
2484
2485 /*
2486  * read an inode from the btree into the in-memory inode
2487  */
2488 static void btrfs_read_locked_inode(struct inode *inode)
2489 {
2490         struct btrfs_path *path;
2491         struct extent_buffer *leaf;
2492         struct btrfs_inode_item *inode_item;
2493         struct btrfs_timespec *tspec;
2494         struct btrfs_root *root = BTRFS_I(inode)->root;
2495         struct btrfs_key location;
2496         int maybe_acls;
2497         u64 alloc_group_block;
2498         u32 rdev;
2499         int ret;
2500
2501         path = btrfs_alloc_path();
2502         BUG_ON(!path);
2503         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2504
2505         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2506         if (ret)
2507                 goto make_bad;
2508
2509         leaf = path->nodes[0];
2510         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2511                                     struct btrfs_inode_item);
2512
2513         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2514         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2515         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2516         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2517         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2518
2519         tspec = btrfs_inode_atime(inode_item);
2520         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2521         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2522
2523         tspec = btrfs_inode_mtime(inode_item);
2524         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2525         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2526
2527         tspec = btrfs_inode_ctime(inode_item);
2528         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2529         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2530
2531         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2532         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2533         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2534         inode->i_generation = BTRFS_I(inode)->generation;
2535         inode->i_rdev = 0;
2536         rdev = btrfs_inode_rdev(leaf, inode_item);
2537
2538         BTRFS_I(inode)->index_cnt = (u64)-1;
2539         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2540
2541         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2542
2543         /*
2544          * try to precache a NULL acl entry for files that don't have
2545          * any xattrs or acls
2546          */
2547         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2548                                            btrfs_ino(inode));
2549         if (!maybe_acls)
2550                 cache_no_acl(inode);
2551
2552         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2553                                                 alloc_group_block, 0);
2554         btrfs_free_path(path);
2555         inode_item = NULL;
2556
2557         switch (inode->i_mode & S_IFMT) {
2558         case S_IFREG:
2559                 inode->i_mapping->a_ops = &btrfs_aops;
2560                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2561                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2562                 inode->i_fop = &btrfs_file_operations;
2563                 inode->i_op = &btrfs_file_inode_operations;
2564                 break;
2565         case S_IFDIR:
2566                 inode->i_fop = &btrfs_dir_file_operations;
2567                 if (root == root->fs_info->tree_root)
2568                         inode->i_op = &btrfs_dir_ro_inode_operations;
2569                 else
2570                         inode->i_op = &btrfs_dir_inode_operations;
2571                 break;
2572         case S_IFLNK:
2573                 inode->i_op = &btrfs_symlink_inode_operations;
2574                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2575                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2576                 break;
2577         default:
2578                 inode->i_op = &btrfs_special_inode_operations;
2579                 init_special_inode(inode, inode->i_mode, rdev);
2580                 break;
2581         }
2582
2583         btrfs_update_iflags(inode);
2584         return;
2585
2586 make_bad:
2587         btrfs_free_path(path);
2588         make_bad_inode(inode);
2589 }
2590
2591 /*
2592  * given a leaf and an inode, copy the inode fields into the leaf
2593  */
2594 static void fill_inode_item(struct btrfs_trans_handle *trans,
2595                             struct extent_buffer *leaf,
2596                             struct btrfs_inode_item *item,
2597                             struct inode *inode)
2598 {
2599         if (!leaf->map_token)
2600                 map_private_extent_buffer(leaf, (unsigned long)item,
2601                                           sizeof(struct btrfs_inode_item),
2602                                           &leaf->map_token, &leaf->kaddr,
2603                                           &leaf->map_start, &leaf->map_len,
2604                                           KM_USER1);
2605
2606         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2607         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2608         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2609         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2610         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2611
2612         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2613                                inode->i_atime.tv_sec);
2614         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2615                                 inode->i_atime.tv_nsec);
2616
2617         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2618                                inode->i_mtime.tv_sec);
2619         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2620                                 inode->i_mtime.tv_nsec);
2621
2622         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2623                                inode->i_ctime.tv_sec);
2624         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2625                                 inode->i_ctime.tv_nsec);
2626
2627         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2628         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2629         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2630         btrfs_set_inode_transid(leaf, item, trans->transid);
2631         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2632         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2633         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2634
2635         if (leaf->map_token) {
2636                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2637                 leaf->map_token = NULL;
2638         }
2639 }
2640
2641 /*
2642  * copy everything in the in-memory inode into the btree.
2643  */
2644 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2645                                 struct btrfs_root *root, struct inode *inode)
2646 {
2647         struct btrfs_inode_item *inode_item;
2648         struct btrfs_path *path;
2649         struct extent_buffer *leaf;
2650         int ret;
2651
2652         path = btrfs_alloc_path();
2653         BUG_ON(!path);
2654         path->leave_spinning = 1;
2655         ret = btrfs_lookup_inode(trans, root, path,
2656                                  &BTRFS_I(inode)->location, 1);
2657         if (ret) {
2658                 if (ret > 0)
2659                         ret = -ENOENT;
2660                 goto failed;
2661         }
2662
2663         btrfs_unlock_up_safe(path, 1);
2664         leaf = path->nodes[0];
2665         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2666                                   struct btrfs_inode_item);
2667
2668         fill_inode_item(trans, leaf, inode_item, inode);
2669         btrfs_mark_buffer_dirty(leaf);
2670         btrfs_set_inode_last_trans(trans, inode);
2671         ret = 0;
2672 failed:
2673         btrfs_free_path(path);
2674         return ret;
2675 }
2676
2677
2678 /*
2679  * unlink helper that gets used here in inode.c and in the tree logging
2680  * recovery code.  It remove a link in a directory with a given name, and
2681  * also drops the back refs in the inode to the directory
2682  */
2683 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2684                                 struct btrfs_root *root,
2685                                 struct inode *dir, struct inode *inode,
2686                                 const char *name, int name_len)
2687 {
2688         struct btrfs_path *path;
2689         int ret = 0;
2690         struct extent_buffer *leaf;
2691         struct btrfs_dir_item *di;
2692         struct btrfs_key key;
2693         u64 index;
2694         u64 ino = btrfs_ino(inode);
2695         u64 dir_ino = btrfs_ino(dir);
2696
2697         path = btrfs_alloc_path();
2698         if (!path) {
2699                 ret = -ENOMEM;
2700                 goto out;
2701         }
2702
2703         path->leave_spinning = 1;
2704         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2705                                     name, name_len, -1);
2706         if (IS_ERR(di)) {
2707                 ret = PTR_ERR(di);
2708                 goto err;
2709         }
2710         if (!di) {
2711                 ret = -ENOENT;
2712                 goto err;
2713         }
2714         leaf = path->nodes[0];
2715         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2716         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2717         if (ret)
2718                 goto err;
2719         btrfs_release_path(root, path);
2720
2721         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2722                                   dir_ino, &index);
2723         if (ret) {
2724                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2725                        "inode %llu parent %llu\n", name_len, name,
2726                        (unsigned long long)ino, (unsigned long long)dir_ino);
2727                 goto err;
2728         }
2729
2730         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino,
2731                                          index, name, name_len, -1);
2732         if (IS_ERR(di)) {
2733                 ret = PTR_ERR(di);
2734                 goto err;
2735         }
2736         if (!di) {
2737                 ret = -ENOENT;
2738                 goto err;
2739         }
2740         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2741         btrfs_release_path(root, path);
2742
2743         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2744                                          inode, dir_ino);
2745         BUG_ON(ret != 0 && ret != -ENOENT);
2746
2747         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2748                                            dir, index);
2749         if (ret == -ENOENT)
2750                 ret = 0;
2751 err:
2752         btrfs_free_path(path);
2753         if (ret)
2754                 goto out;
2755
2756         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2757         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2758         btrfs_update_inode(trans, root, dir);
2759 out:
2760         return ret;
2761 }
2762
2763 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2764                        struct btrfs_root *root,
2765                        struct inode *dir, struct inode *inode,
2766                        const char *name, int name_len)
2767 {
2768         int ret;
2769         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2770         if (!ret) {
2771                 btrfs_drop_nlink(inode);
2772                 ret = btrfs_update_inode(trans, root, inode);
2773         }
2774         return ret;
2775 }
2776
2777
2778 /* helper to check if there is any shared block in the path */
2779 static int check_path_shared(struct btrfs_root *root,
2780                              struct btrfs_path *path)
2781 {
2782         struct extent_buffer *eb;
2783         int level;
2784         u64 refs = 1;
2785
2786         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2787                 int ret;
2788
2789                 if (!path->nodes[level])
2790                         break;
2791                 eb = path->nodes[level];
2792                 if (!btrfs_block_can_be_shared(root, eb))
2793                         continue;
2794                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2795                                                &refs, NULL);
2796                 if (refs > 1)
2797                         return 1;
2798         }
2799         return 0;
2800 }
2801
2802 /*
2803  * helper to start transaction for unlink and rmdir.
2804  *
2805  * unlink and rmdir are special in btrfs, they do not always free space.
2806  * so in enospc case, we should make sure they will free space before
2807  * allowing them to use the global metadata reservation.
2808  */
2809 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2810                                                        struct dentry *dentry)
2811 {
2812         struct btrfs_trans_handle *trans;
2813         struct btrfs_root *root = BTRFS_I(dir)->root;
2814         struct btrfs_path *path;
2815         struct btrfs_inode_ref *ref;
2816         struct btrfs_dir_item *di;
2817         struct inode *inode = dentry->d_inode;
2818         u64 index;
2819         int check_link = 1;
2820         int err = -ENOSPC;
2821         int ret;
2822         u64 ino = btrfs_ino(inode);
2823         u64 dir_ino = btrfs_ino(dir);
2824
2825         trans = btrfs_start_transaction(root, 10);
2826         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2827                 return trans;
2828
2829         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2830                 return ERR_PTR(-ENOSPC);
2831
2832         /* check if there is someone else holds reference */
2833         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2834                 return ERR_PTR(-ENOSPC);
2835
2836         if (atomic_read(&inode->i_count) > 2)
2837                 return ERR_PTR(-ENOSPC);
2838
2839         if (xchg(&root->fs_info->enospc_unlink, 1))
2840                 return ERR_PTR(-ENOSPC);
2841
2842         path = btrfs_alloc_path();
2843         if (!path) {
2844                 root->fs_info->enospc_unlink = 0;
2845                 return ERR_PTR(-ENOMEM);
2846         }
2847
2848         trans = btrfs_start_transaction(root, 0);
2849         if (IS_ERR(trans)) {
2850                 btrfs_free_path(path);
2851                 root->fs_info->enospc_unlink = 0;
2852                 return trans;
2853         }
2854
2855         path->skip_locking = 1;
2856         path->search_commit_root = 1;
2857
2858         ret = btrfs_lookup_inode(trans, root, path,
2859                                 &BTRFS_I(dir)->location, 0);
2860         if (ret < 0) {
2861                 err = ret;
2862                 goto out;
2863         }
2864         if (ret == 0) {
2865                 if (check_path_shared(root, path))
2866                         goto out;
2867         } else {
2868                 check_link = 0;
2869         }
2870         btrfs_release_path(root, path);
2871
2872         ret = btrfs_lookup_inode(trans, root, path,
2873                                 &BTRFS_I(inode)->location, 0);
2874         if (ret < 0) {
2875                 err = ret;
2876                 goto out;
2877         }
2878         if (ret == 0) {
2879                 if (check_path_shared(root, path))
2880                         goto out;
2881         } else {
2882                 check_link = 0;
2883         }
2884         btrfs_release_path(root, path);
2885
2886         if (ret == 0 && S_ISREG(inode->i_mode)) {
2887                 ret = btrfs_lookup_file_extent(trans, root, path,
2888                                                ino, (u64)-1, 0);
2889                 if (ret < 0) {
2890                         err = ret;
2891                         goto out;
2892                 }
2893                 BUG_ON(ret == 0);
2894                 if (check_path_shared(root, path))
2895                         goto out;
2896                 btrfs_release_path(root, path);
2897         }
2898
2899         if (!check_link) {
2900                 err = 0;
2901                 goto out;
2902         }
2903
2904         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2905                                 dentry->d_name.name, dentry->d_name.len, 0);
2906         if (IS_ERR(di)) {
2907                 err = PTR_ERR(di);
2908                 goto out;
2909         }
2910         if (di) {
2911                 if (check_path_shared(root, path))
2912                         goto out;
2913         } else {
2914                 err = 0;
2915                 goto out;
2916         }
2917         btrfs_release_path(root, path);
2918
2919         ref = btrfs_lookup_inode_ref(trans, root, path,
2920                                 dentry->d_name.name, dentry->d_name.len,
2921                                 ino, dir_ino, 0);
2922         if (IS_ERR(ref)) {
2923                 err = PTR_ERR(ref);
2924                 goto out;
2925         }
2926         BUG_ON(!ref);
2927         if (check_path_shared(root, path))
2928                 goto out;
2929         index = btrfs_inode_ref_index(path->nodes[0], ref);
2930         btrfs_release_path(root, path);
2931
2932         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2933                                 dentry->d_name.name, dentry->d_name.len, 0);
2934         if (IS_ERR(di)) {
2935                 err = PTR_ERR(di);
2936                 goto out;
2937         }
2938         BUG_ON(ret == -ENOENT);
2939         if (check_path_shared(root, path))
2940                 goto out;
2941
2942         err = 0;
2943 out:
2944         btrfs_free_path(path);
2945         if (err) {
2946                 btrfs_end_transaction(trans, root);
2947                 root->fs_info->enospc_unlink = 0;
2948                 return ERR_PTR(err);
2949         }
2950
2951         trans->block_rsv = &root->fs_info->global_block_rsv;
2952         return trans;
2953 }
2954
2955 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2956                                struct btrfs_root *root)
2957 {
2958         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2959                 BUG_ON(!root->fs_info->enospc_unlink);
2960                 root->fs_info->enospc_unlink = 0;
2961         }
2962         btrfs_end_transaction_throttle(trans, root);
2963 }
2964
2965 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2966 {
2967         struct btrfs_root *root = BTRFS_I(dir)->root;
2968         struct btrfs_trans_handle *trans;
2969         struct inode *inode = dentry->d_inode;
2970         int ret;
2971         unsigned long nr = 0;
2972
2973         trans = __unlink_start_trans(dir, dentry);
2974         if (IS_ERR(trans))
2975                 return PTR_ERR(trans);
2976
2977         btrfs_set_trans_block_group(trans, dir);
2978
2979         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2980
2981         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2982                                  dentry->d_name.name, dentry->d_name.len);
2983         BUG_ON(ret);
2984
2985         if (inode->i_nlink == 0) {
2986                 ret = btrfs_orphan_add(trans, inode);
2987                 BUG_ON(ret);
2988         }
2989
2990         nr = trans->blocks_used;
2991         __unlink_end_trans(trans, root);
2992         btrfs_btree_balance_dirty(root, nr);
2993         return ret;
2994 }
2995
2996 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2997                         struct btrfs_root *root,
2998                         struct inode *dir, u64 objectid,
2999                         const char *name, int name_len)
3000 {
3001         struct btrfs_path *path;
3002         struct extent_buffer *leaf;
3003         struct btrfs_dir_item *di;
3004         struct btrfs_key key;
3005         u64 index;
3006         int ret;
3007         u64 dir_ino = btrfs_ino(dir);
3008
3009         path = btrfs_alloc_path();
3010         if (!path)
3011                 return -ENOMEM;
3012
3013         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3014                                    name, name_len, -1);
3015         BUG_ON(!di || IS_ERR(di));
3016
3017         leaf = path->nodes[0];
3018         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3019         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3020         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3021         BUG_ON(ret);
3022         btrfs_release_path(root, path);
3023
3024         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3025                                  objectid, root->root_key.objectid,
3026                                  dir_ino, &index, name, name_len);
3027         if (ret < 0) {
3028                 BUG_ON(ret != -ENOENT);
3029                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3030                                                  name, name_len);
3031                 BUG_ON(!di || IS_ERR(di));
3032
3033                 leaf = path->nodes[0];
3034                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3035                 btrfs_release_path(root, path);
3036                 index = key.offset;
3037         }
3038
3039         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino,
3040                                          index, name, name_len, -1);
3041         BUG_ON(!di || IS_ERR(di));
3042
3043         leaf = path->nodes[0];
3044         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3045         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3046         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3047         BUG_ON(ret);
3048         btrfs_release_path(root, path);
3049
3050         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3051         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3052         ret = btrfs_update_inode(trans, root, dir);
3053         BUG_ON(ret);
3054
3055         btrfs_free_path(path);
3056         return 0;
3057 }
3058
3059 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3060 {
3061         struct inode *inode = dentry->d_inode;
3062         int err = 0;
3063         struct btrfs_root *root = BTRFS_I(dir)->root;
3064         struct btrfs_trans_handle *trans;
3065         unsigned long nr = 0;
3066
3067         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3068             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3069                 return -ENOTEMPTY;
3070
3071         trans = __unlink_start_trans(dir, dentry);
3072         if (IS_ERR(trans))
3073                 return PTR_ERR(trans);
3074
3075         btrfs_set_trans_block_group(trans, dir);
3076
3077         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3078                 err = btrfs_unlink_subvol(trans, root, dir,
3079                                           BTRFS_I(inode)->location.objectid,
3080                                           dentry->d_name.name,
3081                                           dentry->d_name.len);
3082                 goto out;
3083         }
3084
3085         err = btrfs_orphan_add(trans, inode);
3086         if (err)
3087                 goto out;
3088
3089         /* now the directory is empty */
3090         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3091                                  dentry->d_name.name, dentry->d_name.len);
3092         if (!err)
3093                 btrfs_i_size_write(inode, 0);
3094 out:
3095         nr = trans->blocks_used;
3096         __unlink_end_trans(trans, root);
3097         btrfs_btree_balance_dirty(root, nr);
3098
3099         return err;
3100 }
3101
3102 #if 0
3103 /*
3104  * when truncating bytes in a file, it is possible to avoid reading
3105  * the leaves that contain only checksum items.  This can be the
3106  * majority of the IO required to delete a large file, but it must
3107  * be done carefully.
3108  *
3109  * The keys in the level just above the leaves are checked to make sure
3110  * the lowest key in a given leaf is a csum key, and starts at an offset
3111  * after the new  size.
3112  *
3113  * Then the key for the next leaf is checked to make sure it also has
3114  * a checksum item for the same file.  If it does, we know our target leaf
3115  * contains only checksum items, and it can be safely freed without reading
3116  * it.
3117  *
3118  * This is just an optimization targeted at large files.  It may do
3119  * nothing.  It will return 0 unless things went badly.
3120  */
3121 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3122                                      struct btrfs_root *root,
3123                                      struct btrfs_path *path,
3124                                      struct inode *inode, u64 new_size)
3125 {
3126         struct btrfs_key key;
3127         int ret;
3128         int nritems;
3129         struct btrfs_key found_key;
3130         struct btrfs_key other_key;
3131         struct btrfs_leaf_ref *ref;
3132         u64 leaf_gen;
3133         u64 leaf_start;
3134
3135         path->lowest_level = 1;
3136         key.objectid = inode->i_ino;
3137         key.type = BTRFS_CSUM_ITEM_KEY;
3138         key.offset = new_size;
3139 again:
3140         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3141         if (ret < 0)
3142                 goto out;
3143
3144         if (path->nodes[1] == NULL) {
3145                 ret = 0;
3146                 goto out;
3147         }
3148         ret = 0;
3149         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3150         nritems = btrfs_header_nritems(path->nodes[1]);
3151
3152         if (!nritems)
3153                 goto out;
3154
3155         if (path->slots[1] >= nritems)
3156                 goto next_node;
3157
3158         /* did we find a key greater than anything we want to delete? */
3159         if (found_key.objectid > inode->i_ino ||
3160            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3161                 goto out;
3162
3163         /* we check the next key in the node to make sure the leave contains
3164          * only checksum items.  This comparison doesn't work if our
3165          * leaf is the last one in the node
3166          */
3167         if (path->slots[1] + 1 >= nritems) {
3168 next_node:
3169                 /* search forward from the last key in the node, this
3170                  * will bring us into the next node in the tree
3171                  */
3172                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3173
3174                 /* unlikely, but we inc below, so check to be safe */
3175                 if (found_key.offset == (u64)-1)
3176                         goto out;
3177
3178                 /* search_forward needs a path with locks held, do the
3179                  * search again for the original key.  It is possible
3180                  * this will race with a balance and return a path that
3181                  * we could modify, but this drop is just an optimization
3182                  * and is allowed to miss some leaves.
3183                  */
3184                 btrfs_release_path(root, path);
3185                 found_key.offset++;
3186
3187                 /* setup a max key for search_forward */
3188                 other_key.offset = (u64)-1;
3189                 other_key.type = key.type;
3190                 other_key.objectid = key.objectid;
3191
3192                 path->keep_locks = 1;
3193                 ret = btrfs_search_forward(root, &found_key, &other_key,
3194                                            path, 0, 0);
3195                 path->keep_locks = 0;
3196                 if (ret || found_key.objectid != key.objectid ||
3197                     found_key.type != key.type) {
3198                         ret = 0;
3199                         goto out;
3200                 }
3201
3202                 key.offset = found_key.offset;
3203                 btrfs_release_path(root, path);
3204                 cond_resched();
3205                 goto again;
3206         }
3207
3208         /* we know there's one more slot after us in the tree,
3209          * read that key so we can verify it is also a checksum item
3210          */
3211         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3212
3213         if (found_key.objectid < inode->i_ino)
3214                 goto next_key;
3215
3216         if (found_key.type != key.type || found_key.offset < new_size)
3217                 goto next_key;
3218
3219         /*
3220          * if the key for the next leaf isn't a csum key from this objectid,
3221          * we can't be sure there aren't good items inside this leaf.
3222          * Bail out
3223          */
3224         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3225                 goto out;
3226
3227         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3228         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3229         /*
3230          * it is safe to delete this leaf, it contains only
3231          * csum items from this inode at an offset >= new_size
3232          */
3233         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3234         BUG_ON(ret);
3235
3236         if (root->ref_cows && leaf_gen < trans->transid) {
3237                 ref = btrfs_alloc_leaf_ref(root, 0);
3238                 if (ref) {
3239                         ref->root_gen = root->root_key.offset;
3240                         ref->bytenr = leaf_start;
3241                         ref->owner = 0;
3242                         ref->generation = leaf_gen;
3243                         ref->nritems = 0;
3244
3245                         btrfs_sort_leaf_ref(ref);
3246
3247                         ret = btrfs_add_leaf_ref(root, ref, 0);
3248                         WARN_ON(ret);
3249                         btrfs_free_leaf_ref(root, ref);
3250                 } else {
3251                         WARN_ON(1);
3252                 }
3253         }
3254 next_key:
3255         btrfs_release_path(root, path);
3256
3257         if (other_key.objectid == inode->i_ino &&
3258             other_key.type == key.type && other_key.offset > key.offset) {
3259                 key.offset = other_key.offset;
3260                 cond_resched();
3261                 goto again;
3262         }
3263         ret = 0;
3264 out:
3265         /* fixup any changes we've made to the path */
3266         path->lowest_level = 0;
3267         path->keep_locks = 0;
3268         btrfs_release_path(root, path);
3269         return ret;
3270 }
3271
3272 #endif
3273
3274 /*
3275  * this can truncate away extent items, csum items and directory items.
3276  * It starts at a high offset and removes keys until it can't find
3277  * any higher than new_size
3278  *
3279  * csum items that cross the new i_size are truncated to the new size
3280  * as well.
3281  *
3282  * min_type is the minimum key type to truncate down to.  If set to 0, this
3283  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3284  */
3285 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3286                                struct btrfs_root *root,
3287                                struct inode *inode,
3288                                u64 new_size, u32 min_type)
3289 {
3290         struct btrfs_path *path;
3291         struct extent_buffer *leaf;
3292         struct btrfs_file_extent_item *fi;
3293         struct btrfs_key key;
3294         struct btrfs_key found_key;
3295         u64 extent_start = 0;
3296         u64 extent_num_bytes = 0;
3297         u64 extent_offset = 0;
3298         u64 item_end = 0;
3299         u64 mask = root->sectorsize - 1;
3300         u32 found_type = (u8)-1;
3301         int found_extent;
3302         int del_item;
3303         int pending_del_nr = 0;
3304         int pending_del_slot = 0;
3305         int extent_type = -1;
3306         int encoding;
3307         int ret;
3308         int err = 0;
3309         u64 ino = btrfs_ino(inode);
3310
3311         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3312
3313         if (root->ref_cows || root == root->fs_info->tree_root)
3314                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3315
3316         path = btrfs_alloc_path();
3317         BUG_ON(!path);
3318         path->reada = -1;
3319
3320         key.objectid = ino;
3321         key.offset = (u64)-1;
3322         key.type = (u8)-1;
3323
3324 search_again:
3325         path->leave_spinning = 1;
3326         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3327         if (ret < 0) {
3328                 err = ret;
3329                 goto out;
3330         }
3331
3332         if (ret > 0) {
3333                 /* there are no items in the tree for us to truncate, we're
3334                  * done
3335                  */
3336                 if (path->slots[0] == 0)
3337                         goto out;
3338                 path->slots[0]--;
3339         }
3340
3341         while (1) {
3342                 fi = NULL;
3343                 leaf = path->nodes[0];
3344                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3345                 found_type = btrfs_key_type(&found_key);
3346                 encoding = 0;
3347
3348                 if (found_key.objectid != ino)
3349                         break;
3350
3351                 if (found_type < min_type)
3352                         break;
3353
3354                 item_end = found_key.offset;
3355                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3356                         fi = btrfs_item_ptr(leaf, path->slots[0],
3357                                             struct btrfs_file_extent_item);
3358                         extent_type = btrfs_file_extent_type(leaf, fi);
3359                         encoding = btrfs_file_extent_compression(leaf, fi);
3360                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3361                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3362
3363                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3364                                 item_end +=
3365                                     btrfs_file_extent_num_bytes(leaf, fi);
3366                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3367                                 item_end += btrfs_file_extent_inline_len(leaf,
3368                                                                          fi);
3369                         }
3370                         item_end--;
3371                 }
3372                 if (found_type > min_type) {
3373                         del_item = 1;
3374                 } else {
3375                         if (item_end < new_size)
3376                                 break;
3377                         if (found_key.offset >= new_size)
3378                                 del_item = 1;
3379                         else
3380                                 del_item = 0;
3381                 }
3382                 found_extent = 0;
3383                 /* FIXME, shrink the extent if the ref count is only 1 */
3384                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3385                         goto delete;
3386
3387                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3388                         u64 num_dec;
3389                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3390                         if (!del_item && !encoding) {
3391                                 u64 orig_num_bytes =
3392                                         btrfs_file_extent_num_bytes(leaf, fi);
3393                                 extent_num_bytes = new_size -
3394                                         found_key.offset + root->sectorsize - 1;
3395                                 extent_num_bytes = extent_num_bytes &
3396                                         ~((u64)root->sectorsize - 1);
3397                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3398                                                          extent_num_bytes);
3399                                 num_dec = (orig_num_bytes -
3400                                            extent_num_bytes);
3401                                 if (root->ref_cows && extent_start != 0)
3402                                         inode_sub_bytes(inode, num_dec);
3403                                 btrfs_mark_buffer_dirty(leaf);
3404                         } else {
3405                                 extent_num_bytes =
3406                                         btrfs_file_extent_disk_num_bytes(leaf,
3407                                                                          fi);
3408                                 extent_offset = found_key.offset -
3409                                         btrfs_file_extent_offset(leaf, fi);
3410
3411                                 /* FIXME blocksize != 4096 */
3412                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3413                                 if (extent_start != 0) {
3414                                         found_extent = 1;
3415                                         if (root->ref_cows)
3416                                                 inode_sub_bytes(inode, num_dec);
3417                                 }
3418                         }
3419                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3420                         /*
3421                          * we can't truncate inline items that have had
3422                          * special encodings
3423                          */
3424                         if (!del_item &&
3425                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3426                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3427                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3428                                 u32 size = new_size - found_key.offset;
3429
3430                                 if (root->ref_cows) {
3431                                         inode_sub_bytes(inode, item_end + 1 -
3432                                                         new_size);
3433                                 }
3434                                 size =
3435                                     btrfs_file_extent_calc_inline_size(size);
3436                                 ret = btrfs_truncate_item(trans, root, path,
3437                                                           size, 1);
3438                                 BUG_ON(ret);
3439                         } else if (root->ref_cows) {
3440                                 inode_sub_bytes(inode, item_end + 1 -
3441                                                 found_key.offset);
3442                         }
3443                 }
3444 delete:
3445                 if (del_item) {
3446                         if (!pending_del_nr) {
3447                                 /* no pending yet, add ourselves */
3448                                 pending_del_slot = path->slots[0];
3449                                 pending_del_nr = 1;
3450                         } else if (pending_del_nr &&
3451                                    path->slots[0] + 1 == pending_del_slot) {
3452                                 /* hop on the pending chunk */
3453                                 pending_del_nr++;
3454                                 pending_del_slot = path->slots[0];
3455                         } else {
3456                                 BUG();
3457                         }
3458                 } else {
3459                         break;
3460                 }
3461                 if (found_extent && (root->ref_cows ||
3462                                      root == root->fs_info->tree_root)) {
3463                         btrfs_set_path_blocking(path);
3464                         ret = btrfs_free_extent(trans, root, extent_start,
3465                                                 extent_num_bytes, 0,
3466                                                 btrfs_header_owner(leaf),
3467                                                 ino, extent_offset);
3468                         BUG_ON(ret);
3469                 }
3470
3471                 if (found_type == BTRFS_INODE_ITEM_KEY)
3472                         break;
3473
3474                 if (path->slots[0] == 0 ||
3475                     path->slots[0] != pending_del_slot) {
3476                         if (root->ref_cows) {
3477                                 err = -EAGAIN;
3478                                 goto out;
3479                         }
3480                         if (pending_del_nr) {
3481                                 ret = btrfs_del_items(trans, root, path,
3482                                                 pending_del_slot,
3483                                                 pending_del_nr);
3484                                 BUG_ON(ret);
3485                                 pending_del_nr = 0;
3486                         }
3487                         btrfs_release_path(root, path);
3488                         goto search_again;
3489                 } else {
3490                         path->slots[0]--;
3491                 }
3492         }
3493 out:
3494         if (pending_del_nr) {
3495                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3496                                       pending_del_nr);
3497                 BUG_ON(ret);
3498         }
3499         btrfs_free_path(path);
3500         return err;
3501 }
3502
3503 /*
3504  * taken from block_truncate_page, but does cow as it zeros out
3505  * any bytes left in the last page in the file.
3506  */
3507 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3508 {
3509         struct inode *inode = mapping->host;
3510         struct btrfs_root *root = BTRFS_I(inode)->root;
3511         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3512         struct btrfs_ordered_extent *ordered;
3513         struct extent_state *cached_state = NULL;
3514         char *kaddr;
3515         u32 blocksize = root->sectorsize;
3516         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3517         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3518         struct page *page;
3519         int ret = 0;
3520         u64 page_start;
3521         u64 page_end;
3522
3523         if ((offset & (blocksize - 1)) == 0)
3524                 goto out;
3525         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3526         if (ret)
3527                 goto out;
3528
3529         ret = -ENOMEM;
3530 again:
3531         page = grab_cache_page(mapping, index);
3532         if (!page) {
3533                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3534                 goto out;
3535         }
3536
3537         page_start = page_offset(page);
3538         page_end = page_start + PAGE_CACHE_SIZE - 1;
3539
3540         if (!PageUptodate(page)) {
3541                 ret = btrfs_readpage(NULL, page);
3542                 lock_page(page);
3543                 if (page->mapping != mapping) {
3544                         unlock_page(page);
3545                         page_cache_release(page);
3546                         goto again;
3547                 }
3548                 if (!PageUptodate(page)) {
3549                         ret = -EIO;
3550                         goto out_unlock;
3551                 }
3552         }
3553         wait_on_page_writeback(page);
3554
3555         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3556                          GFP_NOFS);
3557         set_page_extent_mapped(page);
3558
3559         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3560         if (ordered) {
3561                 unlock_extent_cached(io_tree, page_start, page_end,
3562                                      &cached_state, GFP_NOFS);
3563                 unlock_page(page);
3564                 page_cache_release(page);
3565                 btrfs_start_ordered_extent(inode, ordered, 1);
3566                 btrfs_put_ordered_extent(ordered);
3567                 goto again;
3568         }
3569
3570         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3571                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3572                           0, 0, &cached_state, GFP_NOFS);
3573
3574         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3575                                         &cached_state);
3576         if (ret) {
3577                 unlock_extent_cached(io_tree, page_start, page_end,
3578                                      &cached_state, GFP_NOFS);
3579                 goto out_unlock;
3580         }
3581
3582         ret = 0;
3583         if (offset != PAGE_CACHE_SIZE) {
3584                 kaddr = kmap(page);
3585                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3586                 flush_dcache_page(page);
3587                 kunmap(page);
3588         }
3589         ClearPageChecked(page);
3590         set_page_dirty(page);
3591         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3592                              GFP_NOFS);
3593
3594 out_unlock:
3595         if (ret)
3596                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3597         unlock_page(page);
3598         page_cache_release(page);
3599 out:
3600         return ret;
3601 }
3602
3603 /*
3604  * This function puts in dummy file extents for the area we're creating a hole
3605  * for.  So if we are truncating this file to a larger size we need to insert
3606  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3607  * the range between oldsize and size
3608  */
3609 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3610 {
3611         struct btrfs_trans_handle *trans;
3612         struct btrfs_root *root = BTRFS_I(inode)->root;
3613         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3614         struct extent_map *em = NULL;
3615         struct extent_state *cached_state = NULL;
3616         u64 mask = root->sectorsize - 1;
3617         u64 hole_start = (oldsize + mask) & ~mask;
3618         u64 block_end = (size + mask) & ~mask;
3619         u64 last_byte;
3620         u64 cur_offset;
3621         u64 hole_size;
3622         int err = 0;
3623
3624         if (size <= hole_start)
3625                 return 0;
3626
3627         while (1) {
3628                 struct btrfs_ordered_extent *ordered;
3629                 btrfs_wait_ordered_range(inode, hole_start,
3630                                          block_end - hole_start);
3631                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3632                                  &cached_state, GFP_NOFS);
3633                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3634                 if (!ordered)
3635                         break;
3636                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3637                                      &cached_state, GFP_NOFS);
3638                 btrfs_put_ordered_extent(ordered);
3639         }
3640
3641         cur_offset = hole_start;
3642         while (1) {
3643                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3644                                 block_end - cur_offset, 0);
3645                 BUG_ON(IS_ERR(em) || !em);
3646                 last_byte = min(extent_map_end(em), block_end);
3647                 last_byte = (last_byte + mask) & ~mask;
3648                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3649                         u64 hint_byte = 0;
3650                         hole_size = last_byte - cur_offset;
3651
3652                         trans = btrfs_start_transaction(root, 2);
3653                         if (IS_ERR(trans)) {
3654                                 err = PTR_ERR(trans);
3655                                 break;
3656                         }
3657                         btrfs_set_trans_block_group(trans, inode);
3658
3659                         err = btrfs_drop_extents(trans, inode, cur_offset,
3660                                                  cur_offset + hole_size,
3661                                                  &hint_byte, 1);
3662                         if (err)
3663                                 break;
3664
3665                         err = btrfs_insert_file_extent(trans, root,
3666                                         btrfs_ino(inode), cur_offset, 0,
3667                                         0, hole_size, 0, hole_size,
3668                                         0, 0, 0);
3669                         if (err)
3670                                 break;
3671
3672                         btrfs_drop_extent_cache(inode, hole_start,
3673                                         last_byte - 1, 0);
3674
3675                         btrfs_end_transaction(trans, root);
3676                 }
3677                 free_extent_map(em);
3678                 em = NULL;
3679                 cur_offset = last_byte;
3680                 if (cur_offset >= block_end)
3681                         break;
3682         }
3683
3684         free_extent_map(em);
3685         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3686                              GFP_NOFS);
3687         return err;
3688 }
3689
3690 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3691 {
3692         loff_t oldsize = i_size_read(inode);
3693         int ret;
3694
3695         if (newsize == oldsize)
3696                 return 0;
3697
3698         if (newsize > oldsize) {
3699                 i_size_write(inode, newsize);
3700                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3701                 truncate_pagecache(inode, oldsize, newsize);
3702                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3703                 if (ret) {
3704                         btrfs_setsize(inode, oldsize);
3705                         return ret;
3706                 }
3707
3708                 mark_inode_dirty(inode);
3709         } else {
3710
3711                 /*
3712                  * We're truncating a file that used to have good data down to
3713                  * zero. Make sure it gets into the ordered flush list so that
3714                  * any new writes get down to disk quickly.
3715                  */
3716                 if (newsize == 0)
3717                         BTRFS_I(inode)->ordered_data_close = 1;
3718
3719                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3720                 truncate_setsize(inode, newsize);
3721                 ret = btrfs_truncate(inode);
3722         }
3723
3724         return ret;
3725 }
3726
3727 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3728 {
3729         struct inode *inode = dentry->d_inode;
3730         struct btrfs_root *root = BTRFS_I(inode)->root;
3731         int err;
3732
3733         if (btrfs_root_readonly(root))
3734                 return -EROFS;
3735
3736         err = inode_change_ok(inode, attr);
3737         if (err)
3738                 return err;
3739
3740         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3741                 err = btrfs_setsize(inode, attr->ia_size);
3742                 if (err)
3743                         return err;
3744         }
3745
3746         if (attr->ia_valid) {
3747                 setattr_copy(inode, attr);
3748                 mark_inode_dirty(inode);
3749
3750                 if (attr->ia_valid & ATTR_MODE)
3751                         err = btrfs_acl_chmod(inode);
3752         }
3753
3754         return err;
3755 }
3756
3757 void btrfs_evict_inode(struct inode *inode)
3758 {
3759         struct btrfs_trans_handle *trans;
3760         struct btrfs_root *root = BTRFS_I(inode)->root;
3761         unsigned long nr;
3762         int ret;
3763
3764         trace_btrfs_inode_evict(inode);
3765
3766         truncate_inode_pages(&inode->i_data, 0);
3767         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3768                                root == root->fs_info->tree_root))
3769                 goto no_delete;
3770
3771         if (is_bad_inode(inode)) {
3772                 btrfs_orphan_del(NULL, inode);
3773                 goto no_delete;
3774         }
3775         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3776         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3777
3778         if (root->fs_info->log_root_recovering) {
3779                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3780                 goto no_delete;
3781         }
3782
3783         if (inode->i_nlink > 0) {
3784                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3785                 goto no_delete;
3786         }
3787
3788         btrfs_i_size_write(inode, 0);
3789
3790         while (1) {
3791                 trans = btrfs_start_transaction(root, 0);
3792                 BUG_ON(IS_ERR(trans));
3793                 btrfs_set_trans_block_group(trans, inode);
3794                 trans->block_rsv = root->orphan_block_rsv;
3795
3796                 ret = btrfs_block_rsv_check(trans, root,
3797                                             root->orphan_block_rsv, 0, 5);
3798                 if (ret) {
3799                         BUG_ON(ret != -EAGAIN);
3800                         ret = btrfs_commit_transaction(trans, root);
3801                         BUG_ON(ret);
3802                         continue;
3803                 }
3804
3805                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3806                 if (ret != -EAGAIN)
3807                         break;
3808
3809                 nr = trans->blocks_used;
3810                 btrfs_end_transaction(trans, root);
3811                 trans = NULL;
3812                 btrfs_btree_balance_dirty(root, nr);
3813
3814         }
3815
3816         if (ret == 0) {
3817                 ret = btrfs_orphan_del(trans, inode);
3818                 BUG_ON(ret);
3819         }
3820
3821         if (!(root == root->fs_info->tree_root ||
3822               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3823                 btrfs_return_ino(root, btrfs_ino(inode));
3824
3825         nr = trans->blocks_used;
3826         btrfs_end_transaction(trans, root);
3827         btrfs_btree_balance_dirty(root, nr);
3828 no_delete:
3829         end_writeback(inode);
3830         return;
3831 }
3832
3833 /*
3834  * this returns the key found in the dir entry in the location pointer.
3835  * If no dir entries were found, location->objectid is 0.
3836  */
3837 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3838                                struct btrfs_key *location)
3839 {
3840         const char *name = dentry->d_name.name;
3841         int namelen = dentry->d_name.len;
3842         struct btrfs_dir_item *di;
3843         struct btrfs_path *path;
3844         struct btrfs_root *root = BTRFS_I(dir)->root;
3845         int ret = 0;
3846
3847         path = btrfs_alloc_path();
3848         BUG_ON(!path);
3849
3850         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3851                                     namelen, 0);
3852         if (IS_ERR(di))
3853                 ret = PTR_ERR(di);
3854
3855         if (!di || IS_ERR(di))
3856                 goto out_err;
3857
3858         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3859 out:
3860         btrfs_free_path(path);
3861         return ret;
3862 out_err:
3863         location->objectid = 0;
3864         goto out;
3865 }
3866
3867 /*
3868  * when we hit a tree root in a directory, the btrfs part of the inode
3869  * needs to be changed to reflect the root directory of the tree root.  This
3870  * is kind of like crossing a mount point.
3871  */
3872 static int fixup_tree_root_location(struct btrfs_root *root,
3873                                     struct inode *dir,
3874                                     struct dentry *dentry,
3875                                     struct btrfs_key *location,
3876                                     struct btrfs_root **sub_root)
3877 {
3878         struct btrfs_path *path;
3879         struct btrfs_root *new_root;
3880         struct btrfs_root_ref *ref;
3881         struct extent_buffer *leaf;
3882         int ret;
3883         int err = 0;
3884
3885         path = btrfs_alloc_path();
3886         if (!path) {
3887                 err = -ENOMEM;
3888                 goto out;
3889         }
3890
3891         err = -ENOENT;
3892         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3893                                   BTRFS_I(dir)->root->root_key.objectid,
3894                                   location->objectid);
3895         if (ret) {
3896                 if (ret < 0)
3897                         err = ret;
3898                 goto out;
3899         }
3900
3901         leaf = path->nodes[0];
3902         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3903         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3904             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3905                 goto out;
3906
3907         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3908                                    (unsigned long)(ref + 1),
3909                                    dentry->d_name.len);
3910         if (ret)
3911                 goto out;
3912
3913         btrfs_release_path(root->fs_info->tree_root, path);
3914
3915         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3916         if (IS_ERR(new_root)) {
3917                 err = PTR_ERR(new_root);
3918                 goto out;
3919         }
3920
3921         if (btrfs_root_refs(&new_root->root_item) == 0) {
3922                 err = -ENOENT;
3923                 goto out;
3924         }
3925
3926         *sub_root = new_root;
3927         location->objectid = btrfs_root_dirid(&new_root->root_item);
3928         location->type = BTRFS_INODE_ITEM_KEY;
3929         location->offset = 0;
3930         err = 0;
3931 out:
3932         btrfs_free_path(path);
3933         return err;
3934 }
3935
3936 static void inode_tree_add(struct inode *inode)
3937 {
3938         struct btrfs_root *root = BTRFS_I(inode)->root;
3939         struct btrfs_inode *entry;
3940         struct rb_node **p;
3941         struct rb_node *parent;
3942         u64 ino = btrfs_ino(inode);
3943 again:
3944         p = &root->inode_tree.rb_node;
3945         parent = NULL;
3946
3947         if (inode_unhashed(inode))
3948                 return;
3949
3950         spin_lock(&root->inode_lock);
3951         while (*p) {
3952                 parent = *p;
3953                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3954
3955                 if (ino < btrfs_ino(&entry->vfs_inode))
3956                         p = &parent->rb_left;
3957                 else if (ino > btrfs_ino(&entry->vfs_inode))
3958                         p = &parent->rb_right;
3959                 else {
3960                         WARN_ON(!(entry->vfs_inode.i_state &
3961                                   (I_WILL_FREE | I_FREEING)));
3962                         rb_erase(parent, &root->inode_tree);
3963                         RB_CLEAR_NODE(parent);
3964                         spin_unlock(&root->inode_lock);
3965                         goto again;
3966                 }
3967         }
3968         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3969         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3970         spin_unlock(&root->inode_lock);
3971 }
3972
3973 static void inode_tree_del(struct inode *inode)
3974 {
3975         struct btrfs_root *root = BTRFS_I(inode)->root;
3976         int empty = 0;
3977
3978         spin_lock(&root->inode_lock);
3979         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3980                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3981                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3982                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3983         }
3984         spin_unlock(&root->inode_lock);
3985
3986         /*
3987          * Free space cache has inodes in the tree root, but the tree root has a
3988          * root_refs of 0, so this could end up dropping the tree root as a
3989          * snapshot, so we need the extra !root->fs_info->tree_root check to
3990          * make sure we don't drop it.
3991          */
3992         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3993             root != root->fs_info->tree_root) {
3994                 synchronize_srcu(&root->fs_info->subvol_srcu);
3995                 spin_lock(&root->inode_lock);
3996                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3997                 spin_unlock(&root->inode_lock);
3998                 if (empty)
3999                         btrfs_add_dead_root(root);
4000         }
4001 }
4002
4003 int btrfs_invalidate_inodes(struct btrfs_root *root)
4004 {
4005         struct rb_node *node;
4006         struct rb_node *prev;
4007         struct btrfs_inode *entry;
4008         struct inode *inode;
4009         u64 objectid = 0;
4010
4011         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4012
4013         spin_lock(&root->inode_lock);
4014 again:
4015         node = root->inode_tree.rb_node;
4016         prev = NULL;
4017         while (node) {
4018                 prev = node;
4019                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4020
4021                 if (objectid < btrfs_ino(&entry->vfs_inode))
4022                         node = node->rb_left;
4023                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4024                         node = node->rb_right;
4025                 else
4026                         break;
4027         }
4028         if (!node) {
4029                 while (prev) {
4030                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4031                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4032                                 node = prev;
4033                                 break;
4034                         }
4035                         prev = rb_next(prev);
4036                 }
4037         }
4038         while (node) {
4039                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4040                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4041                 inode = igrab(&entry->vfs_inode);
4042                 if (inode) {
4043                         spin_unlock(&root->inode_lock);
4044                         if (atomic_read(&inode->i_count) > 1)
4045                                 d_prune_aliases(inode);
4046                         /*
4047                          * btrfs_drop_inode will have it removed from
4048                          * the inode cache when its usage count
4049                          * hits zero.
4050                          */
4051                         iput(inode);
4052                         cond_resched();
4053                         spin_lock(&root->inode_lock);
4054                         goto again;
4055                 }
4056
4057                 if (cond_resched_lock(&root->inode_lock))
4058                         goto again;
4059
4060                 node = rb_next(node);
4061         }
4062         spin_unlock(&root->inode_lock);
4063         return 0;
4064 }
4065
4066 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4067 {
4068         struct btrfs_iget_args *args = p;
4069         inode->i_ino = args->ino;
4070         BTRFS_I(inode)->root = args->root;
4071         btrfs_set_inode_space_info(args->root, inode);
4072         return 0;
4073 }
4074
4075 static int btrfs_find_actor(struct inode *inode, void *opaque)
4076 {
4077         struct btrfs_iget_args *args = opaque;
4078         return args->ino == btrfs_ino(inode) &&
4079                 args->root == BTRFS_I(inode)->root;
4080 }
4081
4082 static struct inode *btrfs_iget_locked(struct super_block *s,
4083                                        u64 objectid,
4084                                        struct btrfs_root *root)
4085 {
4086         struct inode *inode;
4087         struct btrfs_iget_args args;
4088         args.ino = objectid;
4089         args.root = root;
4090
4091         inode = iget5_locked(s, objectid, btrfs_find_actor,
4092                              btrfs_init_locked_inode,
4093                              (void *)&args);
4094         return inode;
4095 }
4096
4097 /* Get an inode object given its location and corresponding root.
4098  * Returns in *is_new if the inode was read from disk
4099  */
4100 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4101                          struct btrfs_root *root, int *new)
4102 {
4103         struct inode *inode;
4104
4105         inode = btrfs_iget_locked(s, location->objectid, root);
4106         if (!inode)
4107                 return ERR_PTR(-ENOMEM);
4108
4109         if (inode->i_state & I_NEW) {
4110                 BTRFS_I(inode)->root = root;
4111                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4112                 btrfs_read_locked_inode(inode);
4113                 inode_tree_add(inode);
4114                 unlock_new_inode(inode);
4115                 if (new)
4116                         *new = 1;
4117         }
4118
4119         return inode;
4120 }
4121
4122 static struct inode *new_simple_dir(struct super_block *s,
4123                                     struct btrfs_key *key,
4124                                     struct btrfs_root *root)
4125 {
4126         struct inode *inode = new_inode(s);
4127
4128         if (!inode)
4129                 return ERR_PTR(-ENOMEM);
4130
4131         BTRFS_I(inode)->root = root;
4132         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4133         BTRFS_I(inode)->dummy_inode = 1;
4134
4135         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4136         inode->i_op = &simple_dir_inode_operations;
4137         inode->i_fop = &simple_dir_operations;
4138         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4139         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4140
4141         return inode;
4142 }
4143
4144 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4145 {
4146         struct inode *inode;
4147         struct btrfs_root *root = BTRFS_I(dir)->root;
4148         struct btrfs_root *sub_root = root;
4149         struct btrfs_key location;
4150         int index;
4151         int ret;
4152
4153         if (dentry->d_name.len > BTRFS_NAME_LEN)
4154                 return ERR_PTR(-ENAMETOOLONG);
4155
4156         ret = btrfs_inode_by_name(dir, dentry, &location);
4157
4158         if (ret < 0)
4159                 return ERR_PTR(ret);
4160
4161         if (location.objectid == 0)
4162                 return NULL;
4163
4164         if (location.type == BTRFS_INODE_ITEM_KEY) {
4165                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4166                 return inode;
4167         }
4168
4169         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4170
4171         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4172         ret = fixup_tree_root_location(root, dir, dentry,
4173                                        &location, &sub_root);
4174         if (ret < 0) {
4175                 if (ret != -ENOENT)
4176                         inode = ERR_PTR(ret);
4177                 else
4178                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4179         } else {
4180                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4181         }
4182         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4183
4184         if (!IS_ERR(inode) && root != sub_root) {
4185                 down_read(&root->fs_info->cleanup_work_sem);
4186                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4187                         ret = btrfs_orphan_cleanup(sub_root);
4188                 up_read(&root->fs_info->cleanup_work_sem);
4189                 if (ret)
4190                         inode = ERR_PTR(ret);
4191         }
4192
4193         return inode;
4194 }
4195
4196 static int btrfs_dentry_delete(const struct dentry *dentry)
4197 {
4198         struct btrfs_root *root;
4199
4200         if (!dentry->d_inode && !IS_ROOT(dentry))
4201                 dentry = dentry->d_parent;
4202
4203         if (dentry->d_inode) {
4204                 root = BTRFS_I(dentry->d_inode)->root;
4205                 if (btrfs_root_refs(&root->root_item) == 0)
4206                         return 1;
4207         }
4208         return 0;
4209 }
4210
4211 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4212                                    struct nameidata *nd)
4213 {
4214         struct inode *inode;
4215
4216         inode = btrfs_lookup_dentry(dir, dentry);
4217         if (IS_ERR(inode))
4218                 return ERR_CAST(inode);
4219
4220         return d_splice_alias(inode, dentry);
4221 }
4222
4223 static unsigned char btrfs_filetype_table[] = {
4224         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4225 };
4226
4227 static int btrfs_real_readdir(struct file *filp, void *dirent,
4228                               filldir_t filldir)
4229 {
4230         struct inode *inode = filp->f_dentry->d_inode;
4231         struct btrfs_root *root = BTRFS_I(inode)->root;
4232         struct btrfs_item *item;
4233         struct btrfs_dir_item *di;
4234         struct btrfs_key key;
4235         struct btrfs_key found_key;
4236         struct btrfs_path *path;
4237         int ret;
4238         struct extent_buffer *leaf;
4239         int slot;
4240         unsigned char d_type;
4241         int over = 0;
4242         u32 di_cur;
4243         u32 di_total;
4244         u32 di_len;
4245         int key_type = BTRFS_DIR_INDEX_KEY;
4246         char tmp_name[32];
4247         char *name_ptr;
4248         int name_len;
4249
4250         /* FIXME, use a real flag for deciding about the key type */
4251         if (root->fs_info->tree_root == root)
4252                 key_type = BTRFS_DIR_ITEM_KEY;
4253
4254         /* special case for "." */
4255         if (filp->f_pos == 0) {
4256                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4257                 if (over)
4258                         return 0;
4259                 filp->f_pos = 1;
4260         }
4261         /* special case for .., just use the back ref */
4262         if (filp->f_pos == 1) {
4263                 u64 pino = parent_ino(filp->f_path.dentry);
4264                 over = filldir(dirent, "..", 2,
4265                                2, pino, DT_DIR);
4266                 if (over)
4267                         return 0;
4268                 filp->f_pos = 2;
4269         }
4270         path = btrfs_alloc_path();
4271         path->reada = 2;
4272
4273         btrfs_set_key_type(&key, key_type);
4274         key.offset = filp->f_pos;
4275         key.objectid = btrfs_ino(inode);
4276
4277         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4278         if (ret < 0)
4279                 goto err;
4280
4281         while (1) {
4282                 leaf = path->nodes[0];
4283                 slot = path->slots[0];
4284                 if (slot >= btrfs_header_nritems(leaf)) {
4285                         ret = btrfs_next_leaf(root, path);
4286                         if (ret < 0)
4287                                 goto err;
4288                         else if (ret > 0)
4289                                 break;
4290                         continue;
4291                 }
4292
4293                 item = btrfs_item_nr(leaf, slot);
4294                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4295
4296                 if (found_key.objectid != key.objectid)
4297                         break;
4298                 if (btrfs_key_type(&found_key) != key_type)
4299                         break;
4300                 if (found_key.offset < filp->f_pos)
4301                         goto next;
4302
4303                 filp->f_pos = found_key.offset;
4304
4305                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4306                 di_cur = 0;
4307                 di_total = btrfs_item_size(leaf, item);
4308
4309                 while (di_cur < di_total) {
4310                         struct btrfs_key location;
4311
4312                         if (verify_dir_item(root, leaf, di))
4313                                 break;
4314
4315                         name_len = btrfs_dir_name_len(leaf, di);
4316                         if (name_len <= sizeof(tmp_name)) {
4317                                 name_ptr = tmp_name;
4318                         } else {
4319                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4320                                 if (!name_ptr) {
4321                                         ret = -ENOMEM;
4322                                         goto err;
4323                                 }
4324                         }
4325                         read_extent_buffer(leaf, name_ptr,
4326                                            (unsigned long)(di + 1), name_len);
4327
4328                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4329                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4330
4331                         /* is this a reference to our own snapshot? If so
4332                          * skip it
4333                          */
4334                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4335                             location.objectid == root->root_key.objectid) {
4336                                 over = 0;
4337                                 goto skip;
4338                         }
4339                         over = filldir(dirent, name_ptr, name_len,
4340                                        found_key.offset, location.objectid,
4341                                        d_type);
4342
4343 skip:
4344                         if (name_ptr != tmp_name)
4345                                 kfree(name_ptr);
4346
4347                         if (over)
4348                                 goto nopos;
4349                         di_len = btrfs_dir_name_len(leaf, di) +
4350                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4351                         di_cur += di_len;
4352                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4353                 }
4354 next:
4355                 path->slots[0]++;
4356         }
4357
4358         /* Reached end of directory/root. Bump pos past the last item. */
4359         if (key_type == BTRFS_DIR_INDEX_KEY)
4360                 /*
4361                  * 32-bit glibc will use getdents64, but then strtol -
4362                  * so the last number we can serve is this.
4363                  */
4364                 filp->f_pos = 0x7fffffff;
4365         else
4366                 filp->f_pos++;
4367 nopos:
4368         ret = 0;
4369 err:
4370         btrfs_free_path(path);
4371         return ret;
4372 }
4373
4374 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4375 {
4376         struct btrfs_root *root = BTRFS_I(inode)->root;
4377         struct btrfs_trans_handle *trans;
4378         int ret = 0;
4379         bool nolock = false;
4380
4381         if (BTRFS_I(inode)->dummy_inode)
4382                 return 0;
4383
4384         smp_mb();
4385         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4386
4387         if (wbc->sync_mode == WB_SYNC_ALL) {
4388                 if (nolock)
4389                         trans = btrfs_join_transaction_nolock(root, 1);
4390                 else
4391                         trans = btrfs_join_transaction(root, 1);
4392                 if (IS_ERR(trans))
4393                         return PTR_ERR(trans);
4394                 btrfs_set_trans_block_group(trans, inode);
4395                 if (nolock)
4396                         ret = btrfs_end_transaction_nolock(trans, root);
4397                 else
4398                         ret = btrfs_commit_transaction(trans, root);
4399         }
4400         return ret;
4401 }
4402
4403 /*
4404  * This is somewhat expensive, updating the tree every time the
4405  * inode changes.  But, it is most likely to find the inode in cache.
4406  * FIXME, needs more benchmarking...there are no reasons other than performance
4407  * to keep or drop this code.
4408  */
4409 void btrfs_dirty_inode(struct inode *inode)
4410 {
4411         struct btrfs_root *root = BTRFS_I(inode)->root;
4412         struct btrfs_trans_handle *trans;
4413         int ret;
4414
4415         if (BTRFS_I(inode)->dummy_inode)
4416                 return;
4417
4418         trans = btrfs_join_transaction(root, 1);
4419         BUG_ON(IS_ERR(trans));
4420         btrfs_set_trans_block_group(trans, inode);
4421
4422         ret = btrfs_update_inode(trans, root, inode);
4423         if (ret && ret == -ENOSPC) {
4424                 /* whoops, lets try again with the full transaction */
4425                 btrfs_end_transaction(trans, root);
4426                 trans = btrfs_start_transaction(root, 1);
4427                 if (IS_ERR(trans)) {
4428                         if (printk_ratelimit()) {
4429                                 printk(KERN_ERR "btrfs: fail to "
4430                                        "dirty  inode %llu error %ld\n",
4431                                        (unsigned long long)btrfs_ino(inode),
4432                                        PTR_ERR(trans));
4433                         }
4434                         return;
4435                 }
4436                 btrfs_set_trans_block_group(trans, inode);
4437
4438                 ret = btrfs_update_inode(trans, root, inode);
4439                 if (ret) {
4440                         if (printk_ratelimit()) {
4441                                 printk(KERN_ERR "btrfs: fail to "
4442                                        "dirty  inode %llu error %d\n",
4443                                        (unsigned long long)btrfs_ino(inode),
4444                                        ret);
4445                         }
4446                 }
4447         }
4448         btrfs_end_transaction(trans, root);
4449 }
4450
4451 /*
4452  * find the highest existing sequence number in a directory
4453  * and then set the in-memory index_cnt variable to reflect
4454  * free sequence numbers
4455  */
4456 static int btrfs_set_inode_index_count(struct inode *inode)
4457 {
4458         struct btrfs_root *root = BTRFS_I(inode)->root;
4459         struct btrfs_key key, found_key;
4460         struct btrfs_path *path;
4461         struct extent_buffer *leaf;
4462         int ret;
4463
4464         key.objectid = btrfs_ino(inode);
4465         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4466         key.offset = (u64)-1;
4467
4468         path = btrfs_alloc_path();
4469         if (!path)
4470                 return -ENOMEM;
4471
4472         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4473         if (ret < 0)
4474                 goto out;
4475         /* FIXME: we should be able to handle this */
4476         if (ret == 0)
4477                 goto out;
4478         ret = 0;
4479
4480         /*
4481          * MAGIC NUMBER EXPLANATION:
4482          * since we search a directory based on f_pos we have to start at 2
4483          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4484          * else has to start at 2
4485          */
4486         if (path->slots[0] == 0) {
4487                 BTRFS_I(inode)->index_cnt = 2;
4488                 goto out;
4489         }
4490
4491         path->slots[0]--;
4492
4493         leaf = path->nodes[0];
4494         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4495
4496         if (found_key.objectid != btrfs_ino(inode) ||
4497             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4498                 BTRFS_I(inode)->index_cnt = 2;
4499                 goto out;
4500         }
4501
4502         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4503 out:
4504         btrfs_free_path(path);
4505         return ret;
4506 }
4507
4508 /*
4509  * helper to find a free sequence number in a given directory.  This current
4510  * code is very simple, later versions will do smarter things in the btree
4511  */
4512 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4513 {
4514         int ret = 0;
4515
4516         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4517                 ret = btrfs_set_inode_index_count(dir);
4518                 if (ret)
4519                         return ret;
4520         }
4521
4522         *index = BTRFS_I(dir)->index_cnt;
4523         BTRFS_I(dir)->index_cnt++;
4524
4525         return ret;
4526 }
4527
4528 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4529                                      struct btrfs_root *root,
4530                                      struct inode *dir,
4531                                      const char *name, int name_len,
4532                                      u64 ref_objectid, u64 objectid,
4533                                      u64 alloc_hint, int mode, u64 *index)
4534 {
4535         struct inode *inode;
4536         struct btrfs_inode_item *inode_item;
4537         struct btrfs_key *location;
4538         struct btrfs_path *path;
4539         struct btrfs_inode_ref *ref;
4540         struct btrfs_key key[2];
4541         u32 sizes[2];
4542         unsigned long ptr;
4543         int ret;
4544         int owner;
4545
4546         path = btrfs_alloc_path();
4547         BUG_ON(!path);
4548
4549         inode = new_inode(root->fs_info->sb);
4550         if (!inode) {
4551                 btrfs_free_path(path);
4552                 return ERR_PTR(-ENOMEM);
4553         }
4554
4555         /*
4556          * we have to initialize this early, so we can reclaim the inode
4557          * number if we fail afterwards in this function.
4558          */
4559         inode->i_ino = objectid;
4560
4561         if (dir) {
4562                 trace_btrfs_inode_request(dir);
4563
4564                 ret = btrfs_set_inode_index(dir, index);
4565                 if (ret) {
4566                         btrfs_free_path(path);
4567                         iput(inode);
4568                         return ERR_PTR(ret);
4569                 }
4570         }
4571         /*
4572          * index_cnt is ignored for everything but a dir,
4573          * btrfs_get_inode_index_count has an explanation for the magic
4574          * number
4575          */
4576         BTRFS_I(inode)->index_cnt = 2;
4577         BTRFS_I(inode)->root = root;
4578         BTRFS_I(inode)->generation = trans->transid;
4579         inode->i_generation = BTRFS_I(inode)->generation;
4580         btrfs_set_inode_space_info(root, inode);
4581
4582         if (mode & S_IFDIR)
4583                 owner = 0;
4584         else
4585                 owner = 1;
4586         BTRFS_I(inode)->block_group =
4587                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4588
4589         key[0].objectid = objectid;
4590         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4591         key[0].offset = 0;
4592
4593         key[1].objectid = objectid;
4594         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4595         key[1].offset = ref_objectid;
4596
4597         sizes[0] = sizeof(struct btrfs_inode_item);
4598         sizes[1] = name_len + sizeof(*ref);
4599
4600         path->leave_spinning = 1;
4601         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4602         if (ret != 0)
4603                 goto fail;
4604
4605         inode_init_owner(inode, dir, mode);
4606         inode_set_bytes(inode, 0);
4607         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4608         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4609                                   struct btrfs_inode_item);
4610         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4611
4612         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4613                              struct btrfs_inode_ref);
4614         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4615         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4616         ptr = (unsigned long)(ref + 1);
4617         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4618
4619         btrfs_mark_buffer_dirty(path->nodes[0]);
4620         btrfs_free_path(path);
4621
4622         location = &BTRFS_I(inode)->location;
4623         location->objectid = objectid;
4624         location->offset = 0;
4625         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4626
4627         btrfs_inherit_iflags(inode, dir);
4628
4629         if ((mode & S_IFREG)) {
4630                 if (btrfs_test_opt(root, NODATASUM))
4631                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4632                 if (btrfs_test_opt(root, NODATACOW) ||
4633                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4634                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4635         }
4636
4637         insert_inode_hash(inode);
4638         inode_tree_add(inode);
4639
4640         trace_btrfs_inode_new(inode);
4641
4642         return inode;
4643 fail:
4644         if (dir)
4645                 BTRFS_I(dir)->index_cnt--;
4646         btrfs_free_path(path);
4647         iput(inode);
4648         return ERR_PTR(ret);
4649 }
4650
4651 static inline u8 btrfs_inode_type(struct inode *inode)
4652 {
4653         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4654 }
4655
4656 /*
4657  * utility function to add 'inode' into 'parent_inode' with
4658  * a give name and a given sequence number.
4659  * if 'add_backref' is true, also insert a backref from the
4660  * inode to the parent directory.
4661  */
4662 int btrfs_add_link(struct btrfs_trans_handle *trans,
4663                    struct inode *parent_inode, struct inode *inode,
4664                    const char *name, int name_len, int add_backref, u64 index)
4665 {
4666         int ret = 0;
4667         struct btrfs_key key;
4668         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4669         u64 ino = btrfs_ino(inode);
4670         u64 parent_ino = btrfs_ino(parent_inode);
4671
4672         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4673                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4674         } else {
4675                 key.objectid = ino;
4676                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4677                 key.offset = 0;
4678         }
4679
4680         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4681                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4682                                          key.objectid, root->root_key.objectid,
4683                                          parent_ino, index, name, name_len);
4684         } else if (add_backref) {
4685                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4686                                              parent_ino, index);
4687         }
4688
4689         if (ret == 0) {
4690                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4691                                             parent_ino, &key,
4692                                             btrfs_inode_type(inode), index);
4693                 BUG_ON(ret);
4694
4695                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4696                                    name_len * 2);
4697                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4698                 ret = btrfs_update_inode(trans, root, parent_inode);
4699         }
4700         return ret;
4701 }
4702
4703 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4704                             struct inode *dir, struct dentry *dentry,
4705                             struct inode *inode, int backref, u64 index)
4706 {
4707         int err = btrfs_add_link(trans, dir, inode,
4708                                  dentry->d_name.name, dentry->d_name.len,
4709                                  backref, index);
4710         if (!err) {
4711                 d_instantiate(dentry, inode);
4712                 return 0;
4713         }
4714         if (err > 0)
4715                 err = -EEXIST;
4716         return err;
4717 }
4718
4719 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4720                         int mode, dev_t rdev)
4721 {
4722         struct btrfs_trans_handle *trans;
4723         struct btrfs_root *root = BTRFS_I(dir)->root;
4724         struct inode *inode = NULL;
4725         int err;
4726         int drop_inode = 0;
4727         u64 objectid;
4728         unsigned long nr = 0;
4729         u64 index = 0;
4730
4731         if (!new_valid_dev(rdev))
4732                 return -EINVAL;
4733
4734         /*
4735          * 2 for inode item and ref
4736          * 2 for dir items
4737          * 1 for xattr if selinux is on
4738          */
4739         trans = btrfs_start_transaction(root, 5);
4740         if (IS_ERR(trans))
4741                 return PTR_ERR(trans);
4742
4743         btrfs_set_trans_block_group(trans, dir);
4744
4745         err = btrfs_find_free_ino(root, &objectid);
4746         if (err)
4747                 goto out_unlock;
4748
4749         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4750                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4751                                 BTRFS_I(dir)->block_group, mode, &index);
4752         err = PTR_ERR(inode);
4753         if (IS_ERR(inode))
4754                 goto out_unlock;
4755
4756         err = btrfs_init_inode_security(trans, inode, dir);
4757         if (err) {
4758                 drop_inode = 1;
4759                 goto out_unlock;
4760         }
4761
4762         btrfs_set_trans_block_group(trans, inode);
4763         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4764         if (err)
4765                 drop_inode = 1;
4766         else {
4767                 inode->i_op = &btrfs_special_inode_operations;
4768                 init_special_inode(inode, inode->i_mode, rdev);
4769                 btrfs_update_inode(trans, root, inode);
4770         }
4771         btrfs_update_inode_block_group(trans, inode);
4772         btrfs_update_inode_block_group(trans, dir);
4773 out_unlock:
4774         nr = trans->blocks_used;
4775         btrfs_end_transaction_throttle(trans, root);
4776         btrfs_btree_balance_dirty(root, nr);
4777         if (drop_inode) {
4778                 inode_dec_link_count(inode);
4779                 iput(inode);
4780         }
4781         return err;
4782 }
4783
4784 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4785                         int mode, struct nameidata *nd)
4786 {
4787         struct btrfs_trans_handle *trans;
4788         struct btrfs_root *root = BTRFS_I(dir)->root;
4789         struct inode *inode = NULL;
4790         int drop_inode = 0;
4791         int err;
4792         unsigned long nr = 0;
4793         u64 objectid;
4794         u64 index = 0;
4795
4796         /*
4797          * 2 for inode item and ref
4798          * 2 for dir items
4799          * 1 for xattr if selinux is on
4800          */
4801         trans = btrfs_start_transaction(root, 5);
4802         if (IS_ERR(trans))
4803                 return PTR_ERR(trans);
4804
4805         btrfs_set_trans_block_group(trans, dir);
4806
4807         err = btrfs_find_free_ino(root, &objectid);
4808         if (err)
4809                 goto out_unlock;
4810
4811         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4812                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4813                                 BTRFS_I(dir)->block_group, mode, &index);
4814         err = PTR_ERR(inode);
4815         if (IS_ERR(inode))
4816                 goto out_unlock;
4817
4818         err = btrfs_init_inode_security(trans, inode, dir);
4819         if (err) {
4820                 drop_inode = 1;
4821                 goto out_unlock;
4822         }
4823
4824         btrfs_set_trans_block_group(trans, inode);
4825         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4826         if (err)
4827                 drop_inode = 1;
4828         else {
4829                 inode->i_mapping->a_ops = &btrfs_aops;
4830                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4831                 inode->i_fop = &btrfs_file_operations;
4832                 inode->i_op = &btrfs_file_inode_operations;
4833                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4834         }
4835         btrfs_update_inode_block_group(trans, inode);
4836         btrfs_update_inode_block_group(trans, dir);
4837 out_unlock:
4838         nr = trans->blocks_used;
4839         btrfs_end_transaction_throttle(trans, root);
4840         if (drop_inode) {
4841                 inode_dec_link_count(inode);
4842                 iput(inode);
4843         }
4844         btrfs_btree_balance_dirty(root, nr);
4845         return err;
4846 }
4847
4848 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4849                       struct dentry *dentry)
4850 {
4851         struct btrfs_trans_handle *trans;
4852         struct btrfs_root *root = BTRFS_I(dir)->root;
4853         struct inode *inode = old_dentry->d_inode;
4854         u64 index;
4855         unsigned long nr = 0;
4856         int err;
4857         int drop_inode = 0;
4858
4859         if (inode->i_nlink == 0)
4860                 return -ENOENT;
4861
4862         /* do not allow sys_link's with other subvols of the same device */
4863         if (root->objectid != BTRFS_I(inode)->root->objectid)
4864                 return -EXDEV;
4865
4866         if (inode->i_nlink == ~0U)
4867                 return -EMLINK;
4868
4869         err = btrfs_set_inode_index(dir, &index);
4870         if (err)
4871                 goto fail;
4872
4873         /*
4874          * 2 items for inode and inode ref
4875          * 2 items for dir items
4876          * 1 item for parent inode
4877          */
4878         trans = btrfs_start_transaction(root, 5);
4879         if (IS_ERR(trans)) {
4880                 err = PTR_ERR(trans);
4881                 goto fail;
4882         }
4883
4884         btrfs_inc_nlink(inode);
4885         inode->i_ctime = CURRENT_TIME;
4886
4887         btrfs_set_trans_block_group(trans, dir);
4888         ihold(inode);
4889
4890         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4891
4892         if (err) {
4893                 drop_inode = 1;
4894         } else {
4895                 struct dentry *parent = dget_parent(dentry);
4896                 btrfs_update_inode_block_group(trans, dir);
4897                 err = btrfs_update_inode(trans, root, inode);
4898                 BUG_ON(err);
4899                 btrfs_log_new_name(trans, inode, NULL, parent);
4900                 dput(parent);
4901         }
4902
4903         nr = trans->blocks_used;
4904         btrfs_end_transaction_throttle(trans, root);
4905 fail:
4906         if (drop_inode) {
4907                 inode_dec_link_count(inode);
4908                 iput(inode);
4909         }
4910         btrfs_btree_balance_dirty(root, nr);
4911         return err;
4912 }
4913
4914 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4915 {
4916         struct inode *inode = NULL;
4917         struct btrfs_trans_handle *trans;
4918         struct btrfs_root *root = BTRFS_I(dir)->root;
4919         int err = 0;
4920         int drop_on_err = 0;
4921         u64 objectid = 0;
4922         u64 index = 0;
4923         unsigned long nr = 1;
4924
4925         /*
4926          * 2 items for inode and ref
4927          * 2 items for dir items
4928          * 1 for xattr if selinux is on
4929          */
4930         trans = btrfs_start_transaction(root, 5);
4931         if (IS_ERR(trans))
4932                 return PTR_ERR(trans);
4933         btrfs_set_trans_block_group(trans, dir);
4934
4935         err = btrfs_find_free_ino(root, &objectid);
4936         if (err)
4937                 goto out_fail;
4938
4939         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4940                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4941                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4942                                 &index);
4943         if (IS_ERR(inode)) {
4944                 err = PTR_ERR(inode);
4945                 goto out_fail;
4946         }
4947
4948         drop_on_err = 1;
4949
4950         err = btrfs_init_inode_security(trans, inode, dir);
4951         if (err)
4952                 goto out_fail;
4953
4954         inode->i_op = &btrfs_dir_inode_operations;
4955         inode->i_fop = &btrfs_dir_file_operations;
4956         btrfs_set_trans_block_group(trans, inode);
4957
4958         btrfs_i_size_write(inode, 0);
4959         err = btrfs_update_inode(trans, root, inode);
4960         if (err)
4961                 goto out_fail;
4962
4963         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4964                              dentry->d_name.len, 0, index);
4965         if (err)
4966                 goto out_fail;
4967
4968         d_instantiate(dentry, inode);
4969         drop_on_err = 0;
4970         btrfs_update_inode_block_group(trans, inode);
4971         btrfs_update_inode_block_group(trans, dir);
4972
4973 out_fail:
4974         nr = trans->blocks_used;
4975         btrfs_end_transaction_throttle(trans, root);
4976         if (drop_on_err)
4977                 iput(inode);
4978         btrfs_btree_balance_dirty(root, nr);
4979         return err;
4980 }
4981
4982 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4983  * and an extent that you want to insert, deal with overlap and insert
4984  * the new extent into the tree.
4985  */
4986 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4987                                 struct extent_map *existing,
4988                                 struct extent_map *em,
4989                                 u64 map_start, u64 map_len)
4990 {
4991         u64 start_diff;
4992
4993         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4994         start_diff = map_start - em->start;
4995         em->start = map_start;
4996         em->len = map_len;
4997         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4998             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4999                 em->block_start += start_diff;
5000                 em->block_len -= start_diff;
5001         }
5002         return add_extent_mapping(em_tree, em);
5003 }
5004
5005 static noinline int uncompress_inline(struct btrfs_path *path,
5006                                       struct inode *inode, struct page *page,
5007                                       size_t pg_offset, u64 extent_offset,
5008                                       struct btrfs_file_extent_item *item)
5009 {
5010         int ret;
5011         struct extent_buffer *leaf = path->nodes[0];
5012         char *tmp;
5013         size_t max_size;
5014         unsigned long inline_size;
5015         unsigned long ptr;
5016         int compress_type;
5017
5018         WARN_ON(pg_offset != 0);
5019         compress_type = btrfs_file_extent_compression(leaf, item);
5020         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5021         inline_size = btrfs_file_extent_inline_item_len(leaf,
5022                                         btrfs_item_nr(leaf, path->slots[0]));
5023         tmp = kmalloc(inline_size, GFP_NOFS);
5024         ptr = btrfs_file_extent_inline_start(item);
5025
5026         read_extent_buffer(leaf, tmp, ptr, inline_size);
5027
5028         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5029         ret = btrfs_decompress(compress_type, tmp, page,
5030                                extent_offset, inline_size, max_size);
5031         if (ret) {
5032                 char *kaddr = kmap_atomic(page, KM_USER0);
5033                 unsigned long copy_size = min_t(u64,
5034                                   PAGE_CACHE_SIZE - pg_offset,
5035                                   max_size - extent_offset);
5036                 memset(kaddr + pg_offset, 0, copy_size);
5037                 kunmap_atomic(kaddr, KM_USER0);
5038         }
5039         kfree(tmp);
5040         return 0;
5041 }
5042
5043 /*
5044  * a bit scary, this does extent mapping from logical file offset to the disk.
5045  * the ugly parts come from merging extents from the disk with the in-ram
5046  * representation.  This gets more complex because of the data=ordered code,
5047  * where the in-ram extents might be locked pending data=ordered completion.
5048  *
5049  * This also copies inline extents directly into the page.
5050  */
5051
5052 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5053                                     size_t pg_offset, u64 start, u64 len,
5054                                     int create)
5055 {
5056         int ret;
5057         int err = 0;
5058         u64 bytenr;
5059         u64 extent_start = 0;
5060         u64 extent_end = 0;
5061         u64 objectid = btrfs_ino(inode);
5062         u32 found_type;
5063         struct btrfs_path *path = NULL;
5064         struct btrfs_root *root = BTRFS_I(inode)->root;
5065         struct btrfs_file_extent_item *item;
5066         struct extent_buffer *leaf;
5067         struct btrfs_key found_key;
5068         struct extent_map *em = NULL;
5069         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5070         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5071         struct btrfs_trans_handle *trans = NULL;
5072         int compress_type;
5073
5074 again:
5075         read_lock(&em_tree->lock);
5076         em = lookup_extent_mapping(em_tree, start, len);
5077         if (em)
5078                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5079         read_unlock(&em_tree->lock);
5080
5081         if (em) {
5082                 if (em->start > start || em->start + em->len <= start)
5083                         free_extent_map(em);
5084                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5085                         free_extent_map(em);
5086                 else
5087                         goto out;
5088         }
5089         em = alloc_extent_map(GFP_NOFS);
5090         if (!em) {
5091                 err = -ENOMEM;
5092                 goto out;
5093         }
5094         em->bdev = root->fs_info->fs_devices->latest_bdev;
5095         em->start = EXTENT_MAP_HOLE;
5096         em->orig_start = EXTENT_MAP_HOLE;
5097         em->len = (u64)-1;
5098         em->block_len = (u64)-1;
5099
5100         if (!path) {
5101                 path = btrfs_alloc_path();
5102                 BUG_ON(!path);
5103         }
5104
5105         ret = btrfs_lookup_file_extent(trans, root, path,
5106                                        objectid, start, trans != NULL);
5107         if (ret < 0) {
5108                 err = ret;
5109                 goto out;
5110         }
5111
5112         if (ret != 0) {
5113                 if (path->slots[0] == 0)
5114                         goto not_found;
5115                 path->slots[0]--;
5116         }
5117
5118         leaf = path->nodes[0];
5119         item = btrfs_item_ptr(leaf, path->slots[0],
5120                               struct btrfs_file_extent_item);
5121         /* are we inside the extent that was found? */
5122         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5123         found_type = btrfs_key_type(&found_key);
5124         if (found_key.objectid != objectid ||
5125             found_type != BTRFS_EXTENT_DATA_KEY) {
5126                 goto not_found;
5127         }
5128
5129         found_type = btrfs_file_extent_type(leaf, item);
5130         extent_start = found_key.offset;
5131         compress_type = btrfs_file_extent_compression(leaf, item);
5132         if (found_type == BTRFS_FILE_EXTENT_REG ||
5133             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5134                 extent_end = extent_start +
5135                        btrfs_file_extent_num_bytes(leaf, item);
5136         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5137                 size_t size;
5138                 size = btrfs_file_extent_inline_len(leaf, item);
5139                 extent_end = (extent_start + size + root->sectorsize - 1) &
5140                         ~((u64)root->sectorsize - 1);
5141         }
5142
5143         if (start >= extent_end) {
5144                 path->slots[0]++;
5145                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5146                         ret = btrfs_next_leaf(root, path);
5147                         if (ret < 0) {
5148                                 err = ret;
5149                                 goto out;
5150                         }
5151                         if (ret > 0)
5152                                 goto not_found;
5153                         leaf = path->nodes[0];
5154                 }
5155                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5156                 if (found_key.objectid != objectid ||
5157                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5158                         goto not_found;
5159                 if (start + len <= found_key.offset)
5160                         goto not_found;
5161                 em->start = start;
5162                 em->len = found_key.offset - start;
5163                 goto not_found_em;
5164         }
5165
5166         if (found_type == BTRFS_FILE_EXTENT_REG ||
5167             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5168                 em->start = extent_start;
5169                 em->len = extent_end - extent_start;
5170                 em->orig_start = extent_start -
5171                                  btrfs_file_extent_offset(leaf, item);
5172                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5173                 if (bytenr == 0) {
5174                         em->block_start = EXTENT_MAP_HOLE;
5175                         goto insert;
5176                 }
5177                 if (compress_type != BTRFS_COMPRESS_NONE) {
5178                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5179                         em->compress_type = compress_type;
5180                         em->block_start = bytenr;
5181                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5182                                                                          item);
5183                 } else {
5184                         bytenr += btrfs_file_extent_offset(leaf, item);
5185                         em->block_start = bytenr;
5186                         em->block_len = em->len;
5187                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5188                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5189                 }
5190                 goto insert;
5191         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5192                 unsigned long ptr;
5193                 char *map;
5194                 size_t size;
5195                 size_t extent_offset;
5196                 size_t copy_size;
5197
5198                 em->block_start = EXTENT_MAP_INLINE;
5199                 if (!page || create) {
5200                         em->start = extent_start;
5201                         em->len = extent_end - extent_start;
5202                         goto out;
5203                 }
5204
5205                 size = btrfs_file_extent_inline_len(leaf, item);
5206                 extent_offset = page_offset(page) + pg_offset - extent_start;
5207                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5208                                 size - extent_offset);
5209                 em->start = extent_start + extent_offset;
5210                 em->len = (copy_size + root->sectorsize - 1) &
5211                         ~((u64)root->sectorsize - 1);
5212                 em->orig_start = EXTENT_MAP_INLINE;
5213                 if (compress_type) {
5214                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5215                         em->compress_type = compress_type;
5216                 }
5217                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5218                 if (create == 0 && !PageUptodate(page)) {
5219                         if (btrfs_file_extent_compression(leaf, item) !=
5220                             BTRFS_COMPRESS_NONE) {
5221                                 ret = uncompress_inline(path, inode, page,
5222                                                         pg_offset,
5223                                                         extent_offset, item);
5224                                 BUG_ON(ret);
5225                         } else {
5226                                 map = kmap(page);
5227                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5228                                                    copy_size);
5229                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5230                                         memset(map + pg_offset + copy_size, 0,
5231                                                PAGE_CACHE_SIZE - pg_offset -
5232                                                copy_size);
5233                                 }
5234                                 kunmap(page);
5235                         }
5236                         flush_dcache_page(page);
5237                 } else if (create && PageUptodate(page)) {
5238                         WARN_ON(1);
5239                         if (!trans) {
5240                                 kunmap(page);
5241                                 free_extent_map(em);
5242                                 em = NULL;
5243                                 btrfs_release_path(root, path);
5244                                 trans = btrfs_join_transaction(root, 1);
5245                                 if (IS_ERR(trans))
5246                                         return ERR_CAST(trans);
5247                                 goto again;
5248                         }
5249                         map = kmap(page);
5250                         write_extent_buffer(leaf, map + pg_offset, ptr,
5251                                             copy_size);
5252                         kunmap(page);
5253                         btrfs_mark_buffer_dirty(leaf);
5254                 }
5255                 set_extent_uptodate(io_tree, em->start,
5256                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5257                 goto insert;
5258         } else {
5259                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5260                 WARN_ON(1);
5261         }
5262 not_found:
5263         em->start = start;
5264         em->len = len;
5265 not_found_em:
5266         em->block_start = EXTENT_MAP_HOLE;
5267         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5268 insert:
5269         btrfs_release_path(root, path);
5270         if (em->start > start || extent_map_end(em) <= start) {
5271                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5272                        "[%llu %llu]\n", (unsigned long long)em->start,
5273                        (unsigned long long)em->len,
5274                        (unsigned long long)start,
5275                        (unsigned long long)len);
5276                 err = -EIO;
5277                 goto out;
5278         }
5279
5280         err = 0;
5281         write_lock(&em_tree->lock);
5282         ret = add_extent_mapping(em_tree, em);
5283         /* it is possible that someone inserted the extent into the tree
5284          * while we had the lock dropped.  It is also possible that
5285          * an overlapping map exists in the tree
5286          */
5287         if (ret == -EEXIST) {
5288                 struct extent_map *existing;
5289
5290                 ret = 0;
5291
5292                 existing = lookup_extent_mapping(em_tree, start, len);
5293                 if (existing && (existing->start > start ||
5294                     existing->start + existing->len <= start)) {
5295                         free_extent_map(existing);
5296                         existing = NULL;
5297                 }
5298                 if (!existing) {
5299                         existing = lookup_extent_mapping(em_tree, em->start,
5300                                                          em->len);
5301                         if (existing) {
5302                                 err = merge_extent_mapping(em_tree, existing,
5303                                                            em, start,
5304                                                            root->sectorsize);
5305                                 free_extent_map(existing);
5306                                 if (err) {
5307                                         free_extent_map(em);
5308                                         em = NULL;
5309                                 }
5310                         } else {
5311                                 err = -EIO;
5312                                 free_extent_map(em);
5313                                 em = NULL;
5314                         }
5315                 } else {
5316                         free_extent_map(em);
5317                         em = existing;
5318                         err = 0;
5319                 }
5320         }
5321         write_unlock(&em_tree->lock);
5322 out:
5323
5324         trace_btrfs_get_extent(root, em);
5325
5326         if (path)
5327                 btrfs_free_path(path);
5328         if (trans) {
5329                 ret = btrfs_end_transaction(trans, root);
5330                 if (!err)
5331                         err = ret;
5332         }
5333         if (err) {
5334                 free_extent_map(em);
5335                 return ERR_PTR(err);
5336         }
5337         return em;
5338 }
5339
5340 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5341                                            size_t pg_offset, u64 start, u64 len,
5342                                            int create)
5343 {
5344         struct extent_map *em;
5345         struct extent_map *hole_em = NULL;
5346         u64 range_start = start;
5347         u64 end;
5348         u64 found;
5349         u64 found_end;
5350         int err = 0;
5351
5352         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5353         if (IS_ERR(em))
5354                 return em;
5355         if (em) {
5356                 /*
5357                  * if our em maps to a hole, there might
5358                  * actually be delalloc bytes behind it
5359                  */
5360                 if (em->block_start != EXTENT_MAP_HOLE)
5361                         return em;
5362                 else
5363                         hole_em = em;
5364         }
5365
5366         /* check to see if we've wrapped (len == -1 or similar) */
5367         end = start + len;
5368         if (end < start)
5369                 end = (u64)-1;
5370         else
5371                 end -= 1;
5372
5373         em = NULL;
5374
5375         /* ok, we didn't find anything, lets look for delalloc */
5376         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5377                                  end, len, EXTENT_DELALLOC, 1);
5378         found_end = range_start + found;
5379         if (found_end < range_start)
5380                 found_end = (u64)-1;
5381
5382         /*
5383          * we didn't find anything useful, return
5384          * the original results from get_extent()
5385          */
5386         if (range_start > end || found_end <= start) {
5387                 em = hole_em;
5388                 hole_em = NULL;
5389                 goto out;
5390         }
5391
5392         /* adjust the range_start to make sure it doesn't
5393          * go backwards from the start they passed in
5394          */
5395         range_start = max(start,range_start);
5396         found = found_end - range_start;
5397
5398         if (found > 0) {
5399                 u64 hole_start = start;
5400                 u64 hole_len = len;
5401
5402                 em = alloc_extent_map(GFP_NOFS);
5403                 if (!em) {
5404                         err = -ENOMEM;
5405                         goto out;
5406                 }
5407                 /*
5408                  * when btrfs_get_extent can't find anything it
5409                  * returns one huge hole
5410                  *
5411                  * make sure what it found really fits our range, and
5412                  * adjust to make sure it is based on the start from
5413                  * the caller
5414                  */
5415                 if (hole_em) {
5416                         u64 calc_end = extent_map_end(hole_em);
5417
5418                         if (calc_end <= start || (hole_em->start > end)) {
5419                                 free_extent_map(hole_em);
5420                                 hole_em = NULL;
5421                         } else {
5422                                 hole_start = max(hole_em->start, start);
5423                                 hole_len = calc_end - hole_start;
5424                         }
5425                 }
5426                 em->bdev = NULL;
5427                 if (hole_em && range_start > hole_start) {
5428                         /* our hole starts before our delalloc, so we
5429                          * have to return just the parts of the hole
5430                          * that go until  the delalloc starts
5431                          */
5432                         em->len = min(hole_len,
5433                                       range_start - hole_start);
5434                         em->start = hole_start;
5435                         em->orig_start = hole_start;
5436                         /*
5437                          * don't adjust block start at all,
5438                          * it is fixed at EXTENT_MAP_HOLE
5439                          */
5440                         em->block_start = hole_em->block_start;
5441                         em->block_len = hole_len;
5442                 } else {
5443                         em->start = range_start;
5444                         em->len = found;
5445                         em->orig_start = range_start;
5446                         em->block_start = EXTENT_MAP_DELALLOC;
5447                         em->block_len = found;
5448                 }
5449         } else if (hole_em) {
5450                 return hole_em;
5451         }
5452 out:
5453
5454         free_extent_map(hole_em);
5455         if (err) {
5456                 free_extent_map(em);
5457                 return ERR_PTR(err);
5458         }
5459         return em;
5460 }
5461
5462 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5463                                                   struct extent_map *em,
5464                                                   u64 start, u64 len)
5465 {
5466         struct btrfs_root *root = BTRFS_I(inode)->root;
5467         struct btrfs_trans_handle *trans;
5468         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5469         struct btrfs_key ins;
5470         u64 alloc_hint;
5471         int ret;
5472         bool insert = false;
5473
5474         /*
5475          * Ok if the extent map we looked up is a hole and is for the exact
5476          * range we want, there is no reason to allocate a new one, however if
5477          * it is not right then we need to free this one and drop the cache for
5478          * our range.
5479          */
5480         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5481             em->len != len) {
5482                 free_extent_map(em);
5483                 em = NULL;
5484                 insert = true;
5485                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5486         }
5487
5488         trans = btrfs_join_transaction(root, 0);
5489         if (IS_ERR(trans))
5490                 return ERR_CAST(trans);
5491
5492         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5493
5494         alloc_hint = get_extent_allocation_hint(inode, start, len);
5495         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5496                                    alloc_hint, (u64)-1, &ins, 1);
5497         if (ret) {
5498                 em = ERR_PTR(ret);
5499                 goto out;
5500         }
5501
5502         if (!em) {
5503                 em = alloc_extent_map(GFP_NOFS);
5504                 if (!em) {
5505                         em = ERR_PTR(-ENOMEM);
5506                         goto out;
5507                 }
5508         }
5509
5510         em->start = start;
5511         em->orig_start = em->start;
5512         em->len = ins.offset;
5513
5514         em->block_start = ins.objectid;
5515         em->block_len = ins.offset;
5516         em->bdev = root->fs_info->fs_devices->latest_bdev;
5517
5518         /*
5519          * We need to do this because if we're using the original em we searched
5520          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5521          */
5522         em->flags = 0;
5523         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5524
5525         while (insert) {
5526                 write_lock(&em_tree->lock);
5527                 ret = add_extent_mapping(em_tree, em);
5528                 write_unlock(&em_tree->lock);
5529                 if (ret != -EEXIST)
5530                         break;
5531                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5532         }
5533
5534         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5535                                            ins.offset, ins.offset, 0);
5536         if (ret) {
5537                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5538                 em = ERR_PTR(ret);
5539         }
5540 out:
5541         btrfs_end_transaction(trans, root);
5542         return em;
5543 }
5544
5545 /*
5546  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5547  * block must be cow'd
5548  */
5549 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5550                                       struct inode *inode, u64 offset, u64 len)
5551 {
5552         struct btrfs_path *path;
5553         int ret;
5554         struct extent_buffer *leaf;
5555         struct btrfs_root *root = BTRFS_I(inode)->root;
5556         struct btrfs_file_extent_item *fi;
5557         struct btrfs_key key;
5558         u64 disk_bytenr;
5559         u64 backref_offset;
5560         u64 extent_end;
5561         u64 num_bytes;
5562         int slot;
5563         int found_type;
5564
5565         path = btrfs_alloc_path();
5566         if (!path)
5567                 return -ENOMEM;
5568
5569         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5570                                        offset, 0);
5571         if (ret < 0)
5572                 goto out;
5573
5574         slot = path->slots[0];
5575         if (ret == 1) {
5576                 if (slot == 0) {
5577                         /* can't find the item, must cow */
5578                         ret = 0;
5579                         goto out;
5580                 }
5581                 slot--;
5582         }
5583         ret = 0;
5584         leaf = path->nodes[0];
5585         btrfs_item_key_to_cpu(leaf, &key, slot);
5586         if (key.objectid != btrfs_ino(inode) ||
5587             key.type != BTRFS_EXTENT_DATA_KEY) {
5588                 /* not our file or wrong item type, must cow */
5589                 goto out;
5590         }
5591
5592         if (key.offset > offset) {
5593                 /* Wrong offset, must cow */
5594                 goto out;
5595         }
5596
5597         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5598         found_type = btrfs_file_extent_type(leaf, fi);
5599         if (found_type != BTRFS_FILE_EXTENT_REG &&
5600             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5601                 /* not a regular extent, must cow */
5602                 goto out;
5603         }
5604         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5605         backref_offset = btrfs_file_extent_offset(leaf, fi);
5606
5607         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5608         if (extent_end < offset + len) {
5609                 /* extent doesn't include our full range, must cow */
5610                 goto out;
5611         }
5612
5613         if (btrfs_extent_readonly(root, disk_bytenr))
5614                 goto out;
5615
5616         /*
5617          * look for other files referencing this extent, if we
5618          * find any we must cow
5619          */
5620         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5621                                   key.offset - backref_offset, disk_bytenr))
5622                 goto out;
5623
5624         /*
5625          * adjust disk_bytenr and num_bytes to cover just the bytes
5626          * in this extent we are about to write.  If there
5627          * are any csums in that range we have to cow in order
5628          * to keep the csums correct
5629          */
5630         disk_bytenr += backref_offset;
5631         disk_bytenr += offset - key.offset;
5632         num_bytes = min(offset + len, extent_end) - offset;
5633         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5634                                 goto out;
5635         /*
5636          * all of the above have passed, it is safe to overwrite this extent
5637          * without cow
5638          */
5639         ret = 1;
5640 out:
5641         btrfs_free_path(path);
5642         return ret;
5643 }
5644
5645 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5646                                    struct buffer_head *bh_result, int create)
5647 {
5648         struct extent_map *em;
5649         struct btrfs_root *root = BTRFS_I(inode)->root;
5650         u64 start = iblock << inode->i_blkbits;
5651         u64 len = bh_result->b_size;
5652         struct btrfs_trans_handle *trans;
5653
5654         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5655         if (IS_ERR(em))
5656                 return PTR_ERR(em);
5657
5658         /*
5659          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5660          * io.  INLINE is special, and we could probably kludge it in here, but
5661          * it's still buffered so for safety lets just fall back to the generic
5662          * buffered path.
5663          *
5664          * For COMPRESSED we _have_ to read the entire extent in so we can
5665          * decompress it, so there will be buffering required no matter what we
5666          * do, so go ahead and fallback to buffered.
5667          *
5668          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5669          * to buffered IO.  Don't blame me, this is the price we pay for using
5670          * the generic code.
5671          */
5672         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5673             em->block_start == EXTENT_MAP_INLINE) {
5674                 free_extent_map(em);
5675                 return -ENOTBLK;
5676         }
5677
5678         /* Just a good old fashioned hole, return */
5679         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5680                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5681                 free_extent_map(em);
5682                 /* DIO will do one hole at a time, so just unlock a sector */
5683                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5684                               start + root->sectorsize - 1, GFP_NOFS);
5685                 return 0;
5686         }
5687
5688         /*
5689          * We don't allocate a new extent in the following cases
5690          *
5691          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5692          * existing extent.
5693          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5694          * just use the extent.
5695          *
5696          */
5697         if (!create) {
5698                 len = em->len - (start - em->start);
5699                 goto map;
5700         }
5701
5702         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5703             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5704              em->block_start != EXTENT_MAP_HOLE)) {
5705                 int type;
5706                 int ret;
5707                 u64 block_start;
5708
5709                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5710                         type = BTRFS_ORDERED_PREALLOC;
5711                 else
5712                         type = BTRFS_ORDERED_NOCOW;
5713                 len = min(len, em->len - (start - em->start));
5714                 block_start = em->block_start + (start - em->start);
5715
5716                 /*
5717                  * we're not going to log anything, but we do need
5718                  * to make sure the current transaction stays open
5719                  * while we look for nocow cross refs
5720                  */
5721                 trans = btrfs_join_transaction(root, 0);
5722                 if (IS_ERR(trans))
5723                         goto must_cow;
5724
5725                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5726                         ret = btrfs_add_ordered_extent_dio(inode, start,
5727                                            block_start, len, len, type);
5728                         btrfs_end_transaction(trans, root);
5729                         if (ret) {
5730                                 free_extent_map(em);
5731                                 return ret;
5732                         }
5733                         goto unlock;
5734                 }
5735                 btrfs_end_transaction(trans, root);
5736         }
5737 must_cow:
5738         /*
5739          * this will cow the extent, reset the len in case we changed
5740          * it above
5741          */
5742         len = bh_result->b_size;
5743         em = btrfs_new_extent_direct(inode, em, start, len);
5744         if (IS_ERR(em))
5745                 return PTR_ERR(em);
5746         len = min(len, em->len - (start - em->start));
5747 unlock:
5748         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5749                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5750                           0, NULL, GFP_NOFS);
5751 map:
5752         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5753                 inode->i_blkbits;
5754         bh_result->b_size = len;
5755         bh_result->b_bdev = em->bdev;
5756         set_buffer_mapped(bh_result);
5757         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5758                 set_buffer_new(bh_result);
5759
5760         free_extent_map(em);
5761
5762         return 0;
5763 }
5764
5765 struct btrfs_dio_private {
5766         struct inode *inode;
5767         u64 logical_offset;
5768         u64 disk_bytenr;
5769         u64 bytes;
5770         u32 *csums;
5771         void *private;
5772
5773         /* number of bios pending for this dio */
5774         atomic_t pending_bios;
5775
5776         /* IO errors */
5777         int errors;
5778
5779         struct bio *orig_bio;
5780 };
5781
5782 static void btrfs_endio_direct_read(struct bio *bio, int err)
5783 {
5784         struct btrfs_dio_private *dip = bio->bi_private;
5785         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5786         struct bio_vec *bvec = bio->bi_io_vec;
5787         struct inode *inode = dip->inode;
5788         struct btrfs_root *root = BTRFS_I(inode)->root;
5789         u64 start;
5790         u32 *private = dip->csums;
5791
5792         start = dip->logical_offset;
5793         do {
5794                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5795                         struct page *page = bvec->bv_page;
5796                         char *kaddr;
5797                         u32 csum = ~(u32)0;
5798                         unsigned long flags;
5799
5800                         local_irq_save(flags);
5801                         kaddr = kmap_atomic(page, KM_IRQ0);
5802                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5803                                                csum, bvec->bv_len);
5804                         btrfs_csum_final(csum, (char *)&csum);
5805                         kunmap_atomic(kaddr, KM_IRQ0);
5806                         local_irq_restore(flags);
5807
5808                         flush_dcache_page(bvec->bv_page);
5809                         if (csum != *private) {
5810                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5811                                       " %llu csum %u private %u\n",
5812                                       (unsigned long long)btrfs_ino(inode),
5813                                       (unsigned long long)start,
5814                                       csum, *private);
5815                                 err = -EIO;
5816                         }
5817                 }
5818
5819                 start += bvec->bv_len;
5820                 private++;
5821                 bvec++;
5822         } while (bvec <= bvec_end);
5823
5824         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5825                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5826         bio->bi_private = dip->private;
5827
5828         kfree(dip->csums);
5829         kfree(dip);
5830
5831         /* If we had a csum failure make sure to clear the uptodate flag */
5832         if (err)
5833                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5834         dio_end_io(bio, err);
5835 }
5836
5837 static void btrfs_endio_direct_write(struct bio *bio, int err)
5838 {
5839         struct btrfs_dio_private *dip = bio->bi_private;
5840         struct inode *inode = dip->inode;
5841         struct btrfs_root *root = BTRFS_I(inode)->root;
5842         struct btrfs_trans_handle *trans;
5843         struct btrfs_ordered_extent *ordered = NULL;
5844         struct extent_state *cached_state = NULL;
5845         u64 ordered_offset = dip->logical_offset;
5846         u64 ordered_bytes = dip->bytes;
5847         int ret;
5848
5849         if (err)
5850                 goto out_done;
5851 again:
5852         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5853                                                    &ordered_offset,
5854                                                    ordered_bytes);
5855         if (!ret)
5856                 goto out_test;
5857
5858         BUG_ON(!ordered);
5859
5860         trans = btrfs_join_transaction(root, 1);
5861         if (IS_ERR(trans)) {
5862                 err = -ENOMEM;
5863                 goto out;
5864         }
5865         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5866
5867         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5868                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5869                 if (!ret)
5870                         ret = btrfs_update_inode(trans, root, inode);
5871                 err = ret;
5872                 goto out;
5873         }
5874
5875         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5876                          ordered->file_offset + ordered->len - 1, 0,
5877                          &cached_state, GFP_NOFS);
5878
5879         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5880                 ret = btrfs_mark_extent_written(trans, inode,
5881                                                 ordered->file_offset,
5882                                                 ordered->file_offset +
5883                                                 ordered->len);
5884                 if (ret) {
5885                         err = ret;
5886                         goto out_unlock;
5887                 }
5888         } else {
5889                 ret = insert_reserved_file_extent(trans, inode,
5890                                                   ordered->file_offset,
5891                                                   ordered->start,
5892                                                   ordered->disk_len,
5893                                                   ordered->len,
5894                                                   ordered->len,
5895                                                   0, 0, 0,
5896                                                   BTRFS_FILE_EXTENT_REG);
5897                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5898                                    ordered->file_offset, ordered->len);
5899                 if (ret) {
5900                         err = ret;
5901                         WARN_ON(1);
5902                         goto out_unlock;
5903                 }
5904         }
5905
5906         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5907         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5908         if (!ret)
5909                 btrfs_update_inode(trans, root, inode);
5910         ret = 0;
5911 out_unlock:
5912         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5913                              ordered->file_offset + ordered->len - 1,
5914                              &cached_state, GFP_NOFS);
5915 out:
5916         btrfs_delalloc_release_metadata(inode, ordered->len);
5917         btrfs_end_transaction(trans, root);
5918         ordered_offset = ordered->file_offset + ordered->len;
5919         btrfs_put_ordered_extent(ordered);
5920         btrfs_put_ordered_extent(ordered);
5921
5922 out_test:
5923         /*
5924          * our bio might span multiple ordered extents.  If we haven't
5925          * completed the accounting for the whole dio, go back and try again
5926          */
5927         if (ordered_offset < dip->logical_offset + dip->bytes) {
5928                 ordered_bytes = dip->logical_offset + dip->bytes -
5929                         ordered_offset;
5930                 goto again;
5931         }
5932 out_done:
5933         bio->bi_private = dip->private;
5934
5935         kfree(dip->csums);
5936         kfree(dip);
5937
5938         /* If we had an error make sure to clear the uptodate flag */
5939         if (err)
5940                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5941         dio_end_io(bio, err);
5942 }
5943
5944 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5945                                     struct bio *bio, int mirror_num,
5946                                     unsigned long bio_flags, u64 offset)
5947 {
5948         int ret;
5949         struct btrfs_root *root = BTRFS_I(inode)->root;
5950         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5951         BUG_ON(ret);
5952         return 0;
5953 }
5954
5955 static void btrfs_end_dio_bio(struct bio *bio, int err)
5956 {
5957         struct btrfs_dio_private *dip = bio->bi_private;
5958
5959         if (err) {
5960                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5961                       "sector %#Lx len %u err no %d\n",
5962                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5963                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5964                 dip->errors = 1;
5965
5966                 /*
5967                  * before atomic variable goto zero, we must make sure
5968                  * dip->errors is perceived to be set.
5969                  */
5970                 smp_mb__before_atomic_dec();
5971         }
5972
5973         /* if there are more bios still pending for this dio, just exit */
5974         if (!atomic_dec_and_test(&dip->pending_bios))
5975                 goto out;
5976
5977         if (dip->errors)
5978                 bio_io_error(dip->orig_bio);
5979         else {
5980                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5981                 bio_endio(dip->orig_bio, 0);
5982         }
5983 out:
5984         bio_put(bio);
5985 }
5986
5987 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5988                                        u64 first_sector, gfp_t gfp_flags)
5989 {
5990         int nr_vecs = bio_get_nr_vecs(bdev);
5991         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5992 }
5993
5994 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5995                                          int rw, u64 file_offset, int skip_sum,
5996                                          u32 *csums, int async_submit)
5997 {
5998         int write = rw & REQ_WRITE;
5999         struct btrfs_root *root = BTRFS_I(inode)->root;
6000         int ret;
6001
6002         bio_get(bio);
6003         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6004         if (ret)
6005                 goto err;
6006
6007         if (skip_sum)
6008                 goto map;
6009
6010         if (write && async_submit) {
6011                 ret = btrfs_wq_submit_bio(root->fs_info,
6012                                    inode, rw, bio, 0, 0,
6013                                    file_offset,
6014                                    __btrfs_submit_bio_start_direct_io,
6015                                    __btrfs_submit_bio_done);
6016                 goto err;
6017         } else if (write) {
6018                 /*
6019                  * If we aren't doing async submit, calculate the csum of the
6020                  * bio now.
6021                  */
6022                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6023                 if (ret)
6024                         goto err;
6025         } else if (!skip_sum) {
6026                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6027                                           file_offset, csums);
6028                 if (ret)
6029                         goto err;
6030         }
6031
6032 map:
6033         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6034 err:
6035         bio_put(bio);
6036         return ret;
6037 }
6038
6039 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6040                                     int skip_sum)
6041 {
6042         struct inode *inode = dip->inode;
6043         struct btrfs_root *root = BTRFS_I(inode)->root;
6044         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6045         struct bio *bio;
6046         struct bio *orig_bio = dip->orig_bio;
6047         struct bio_vec *bvec = orig_bio->bi_io_vec;
6048         u64 start_sector = orig_bio->bi_sector;
6049         u64 file_offset = dip->logical_offset;
6050         u64 submit_len = 0;
6051         u64 map_length;
6052         int nr_pages = 0;
6053         u32 *csums = dip->csums;
6054         int ret = 0;
6055         int async_submit = 0;
6056         int write = rw & REQ_WRITE;
6057
6058         map_length = orig_bio->bi_size;
6059         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6060                               &map_length, NULL, 0);
6061         if (ret) {
6062                 bio_put(bio);
6063                 return -EIO;
6064         }
6065
6066         if (map_length >= orig_bio->bi_size) {
6067                 bio = orig_bio;
6068                 goto submit;
6069         }
6070
6071         async_submit = 1;
6072         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6073         if (!bio)
6074                 return -ENOMEM;
6075         bio->bi_private = dip;
6076         bio->bi_end_io = btrfs_end_dio_bio;
6077         atomic_inc(&dip->pending_bios);
6078
6079         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6080                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6081                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6082                                  bvec->bv_offset) < bvec->bv_len)) {
6083                         /*
6084                          * inc the count before we submit the bio so
6085                          * we know the end IO handler won't happen before
6086                          * we inc the count. Otherwise, the dip might get freed
6087                          * before we're done setting it up
6088                          */
6089                         atomic_inc(&dip->pending_bios);
6090                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6091                                                      file_offset, skip_sum,
6092                                                      csums, async_submit);
6093                         if (ret) {
6094                                 bio_put(bio);
6095                                 atomic_dec(&dip->pending_bios);
6096                                 goto out_err;
6097                         }
6098
6099                         /* Write's use the ordered csums */
6100                         if (!write && !skip_sum)
6101                                 csums = csums + nr_pages;
6102                         start_sector += submit_len >> 9;
6103                         file_offset += submit_len;
6104
6105                         submit_len = 0;
6106                         nr_pages = 0;
6107
6108                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6109                                                   start_sector, GFP_NOFS);
6110                         if (!bio)
6111                                 goto out_err;
6112                         bio->bi_private = dip;
6113                         bio->bi_end_io = btrfs_end_dio_bio;
6114
6115                         map_length = orig_bio->bi_size;
6116                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6117                                               &map_length, NULL, 0);
6118                         if (ret) {
6119                                 bio_put(bio);
6120                                 goto out_err;
6121                         }
6122                 } else {
6123                         submit_len += bvec->bv_len;
6124                         nr_pages ++;
6125                         bvec++;
6126                 }
6127         }
6128
6129 submit:
6130         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6131                                      csums, async_submit);
6132         if (!ret)
6133                 return 0;
6134
6135         bio_put(bio);
6136 out_err:
6137         dip->errors = 1;
6138         /*
6139          * before atomic variable goto zero, we must
6140          * make sure dip->errors is perceived to be set.
6141          */
6142         smp_mb__before_atomic_dec();
6143         if (atomic_dec_and_test(&dip->pending_bios))
6144                 bio_io_error(dip->orig_bio);
6145
6146         /* bio_end_io() will handle error, so we needn't return it */
6147         return 0;
6148 }
6149
6150 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6151                                 loff_t file_offset)
6152 {
6153         struct btrfs_root *root = BTRFS_I(inode)->root;
6154         struct btrfs_dio_private *dip;
6155         struct bio_vec *bvec = bio->bi_io_vec;
6156         int skip_sum;
6157         int write = rw & REQ_WRITE;
6158         int ret = 0;
6159
6160         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6161
6162         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6163         if (!dip) {
6164                 ret = -ENOMEM;
6165                 goto free_ordered;
6166         }
6167         dip->csums = NULL;
6168
6169         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6170         if (!write && !skip_sum) {
6171                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6172                 if (!dip->csums) {
6173                         kfree(dip);
6174                         ret = -ENOMEM;
6175                         goto free_ordered;
6176                 }
6177         }
6178
6179         dip->private = bio->bi_private;
6180         dip->inode = inode;
6181         dip->logical_offset = file_offset;
6182
6183         dip->bytes = 0;
6184         do {
6185                 dip->bytes += bvec->bv_len;
6186                 bvec++;
6187         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6188
6189         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6190         bio->bi_private = dip;
6191         dip->errors = 0;
6192         dip->orig_bio = bio;
6193         atomic_set(&dip->pending_bios, 0);
6194
6195         if (write)
6196                 bio->bi_end_io = btrfs_endio_direct_write;
6197         else
6198                 bio->bi_end_io = btrfs_endio_direct_read;
6199
6200         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6201         if (!ret)
6202                 return;
6203 free_ordered:
6204         /*
6205          * If this is a write, we need to clean up the reserved space and kill
6206          * the ordered extent.
6207          */
6208         if (write) {
6209                 struct btrfs_ordered_extent *ordered;
6210                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6211                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6212                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6213                         btrfs_free_reserved_extent(root, ordered->start,
6214                                                    ordered->disk_len);
6215                 btrfs_put_ordered_extent(ordered);
6216                 btrfs_put_ordered_extent(ordered);
6217         }
6218         bio_endio(bio, ret);
6219 }
6220
6221 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6222                         const struct iovec *iov, loff_t offset,
6223                         unsigned long nr_segs)
6224 {
6225         int seg;
6226         int i;
6227         size_t size;
6228         unsigned long addr;
6229         unsigned blocksize_mask = root->sectorsize - 1;
6230         ssize_t retval = -EINVAL;
6231         loff_t end = offset;
6232
6233         if (offset & blocksize_mask)
6234                 goto out;
6235
6236         /* Check the memory alignment.  Blocks cannot straddle pages */
6237         for (seg = 0; seg < nr_segs; seg++) {
6238                 addr = (unsigned long)iov[seg].iov_base;
6239                 size = iov[seg].iov_len;
6240                 end += size;
6241                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6242                         goto out;
6243
6244                 /* If this is a write we don't need to check anymore */
6245                 if (rw & WRITE)
6246                         continue;
6247
6248                 /*
6249                  * Check to make sure we don't have duplicate iov_base's in this
6250                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6251                  * when reading back.
6252                  */
6253                 for (i = seg + 1; i < nr_segs; i++) {
6254                         if (iov[seg].iov_base == iov[i].iov_base)
6255                                 goto out;
6256                 }
6257         }
6258         retval = 0;
6259 out:
6260         return retval;
6261 }
6262 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6263                         const struct iovec *iov, loff_t offset,
6264                         unsigned long nr_segs)
6265 {
6266         struct file *file = iocb->ki_filp;
6267         struct inode *inode = file->f_mapping->host;
6268         struct btrfs_ordered_extent *ordered;
6269         struct extent_state *cached_state = NULL;
6270         u64 lockstart, lockend;
6271         ssize_t ret;
6272         int writing = rw & WRITE;
6273         int write_bits = 0;
6274         size_t count = iov_length(iov, nr_segs);
6275
6276         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6277                             offset, nr_segs)) {
6278                 return 0;
6279         }
6280
6281         lockstart = offset;
6282         lockend = offset + count - 1;
6283
6284         if (writing) {
6285                 ret = btrfs_delalloc_reserve_space(inode, count);
6286                 if (ret)
6287                         goto out;
6288         }
6289
6290         while (1) {
6291                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6292                                  0, &cached_state, GFP_NOFS);
6293                 /*
6294                  * We're concerned with the entire range that we're going to be
6295                  * doing DIO to, so we need to make sure theres no ordered
6296                  * extents in this range.
6297                  */
6298                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6299                                                      lockend - lockstart + 1);
6300                 if (!ordered)
6301                         break;
6302                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6303                                      &cached_state, GFP_NOFS);
6304                 btrfs_start_ordered_extent(inode, ordered, 1);
6305                 btrfs_put_ordered_extent(ordered);
6306                 cond_resched();
6307         }
6308
6309         /*
6310          * we don't use btrfs_set_extent_delalloc because we don't want
6311          * the dirty or uptodate bits
6312          */
6313         if (writing) {
6314                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6315                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6316                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6317                                      GFP_NOFS);
6318                 if (ret) {
6319                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6320                                          lockend, EXTENT_LOCKED | write_bits,
6321                                          1, 0, &cached_state, GFP_NOFS);
6322                         goto out;
6323                 }
6324         }
6325
6326         free_extent_state(cached_state);
6327         cached_state = NULL;
6328
6329         ret = __blockdev_direct_IO(rw, iocb, inode,
6330                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6331                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6332                    btrfs_submit_direct, 0);
6333
6334         if (ret < 0 && ret != -EIOCBQUEUED) {
6335                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6336                               offset + iov_length(iov, nr_segs) - 1,
6337                               EXTENT_LOCKED | write_bits, 1, 0,
6338                               &cached_state, GFP_NOFS);
6339         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6340                 /*
6341                  * We're falling back to buffered, unlock the section we didn't
6342                  * do IO on.
6343                  */
6344                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6345                               offset + iov_length(iov, nr_segs) - 1,
6346                               EXTENT_LOCKED | write_bits, 1, 0,
6347                               &cached_state, GFP_NOFS);
6348         }
6349 out:
6350         free_extent_state(cached_state);
6351         return ret;
6352 }
6353
6354 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6355                 __u64 start, __u64 len)
6356 {
6357         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6358 }
6359
6360 int btrfs_readpage(struct file *file, struct page *page)
6361 {
6362         struct extent_io_tree *tree;
6363         tree = &BTRFS_I(page->mapping->host)->io_tree;
6364         return extent_read_full_page(tree, page, btrfs_get_extent);
6365 }
6366
6367 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6368 {
6369         struct extent_io_tree *tree;
6370
6371
6372         if (current->flags & PF_MEMALLOC) {
6373                 redirty_page_for_writepage(wbc, page);
6374                 unlock_page(page);
6375                 return 0;
6376         }
6377         tree = &BTRFS_I(page->mapping->host)->io_tree;
6378         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6379 }
6380
6381 int btrfs_writepages(struct address_space *mapping,
6382                      struct writeback_control *wbc)
6383 {
6384         struct extent_io_tree *tree;
6385
6386         tree = &BTRFS_I(mapping->host)->io_tree;
6387         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6388 }
6389
6390 static int
6391 btrfs_readpages(struct file *file, struct address_space *mapping,
6392                 struct list_head *pages, unsigned nr_pages)
6393 {
6394         struct extent_io_tree *tree;
6395         tree = &BTRFS_I(mapping->host)->io_tree;
6396         return extent_readpages(tree, mapping, pages, nr_pages,
6397                                 btrfs_get_extent);
6398 }
6399 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6400 {
6401         struct extent_io_tree *tree;
6402         struct extent_map_tree *map;
6403         int ret;
6404
6405         tree = &BTRFS_I(page->mapping->host)->io_tree;
6406         map = &BTRFS_I(page->mapping->host)->extent_tree;
6407         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6408         if (ret == 1) {
6409                 ClearPagePrivate(page);
6410                 set_page_private(page, 0);
6411                 page_cache_release(page);
6412         }
6413         return ret;
6414 }
6415
6416 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6417 {
6418         if (PageWriteback(page) || PageDirty(page))
6419                 return 0;
6420         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6421 }
6422
6423 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6424 {
6425         struct extent_io_tree *tree;
6426         struct btrfs_ordered_extent *ordered;
6427         struct extent_state *cached_state = NULL;
6428         u64 page_start = page_offset(page);
6429         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6430
6431
6432         /*
6433          * we have the page locked, so new writeback can't start,
6434          * and the dirty bit won't be cleared while we are here.
6435          *
6436          * Wait for IO on this page so that we can safely clear
6437          * the PagePrivate2 bit and do ordered accounting
6438          */
6439         wait_on_page_writeback(page);
6440
6441         tree = &BTRFS_I(page->mapping->host)->io_tree;
6442         if (offset) {
6443                 btrfs_releasepage(page, GFP_NOFS);
6444                 return;
6445         }
6446         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6447                          GFP_NOFS);
6448         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6449                                            page_offset(page));
6450         if (ordered) {
6451                 /*
6452                  * IO on this page will never be started, so we need
6453                  * to account for any ordered extents now
6454                  */
6455                 clear_extent_bit(tree, page_start, page_end,
6456                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6457                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6458                                  &cached_state, GFP_NOFS);
6459                 /*
6460                  * whoever cleared the private bit is responsible
6461                  * for the finish_ordered_io
6462                  */
6463                 if (TestClearPagePrivate2(page)) {
6464                         btrfs_finish_ordered_io(page->mapping->host,
6465                                                 page_start, page_end);
6466                 }
6467                 btrfs_put_ordered_extent(ordered);
6468                 cached_state = NULL;
6469                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6470                                  GFP_NOFS);
6471         }
6472         clear_extent_bit(tree, page_start, page_end,
6473                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6474                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6475         __btrfs_releasepage(page, GFP_NOFS);
6476
6477         ClearPageChecked(page);
6478         if (PagePrivate(page)) {
6479                 ClearPagePrivate(page);
6480                 set_page_private(page, 0);
6481                 page_cache_release(page);
6482         }
6483 }
6484
6485 /*
6486  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6487  * called from a page fault handler when a page is first dirtied. Hence we must
6488  * be careful to check for EOF conditions here. We set the page up correctly
6489  * for a written page which means we get ENOSPC checking when writing into
6490  * holes and correct delalloc and unwritten extent mapping on filesystems that
6491  * support these features.
6492  *
6493  * We are not allowed to take the i_mutex here so we have to play games to
6494  * protect against truncate races as the page could now be beyond EOF.  Because
6495  * vmtruncate() writes the inode size before removing pages, once we have the
6496  * page lock we can determine safely if the page is beyond EOF. If it is not
6497  * beyond EOF, then the page is guaranteed safe against truncation until we
6498  * unlock the page.
6499  */
6500 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6501 {
6502         struct page *page = vmf->page;
6503         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6504         struct btrfs_root *root = BTRFS_I(inode)->root;
6505         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6506         struct btrfs_ordered_extent *ordered;
6507         struct extent_state *cached_state = NULL;
6508         char *kaddr;
6509         unsigned long zero_start;
6510         loff_t size;
6511         int ret;
6512         u64 page_start;
6513         u64 page_end;
6514
6515         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6516         if (ret) {
6517                 if (ret == -ENOMEM)
6518                         ret = VM_FAULT_OOM;
6519                 else /* -ENOSPC, -EIO, etc */
6520                         ret = VM_FAULT_SIGBUS;
6521                 goto out;
6522         }
6523
6524         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6525 again:
6526         lock_page(page);
6527         size = i_size_read(inode);
6528         page_start = page_offset(page);
6529         page_end = page_start + PAGE_CACHE_SIZE - 1;
6530
6531         if ((page->mapping != inode->i_mapping) ||
6532             (page_start >= size)) {
6533                 /* page got truncated out from underneath us */
6534                 goto out_unlock;
6535         }
6536         wait_on_page_writeback(page);
6537
6538         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6539                          GFP_NOFS);
6540         set_page_extent_mapped(page);
6541
6542         /*
6543          * we can't set the delalloc bits if there are pending ordered
6544          * extents.  Drop our locks and wait for them to finish
6545          */
6546         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6547         if (ordered) {
6548                 unlock_extent_cached(io_tree, page_start, page_end,
6549                                      &cached_state, GFP_NOFS);
6550                 unlock_page(page);
6551                 btrfs_start_ordered_extent(inode, ordered, 1);
6552                 btrfs_put_ordered_extent(ordered);
6553                 goto again;
6554         }
6555
6556         /*
6557          * XXX - page_mkwrite gets called every time the page is dirtied, even
6558          * if it was already dirty, so for space accounting reasons we need to
6559          * clear any delalloc bits for the range we are fixing to save.  There
6560          * is probably a better way to do this, but for now keep consistent with
6561          * prepare_pages in the normal write path.
6562          */
6563         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6564                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6565                           0, 0, &cached_state, GFP_NOFS);
6566
6567         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6568                                         &cached_state);
6569         if (ret) {
6570                 unlock_extent_cached(io_tree, page_start, page_end,
6571                                      &cached_state, GFP_NOFS);
6572                 ret = VM_FAULT_SIGBUS;
6573                 goto out_unlock;
6574         }
6575         ret = 0;
6576
6577         /* page is wholly or partially inside EOF */
6578         if (page_start + PAGE_CACHE_SIZE > size)
6579                 zero_start = size & ~PAGE_CACHE_MASK;
6580         else
6581                 zero_start = PAGE_CACHE_SIZE;
6582
6583         if (zero_start != PAGE_CACHE_SIZE) {
6584                 kaddr = kmap(page);
6585                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6586                 flush_dcache_page(page);
6587                 kunmap(page);
6588         }
6589         ClearPageChecked(page);
6590         set_page_dirty(page);
6591         SetPageUptodate(page);
6592
6593         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6594         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6595
6596         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6597
6598 out_unlock:
6599         if (!ret)
6600                 return VM_FAULT_LOCKED;
6601         unlock_page(page);
6602         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6603 out:
6604         return ret;
6605 }
6606
6607 static int btrfs_truncate(struct inode *inode)
6608 {
6609         struct btrfs_root *root = BTRFS_I(inode)->root;
6610         int ret;
6611         int err = 0;
6612         struct btrfs_trans_handle *trans;
6613         unsigned long nr;
6614         u64 mask = root->sectorsize - 1;
6615
6616         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6617         if (ret)
6618                 return ret;
6619
6620         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6621         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6622
6623         trans = btrfs_start_transaction(root, 5);
6624         if (IS_ERR(trans))
6625                 return PTR_ERR(trans);
6626
6627         btrfs_set_trans_block_group(trans, inode);
6628
6629         ret = btrfs_orphan_add(trans, inode);
6630         if (ret) {
6631                 btrfs_end_transaction(trans, root);
6632                 return ret;
6633         }
6634
6635         nr = trans->blocks_used;
6636         btrfs_end_transaction(trans, root);
6637         btrfs_btree_balance_dirty(root, nr);
6638
6639         /* Now start a transaction for the truncate */
6640         trans = btrfs_start_transaction(root, 0);
6641         if (IS_ERR(trans))
6642                 return PTR_ERR(trans);
6643         btrfs_set_trans_block_group(trans, inode);
6644         trans->block_rsv = root->orphan_block_rsv;
6645
6646         /*
6647          * setattr is responsible for setting the ordered_data_close flag,
6648          * but that is only tested during the last file release.  That
6649          * could happen well after the next commit, leaving a great big
6650          * window where new writes may get lost if someone chooses to write
6651          * to this file after truncating to zero
6652          *
6653          * The inode doesn't have any dirty data here, and so if we commit
6654          * this is a noop.  If someone immediately starts writing to the inode
6655          * it is very likely we'll catch some of their writes in this
6656          * transaction, and the commit will find this file on the ordered
6657          * data list with good things to send down.
6658          *
6659          * This is a best effort solution, there is still a window where
6660          * using truncate to replace the contents of the file will
6661          * end up with a zero length file after a crash.
6662          */
6663         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6664                 btrfs_add_ordered_operation(trans, root, inode);
6665
6666         while (1) {
6667                 if (!trans) {
6668                         trans = btrfs_start_transaction(root, 0);
6669                         if (IS_ERR(trans))
6670                                 return PTR_ERR(trans);
6671                         btrfs_set_trans_block_group(trans, inode);
6672                         trans->block_rsv = root->orphan_block_rsv;
6673                 }
6674
6675                 ret = btrfs_block_rsv_check(trans, root,
6676                                             root->orphan_block_rsv, 0, 5);
6677                 if (ret == -EAGAIN) {
6678                         ret = btrfs_commit_transaction(trans, root);
6679                         if (ret)
6680                                 return ret;
6681                         trans = NULL;
6682                         continue;
6683                 } else if (ret) {
6684                         err = ret;
6685                         break;
6686                 }
6687
6688                 ret = btrfs_truncate_inode_items(trans, root, inode,
6689                                                  inode->i_size,
6690                                                  BTRFS_EXTENT_DATA_KEY);
6691                 if (ret != -EAGAIN) {
6692                         err = ret;
6693                         break;
6694                 }
6695
6696                 ret = btrfs_update_inode(trans, root, inode);
6697                 if (ret) {
6698                         err = ret;
6699                         break;
6700                 }
6701
6702                 nr = trans->blocks_used;
6703                 btrfs_end_transaction(trans, root);
6704                 trans = NULL;
6705                 btrfs_btree_balance_dirty(root, nr);
6706         }
6707
6708         if (ret == 0 && inode->i_nlink > 0) {
6709                 ret = btrfs_orphan_del(trans, inode);
6710                 if (ret)
6711                         err = ret;
6712         } else if (ret && inode->i_nlink > 0) {
6713                 /*
6714                  * Failed to do the truncate, remove us from the in memory
6715                  * orphan list.
6716                  */
6717                 ret = btrfs_orphan_del(NULL, inode);
6718         }
6719
6720         ret = btrfs_update_inode(trans, root, inode);
6721         if (ret && !err)
6722                 err = ret;
6723
6724         nr = trans->blocks_used;
6725         ret = btrfs_end_transaction_throttle(trans, root);
6726         if (ret && !err)
6727                 err = ret;
6728         btrfs_btree_balance_dirty(root, nr);
6729
6730         return err;
6731 }
6732
6733 /*
6734  * create a new subvolume directory/inode (helper for the ioctl).
6735  */
6736 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6737                              struct btrfs_root *new_root,
6738                              u64 new_dirid, u64 alloc_hint)
6739 {
6740         struct inode *inode;
6741         int err;
6742         u64 index = 0;
6743
6744         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6745                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6746         if (IS_ERR(inode))
6747                 return PTR_ERR(inode);
6748         inode->i_op = &btrfs_dir_inode_operations;
6749         inode->i_fop = &btrfs_dir_file_operations;
6750
6751         inode->i_nlink = 1;
6752         btrfs_i_size_write(inode, 0);
6753
6754         err = btrfs_update_inode(trans, new_root, inode);
6755         BUG_ON(err);
6756
6757         iput(inode);
6758         return 0;
6759 }
6760
6761 /* helper function for file defrag and space balancing.  This
6762  * forces readahead on a given range of bytes in an inode
6763  */
6764 unsigned long btrfs_force_ra(struct address_space *mapping,
6765                               struct file_ra_state *ra, struct file *file,
6766                               pgoff_t offset, pgoff_t last_index)
6767 {
6768         pgoff_t req_size = last_index - offset + 1;
6769
6770         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6771         return offset + req_size;
6772 }
6773
6774 struct inode *btrfs_alloc_inode(struct super_block *sb)
6775 {
6776         struct btrfs_inode *ei;
6777         struct inode *inode;
6778
6779         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6780         if (!ei)
6781                 return NULL;
6782
6783         ei->root = NULL;
6784         ei->space_info = NULL;
6785         ei->generation = 0;
6786         ei->sequence = 0;
6787         ei->last_trans = 0;
6788         ei->last_sub_trans = 0;
6789         ei->logged_trans = 0;
6790         ei->delalloc_bytes = 0;
6791         ei->reserved_bytes = 0;
6792         ei->disk_i_size = 0;
6793         ei->flags = 0;
6794         ei->index_cnt = (u64)-1;
6795         ei->last_unlink_trans = 0;
6796
6797         atomic_set(&ei->outstanding_extents, 0);
6798         atomic_set(&ei->reserved_extents, 0);
6799
6800         ei->ordered_data_close = 0;
6801         ei->orphan_meta_reserved = 0;
6802         ei->dummy_inode = 0;
6803         ei->force_compress = BTRFS_COMPRESS_NONE;
6804
6805         inode = &ei->vfs_inode;
6806         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6807         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6808         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6809         mutex_init(&ei->log_mutex);
6810         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6811         INIT_LIST_HEAD(&ei->i_orphan);
6812         INIT_LIST_HEAD(&ei->delalloc_inodes);
6813         INIT_LIST_HEAD(&ei->ordered_operations);
6814         RB_CLEAR_NODE(&ei->rb_node);
6815
6816         return inode;
6817 }
6818
6819 static void btrfs_i_callback(struct rcu_head *head)
6820 {
6821         struct inode *inode = container_of(head, struct inode, i_rcu);
6822         INIT_LIST_HEAD(&inode->i_dentry);
6823         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6824 }
6825
6826 void btrfs_destroy_inode(struct inode *inode)
6827 {
6828         struct btrfs_ordered_extent *ordered;
6829         struct btrfs_root *root = BTRFS_I(inode)->root;
6830
6831         WARN_ON(!list_empty(&inode->i_dentry));
6832         WARN_ON(inode->i_data.nrpages);
6833         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6834         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6835
6836         /*
6837          * This can happen where we create an inode, but somebody else also
6838          * created the same inode and we need to destroy the one we already
6839          * created.
6840          */
6841         if (!root)
6842                 goto free;
6843
6844         /*
6845          * Make sure we're properly removed from the ordered operation
6846          * lists.
6847          */
6848         smp_mb();
6849         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6850                 spin_lock(&root->fs_info->ordered_extent_lock);
6851                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6852                 spin_unlock(&root->fs_info->ordered_extent_lock);
6853         }
6854
6855         if (root == root->fs_info->tree_root) {
6856                 struct btrfs_block_group_cache *block_group;
6857
6858                 block_group = btrfs_lookup_block_group(root->fs_info,
6859                                                 BTRFS_I(inode)->block_group);
6860                 if (block_group && block_group->inode == inode) {
6861                         spin_lock(&block_group->lock);
6862                         block_group->inode = NULL;
6863                         spin_unlock(&block_group->lock);
6864                         btrfs_put_block_group(block_group);
6865                 } else if (block_group) {
6866                         btrfs_put_block_group(block_group);
6867                 }
6868         }
6869
6870         spin_lock(&root->orphan_lock);
6871         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6872                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6873                        (unsigned long long)btrfs_ino(inode));
6874                 list_del_init(&BTRFS_I(inode)->i_orphan);
6875         }
6876         spin_unlock(&root->orphan_lock);
6877
6878         while (1) {
6879                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6880                 if (!ordered)
6881                         break;
6882                 else {
6883                         printk(KERN_ERR "btrfs found ordered "
6884                                "extent %llu %llu on inode cleanup\n",
6885                                (unsigned long long)ordered->file_offset,
6886                                (unsigned long long)ordered->len);
6887                         btrfs_remove_ordered_extent(inode, ordered);
6888                         btrfs_put_ordered_extent(ordered);
6889                         btrfs_put_ordered_extent(ordered);
6890                 }
6891         }
6892         inode_tree_del(inode);
6893         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6894 free:
6895         call_rcu(&inode->i_rcu, btrfs_i_callback);
6896 }
6897
6898 int btrfs_drop_inode(struct inode *inode)
6899 {
6900         struct btrfs_root *root = BTRFS_I(inode)->root;
6901
6902         if (btrfs_root_refs(&root->root_item) == 0 &&
6903             root != root->fs_info->tree_root)
6904                 return 1;
6905         else
6906                 return generic_drop_inode(inode);
6907 }
6908
6909 static void init_once(void *foo)
6910 {
6911         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6912
6913         inode_init_once(&ei->vfs_inode);
6914 }
6915
6916 void btrfs_destroy_cachep(void)
6917 {
6918         if (btrfs_inode_cachep)
6919                 kmem_cache_destroy(btrfs_inode_cachep);
6920         if (btrfs_trans_handle_cachep)
6921                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6922         if (btrfs_transaction_cachep)
6923                 kmem_cache_destroy(btrfs_transaction_cachep);
6924         if (btrfs_path_cachep)
6925                 kmem_cache_destroy(btrfs_path_cachep);
6926         if (btrfs_free_space_cachep)
6927                 kmem_cache_destroy(btrfs_free_space_cachep);
6928 }
6929
6930 int btrfs_init_cachep(void)
6931 {
6932         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6933                         sizeof(struct btrfs_inode), 0,
6934                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6935         if (!btrfs_inode_cachep)
6936                 goto fail;
6937
6938         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6939                         sizeof(struct btrfs_trans_handle), 0,
6940                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6941         if (!btrfs_trans_handle_cachep)
6942                 goto fail;
6943
6944         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6945                         sizeof(struct btrfs_transaction), 0,
6946                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6947         if (!btrfs_transaction_cachep)
6948                 goto fail;
6949
6950         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6951                         sizeof(struct btrfs_path), 0,
6952                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6953         if (!btrfs_path_cachep)
6954                 goto fail;
6955
6956         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6957                         sizeof(struct btrfs_free_space), 0,
6958                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6959         if (!btrfs_free_space_cachep)
6960                 goto fail;
6961
6962         return 0;
6963 fail:
6964         btrfs_destroy_cachep();
6965         return -ENOMEM;
6966 }
6967
6968 static int btrfs_getattr(struct vfsmount *mnt,
6969                          struct dentry *dentry, struct kstat *stat)
6970 {
6971         struct inode *inode = dentry->d_inode;
6972         generic_fillattr(inode, stat);
6973         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6974         stat->blksize = PAGE_CACHE_SIZE;
6975         stat->blocks = (inode_get_bytes(inode) +
6976                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6977         return 0;
6978 }
6979
6980 /*
6981  * If a file is moved, it will inherit the cow and compression flags of the new
6982  * directory.
6983  */
6984 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6985 {
6986         struct btrfs_inode *b_dir = BTRFS_I(dir);
6987         struct btrfs_inode *b_inode = BTRFS_I(inode);
6988
6989         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6990                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6991         else
6992                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6993
6994         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6995                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6996         else
6997                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6998 }
6999
7000 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7001                            struct inode *new_dir, struct dentry *new_dentry)
7002 {
7003         struct btrfs_trans_handle *trans;
7004         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7005         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7006         struct inode *new_inode = new_dentry->d_inode;
7007         struct inode *old_inode = old_dentry->d_inode;
7008         struct timespec ctime = CURRENT_TIME;
7009         u64 index = 0;
7010         u64 root_objectid;
7011         int ret;
7012         u64 old_ino = btrfs_ino(old_inode);
7013
7014         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7015                 return -EPERM;
7016
7017         /* we only allow rename subvolume link between subvolumes */
7018         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7019                 return -EXDEV;
7020
7021         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7022             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7023                 return -ENOTEMPTY;
7024
7025         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7026             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7027                 return -ENOTEMPTY;
7028         /*
7029          * we're using rename to replace one file with another.
7030          * and the replacement file is large.  Start IO on it now so
7031          * we don't add too much work to the end of the transaction
7032          */
7033         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7034             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7035                 filemap_flush(old_inode->i_mapping);
7036
7037         /* close the racy window with snapshot create/destroy ioctl */
7038         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7039                 down_read(&root->fs_info->subvol_sem);
7040         /*
7041          * We want to reserve the absolute worst case amount of items.  So if
7042          * both inodes are subvols and we need to unlink them then that would
7043          * require 4 item modifications, but if they are both normal inodes it
7044          * would require 5 item modifications, so we'll assume their normal
7045          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7046          * should cover the worst case number of items we'll modify.
7047          */
7048         trans = btrfs_start_transaction(root, 20);
7049         if (IS_ERR(trans)) {
7050                 ret = PTR_ERR(trans);
7051                 goto out_notrans;
7052         }
7053
7054         btrfs_set_trans_block_group(trans, new_dir);
7055
7056         if (dest != root)
7057                 btrfs_record_root_in_trans(trans, dest);
7058
7059         ret = btrfs_set_inode_index(new_dir, &index);
7060         if (ret)
7061                 goto out_fail;
7062
7063         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7064                 /* force full log commit if subvolume involved. */
7065                 root->fs_info->last_trans_log_full_commit = trans->transid;
7066         } else {
7067                 ret = btrfs_insert_inode_ref(trans, dest,
7068                                              new_dentry->d_name.name,
7069                                              new_dentry->d_name.len,
7070                                              old_ino,
7071                                              btrfs_ino(new_dir), index);
7072                 if (ret)
7073                         goto out_fail;
7074                 /*
7075                  * this is an ugly little race, but the rename is required
7076                  * to make sure that if we crash, the inode is either at the
7077                  * old name or the new one.  pinning the log transaction lets
7078                  * us make sure we don't allow a log commit to come in after
7079                  * we unlink the name but before we add the new name back in.
7080                  */
7081                 btrfs_pin_log_trans(root);
7082         }
7083         /*
7084          * make sure the inode gets flushed if it is replacing
7085          * something.
7086          */
7087         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7088                 btrfs_add_ordered_operation(trans, root, old_inode);
7089
7090         old_dir->i_ctime = old_dir->i_mtime = ctime;
7091         new_dir->i_ctime = new_dir->i_mtime = ctime;
7092         old_inode->i_ctime = ctime;
7093
7094         if (old_dentry->d_parent != new_dentry->d_parent)
7095                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7096
7097         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7098                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7099                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7100                                         old_dentry->d_name.name,
7101                                         old_dentry->d_name.len);
7102         } else {
7103                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7104                                         old_dentry->d_inode,
7105                                         old_dentry->d_name.name,
7106                                         old_dentry->d_name.len);
7107                 if (!ret)
7108                         ret = btrfs_update_inode(trans, root, old_inode);
7109         }
7110         BUG_ON(ret);
7111
7112         if (new_inode) {
7113                 new_inode->i_ctime = CURRENT_TIME;
7114                 if (unlikely(btrfs_ino(new_inode) ==
7115                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7116                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7117                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7118                                                 root_objectid,
7119                                                 new_dentry->d_name.name,
7120                                                 new_dentry->d_name.len);
7121                         BUG_ON(new_inode->i_nlink == 0);
7122                 } else {
7123                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7124                                                  new_dentry->d_inode,
7125                                                  new_dentry->d_name.name,
7126                                                  new_dentry->d_name.len);
7127                 }
7128                 BUG_ON(ret);
7129                 if (new_inode->i_nlink == 0) {
7130                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7131                         BUG_ON(ret);
7132                 }
7133         }
7134
7135         fixup_inode_flags(new_dir, old_inode);
7136
7137         ret = btrfs_add_link(trans, new_dir, old_inode,
7138                              new_dentry->d_name.name,
7139                              new_dentry->d_name.len, 0, index);
7140         BUG_ON(ret);
7141
7142         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7143                 struct dentry *parent = dget_parent(new_dentry);
7144                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7145                 dput(parent);
7146                 btrfs_end_log_trans(root);
7147         }
7148 out_fail:
7149         btrfs_end_transaction_throttle(trans, root);
7150 out_notrans:
7151         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7152                 up_read(&root->fs_info->subvol_sem);
7153
7154         return ret;
7155 }
7156
7157 /*
7158  * some fairly slow code that needs optimization. This walks the list
7159  * of all the inodes with pending delalloc and forces them to disk.
7160  */
7161 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7162 {
7163         struct list_head *head = &root->fs_info->delalloc_inodes;
7164         struct btrfs_inode *binode;
7165         struct inode *inode;
7166
7167         if (root->fs_info->sb->s_flags & MS_RDONLY)
7168                 return -EROFS;
7169
7170         spin_lock(&root->fs_info->delalloc_lock);
7171         while (!list_empty(head)) {
7172                 binode = list_entry(head->next, struct btrfs_inode,
7173                                     delalloc_inodes);
7174                 inode = igrab(&binode->vfs_inode);
7175                 if (!inode)
7176                         list_del_init(&binode->delalloc_inodes);
7177                 spin_unlock(&root->fs_info->delalloc_lock);
7178                 if (inode) {
7179                         filemap_flush(inode->i_mapping);
7180                         if (delay_iput)
7181                                 btrfs_add_delayed_iput(inode);
7182                         else
7183                                 iput(inode);
7184                 }
7185                 cond_resched();
7186                 spin_lock(&root->fs_info->delalloc_lock);
7187         }
7188         spin_unlock(&root->fs_info->delalloc_lock);
7189
7190         /* the filemap_flush will queue IO into the worker threads, but
7191          * we have to make sure the IO is actually started and that
7192          * ordered extents get created before we return
7193          */
7194         atomic_inc(&root->fs_info->async_submit_draining);
7195         while (atomic_read(&root->fs_info->nr_async_submits) ||
7196               atomic_read(&root->fs_info->async_delalloc_pages)) {
7197                 wait_event(root->fs_info->async_submit_wait,
7198                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7199                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7200         }
7201         atomic_dec(&root->fs_info->async_submit_draining);
7202         return 0;
7203 }
7204
7205 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7206                                    int sync)
7207 {
7208         struct btrfs_inode *binode;
7209         struct inode *inode = NULL;
7210
7211         spin_lock(&root->fs_info->delalloc_lock);
7212         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7213                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7214                                     struct btrfs_inode, delalloc_inodes);
7215                 inode = igrab(&binode->vfs_inode);
7216                 if (inode) {
7217                         list_move_tail(&binode->delalloc_inodes,
7218                                        &root->fs_info->delalloc_inodes);
7219                         break;
7220                 }
7221
7222                 list_del_init(&binode->delalloc_inodes);
7223                 cond_resched_lock(&root->fs_info->delalloc_lock);
7224         }
7225         spin_unlock(&root->fs_info->delalloc_lock);
7226
7227         if (inode) {
7228                 if (sync) {
7229                         filemap_write_and_wait(inode->i_mapping);
7230                         /*
7231                          * We have to do this because compression doesn't
7232                          * actually set PG_writeback until it submits the pages
7233                          * for IO, which happens in an async thread, so we could
7234                          * race and not actually wait for any writeback pages
7235                          * because they've not been submitted yet.  Technically
7236                          * this could still be the case for the ordered stuff
7237                          * since the async thread may not have started to do its
7238                          * work yet.  If this becomes the case then we need to
7239                          * figure out a way to make sure that in writepage we
7240                          * wait for any async pages to be submitted before
7241                          * returning so that fdatawait does what its supposed to
7242                          * do.
7243                          */
7244                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7245                 } else {
7246                         filemap_flush(inode->i_mapping);
7247                 }
7248                 if (delay_iput)
7249                         btrfs_add_delayed_iput(inode);
7250                 else
7251                         iput(inode);
7252                 return 1;
7253         }
7254         return 0;
7255 }
7256
7257 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7258                          const char *symname)
7259 {
7260         struct btrfs_trans_handle *trans;
7261         struct btrfs_root *root = BTRFS_I(dir)->root;
7262         struct btrfs_path *path;
7263         struct btrfs_key key;
7264         struct inode *inode = NULL;
7265         int err;
7266         int drop_inode = 0;
7267         u64 objectid;
7268         u64 index = 0 ;
7269         int name_len;
7270         int datasize;
7271         unsigned long ptr;
7272         struct btrfs_file_extent_item *ei;
7273         struct extent_buffer *leaf;
7274         unsigned long nr = 0;
7275
7276         name_len = strlen(symname) + 1;
7277         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7278                 return -ENAMETOOLONG;
7279
7280         /*
7281          * 2 items for inode item and ref
7282          * 2 items for dir items
7283          * 1 item for xattr if selinux is on
7284          */
7285         trans = btrfs_start_transaction(root, 5);
7286         if (IS_ERR(trans))
7287                 return PTR_ERR(trans);
7288
7289         btrfs_set_trans_block_group(trans, dir);
7290
7291         err = btrfs_find_free_ino(root, &objectid);
7292         if (err)
7293                 goto out_unlock;
7294
7295         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7296                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7297                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7298                                 &index);
7299         err = PTR_ERR(inode);
7300         if (IS_ERR(inode))
7301                 goto out_unlock;
7302
7303         err = btrfs_init_inode_security(trans, inode, dir);
7304         if (err) {
7305                 drop_inode = 1;
7306                 goto out_unlock;
7307         }
7308
7309         btrfs_set_trans_block_group(trans, inode);
7310         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7311         if (err)
7312                 drop_inode = 1;
7313         else {
7314                 inode->i_mapping->a_ops = &btrfs_aops;
7315                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7316                 inode->i_fop = &btrfs_file_operations;
7317                 inode->i_op = &btrfs_file_inode_operations;
7318                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7319         }
7320         btrfs_update_inode_block_group(trans, inode);
7321         btrfs_update_inode_block_group(trans, dir);
7322         if (drop_inode)
7323                 goto out_unlock;
7324
7325         path = btrfs_alloc_path();
7326         BUG_ON(!path);
7327         key.objectid = btrfs_ino(inode);
7328         key.offset = 0;
7329         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7330         datasize = btrfs_file_extent_calc_inline_size(name_len);
7331         err = btrfs_insert_empty_item(trans, root, path, &key,
7332                                       datasize);
7333         if (err) {
7334                 drop_inode = 1;
7335                 goto out_unlock;
7336         }
7337         leaf = path->nodes[0];
7338         ei = btrfs_item_ptr(leaf, path->slots[0],
7339                             struct btrfs_file_extent_item);
7340         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7341         btrfs_set_file_extent_type(leaf, ei,
7342                                    BTRFS_FILE_EXTENT_INLINE);
7343         btrfs_set_file_extent_encryption(leaf, ei, 0);
7344         btrfs_set_file_extent_compression(leaf, ei, 0);
7345         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7346         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7347
7348         ptr = btrfs_file_extent_inline_start(ei);
7349         write_extent_buffer(leaf, symname, ptr, name_len);
7350         btrfs_mark_buffer_dirty(leaf);
7351         btrfs_free_path(path);
7352
7353         inode->i_op = &btrfs_symlink_inode_operations;
7354         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7355         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7356         inode_set_bytes(inode, name_len);
7357         btrfs_i_size_write(inode, name_len - 1);
7358         err = btrfs_update_inode(trans, root, inode);
7359         if (err)
7360                 drop_inode = 1;
7361
7362 out_unlock:
7363         nr = trans->blocks_used;
7364         btrfs_end_transaction_throttle(trans, root);
7365         if (drop_inode) {
7366                 inode_dec_link_count(inode);
7367                 iput(inode);
7368         }
7369         btrfs_btree_balance_dirty(root, nr);
7370         return err;
7371 }
7372
7373 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7374                                        u64 start, u64 num_bytes, u64 min_size,
7375                                        loff_t actual_len, u64 *alloc_hint,
7376                                        struct btrfs_trans_handle *trans)
7377 {
7378         struct btrfs_root *root = BTRFS_I(inode)->root;
7379         struct btrfs_key ins;
7380         u64 cur_offset = start;
7381         u64 i_size;
7382         int ret = 0;
7383         bool own_trans = true;
7384
7385         if (trans)
7386                 own_trans = false;
7387         while (num_bytes > 0) {
7388                 if (own_trans) {
7389                         trans = btrfs_start_transaction(root, 3);
7390                         if (IS_ERR(trans)) {
7391                                 ret = PTR_ERR(trans);
7392                                 break;
7393                         }
7394                 }
7395
7396                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7397                                            0, *alloc_hint, (u64)-1, &ins, 1);
7398                 if (ret) {
7399                         if (own_trans)
7400                                 btrfs_end_transaction(trans, root);
7401                         break;
7402                 }
7403
7404                 ret = insert_reserved_file_extent(trans, inode,
7405                                                   cur_offset, ins.objectid,
7406                                                   ins.offset, ins.offset,
7407                                                   ins.offset, 0, 0, 0,
7408                                                   BTRFS_FILE_EXTENT_PREALLOC);
7409                 BUG_ON(ret);
7410                 btrfs_drop_extent_cache(inode, cur_offset,
7411                                         cur_offset + ins.offset -1, 0);
7412
7413                 num_bytes -= ins.offset;
7414                 cur_offset += ins.offset;
7415                 *alloc_hint = ins.objectid + ins.offset;
7416
7417                 inode->i_ctime = CURRENT_TIME;
7418                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7419                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7420                     (actual_len > inode->i_size) &&
7421                     (cur_offset > inode->i_size)) {
7422                         if (cur_offset > actual_len)
7423                                 i_size = actual_len;
7424                         else
7425                                 i_size = cur_offset;
7426                         i_size_write(inode, i_size);
7427                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7428                 }
7429
7430                 ret = btrfs_update_inode(trans, root, inode);
7431                 BUG_ON(ret);
7432
7433                 if (own_trans)
7434                         btrfs_end_transaction(trans, root);
7435         }
7436         return ret;
7437 }
7438
7439 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7440                               u64 start, u64 num_bytes, u64 min_size,
7441                               loff_t actual_len, u64 *alloc_hint)
7442 {
7443         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7444                                            min_size, actual_len, alloc_hint,
7445                                            NULL);
7446 }
7447
7448 int btrfs_prealloc_file_range_trans(struct inode *inode,
7449                                     struct btrfs_trans_handle *trans, int mode,
7450                                     u64 start, u64 num_bytes, u64 min_size,
7451                                     loff_t actual_len, u64 *alloc_hint)
7452 {
7453         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7454                                            min_size, actual_len, alloc_hint, trans);
7455 }
7456
7457 static int btrfs_set_page_dirty(struct page *page)
7458 {
7459         return __set_page_dirty_nobuffers(page);
7460 }
7461
7462 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7463 {
7464         struct btrfs_root *root = BTRFS_I(inode)->root;
7465
7466         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7467                 return -EROFS;
7468         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7469                 return -EACCES;
7470         return generic_permission(inode, mask, flags, btrfs_check_acl);
7471 }
7472
7473 static const struct inode_operations btrfs_dir_inode_operations = {
7474         .getattr        = btrfs_getattr,
7475         .lookup         = btrfs_lookup,
7476         .create         = btrfs_create,
7477         .unlink         = btrfs_unlink,
7478         .link           = btrfs_link,
7479         .mkdir          = btrfs_mkdir,
7480         .rmdir          = btrfs_rmdir,
7481         .rename         = btrfs_rename,
7482         .symlink        = btrfs_symlink,
7483         .setattr        = btrfs_setattr,
7484         .mknod          = btrfs_mknod,
7485         .setxattr       = btrfs_setxattr,
7486         .getxattr       = btrfs_getxattr,
7487         .listxattr      = btrfs_listxattr,
7488         .removexattr    = btrfs_removexattr,
7489         .permission     = btrfs_permission,
7490 };
7491 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7492         .lookup         = btrfs_lookup,
7493         .permission     = btrfs_permission,
7494 };
7495
7496 static const struct file_operations btrfs_dir_file_operations = {
7497         .llseek         = generic_file_llseek,
7498         .read           = generic_read_dir,
7499         .readdir        = btrfs_real_readdir,
7500         .unlocked_ioctl = btrfs_ioctl,
7501 #ifdef CONFIG_COMPAT
7502         .compat_ioctl   = btrfs_ioctl,
7503 #endif
7504         .release        = btrfs_release_file,
7505         .fsync          = btrfs_sync_file,
7506 };
7507
7508 static struct extent_io_ops btrfs_extent_io_ops = {
7509         .fill_delalloc = run_delalloc_range,
7510         .submit_bio_hook = btrfs_submit_bio_hook,
7511         .merge_bio_hook = btrfs_merge_bio_hook,
7512         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7513         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7514         .writepage_start_hook = btrfs_writepage_start_hook,
7515         .readpage_io_failed_hook = btrfs_io_failed_hook,
7516         .set_bit_hook = btrfs_set_bit_hook,
7517         .clear_bit_hook = btrfs_clear_bit_hook,
7518         .merge_extent_hook = btrfs_merge_extent_hook,
7519         .split_extent_hook = btrfs_split_extent_hook,
7520 };
7521
7522 /*
7523  * btrfs doesn't support the bmap operation because swapfiles
7524  * use bmap to make a mapping of extents in the file.  They assume
7525  * these extents won't change over the life of the file and they
7526  * use the bmap result to do IO directly to the drive.
7527  *
7528  * the btrfs bmap call would return logical addresses that aren't
7529  * suitable for IO and they also will change frequently as COW
7530  * operations happen.  So, swapfile + btrfs == corruption.
7531  *
7532  * For now we're avoiding this by dropping bmap.
7533  */
7534 static const struct address_space_operations btrfs_aops = {
7535         .readpage       = btrfs_readpage,
7536         .writepage      = btrfs_writepage,
7537         .writepages     = btrfs_writepages,
7538         .readpages      = btrfs_readpages,
7539         .sync_page      = block_sync_page,
7540         .direct_IO      = btrfs_direct_IO,
7541         .invalidatepage = btrfs_invalidatepage,
7542         .releasepage    = btrfs_releasepage,
7543         .set_page_dirty = btrfs_set_page_dirty,
7544         .error_remove_page = generic_error_remove_page,
7545 };
7546
7547 static const struct address_space_operations btrfs_symlink_aops = {
7548         .readpage       = btrfs_readpage,
7549         .writepage      = btrfs_writepage,
7550         .invalidatepage = btrfs_invalidatepage,
7551         .releasepage    = btrfs_releasepage,
7552 };
7553
7554 static const struct inode_operations btrfs_file_inode_operations = {
7555         .getattr        = btrfs_getattr,
7556         .setattr        = btrfs_setattr,
7557         .setxattr       = btrfs_setxattr,
7558         .getxattr       = btrfs_getxattr,
7559         .listxattr      = btrfs_listxattr,
7560         .removexattr    = btrfs_removexattr,
7561         .permission     = btrfs_permission,
7562         .fiemap         = btrfs_fiemap,
7563 };
7564 static const struct inode_operations btrfs_special_inode_operations = {
7565         .getattr        = btrfs_getattr,
7566         .setattr        = btrfs_setattr,
7567         .permission     = btrfs_permission,
7568         .setxattr       = btrfs_setxattr,
7569         .getxattr       = btrfs_getxattr,
7570         .listxattr      = btrfs_listxattr,
7571         .removexattr    = btrfs_removexattr,
7572 };
7573 static const struct inode_operations btrfs_symlink_inode_operations = {
7574         .readlink       = generic_readlink,
7575         .follow_link    = page_follow_link_light,
7576         .put_link       = page_put_link,
7577         .getattr        = btrfs_getattr,
7578         .permission     = btrfs_permission,
7579         .setxattr       = btrfs_setxattr,
7580         .getxattr       = btrfs_getxattr,
7581         .listxattr      = btrfs_listxattr,
7582         .removexattr    = btrfs_removexattr,
7583 };
7584
7585 const struct dentry_operations btrfs_dentry_operations = {
7586         .d_delete       = btrfs_dentry_delete,
7587 };