fs/btrfs/inode.c

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