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 || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1790                 ret = btrfs_update_inode(trans, root, inode);
1791                 BUG_ON(ret);
1792         }
1793         ret = 0;
1794 out:
1795         btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1796         if (nolock) {
1797                 if (trans)
1798                         btrfs_end_transaction_nolock(trans, root);
1799         } else {
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->start > start || em->start + em->len < start) {
1867                         free_extent_map(em);
1868                         em = NULL;
1869                 }
1870                 read_unlock(&em_tree->lock);
1871
1872                 if (IS_ERR_OR_NULL(em)) {
1873                         kfree(failrec);
1874                         return -EIO;
1875                 }
1876                 logical = start - em->start;
1877                 logical = em->block_start + logical;
1878                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879                         logical = em->block_start;
1880                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1881                         extent_set_compress_type(&failrec->bio_flags,
1882                                                  em->compress_type);
1883                 }
1884                 failrec->logical = logical;
1885                 free_extent_map(em);
1886                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887                                 EXTENT_DIRTY, GFP_NOFS);
1888                 set_state_private(failure_tree, start,
1889                                  (u64)(unsigned long)failrec);
1890         } else {
1891                 failrec = (struct io_failure_record *)(unsigned long)private;
1892         }
1893         num_copies = btrfs_num_copies(
1894                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895                               failrec->logical, failrec->len);
1896         failrec->last_mirror++;
1897         if (!state) {
1898                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1899                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1900                                                     failrec->start,
1901                                                     EXTENT_LOCKED);
1902                 if (state && state->start != failrec->start)
1903                         state = NULL;
1904                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1905         }
1906         if (!state || failrec->last_mirror > num_copies) {
1907                 set_state_private(failure_tree, failrec->start, 0);
1908                 clear_extent_bits(failure_tree, failrec->start,
1909                                   failrec->start + failrec->len - 1,
1910                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1911                 kfree(failrec);
1912                 return -EIO;
1913         }
1914         bio = bio_alloc(GFP_NOFS, 1);
1915         bio->bi_private = state;
1916         bio->bi_end_io = failed_bio->bi_end_io;
1917         bio->bi_sector = failrec->logical >> 9;
1918         bio->bi_bdev = failed_bio->bi_bdev;
1919         bio->bi_size = 0;
1920
1921         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1922         if (failed_bio->bi_rw & REQ_WRITE)
1923                 rw = WRITE;
1924         else
1925                 rw = READ;
1926
1927         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1928                                                       failrec->last_mirror,
1929                                                       failrec->bio_flags, 0);
1930         return ret;
1931 }
1932
1933 /*
1934  * each time an IO finishes, we do a fast check in the IO failure tree
1935  * to see if we need to process or clean up an io_failure_record
1936  */
1937 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1938 {
1939         u64 private;
1940         u64 private_failure;
1941         struct io_failure_record *failure;
1942         int ret;
1943
1944         private = 0;
1945         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1946                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1947                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948                                         start, &private_failure);
1949                 if (ret == 0) {
1950                         failure = (struct io_failure_record *)(unsigned long)
1951                                    private_failure;
1952                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1953                                           failure->start, 0);
1954                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1955                                           failure->start,
1956                                           failure->start + failure->len - 1,
1957                                           EXTENT_DIRTY | EXTENT_LOCKED,
1958                                           GFP_NOFS);
1959                         kfree(failure);
1960                 }
1961         }
1962         return 0;
1963 }
1964
1965 /*
1966  * when reads are done, we need to check csums to verify the data is correct
1967  * if there's a match, we allow the bio to finish.  If not, we go through
1968  * the io_failure_record routines to find good copies
1969  */
1970 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1971                                struct extent_state *state)
1972 {
1973         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1974         struct inode *inode = page->mapping->host;
1975         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1976         char *kaddr;
1977         u64 private = ~(u32)0;
1978         int ret;
1979         struct btrfs_root *root = BTRFS_I(inode)->root;
1980         u32 csum = ~(u32)0;
1981
1982         if (PageChecked(page)) {
1983                 ClearPageChecked(page);
1984                 goto good;
1985         }
1986
1987         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1988                 goto good;
1989
1990         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1991             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1992                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1993                                   GFP_NOFS);
1994                 return 0;
1995         }
1996
1997         if (state && state->start == start) {
1998                 private = state->private;
1999                 ret = 0;
2000         } else {
2001                 ret = get_state_private(io_tree, start, &private);
2002         }
2003         kaddr = kmap_atomic(page, KM_USER0);
2004         if (ret)
2005                 goto zeroit;
2006
2007         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2008         btrfs_csum_final(csum, (char *)&csum);
2009         if (csum != private)
2010                 goto zeroit;
2011
2012         kunmap_atomic(kaddr, KM_USER0);
2013 good:
2014         /* if the io failure tree for this inode is non-empty,
2015          * check to see if we've recovered from a failed IO
2016          */
2017         btrfs_clean_io_failures(inode, start);
2018         return 0;
2019
2020 zeroit:
2021         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2022                        "private %llu\n",
2023                        (unsigned long long)btrfs_ino(page->mapping->host),
2024                        (unsigned long long)start, csum,
2025                        (unsigned long long)private);
2026         memset(kaddr + offset, 1, end - start + 1);
2027         flush_dcache_page(page);
2028         kunmap_atomic(kaddr, KM_USER0);
2029         if (private == 0)
2030                 return 0;
2031         return -EIO;
2032 }
2033
2034 struct delayed_iput {
2035         struct list_head list;
2036         struct inode *inode;
2037 };
2038
2039 void btrfs_add_delayed_iput(struct inode *inode)
2040 {
2041         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042         struct delayed_iput *delayed;
2043
2044         if (atomic_add_unless(&inode->i_count, -1, 1))
2045                 return;
2046
2047         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048         delayed->inode = inode;
2049
2050         spin_lock(&fs_info->delayed_iput_lock);
2051         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052         spin_unlock(&fs_info->delayed_iput_lock);
2053 }
2054
2055 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2056 {
2057         LIST_HEAD(list);
2058         struct btrfs_fs_info *fs_info = root->fs_info;
2059         struct delayed_iput *delayed;
2060         int empty;
2061
2062         spin_lock(&fs_info->delayed_iput_lock);
2063         empty = list_empty(&fs_info->delayed_iputs);
2064         spin_unlock(&fs_info->delayed_iput_lock);
2065         if (empty)
2066                 return;
2067
2068         down_read(&root->fs_info->cleanup_work_sem);
2069         spin_lock(&fs_info->delayed_iput_lock);
2070         list_splice_init(&fs_info->delayed_iputs, &list);
2071         spin_unlock(&fs_info->delayed_iput_lock);
2072
2073         while (!list_empty(&list)) {
2074                 delayed = list_entry(list.next, struct delayed_iput, list);
2075                 list_del(&delayed->list);
2076                 iput(delayed->inode);
2077                 kfree(delayed);
2078         }
2079         up_read(&root->fs_info->cleanup_work_sem);
2080 }
2081
2082 enum btrfs_orphan_cleanup_state {
2083         ORPHAN_CLEANUP_STARTED  = 1,
2084         ORPHAN_CLEANUP_DONE     = 2,
2085 };
2086
2087 /*
2088  * This is called in transaction commmit time. If there are no orphan
2089  * files in the subvolume, it removes orphan item and frees block_rsv
2090  * structure.
2091  */
2092 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2093                               struct btrfs_root *root)
2094 {
2095         int ret;
2096
2097         if (!list_empty(&root->orphan_list) ||
2098             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2099                 return;
2100
2101         if (root->orphan_item_inserted &&
2102             btrfs_root_refs(&root->root_item) > 0) {
2103                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2104                                             root->root_key.objectid);
2105                 BUG_ON(ret);
2106                 root->orphan_item_inserted = 0;
2107         }
2108
2109         if (root->orphan_block_rsv) {
2110                 WARN_ON(root->orphan_block_rsv->size > 0);
2111                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2112                 root->orphan_block_rsv = NULL;
2113         }
2114 }
2115
2116 /*
2117  * This creates an orphan entry for the given inode in case something goes
2118  * wrong in the middle of an unlink/truncate.
2119  *
2120  * NOTE: caller of this function should reserve 5 units of metadata for
2121  *       this function.
2122  */
2123 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2124 {
2125         struct btrfs_root *root = BTRFS_I(inode)->root;
2126         struct btrfs_block_rsv *block_rsv = NULL;
2127         int reserve = 0;
2128         int insert = 0;
2129         int ret;
2130
2131         if (!root->orphan_block_rsv) {
2132                 block_rsv = btrfs_alloc_block_rsv(root);
2133                 if (!block_rsv)
2134                         return -ENOMEM;
2135         }
2136
2137         spin_lock(&root->orphan_lock);
2138         if (!root->orphan_block_rsv) {
2139                 root->orphan_block_rsv = block_rsv;
2140         } else if (block_rsv) {
2141                 btrfs_free_block_rsv(root, block_rsv);
2142                 block_rsv = NULL;
2143         }
2144
2145         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2146                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2147 #if 0
2148                 /*
2149                  * For proper ENOSPC handling, we should do orphan
2150                  * cleanup when mounting. But this introduces backward
2151                  * compatibility issue.
2152                  */
2153                 if (!xchg(&root->orphan_item_inserted, 1))
2154                         insert = 2;
2155                 else
2156                         insert = 1;
2157 #endif
2158                 insert = 1;
2159         }
2160
2161         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2162                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2163                 reserve = 1;
2164         }
2165         spin_unlock(&root->orphan_lock);
2166
2167         /* grab metadata reservation from transaction handle */
2168         if (reserve) {
2169                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2170                 BUG_ON(ret);
2171         }
2172
2173         /* insert an orphan item to track this unlinked/truncated file */
2174         if (insert >= 1) {
2175                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2176                 BUG_ON(ret);
2177         }
2178
2179         /* insert an orphan item to track subvolume contains orphan files */
2180         if (insert >= 2) {
2181                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2182                                                root->root_key.objectid);
2183                 BUG_ON(ret);
2184         }
2185         return 0;
2186 }
2187
2188 /*
2189  * We have done the truncate/delete so we can go ahead and remove the orphan
2190  * item for this particular inode.
2191  */
2192 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2193 {
2194         struct btrfs_root *root = BTRFS_I(inode)->root;
2195         int delete_item = 0;
2196         int release_rsv = 0;
2197         int ret = 0;
2198
2199         spin_lock(&root->orphan_lock);
2200         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2201                 list_del_init(&BTRFS_I(inode)->i_orphan);
2202                 delete_item = 1;
2203         }
2204
2205         if (BTRFS_I(inode)->orphan_meta_reserved) {
2206                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2207                 release_rsv = 1;
2208         }
2209         spin_unlock(&root->orphan_lock);
2210
2211         if (trans && delete_item) {
2212                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2213                 BUG_ON(ret);
2214         }
2215
2216         if (release_rsv)
2217                 btrfs_orphan_release_metadata(inode);
2218
2219         return 0;
2220 }
2221
2222 /*
2223  * this cleans up any orphans that may be left on the list from the last use
2224  * of this root.
2225  */
2226 int btrfs_orphan_cleanup(struct btrfs_root *root)
2227 {
2228         struct btrfs_path *path;
2229         struct extent_buffer *leaf;
2230         struct btrfs_key key, found_key;
2231         struct btrfs_trans_handle *trans;
2232         struct inode *inode;
2233         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2234
2235         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2236                 return 0;
2237
2238         path = btrfs_alloc_path();
2239         if (!path) {
2240                 ret = -ENOMEM;
2241                 goto out;
2242         }
2243         path->reada = -1;
2244
2245         key.objectid = BTRFS_ORPHAN_OBJECTID;
2246         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2247         key.offset = (u64)-1;
2248
2249         while (1) {
2250                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2251                 if (ret < 0)
2252                         goto out;
2253
2254                 /*
2255                  * if ret == 0 means we found what we were searching for, which
2256                  * is weird, but possible, so only screw with path if we didn't
2257                  * find the key and see if we have stuff that matches
2258                  */
2259                 if (ret > 0) {
2260                         ret = 0;
2261                         if (path->slots[0] == 0)
2262                                 break;
2263                         path->slots[0]--;
2264                 }
2265
2266                 /* pull out the item */
2267                 leaf = path->nodes[0];
2268                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2269
2270                 /* make sure the item matches what we want */
2271                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2272                         break;
2273                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2274                         break;
2275
2276                 /* release the path since we're done with it */
2277                 btrfs_release_path(path);
2278
2279                 /*
2280                  * this is where we are basically btrfs_lookup, without the
2281                  * crossing root thing.  we store the inode number in the
2282                  * offset of the orphan item.
2283                  */
2284                 found_key.objectid = found_key.offset;
2285                 found_key.type = BTRFS_INODE_ITEM_KEY;
2286                 found_key.offset = 0;
2287                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2288                 ret = PTR_RET(inode);
2289                 if (ret && ret != -ESTALE)
2290                         goto out;
2291
2292                 /*
2293                  * Inode is already gone but the orphan item is still there,
2294                  * kill the orphan item.
2295                  */
2296                 if (ret == -ESTALE) {
2297                         trans = btrfs_start_transaction(root, 1);
2298                         if (IS_ERR(trans)) {
2299                                 ret = PTR_ERR(trans);
2300                                 goto out;
2301                         }
2302                         ret = btrfs_del_orphan_item(trans, root,
2303                                                     found_key.objectid);
2304                         BUG_ON(ret);
2305                         btrfs_end_transaction(trans, root);
2306                         continue;
2307                 }
2308
2309                 /*
2310                  * add this inode to the orphan list so btrfs_orphan_del does
2311                  * the proper thing when we hit it
2312                  */
2313                 spin_lock(&root->orphan_lock);
2314                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2315                 spin_unlock(&root->orphan_lock);
2316
2317                 /* if we have links, this was a truncate, lets do that */
2318                 if (inode->i_nlink) {
2319                         if (!S_ISREG(inode->i_mode)) {
2320                                 WARN_ON(1);
2321                                 iput(inode);
2322                                 continue;
2323                         }
2324                         nr_truncate++;
2325                         ret = btrfs_truncate(inode);
2326                 } else {
2327                         nr_unlink++;
2328                 }
2329
2330                 /* this will do delete_inode and everything for us */
2331                 iput(inode);
2332                 if (ret)
2333                         goto out;
2334         }
2335         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2336
2337         if (root->orphan_block_rsv)
2338                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2339                                         (u64)-1);
2340
2341         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2342                 trans = btrfs_join_transaction(root);
2343                 if (!IS_ERR(trans))
2344                         btrfs_end_transaction(trans, root);
2345         }
2346
2347         if (nr_unlink)
2348                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2349         if (nr_truncate)
2350                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2351
2352 out:
2353         if (ret)
2354                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2355         btrfs_free_path(path);
2356         return ret;
2357 }
2358
2359 /*
2360  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2361  * don't find any xattrs, we know there can't be any acls.
2362  *
2363  * slot is the slot the inode is in, objectid is the objectid of the inode
2364  */
2365 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2366                                           int slot, u64 objectid)
2367 {
2368         u32 nritems = btrfs_header_nritems(leaf);
2369         struct btrfs_key found_key;
2370         int scanned = 0;
2371
2372         slot++;
2373         while (slot < nritems) {
2374                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2375
2376                 /* we found a different objectid, there must not be acls */
2377                 if (found_key.objectid != objectid)
2378                         return 0;
2379
2380                 /* we found an xattr, assume we've got an acl */
2381                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2382                         return 1;
2383
2384                 /*
2385                  * we found a key greater than an xattr key, there can't
2386                  * be any acls later on
2387                  */
2388                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2389                         return 0;
2390
2391                 slot++;
2392                 scanned++;
2393
2394                 /*
2395                  * it goes inode, inode backrefs, xattrs, extents,
2396                  * so if there are a ton of hard links to an inode there can
2397                  * be a lot of backrefs.  Don't waste time searching too hard,
2398                  * this is just an optimization
2399                  */
2400                 if (scanned >= 8)
2401                         break;
2402         }
2403         /* we hit the end of the leaf before we found an xattr or
2404          * something larger than an xattr.  We have to assume the inode
2405          * has acls
2406          */
2407         return 1;
2408 }
2409
2410 /*
2411  * read an inode from the btree into the in-memory inode
2412  */
2413 static void btrfs_read_locked_inode(struct inode *inode)
2414 {
2415         struct btrfs_path *path;
2416         struct extent_buffer *leaf;
2417         struct btrfs_inode_item *inode_item;
2418         struct btrfs_timespec *tspec;
2419         struct btrfs_root *root = BTRFS_I(inode)->root;
2420         struct btrfs_key location;
2421         int maybe_acls;
2422         u32 rdev;
2423         int ret;
2424         bool filled = false;
2425
2426         ret = btrfs_fill_inode(inode, &rdev);
2427         if (!ret)
2428                 filled = true;
2429
2430         path = btrfs_alloc_path();
2431         if (!path)
2432                 goto make_bad;
2433
2434         path->leave_spinning = 1;
2435         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2436
2437         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2438         if (ret)
2439                 goto make_bad;
2440
2441         leaf = path->nodes[0];
2442
2443         if (filled)
2444                 goto cache_acl;
2445
2446         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2447                                     struct btrfs_inode_item);
2448         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2449         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2450         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2451         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2452         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2453
2454         tspec = btrfs_inode_atime(inode_item);
2455         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2456         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2457
2458         tspec = btrfs_inode_mtime(inode_item);
2459         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2460         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2461
2462         tspec = btrfs_inode_ctime(inode_item);
2463         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2464         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2465
2466         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2467         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2468         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2469         inode->i_generation = BTRFS_I(inode)->generation;
2470         inode->i_rdev = 0;
2471         rdev = btrfs_inode_rdev(leaf, inode_item);
2472
2473         BTRFS_I(inode)->index_cnt = (u64)-1;
2474         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2475 cache_acl:
2476         /*
2477          * try to precache a NULL acl entry for files that don't have
2478          * any xattrs or acls
2479          */
2480         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2481                                            btrfs_ino(inode));
2482         if (!maybe_acls)
2483                 cache_no_acl(inode);
2484
2485         btrfs_free_path(path);
2486
2487         switch (inode->i_mode & S_IFMT) {
2488         case S_IFREG:
2489                 inode->i_mapping->a_ops = &btrfs_aops;
2490                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2491                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2492                 inode->i_fop = &btrfs_file_operations;
2493                 inode->i_op = &btrfs_file_inode_operations;
2494                 break;
2495         case S_IFDIR:
2496                 inode->i_fop = &btrfs_dir_file_operations;
2497                 if (root == root->fs_info->tree_root)
2498                         inode->i_op = &btrfs_dir_ro_inode_operations;
2499                 else
2500                         inode->i_op = &btrfs_dir_inode_operations;
2501                 break;
2502         case S_IFLNK:
2503                 inode->i_op = &btrfs_symlink_inode_operations;
2504                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2505                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2506                 break;
2507         default:
2508                 inode->i_op = &btrfs_special_inode_operations;
2509                 init_special_inode(inode, inode->i_mode, rdev);
2510                 break;
2511         }
2512
2513         btrfs_update_iflags(inode);
2514         return;
2515
2516 make_bad:
2517         btrfs_free_path(path);
2518         make_bad_inode(inode);
2519 }
2520
2521 /*
2522  * given a leaf and an inode, copy the inode fields into the leaf
2523  */
2524 static void fill_inode_item(struct btrfs_trans_handle *trans,
2525                             struct extent_buffer *leaf,
2526                             struct btrfs_inode_item *item,
2527                             struct inode *inode)
2528 {
2529         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2530         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2531         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2532         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2533         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2534
2535         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2536                                inode->i_atime.tv_sec);
2537         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2538                                 inode->i_atime.tv_nsec);
2539
2540         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2541                                inode->i_mtime.tv_sec);
2542         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2543                                 inode->i_mtime.tv_nsec);
2544
2545         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2546                                inode->i_ctime.tv_sec);
2547         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2548                                 inode->i_ctime.tv_nsec);
2549
2550         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2551         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2552         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2553         btrfs_set_inode_transid(leaf, item, trans->transid);
2554         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2555         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2556         btrfs_set_inode_block_group(leaf, item, 0);
2557 }
2558
2559 /*
2560  * copy everything in the in-memory inode into the btree.
2561  */
2562 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2563                                 struct btrfs_root *root, struct inode *inode)
2564 {
2565         struct btrfs_inode_item *inode_item;
2566         struct btrfs_path *path;
2567         struct extent_buffer *leaf;
2568         int ret;
2569
2570         /*
2571          * If the inode is a free space inode, we can deadlock during commit
2572          * if we put it into the delayed code.
2573          *
2574          * The data relocation inode should also be directly updated
2575          * without delay
2576          */
2577         if (!btrfs_is_free_space_inode(root, inode)
2578             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2579                 ret = btrfs_delayed_update_inode(trans, root, inode);
2580                 if (!ret)
2581                         btrfs_set_inode_last_trans(trans, inode);
2582                 return ret;
2583         }
2584
2585         path = btrfs_alloc_path();
2586         if (!path)
2587                 return -ENOMEM;
2588
2589         path->leave_spinning = 1;
2590         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2591                                  1);
2592         if (ret) {
2593                 if (ret > 0)
2594                         ret = -ENOENT;
2595                 goto failed;
2596         }
2597
2598         btrfs_unlock_up_safe(path, 1);
2599         leaf = path->nodes[0];
2600         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2601                                     struct btrfs_inode_item);
2602
2603         fill_inode_item(trans, leaf, inode_item, inode);
2604         btrfs_mark_buffer_dirty(leaf);
2605         btrfs_set_inode_last_trans(trans, inode);
2606         ret = 0;
2607 failed:
2608         btrfs_free_path(path);
2609         return ret;
2610 }
2611
2612 /*
2613  * unlink helper that gets used here in inode.c and in the tree logging
2614  * recovery code.  It remove a link in a directory with a given name, and
2615  * also drops the back refs in the inode to the directory
2616  */
2617 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2618                                 struct btrfs_root *root,
2619                                 struct inode *dir, struct inode *inode,
2620                                 const char *name, int name_len)
2621 {
2622         struct btrfs_path *path;
2623         int ret = 0;
2624         struct extent_buffer *leaf;
2625         struct btrfs_dir_item *di;
2626         struct btrfs_key key;
2627         u64 index;
2628         u64 ino = btrfs_ino(inode);
2629         u64 dir_ino = btrfs_ino(dir);
2630
2631         path = btrfs_alloc_path();
2632         if (!path) {
2633                 ret = -ENOMEM;
2634                 goto out;
2635         }
2636
2637         path->leave_spinning = 1;
2638         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2639                                     name, name_len, -1);
2640         if (IS_ERR(di)) {
2641                 ret = PTR_ERR(di);
2642                 goto err;
2643         }
2644         if (!di) {
2645                 ret = -ENOENT;
2646                 goto err;
2647         }
2648         leaf = path->nodes[0];
2649         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2650         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2651         if (ret)
2652                 goto err;
2653         btrfs_release_path(path);
2654
2655         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2656                                   dir_ino, &index);
2657         if (ret) {
2658                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2659                        "inode %llu parent %llu\n", name_len, name,
2660                        (unsigned long long)ino, (unsigned long long)dir_ino);
2661                 goto err;
2662         }
2663
2664         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2665         if (ret)
2666                 goto err;
2667
2668         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2669                                          inode, dir_ino);
2670         BUG_ON(ret != 0 && ret != -ENOENT);
2671
2672         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2673                                            dir, index);
2674         if (ret == -ENOENT)
2675                 ret = 0;
2676 err:
2677         btrfs_free_path(path);
2678         if (ret)
2679                 goto out;
2680
2681         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2682         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2683         btrfs_update_inode(trans, root, dir);
2684 out:
2685         return ret;
2686 }
2687
2688 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2689                        struct btrfs_root *root,
2690                        struct inode *dir, struct inode *inode,
2691                        const char *name, int name_len)
2692 {
2693         int ret;
2694         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2695         if (!ret) {
2696                 btrfs_drop_nlink(inode);
2697                 ret = btrfs_update_inode(trans, root, inode);
2698         }
2699         return ret;
2700 }
2701
2702
2703 /* helper to check if there is any shared block in the path */
2704 static int check_path_shared(struct btrfs_root *root,
2705                              struct btrfs_path *path)
2706 {
2707         struct extent_buffer *eb;
2708         int level;
2709         u64 refs = 1;
2710
2711         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2712                 int ret;
2713
2714                 if (!path->nodes[level])
2715                         break;
2716                 eb = path->nodes[level];
2717                 if (!btrfs_block_can_be_shared(root, eb))
2718                         continue;
2719                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2720                                                &refs, NULL);
2721                 if (refs > 1)
2722                         return 1;
2723         }
2724         return 0;
2725 }
2726
2727 /*
2728  * helper to start transaction for unlink and rmdir.
2729  *
2730  * unlink and rmdir are special in btrfs, they do not always free space.
2731  * so in enospc case, we should make sure they will free space before
2732  * allowing them to use the global metadata reservation.
2733  */
2734 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2735                                                        struct dentry *dentry)
2736 {
2737         struct btrfs_trans_handle *trans;
2738         struct btrfs_root *root = BTRFS_I(dir)->root;
2739         struct btrfs_path *path;
2740         struct btrfs_inode_ref *ref;
2741         struct btrfs_dir_item *di;
2742         struct inode *inode = dentry->d_inode;
2743         u64 index;
2744         int check_link = 1;
2745         int err = -ENOSPC;
2746         int ret;
2747         u64 ino = btrfs_ino(inode);
2748         u64 dir_ino = btrfs_ino(dir);
2749
2750         trans = btrfs_start_transaction(root, 10);
2751         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2752                 return trans;
2753
2754         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2755                 return ERR_PTR(-ENOSPC);
2756
2757         /* check if there is someone else holds reference */
2758         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2759                 return ERR_PTR(-ENOSPC);
2760
2761         if (atomic_read(&inode->i_count) > 2)
2762                 return ERR_PTR(-ENOSPC);
2763
2764         if (xchg(&root->fs_info->enospc_unlink, 1))
2765                 return ERR_PTR(-ENOSPC);
2766
2767         path = btrfs_alloc_path();
2768         if (!path) {
2769                 root->fs_info->enospc_unlink = 0;
2770                 return ERR_PTR(-ENOMEM);
2771         }
2772
2773         trans = btrfs_start_transaction(root, 0);
2774         if (IS_ERR(trans)) {
2775                 btrfs_free_path(path);
2776                 root->fs_info->enospc_unlink = 0;
2777                 return trans;
2778         }
2779
2780         path->skip_locking = 1;
2781         path->search_commit_root = 1;
2782
2783         ret = btrfs_lookup_inode(trans, root, path,
2784                                 &BTRFS_I(dir)->location, 0);
2785         if (ret < 0) {
2786                 err = ret;
2787                 goto out;
2788         }
2789         if (ret == 0) {
2790                 if (check_path_shared(root, path))
2791                         goto out;
2792         } else {
2793                 check_link = 0;
2794         }
2795         btrfs_release_path(path);
2796
2797         ret = btrfs_lookup_inode(trans, root, path,
2798                                 &BTRFS_I(inode)->location, 0);
2799         if (ret < 0) {
2800                 err = ret;
2801                 goto out;
2802         }
2803         if (ret == 0) {
2804                 if (check_path_shared(root, path))
2805                         goto out;
2806         } else {
2807                 check_link = 0;
2808         }
2809         btrfs_release_path(path);
2810
2811         if (ret == 0 && S_ISREG(inode->i_mode)) {
2812                 ret = btrfs_lookup_file_extent(trans, root, path,
2813                                                ino, (u64)-1, 0);
2814                 if (ret < 0) {
2815                         err = ret;
2816                         goto out;
2817                 }
2818                 BUG_ON(ret == 0);
2819                 if (check_path_shared(root, path))
2820                         goto out;
2821                 btrfs_release_path(path);
2822         }
2823
2824         if (!check_link) {
2825                 err = 0;
2826                 goto out;
2827         }
2828
2829         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2830                                 dentry->d_name.name, dentry->d_name.len, 0);
2831         if (IS_ERR(di)) {
2832                 err = PTR_ERR(di);
2833                 goto out;
2834         }
2835         if (di) {
2836                 if (check_path_shared(root, path))
2837                         goto out;
2838         } else {
2839                 err = 0;
2840                 goto out;
2841         }
2842         btrfs_release_path(path);
2843
2844         ref = btrfs_lookup_inode_ref(trans, root, path,
2845                                 dentry->d_name.name, dentry->d_name.len,
2846                                 ino, dir_ino, 0);
2847         if (IS_ERR(ref)) {
2848                 err = PTR_ERR(ref);
2849                 goto out;
2850         }
2851         BUG_ON(!ref);
2852         if (check_path_shared(root, path))
2853                 goto out;
2854         index = btrfs_inode_ref_index(path->nodes[0], ref);
2855         btrfs_release_path(path);
2856
2857         /*
2858          * This is a commit root search, if we can lookup inode item and other
2859          * relative items in the commit root, it means the transaction of
2860          * dir/file creation has been committed, and the dir index item that we
2861          * delay to insert has also been inserted into the commit root. So
2862          * we needn't worry about the delayed insertion of the dir index item
2863          * here.
2864          */
2865         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2866                                 dentry->d_name.name, dentry->d_name.len, 0);
2867         if (IS_ERR(di)) {
2868                 err = PTR_ERR(di);
2869                 goto out;
2870         }
2871         BUG_ON(ret == -ENOENT);
2872         if (check_path_shared(root, path))
2873                 goto out;
2874
2875         err = 0;
2876 out:
2877         btrfs_free_path(path);
2878         if (err) {
2879                 btrfs_end_transaction(trans, root);
2880                 root->fs_info->enospc_unlink = 0;
2881                 return ERR_PTR(err);
2882         }
2883
2884         trans->block_rsv = &root->fs_info->global_block_rsv;
2885         return trans;
2886 }
2887
2888 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2889                                struct btrfs_root *root)
2890 {
2891         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2892                 BUG_ON(!root->fs_info->enospc_unlink);
2893                 root->fs_info->enospc_unlink = 0;
2894         }
2895         btrfs_end_transaction_throttle(trans, root);
2896 }
2897
2898 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2899 {
2900         struct btrfs_root *root = BTRFS_I(dir)->root;
2901         struct btrfs_trans_handle *trans;
2902         struct inode *inode = dentry->d_inode;
2903         int ret;
2904         unsigned long nr = 0;
2905
2906         trans = __unlink_start_trans(dir, dentry);
2907         if (IS_ERR(trans))
2908                 return PTR_ERR(trans);
2909
2910         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2911
2912         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2913                                  dentry->d_name.name, dentry->d_name.len);
2914         if (ret)
2915                 goto out;
2916
2917         if (inode->i_nlink == 0) {
2918                 ret = btrfs_orphan_add(trans, inode);
2919                 if (ret)
2920                         goto out;
2921         }
2922
2923 out:
2924         nr = trans->blocks_used;
2925         __unlink_end_trans(trans, root);
2926         btrfs_btree_balance_dirty(root, nr);
2927         return ret;
2928 }
2929
2930 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2931                         struct btrfs_root *root,
2932                         struct inode *dir, u64 objectid,
2933                         const char *name, int name_len)
2934 {
2935         struct btrfs_path *path;
2936         struct extent_buffer *leaf;
2937         struct btrfs_dir_item *di;
2938         struct btrfs_key key;
2939         u64 index;
2940         int ret;
2941         u64 dir_ino = btrfs_ino(dir);
2942
2943         path = btrfs_alloc_path();
2944         if (!path)
2945                 return -ENOMEM;
2946
2947         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2948                                    name, name_len, -1);
2949         BUG_ON(IS_ERR_OR_NULL(di));
2950
2951         leaf = path->nodes[0];
2952         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2953         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2954         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2955         BUG_ON(ret);
2956         btrfs_release_path(path);
2957
2958         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2959                                  objectid, root->root_key.objectid,
2960                                  dir_ino, &index, name, name_len);
2961         if (ret < 0) {
2962                 BUG_ON(ret != -ENOENT);
2963                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2964                                                  name, name_len);
2965                 BUG_ON(IS_ERR_OR_NULL(di));
2966
2967                 leaf = path->nodes[0];
2968                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2969                 btrfs_release_path(path);
2970                 index = key.offset;
2971         }
2972         btrfs_release_path(path);
2973
2974         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2975         BUG_ON(ret);
2976
2977         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2978         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2979         ret = btrfs_update_inode(trans, root, dir);
2980         BUG_ON(ret);
2981
2982         btrfs_free_path(path);
2983         return 0;
2984 }
2985
2986 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2987 {
2988         struct inode *inode = dentry->d_inode;
2989         int err = 0;
2990         struct btrfs_root *root = BTRFS_I(dir)->root;
2991         struct btrfs_trans_handle *trans;
2992         unsigned long nr = 0;
2993
2994         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2995             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2996                 return -ENOTEMPTY;
2997
2998         trans = __unlink_start_trans(dir, dentry);
2999         if (IS_ERR(trans))
3000                 return PTR_ERR(trans);
3001
3002         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3003                 err = btrfs_unlink_subvol(trans, root, dir,
3004                                           BTRFS_I(inode)->location.objectid,
3005                                           dentry->d_name.name,
3006                                           dentry->d_name.len);
3007                 goto out;
3008         }
3009
3010         err = btrfs_orphan_add(trans, inode);
3011         if (err)
3012                 goto out;
3013
3014         /* now the directory is empty */
3015         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3016                                  dentry->d_name.name, dentry->d_name.len);
3017         if (!err)
3018                 btrfs_i_size_write(inode, 0);
3019 out:
3020         nr = trans->blocks_used;
3021         __unlink_end_trans(trans, root);
3022         btrfs_btree_balance_dirty(root, nr);
3023
3024         return err;
3025 }
3026
3027 /*
3028  * this can truncate away extent items, csum items and directory items.
3029  * It starts at a high offset and removes keys until it can't find
3030  * any higher than new_size
3031  *
3032  * csum items that cross the new i_size are truncated to the new size
3033  * as well.
3034  *
3035  * min_type is the minimum key type to truncate down to.  If set to 0, this
3036  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3037  */
3038 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3039                                struct btrfs_root *root,
3040                                struct inode *inode,
3041                                u64 new_size, u32 min_type)
3042 {
3043         struct btrfs_path *path;
3044         struct extent_buffer *leaf;
3045         struct btrfs_file_extent_item *fi;
3046         struct btrfs_key key;
3047         struct btrfs_key found_key;
3048         u64 extent_start = 0;
3049         u64 extent_num_bytes = 0;
3050         u64 extent_offset = 0;
3051         u64 item_end = 0;
3052         u64 mask = root->sectorsize - 1;
3053         u32 found_type = (u8)-1;
3054         int found_extent;
3055         int del_item;
3056         int pending_del_nr = 0;
3057         int pending_del_slot = 0;
3058         int extent_type = -1;
3059         int encoding;
3060         int ret;
3061         int err = 0;
3062         u64 ino = btrfs_ino(inode);
3063
3064         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3065
3066         path = btrfs_alloc_path();
3067         if (!path)
3068                 return -ENOMEM;
3069         path->reada = -1;
3070
3071         if (root->ref_cows || root == root->fs_info->tree_root)
3072                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3073
3074         /*
3075          * This function is also used to drop the items in the log tree before
3076          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3077          * it is used to drop the loged items. So we shouldn't kill the delayed
3078          * items.
3079          */
3080         if (min_type == 0 && root == BTRFS_I(inode)->root)
3081                 btrfs_kill_delayed_inode_items(inode);
3082
3083         key.objectid = ino;
3084         key.offset = (u64)-1;
3085         key.type = (u8)-1;
3086
3087 search_again:
3088         path->leave_spinning = 1;
3089         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3090         if (ret < 0) {
3091                 err = ret;
3092                 goto out;
3093         }
3094
3095         if (ret > 0) {
3096                 /* there are no items in the tree for us to truncate, we're
3097                  * done
3098                  */
3099                 if (path->slots[0] == 0)
3100                         goto out;
3101                 path->slots[0]--;
3102         }
3103
3104         while (1) {
3105                 fi = NULL;
3106                 leaf = path->nodes[0];
3107                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3108                 found_type = btrfs_key_type(&found_key);
3109                 encoding = 0;
3110
3111                 if (found_key.objectid != ino)
3112                         break;
3113
3114                 if (found_type < min_type)
3115                         break;
3116
3117                 item_end = found_key.offset;
3118                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3119                         fi = btrfs_item_ptr(leaf, path->slots[0],
3120                                             struct btrfs_file_extent_item);
3121                         extent_type = btrfs_file_extent_type(leaf, fi);
3122                         encoding = btrfs_file_extent_compression(leaf, fi);
3123                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3124                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3125
3126                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3127                                 item_end +=
3128                                     btrfs_file_extent_num_bytes(leaf, fi);
3129                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3130                                 item_end += btrfs_file_extent_inline_len(leaf,
3131                                                                          fi);
3132                         }
3133                         item_end--;
3134                 }
3135                 if (found_type > min_type) {
3136                         del_item = 1;
3137                 } else {
3138                         if (item_end < new_size)
3139                                 break;
3140                         if (found_key.offset >= new_size)
3141                                 del_item = 1;
3142                         else
3143                                 del_item = 0;
3144                 }
3145                 found_extent = 0;
3146                 /* FIXME, shrink the extent if the ref count is only 1 */
3147                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3148                         goto delete;
3149
3150                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3151                         u64 num_dec;
3152                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3153                         if (!del_item && !encoding) {
3154                                 u64 orig_num_bytes =
3155                                         btrfs_file_extent_num_bytes(leaf, fi);
3156                                 extent_num_bytes = new_size -
3157                                         found_key.offset + root->sectorsize - 1;
3158                                 extent_num_bytes = extent_num_bytes &
3159                                         ~((u64)root->sectorsize - 1);
3160                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3161                                                          extent_num_bytes);
3162                                 num_dec = (orig_num_bytes -
3163                                            extent_num_bytes);
3164                                 if (root->ref_cows && extent_start != 0)
3165                                         inode_sub_bytes(inode, num_dec);
3166                                 btrfs_mark_buffer_dirty(leaf);
3167                         } else {
3168                                 extent_num_bytes =
3169                                         btrfs_file_extent_disk_num_bytes(leaf,
3170                                                                          fi);
3171                                 extent_offset = found_key.offset -
3172                                         btrfs_file_extent_offset(leaf, fi);
3173
3174                                 /* FIXME blocksize != 4096 */
3175                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3176                                 if (extent_start != 0) {
3177                                         found_extent = 1;
3178                                         if (root->ref_cows)
3179                                                 inode_sub_bytes(inode, num_dec);
3180                                 }
3181                         }
3182                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3183                         /*
3184                          * we can't truncate inline items that have had
3185                          * special encodings
3186                          */
3187                         if (!del_item &&
3188                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3189                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3190                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3191                                 u32 size = new_size - found_key.offset;
3192
3193                                 if (root->ref_cows) {
3194                                         inode_sub_bytes(inode, item_end + 1 -
3195                                                         new_size);
3196                                 }
3197                                 size =
3198                                     btrfs_file_extent_calc_inline_size(size);
3199                                 ret = btrfs_truncate_item(trans, root, path,
3200                                                           size, 1);
3201                         } else if (root->ref_cows) {
3202                                 inode_sub_bytes(inode, item_end + 1 -
3203                                                 found_key.offset);
3204                         }
3205                 }
3206 delete:
3207                 if (del_item) {
3208                         if (!pending_del_nr) {
3209                                 /* no pending yet, add ourselves */
3210                                 pending_del_slot = path->slots[0];
3211                                 pending_del_nr = 1;
3212                         } else if (pending_del_nr &&
3213                                    path->slots[0] + 1 == pending_del_slot) {
3214                                 /* hop on the pending chunk */
3215                                 pending_del_nr++;
3216                                 pending_del_slot = path->slots[0];
3217                         } else {
3218                                 BUG();
3219                         }
3220                 } else {
3221                         break;
3222                 }
3223                 if (found_extent && (root->ref_cows ||
3224                                      root == root->fs_info->tree_root)) {
3225                         btrfs_set_path_blocking(path);
3226                         ret = btrfs_free_extent(trans, root, extent_start,
3227                                                 extent_num_bytes, 0,
3228                                                 btrfs_header_owner(leaf),
3229                                                 ino, extent_offset);
3230                         BUG_ON(ret);
3231                 }
3232
3233                 if (found_type == BTRFS_INODE_ITEM_KEY)
3234                         break;
3235
3236                 if (path->slots[0] == 0 ||
3237                     path->slots[0] != pending_del_slot) {
3238                         if (root->ref_cows &&
3239                             BTRFS_I(inode)->location.objectid !=
3240                                                 BTRFS_FREE_INO_OBJECTID) {
3241                                 err = -EAGAIN;
3242                                 goto out;
3243                         }
3244                         if (pending_del_nr) {
3245                                 ret = btrfs_del_items(trans, root, path,
3246                                                 pending_del_slot,
3247                                                 pending_del_nr);
3248                                 BUG_ON(ret);
3249                                 pending_del_nr = 0;
3250                         }
3251                         btrfs_release_path(path);
3252                         goto search_again;
3253                 } else {
3254                         path->slots[0]--;
3255                 }
3256         }
3257 out:
3258         if (pending_del_nr) {
3259                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3260                                       pending_del_nr);
3261                 BUG_ON(ret);
3262         }
3263         btrfs_free_path(path);
3264         return err;
3265 }
3266
3267 /*
3268  * taken from block_truncate_page, but does cow as it zeros out
3269  * any bytes left in the last page in the file.
3270  */
3271 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3272 {
3273         struct inode *inode = mapping->host;
3274         struct btrfs_root *root = BTRFS_I(inode)->root;
3275         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3276         struct btrfs_ordered_extent *ordered;
3277         struct extent_state *cached_state = NULL;
3278         char *kaddr;
3279         u32 blocksize = root->sectorsize;
3280         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3281         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3282         struct page *page;
3283         gfp_t mask = btrfs_alloc_write_mask(mapping);
3284         int ret = 0;
3285         u64 page_start;
3286         u64 page_end;
3287
3288         if ((offset & (blocksize - 1)) == 0)
3289                 goto out;
3290         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3291         if (ret)
3292                 goto out;
3293
3294         ret = -ENOMEM;
3295 again:
3296         page = find_or_create_page(mapping, index, mask);
3297         if (!page) {
3298                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3299                 goto out;
3300         }
3301
3302         page_start = page_offset(page);
3303         page_end = page_start + PAGE_CACHE_SIZE - 1;
3304
3305         if (!PageUptodate(page)) {
3306                 ret = btrfs_readpage(NULL, page);
3307                 lock_page(page);
3308                 if (page->mapping != mapping) {
3309                         unlock_page(page);
3310                         page_cache_release(page);
3311                         goto again;
3312                 }
3313                 if (!PageUptodate(page)) {
3314                         ret = -EIO;
3315                         goto out_unlock;
3316                 }
3317         }
3318         wait_on_page_writeback(page);
3319
3320         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3321                          GFP_NOFS);
3322         set_page_extent_mapped(page);
3323
3324         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3325         if (ordered) {
3326                 unlock_extent_cached(io_tree, page_start, page_end,
3327                                      &cached_state, GFP_NOFS);
3328                 unlock_page(page);
3329                 page_cache_release(page);
3330                 btrfs_start_ordered_extent(inode, ordered, 1);
3331                 btrfs_put_ordered_extent(ordered);
3332                 goto again;
3333         }
3334
3335         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3336                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3337                           0, 0, &cached_state, GFP_NOFS);
3338
3339         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3340                                         &cached_state);
3341         if (ret) {
3342                 unlock_extent_cached(io_tree, page_start, page_end,
3343                                      &cached_state, GFP_NOFS);
3344                 goto out_unlock;
3345         }
3346
3347         ret = 0;
3348         if (offset != PAGE_CACHE_SIZE) {
3349                 kaddr = kmap(page);
3350                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3351                 flush_dcache_page(page);
3352                 kunmap(page);
3353         }
3354         ClearPageChecked(page);
3355         set_page_dirty(page);
3356         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3357                              GFP_NOFS);
3358
3359 out_unlock:
3360         if (ret)
3361                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3362         unlock_page(page);
3363         page_cache_release(page);
3364 out:
3365         return ret;
3366 }
3367
3368 /*
3369  * This function puts in dummy file extents for the area we're creating a hole
3370  * for.  So if we are truncating this file to a larger size we need to insert
3371  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3372  * the range between oldsize and size
3373  */
3374 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3375 {
3376         struct btrfs_trans_handle *trans;
3377         struct btrfs_root *root = BTRFS_I(inode)->root;
3378         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3379         struct extent_map *em = NULL;
3380         struct extent_state *cached_state = NULL;
3381         u64 mask = root->sectorsize - 1;
3382         u64 hole_start = (oldsize + mask) & ~mask;
3383         u64 block_end = (size + mask) & ~mask;
3384         u64 last_byte;
3385         u64 cur_offset;
3386         u64 hole_size;
3387         int err = 0;
3388
3389         if (size <= hole_start)
3390                 return 0;
3391
3392         while (1) {
3393                 struct btrfs_ordered_extent *ordered;
3394                 btrfs_wait_ordered_range(inode, hole_start,
3395                                          block_end - hole_start);
3396                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3397                                  &cached_state, GFP_NOFS);
3398                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3399                 if (!ordered)
3400                         break;
3401                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3402                                      &cached_state, GFP_NOFS);
3403                 btrfs_put_ordered_extent(ordered);
3404         }
3405
3406         cur_offset = hole_start;
3407         while (1) {
3408                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3409                                 block_end - cur_offset, 0);
3410                 BUG_ON(IS_ERR_OR_NULL(em));
3411                 last_byte = min(extent_map_end(em), block_end);
3412                 last_byte = (last_byte + mask) & ~mask;
3413                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3414                         u64 hint_byte = 0;
3415                         hole_size = last_byte - cur_offset;
3416
3417                         trans = btrfs_start_transaction(root, 2);
3418                         if (IS_ERR(trans)) {
3419                                 err = PTR_ERR(trans);
3420                                 break;
3421                         }
3422
3423                         err = btrfs_drop_extents(trans, inode, cur_offset,
3424                                                  cur_offset + hole_size,
3425                                                  &hint_byte, 1);
3426                         if (err) {
3427                                 btrfs_end_transaction(trans, root);
3428                                 break;
3429                         }
3430
3431                         err = btrfs_insert_file_extent(trans, root,
3432                                         btrfs_ino(inode), cur_offset, 0,
3433                                         0, hole_size, 0, hole_size,
3434                                         0, 0, 0);
3435                         if (err) {
3436                                 btrfs_end_transaction(trans, root);
3437                                 break;
3438                         }
3439
3440                         btrfs_drop_extent_cache(inode, hole_start,
3441                                         last_byte - 1, 0);
3442
3443                         btrfs_end_transaction(trans, root);
3444                 }
3445                 free_extent_map(em);
3446                 em = NULL;
3447                 cur_offset = last_byte;
3448                 if (cur_offset >= block_end)
3449                         break;
3450         }
3451
3452         free_extent_map(em);
3453         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3454                              GFP_NOFS);
3455         return err;
3456 }
3457
3458 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3459 {
3460         loff_t oldsize = i_size_read(inode);
3461         int ret;
3462
3463         if (newsize == oldsize)
3464                 return 0;
3465
3466         if (newsize > oldsize) {
3467                 i_size_write(inode, newsize);
3468                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3469                 truncate_pagecache(inode, oldsize, newsize);
3470                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3471                 if (ret) {
3472                         btrfs_setsize(inode, oldsize);
3473                         return ret;
3474                 }
3475
3476                 mark_inode_dirty(inode);
3477         } else {
3478
3479                 /*
3480                  * We're truncating a file that used to have good data down to
3481                  * zero. Make sure it gets into the ordered flush list so that
3482                  * any new writes get down to disk quickly.
3483                  */
3484                 if (newsize == 0)
3485                         BTRFS_I(inode)->ordered_data_close = 1;
3486
3487                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3488                 truncate_setsize(inode, newsize);
3489                 ret = btrfs_truncate(inode);
3490         }
3491
3492         return ret;
3493 }
3494
3495 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3496 {
3497         struct inode *inode = dentry->d_inode;
3498         struct btrfs_root *root = BTRFS_I(inode)->root;
3499         int err;
3500
3501         if (btrfs_root_readonly(root))
3502                 return -EROFS;
3503
3504         err = inode_change_ok(inode, attr);
3505         if (err)
3506                 return err;
3507
3508         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3509                 err = btrfs_setsize(inode, attr->ia_size);
3510                 if (err)
3511                         return err;
3512         }
3513
3514         if (attr->ia_valid) {
3515                 setattr_copy(inode, attr);
3516                 mark_inode_dirty(inode);
3517
3518                 if (attr->ia_valid & ATTR_MODE)
3519                         err = btrfs_acl_chmod(inode);
3520         }
3521
3522         return err;
3523 }
3524
3525 void btrfs_evict_inode(struct inode *inode)
3526 {
3527         struct btrfs_trans_handle *trans;
3528         struct btrfs_root *root = BTRFS_I(inode)->root;
3529         struct btrfs_block_rsv *rsv;
3530         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3531         unsigned long nr;
3532         int ret;
3533
3534         trace_btrfs_inode_evict(inode);
3535
3536         truncate_inode_pages(&inode->i_data, 0);
3537         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3538                                btrfs_is_free_space_inode(root, inode)))
3539                 goto no_delete;
3540
3541         if (is_bad_inode(inode)) {
3542                 btrfs_orphan_del(NULL, inode);
3543                 goto no_delete;
3544         }
3545         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3546         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3547
3548         if (root->fs_info->log_root_recovering) {
3549                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3550                 goto no_delete;
3551         }
3552
3553         if (inode->i_nlink > 0) {
3554                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3555                 goto no_delete;
3556         }
3557
3558         rsv = btrfs_alloc_block_rsv(root);
3559         if (!rsv) {
3560                 btrfs_orphan_del(NULL, inode);
3561                 goto no_delete;
3562         }
3563         rsv->size = min_size;
3564
3565         btrfs_i_size_write(inode, 0);
3566
3567         /*
3568          * This is a bit simpler than btrfs_truncate since
3569          *
3570          * 1) We've already reserved our space for our orphan item in the
3571          *    unlink.
3572          * 2) We're going to delete the inode item, so we don't need to update
3573          *    it at all.
3574          *
3575          * So we just need to reserve some slack space in case we add bytes when
3576          * doing the truncate.
3577          */
3578         while (1) {
3579                 ret = btrfs_block_rsv_check(root, rsv, min_size, 0, 1);
3580                 if (ret) {
3581                         printk(KERN_WARNING "Could not get space for a "
3582                                "delete, will truncate on mount %d\n", ret);
3583                         btrfs_orphan_del(NULL, inode);
3584                         btrfs_free_block_rsv(root, rsv);
3585                         goto no_delete;
3586                 }
3587
3588                 trans = btrfs_start_transaction(root, 0);
3589                 if (IS_ERR(trans)) {
3590                         btrfs_orphan_del(NULL, inode);
3591                         btrfs_free_block_rsv(root, rsv);
3592                         goto no_delete;
3593                 }
3594
3595                 trans->block_rsv = rsv;
3596
3597                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3598                 if (ret != -EAGAIN)
3599                         break;
3600
3601                 nr = trans->blocks_used;
3602                 btrfs_end_transaction(trans, root);
3603                 trans = NULL;
3604                 btrfs_btree_balance_dirty(root, nr);
3605         }
3606
3607         btrfs_free_block_rsv(root, rsv);
3608
3609         if (ret == 0) {
3610                 trans->block_rsv = root->orphan_block_rsv;
3611                 ret = btrfs_orphan_del(trans, inode);
3612                 BUG_ON(ret);
3613         }
3614
3615         trans->block_rsv = &root->fs_info->trans_block_rsv;
3616         if (!(root == root->fs_info->tree_root ||
3617               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3618                 btrfs_return_ino(root, btrfs_ino(inode));
3619
3620         nr = trans->blocks_used;
3621         btrfs_end_transaction(trans, root);
3622         btrfs_btree_balance_dirty(root, nr);
3623 no_delete:
3624         end_writeback(inode);
3625         return;
3626 }
3627
3628 /*
3629  * this returns the key found in the dir entry in the location pointer.
3630  * If no dir entries were found, location->objectid is 0.
3631  */
3632 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3633                                struct btrfs_key *location)
3634 {
3635         const char *name = dentry->d_name.name;
3636         int namelen = dentry->d_name.len;
3637         struct btrfs_dir_item *di;
3638         struct btrfs_path *path;
3639         struct btrfs_root *root = BTRFS_I(dir)->root;
3640         int ret = 0;
3641
3642         path = btrfs_alloc_path();
3643         if (!path)
3644                 return -ENOMEM;
3645
3646         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3647                                     namelen, 0);
3648         if (IS_ERR(di))
3649                 ret = PTR_ERR(di);
3650
3651         if (IS_ERR_OR_NULL(di))
3652                 goto out_err;
3653
3654         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3655 out:
3656         btrfs_free_path(path);
3657         return ret;
3658 out_err:
3659         location->objectid = 0;
3660         goto out;
3661 }
3662
3663 /*
3664  * when we hit a tree root in a directory, the btrfs part of the inode
3665  * needs to be changed to reflect the root directory of the tree root.  This
3666  * is kind of like crossing a mount point.
3667  */
3668 static int fixup_tree_root_location(struct btrfs_root *root,
3669                                     struct inode *dir,
3670                                     struct dentry *dentry,
3671                                     struct btrfs_key *location,
3672                                     struct btrfs_root **sub_root)
3673 {
3674         struct btrfs_path *path;
3675         struct btrfs_root *new_root;
3676         struct btrfs_root_ref *ref;
3677         struct extent_buffer *leaf;
3678         int ret;
3679         int err = 0;
3680
3681         path = btrfs_alloc_path();
3682         if (!path) {
3683                 err = -ENOMEM;
3684                 goto out;
3685         }
3686
3687         err = -ENOENT;
3688         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3689                                   BTRFS_I(dir)->root->root_key.objectid,
3690                                   location->objectid);
3691         if (ret) {
3692                 if (ret < 0)
3693                         err = ret;
3694                 goto out;
3695         }
3696
3697         leaf = path->nodes[0];
3698         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3699         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3700             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3701                 goto out;
3702
3703         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3704                                    (unsigned long)(ref + 1),
3705                                    dentry->d_name.len);
3706         if (ret)
3707                 goto out;
3708
3709         btrfs_release_path(path);
3710
3711         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3712         if (IS_ERR(new_root)) {
3713                 err = PTR_ERR(new_root);
3714                 goto out;
3715         }
3716
3717         if (btrfs_root_refs(&new_root->root_item) == 0) {
3718                 err = -ENOENT;
3719                 goto out;
3720         }
3721
3722         *sub_root = new_root;
3723         location->objectid = btrfs_root_dirid(&new_root->root_item);
3724         location->type = BTRFS_INODE_ITEM_KEY;
3725         location->offset = 0;
3726         err = 0;
3727 out:
3728         btrfs_free_path(path);
3729         return err;
3730 }
3731
3732 static void inode_tree_add(struct inode *inode)
3733 {
3734         struct btrfs_root *root = BTRFS_I(inode)->root;
3735         struct btrfs_inode *entry;
3736         struct rb_node **p;
3737         struct rb_node *parent;
3738         u64 ino = btrfs_ino(inode);
3739 again:
3740         p = &root->inode_tree.rb_node;
3741         parent = NULL;
3742
3743         if (inode_unhashed(inode))
3744                 return;
3745
3746         spin_lock(&root->inode_lock);
3747         while (*p) {
3748                 parent = *p;
3749                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3750
3751                 if (ino < btrfs_ino(&entry->vfs_inode))
3752                         p = &parent->rb_left;
3753                 else if (ino > btrfs_ino(&entry->vfs_inode))
3754                         p = &parent->rb_right;
3755                 else {
3756                         WARN_ON(!(entry->vfs_inode.i_state &
3757                                   (I_WILL_FREE | I_FREEING)));
3758                         rb_erase(parent, &root->inode_tree);
3759                         RB_CLEAR_NODE(parent);
3760                         spin_unlock(&root->inode_lock);
3761                         goto again;
3762                 }
3763         }
3764         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3765         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3766         spin_unlock(&root->inode_lock);
3767 }
3768
3769 static void inode_tree_del(struct inode *inode)
3770 {
3771         struct btrfs_root *root = BTRFS_I(inode)->root;
3772         int empty = 0;
3773
3774         spin_lock(&root->inode_lock);
3775         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3776                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3777                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3778                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3779         }
3780         spin_unlock(&root->inode_lock);
3781
3782         /*
3783          * Free space cache has inodes in the tree root, but the tree root has a
3784          * root_refs of 0, so this could end up dropping the tree root as a
3785          * snapshot, so we need the extra !root->fs_info->tree_root check to
3786          * make sure we don't drop it.
3787          */
3788         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3789             root != root->fs_info->tree_root) {
3790                 synchronize_srcu(&root->fs_info->subvol_srcu);
3791                 spin_lock(&root->inode_lock);
3792                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3793                 spin_unlock(&root->inode_lock);
3794                 if (empty)
3795                         btrfs_add_dead_root(root);
3796         }
3797 }
3798
3799 int btrfs_invalidate_inodes(struct btrfs_root *root)
3800 {
3801         struct rb_node *node;
3802         struct rb_node *prev;
3803         struct btrfs_inode *entry;
3804         struct inode *inode;
3805         u64 objectid = 0;
3806
3807         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3808
3809         spin_lock(&root->inode_lock);
3810 again:
3811         node = root->inode_tree.rb_node;
3812         prev = NULL;
3813         while (node) {
3814                 prev = node;
3815                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3816
3817                 if (objectid < btrfs_ino(&entry->vfs_inode))
3818                         node = node->rb_left;
3819                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3820                         node = node->rb_right;
3821                 else
3822                         break;
3823         }
3824         if (!node) {
3825                 while (prev) {
3826                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3827                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3828                                 node = prev;
3829                                 break;
3830                         }
3831                         prev = rb_next(prev);
3832                 }
3833         }
3834         while (node) {
3835                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3836                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3837                 inode = igrab(&entry->vfs_inode);
3838                 if (inode) {
3839                         spin_unlock(&root->inode_lock);
3840                         if (atomic_read(&inode->i_count) > 1)
3841                                 d_prune_aliases(inode);
3842                         /*
3843                          * btrfs_drop_inode will have it removed from
3844                          * the inode cache when its usage count
3845                          * hits zero.
3846                          */
3847                         iput(inode);
3848                         cond_resched();
3849                         spin_lock(&root->inode_lock);
3850                         goto again;
3851                 }
3852
3853                 if (cond_resched_lock(&root->inode_lock))
3854                         goto again;
3855
3856                 node = rb_next(node);
3857         }
3858         spin_unlock(&root->inode_lock);
3859         return 0;
3860 }
3861
3862 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3863 {
3864         struct btrfs_iget_args *args = p;
3865         inode->i_ino = args->ino;
3866         BTRFS_I(inode)->root = args->root;
3867         btrfs_set_inode_space_info(args->root, inode);
3868         return 0;
3869 }
3870
3871 static int btrfs_find_actor(struct inode *inode, void *opaque)
3872 {
3873         struct btrfs_iget_args *args = opaque;
3874         return args->ino == btrfs_ino(inode) &&
3875                 args->root == BTRFS_I(inode)->root;
3876 }
3877
3878 static struct inode *btrfs_iget_locked(struct super_block *s,
3879                                        u64 objectid,
3880                                        struct btrfs_root *root)
3881 {
3882         struct inode *inode;
3883         struct btrfs_iget_args args;
3884         args.ino = objectid;
3885         args.root = root;
3886
3887         inode = iget5_locked(s, objectid, btrfs_find_actor,
3888                              btrfs_init_locked_inode,
3889                              (void *)&args);
3890         return inode;
3891 }
3892
3893 /* Get an inode object given its location and corresponding root.
3894  * Returns in *is_new if the inode was read from disk
3895  */
3896 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3897                          struct btrfs_root *root, int *new)
3898 {
3899         struct inode *inode;
3900
3901         inode = btrfs_iget_locked(s, location->objectid, root);
3902         if (!inode)
3903                 return ERR_PTR(-ENOMEM);
3904
3905         if (inode->i_state & I_NEW) {
3906                 BTRFS_I(inode)->root = root;
3907                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3908                 btrfs_read_locked_inode(inode);
3909                 if (!is_bad_inode(inode)) {
3910                         inode_tree_add(inode);
3911                         unlock_new_inode(inode);
3912                         if (new)
3913                                 *new = 1;
3914                 } else {
3915                         unlock_new_inode(inode);
3916                         iput(inode);
3917                         inode = ERR_PTR(-ESTALE);
3918                 }
3919         }
3920
3921         return inode;
3922 }
3923
3924 static struct inode *new_simple_dir(struct super_block *s,
3925                                     struct btrfs_key *key,
3926                                     struct btrfs_root *root)
3927 {
3928         struct inode *inode = new_inode(s);
3929
3930         if (!inode)
3931                 return ERR_PTR(-ENOMEM);
3932
3933         BTRFS_I(inode)->root = root;
3934         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3935         BTRFS_I(inode)->dummy_inode = 1;
3936
3937         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3938         inode->i_op = &simple_dir_inode_operations;
3939         inode->i_fop = &simple_dir_operations;
3940         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3941         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3942
3943         return inode;
3944 }
3945
3946 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3947 {
3948         struct inode *inode;
3949         struct btrfs_root *root = BTRFS_I(dir)->root;
3950         struct btrfs_root *sub_root = root;
3951         struct btrfs_key location;
3952         int index;
3953         int ret = 0;
3954
3955         if (dentry->d_name.len > BTRFS_NAME_LEN)
3956                 return ERR_PTR(-ENAMETOOLONG);
3957
3958         if (unlikely(d_need_lookup(dentry))) {
3959                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3960                 kfree(dentry->d_fsdata);
3961                 dentry->d_fsdata = NULL;
3962                 /* This thing is hashed, drop it for now */
3963                 d_drop(dentry);
3964         } else {
3965                 ret = btrfs_inode_by_name(dir, dentry, &location);
3966         }
3967
3968         if (ret < 0)
3969                 return ERR_PTR(ret);
3970
3971         if (location.objectid == 0)
3972                 return NULL;
3973
3974         if (location.type == BTRFS_INODE_ITEM_KEY) {
3975                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3976                 return inode;
3977         }
3978
3979         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3980
3981         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3982         ret = fixup_tree_root_location(root, dir, dentry,
3983                                        &location, &sub_root);
3984         if (ret < 0) {
3985                 if (ret != -ENOENT)
3986                         inode = ERR_PTR(ret);
3987                 else
3988                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3989         } else {
3990                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3991         }
3992         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3993
3994         if (!IS_ERR(inode) && root != sub_root) {
3995                 down_read(&root->fs_info->cleanup_work_sem);
3996                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3997                         ret = btrfs_orphan_cleanup(sub_root);
3998                 up_read(&root->fs_info->cleanup_work_sem);
3999                 if (ret)
4000                         inode = ERR_PTR(ret);
4001         }
4002
4003         return inode;
4004 }
4005
4006 static int btrfs_dentry_delete(const struct dentry *dentry)
4007 {
4008         struct btrfs_root *root;
4009
4010         if (!dentry->d_inode && !IS_ROOT(dentry))
4011                 dentry = dentry->d_parent;
4012
4013         if (dentry->d_inode) {
4014                 root = BTRFS_I(dentry->d_inode)->root;
4015                 if (btrfs_root_refs(&root->root_item) == 0)
4016                         return 1;
4017         }
4018         return 0;
4019 }
4020
4021 static void btrfs_dentry_release(struct dentry *dentry)
4022 {
4023         if (dentry->d_fsdata)
4024                 kfree(dentry->d_fsdata);
4025 }
4026
4027 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4028                                    struct nameidata *nd)
4029 {
4030         struct dentry *ret;
4031
4032         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4033         if (unlikely(d_need_lookup(dentry))) {
4034                 spin_lock(&dentry->d_lock);
4035                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4036                 spin_unlock(&dentry->d_lock);
4037         }
4038         return ret;
4039 }
4040
4041 unsigned char btrfs_filetype_table[] = {
4042         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4043 };
4044
4045 static int btrfs_real_readdir(struct file *filp, void *dirent,
4046                               filldir_t filldir)
4047 {
4048         struct inode *inode = filp->f_dentry->d_inode;
4049         struct btrfs_root *root = BTRFS_I(inode)->root;
4050         struct btrfs_item *item;
4051         struct btrfs_dir_item *di;
4052         struct btrfs_key key;
4053         struct btrfs_key found_key;
4054         struct btrfs_path *path;
4055         struct list_head ins_list;
4056         struct list_head del_list;
4057         struct qstr q;
4058         int ret;
4059         struct extent_buffer *leaf;
4060         int slot;
4061         unsigned char d_type;
4062         int over = 0;
4063         u32 di_cur;
4064         u32 di_total;
4065         u32 di_len;
4066         int key_type = BTRFS_DIR_INDEX_KEY;
4067         char tmp_name[32];
4068         char *name_ptr;
4069         int name_len;
4070         int is_curr = 0;        /* filp->f_pos points to the current index? */
4071
4072         /* FIXME, use a real flag for deciding about the key type */
4073         if (root->fs_info->tree_root == root)
4074                 key_type = BTRFS_DIR_ITEM_KEY;
4075
4076         /* special case for "." */
4077         if (filp->f_pos == 0) {
4078                 over = filldir(dirent, ".", 1,
4079                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4080                 if (over)
4081                         return 0;
4082                 filp->f_pos = 1;
4083         }
4084         /* special case for .., just use the back ref */
4085         if (filp->f_pos == 1) {
4086                 u64 pino = parent_ino(filp->f_path.dentry);
4087                 over = filldir(dirent, "..", 2,
4088                                filp->f_pos, pino, DT_DIR);
4089                 if (over)
4090                         return 0;
4091                 filp->f_pos = 2;
4092         }
4093         path = btrfs_alloc_path();
4094         if (!path)
4095                 return -ENOMEM;
4096
4097         path->reada = 1;
4098
4099         if (key_type == BTRFS_DIR_INDEX_KEY) {
4100                 INIT_LIST_HEAD(&ins_list);
4101                 INIT_LIST_HEAD(&del_list);
4102                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4103         }
4104
4105         btrfs_set_key_type(&key, key_type);
4106         key.offset = filp->f_pos;
4107         key.objectid = btrfs_ino(inode);
4108
4109         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4110         if (ret < 0)
4111                 goto err;
4112
4113         while (1) {
4114                 leaf = path->nodes[0];
4115                 slot = path->slots[0];
4116                 if (slot >= btrfs_header_nritems(leaf)) {
4117                         ret = btrfs_next_leaf(root, path);
4118                         if (ret < 0)
4119                                 goto err;
4120                         else if (ret > 0)
4121                                 break;
4122                         continue;
4123                 }
4124
4125                 item = btrfs_item_nr(leaf, slot);
4126                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4127
4128                 if (found_key.objectid != key.objectid)
4129                         break;
4130                 if (btrfs_key_type(&found_key) != key_type)
4131                         break;
4132                 if (found_key.offset < filp->f_pos)
4133                         goto next;
4134                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4135                     btrfs_should_delete_dir_index(&del_list,
4136                                                   found_key.offset))
4137                         goto next;
4138
4139                 filp->f_pos = found_key.offset;
4140                 is_curr = 1;
4141
4142                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4143                 di_cur = 0;
4144                 di_total = btrfs_item_size(leaf, item);
4145
4146                 while (di_cur < di_total) {
4147                         struct btrfs_key location;
4148                         struct dentry *tmp;
4149
4150                         if (verify_dir_item(root, leaf, di))
4151                                 break;
4152
4153                         name_len = btrfs_dir_name_len(leaf, di);
4154                         if (name_len <= sizeof(tmp_name)) {
4155                                 name_ptr = tmp_name;
4156                         } else {
4157                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4158                                 if (!name_ptr) {
4159                                         ret = -ENOMEM;
4160                                         goto err;
4161                                 }
4162                         }
4163                         read_extent_buffer(leaf, name_ptr,
4164                                            (unsigned long)(di + 1), name_len);
4165
4166                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4167                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4168
4169                         q.name = name_ptr;
4170                         q.len = name_len;
4171                         q.hash = full_name_hash(q.name, q.len);
4172                         tmp = d_lookup(filp->f_dentry, &q);
4173                         if (!tmp) {
4174                                 struct btrfs_key *newkey;
4175
4176                                 newkey = kzalloc(sizeof(struct btrfs_key),
4177                                                  GFP_NOFS);
4178                                 if (!newkey)
4179                                         goto no_dentry;
4180                                 tmp = d_alloc(filp->f_dentry, &q);
4181                                 if (!tmp) {
4182                                         kfree(newkey);
4183                                         dput(tmp);
4184                                         goto no_dentry;
4185                                 }
4186                                 memcpy(newkey, &location,
4187                                        sizeof(struct btrfs_key));
4188                                 tmp->d_fsdata = newkey;
4189                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4190                                 d_rehash(tmp);
4191                                 dput(tmp);
4192                         } else {
4193                                 dput(tmp);
4194                         }
4195 no_dentry:
4196                         /* is this a reference to our own snapshot? If so
4197                          * skip it
4198                          */
4199                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4200                             location.objectid == root->root_key.objectid) {
4201                                 over = 0;
4202                                 goto skip;
4203                         }
4204                         over = filldir(dirent, name_ptr, name_len,
4205                                        found_key.offset, location.objectid,
4206                                        d_type);
4207
4208 skip:
4209                         if (name_ptr != tmp_name)
4210                                 kfree(name_ptr);
4211
4212                         if (over)
4213                                 goto nopos;
4214                         di_len = btrfs_dir_name_len(leaf, di) +
4215                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4216                         di_cur += di_len;
4217                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4218                 }
4219 next:
4220                 path->slots[0]++;
4221         }
4222
4223         if (key_type == BTRFS_DIR_INDEX_KEY) {
4224                 if (is_curr)
4225                         filp->f_pos++;
4226                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4227                                                       &ins_list);
4228                 if (ret)
4229                         goto nopos;
4230         }
4231
4232         /* Reached end of directory/root. Bump pos past the last item. */
4233         if (key_type == BTRFS_DIR_INDEX_KEY)
4234                 /*
4235                  * 32-bit glibc will use getdents64, but then strtol -
4236                  * so the last number we can serve is this.
4237                  */
4238                 filp->f_pos = 0x7fffffff;
4239         else
4240                 filp->f_pos++;
4241 nopos:
4242         ret = 0;
4243 err:
4244         if (key_type == BTRFS_DIR_INDEX_KEY)
4245                 btrfs_put_delayed_items(&ins_list, &del_list);
4246         btrfs_free_path(path);
4247         return ret;
4248 }
4249
4250 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4251 {
4252         struct btrfs_root *root = BTRFS_I(inode)->root;
4253         struct btrfs_trans_handle *trans;
4254         int ret = 0;
4255         bool nolock = false;
4256
4257         if (BTRFS_I(inode)->dummy_inode)
4258                 return 0;
4259
4260         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4261                 nolock = true;
4262
4263         if (wbc->sync_mode == WB_SYNC_ALL) {
4264                 if (nolock)
4265                         trans = btrfs_join_transaction_nolock(root);
4266                 else
4267                         trans = btrfs_join_transaction(root);
4268                 if (IS_ERR(trans))
4269                         return PTR_ERR(trans);
4270                 if (nolock)
4271                         ret = btrfs_end_transaction_nolock(trans, root);
4272                 else
4273                         ret = btrfs_commit_transaction(trans, root);
4274         }
4275         return ret;
4276 }
4277
4278 /*
4279  * This is somewhat expensive, updating the tree every time the
4280  * inode changes.  But, it is most likely to find the inode in cache.
4281  * FIXME, needs more benchmarking...there are no reasons other than performance
4282  * to keep or drop this code.
4283  */
4284 void btrfs_dirty_inode(struct inode *inode, int flags)
4285 {
4286         struct btrfs_root *root = BTRFS_I(inode)->root;
4287         struct btrfs_trans_handle *trans;
4288         int ret;
4289
4290         if (BTRFS_I(inode)->dummy_inode)
4291                 return;
4292
4293         trans = btrfs_join_transaction(root);
4294         BUG_ON(IS_ERR(trans));
4295
4296         ret = btrfs_update_inode(trans, root, inode);
4297         if (ret && ret == -ENOSPC) {
4298                 /* whoops, lets try again with the full transaction */
4299                 btrfs_end_transaction(trans, root);
4300                 trans = btrfs_start_transaction(root, 1);
4301                 if (IS_ERR(trans)) {
4302                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4303                                        "dirty  inode %llu error %ld\n",
4304                                        (unsigned long long)btrfs_ino(inode),
4305                                        PTR_ERR(trans));
4306                         return;
4307                 }
4308
4309                 ret = btrfs_update_inode(trans, root, inode);
4310                 if (ret) {
4311                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4312                                        "dirty  inode %llu error %d\n",
4313                                        (unsigned long long)btrfs_ino(inode),
4314                                        ret);
4315                 }
4316         }
4317         btrfs_end_transaction(trans, root);
4318         if (BTRFS_I(inode)->delayed_node)
4319                 btrfs_balance_delayed_items(root);
4320 }
4321
4322 /*
4323  * find the highest existing sequence number in a directory
4324  * and then set the in-memory index_cnt variable to reflect
4325  * free sequence numbers
4326  */
4327 static int btrfs_set_inode_index_count(struct inode *inode)
4328 {
4329         struct btrfs_root *root = BTRFS_I(inode)->root;
4330         struct btrfs_key key, found_key;
4331         struct btrfs_path *path;
4332         struct extent_buffer *leaf;
4333         int ret;
4334
4335         key.objectid = btrfs_ino(inode);
4336         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4337         key.offset = (u64)-1;
4338
4339         path = btrfs_alloc_path();
4340         if (!path)
4341                 return -ENOMEM;
4342
4343         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4344         if (ret < 0)
4345                 goto out;
4346         /* FIXME: we should be able to handle this */
4347         if (ret == 0)
4348                 goto out;
4349         ret = 0;
4350
4351         /*
4352          * MAGIC NUMBER EXPLANATION:
4353          * since we search a directory based on f_pos we have to start at 2
4354          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4355          * else has to start at 2
4356          */
4357         if (path->slots[0] == 0) {
4358                 BTRFS_I(inode)->index_cnt = 2;
4359                 goto out;
4360         }
4361
4362         path->slots[0]--;
4363
4364         leaf = path->nodes[0];
4365         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4366
4367         if (found_key.objectid != btrfs_ino(inode) ||
4368             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4369                 BTRFS_I(inode)->index_cnt = 2;
4370                 goto out;
4371         }
4372
4373         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4374 out:
4375         btrfs_free_path(path);
4376         return ret;
4377 }
4378
4379 /*
4380  * helper to find a free sequence number in a given directory.  This current
4381  * code is very simple, later versions will do smarter things in the btree
4382  */
4383 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4384 {
4385         int ret = 0;
4386
4387         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4388                 ret = btrfs_inode_delayed_dir_index_count(dir);
4389                 if (ret) {
4390                         ret = btrfs_set_inode_index_count(dir);
4391                         if (ret)
4392                                 return ret;
4393                 }
4394         }
4395
4396         *index = BTRFS_I(dir)->index_cnt;
4397         BTRFS_I(dir)->index_cnt++;
4398
4399         return ret;
4400 }
4401
4402 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4403                                      struct btrfs_root *root,
4404                                      struct inode *dir,
4405                                      const char *name, int name_len,
4406                                      u64 ref_objectid, u64 objectid, int mode,
4407                                      u64 *index)
4408 {
4409         struct inode *inode;
4410         struct btrfs_inode_item *inode_item;
4411         struct btrfs_key *location;
4412         struct btrfs_path *path;
4413         struct btrfs_inode_ref *ref;
4414         struct btrfs_key key[2];
4415         u32 sizes[2];
4416         unsigned long ptr;
4417         int ret;
4418         int owner;
4419
4420         path = btrfs_alloc_path();
4421         if (!path)
4422                 return ERR_PTR(-ENOMEM);
4423
4424         inode = new_inode(root->fs_info->sb);
4425         if (!inode) {
4426                 btrfs_free_path(path);
4427                 return ERR_PTR(-ENOMEM);
4428         }
4429
4430         /*
4431          * we have to initialize this early, so we can reclaim the inode
4432          * number if we fail afterwards in this function.
4433          */
4434         inode->i_ino = objectid;
4435
4436         if (dir) {
4437                 trace_btrfs_inode_request(dir);
4438
4439                 ret = btrfs_set_inode_index(dir, index);
4440                 if (ret) {
4441                         btrfs_free_path(path);
4442                         iput(inode);
4443                         return ERR_PTR(ret);
4444                 }
4445         }
4446         /*
4447          * index_cnt is ignored for everything but a dir,
4448          * btrfs_get_inode_index_count has an explanation for the magic
4449          * number
4450          */
4451         BTRFS_I(inode)->index_cnt = 2;
4452         BTRFS_I(inode)->root = root;
4453         BTRFS_I(inode)->generation = trans->transid;
4454         inode->i_generation = BTRFS_I(inode)->generation;
4455         btrfs_set_inode_space_info(root, inode);
4456
4457         if (S_ISDIR(mode))
4458                 owner = 0;
4459         else
4460                 owner = 1;
4461
4462         key[0].objectid = objectid;
4463         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4464         key[0].offset = 0;
4465
4466         key[1].objectid = objectid;
4467         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4468         key[1].offset = ref_objectid;
4469
4470         sizes[0] = sizeof(struct btrfs_inode_item);
4471         sizes[1] = name_len + sizeof(*ref);
4472
4473         path->leave_spinning = 1;
4474         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4475         if (ret != 0)
4476                 goto fail;
4477
4478         inode_init_owner(inode, dir, mode);
4479         inode_set_bytes(inode, 0);
4480         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4481         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4482                                   struct btrfs_inode_item);
4483         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4484
4485         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4486                              struct btrfs_inode_ref);
4487         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4488         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4489         ptr = (unsigned long)(ref + 1);
4490         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4491
4492         btrfs_mark_buffer_dirty(path->nodes[0]);
4493         btrfs_free_path(path);
4494
4495         location = &BTRFS_I(inode)->location;
4496         location->objectid = objectid;
4497         location->offset = 0;
4498         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4499
4500         btrfs_inherit_iflags(inode, dir);
4501
4502         if (S_ISREG(mode)) {
4503                 if (btrfs_test_opt(root, NODATASUM))
4504                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4505                 if (btrfs_test_opt(root, NODATACOW) ||
4506                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4507                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4508         }
4509
4510         insert_inode_hash(inode);
4511         inode_tree_add(inode);
4512
4513         trace_btrfs_inode_new(inode);
4514         btrfs_set_inode_last_trans(trans, inode);
4515
4516         return inode;
4517 fail:
4518         if (dir)
4519                 BTRFS_I(dir)->index_cnt--;
4520         btrfs_free_path(path);
4521         iput(inode);
4522         return ERR_PTR(ret);
4523 }
4524
4525 static inline u8 btrfs_inode_type(struct inode *inode)
4526 {
4527         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4528 }
4529
4530 /*
4531  * utility function to add 'inode' into 'parent_inode' with
4532  * a give name and a given sequence number.
4533  * if 'add_backref' is true, also insert a backref from the
4534  * inode to the parent directory.
4535  */
4536 int btrfs_add_link(struct btrfs_trans_handle *trans,
4537                    struct inode *parent_inode, struct inode *inode,
4538                    const char *name, int name_len, int add_backref, u64 index)
4539 {
4540         int ret = 0;
4541         struct btrfs_key key;
4542         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4543         u64 ino = btrfs_ino(inode);
4544         u64 parent_ino = btrfs_ino(parent_inode);
4545
4546         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4547                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4548         } else {
4549                 key.objectid = ino;
4550                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4551                 key.offset = 0;
4552         }
4553
4554         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4555                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4556                                          key.objectid, root->root_key.objectid,
4557                                          parent_ino, index, name, name_len);
4558         } else if (add_backref) {
4559                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4560                                              parent_ino, index);
4561         }
4562
4563         if (ret == 0) {
4564                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4565                                             parent_inode, &key,
4566                                             btrfs_inode_type(inode), index);
4567                 BUG_ON(ret);
4568
4569                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4570                                    name_len * 2);
4571                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4572                 ret = btrfs_update_inode(trans, root, parent_inode);
4573         }
4574         return ret;
4575 }
4576
4577 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4578                             struct inode *dir, struct dentry *dentry,
4579                             struct inode *inode, int backref, u64 index)
4580 {
4581         int err = btrfs_add_link(trans, dir, inode,
4582                                  dentry->d_name.name, dentry->d_name.len,
4583                                  backref, index);
4584         if (!err) {
4585                 d_instantiate(dentry, inode);
4586                 return 0;
4587         }
4588         if (err > 0)
4589                 err = -EEXIST;
4590         return err;
4591 }
4592
4593 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4594                         int mode, dev_t rdev)
4595 {
4596         struct btrfs_trans_handle *trans;
4597         struct btrfs_root *root = BTRFS_I(dir)->root;
4598         struct inode *inode = NULL;
4599         int err;
4600         int drop_inode = 0;
4601         u64 objectid;
4602         unsigned long nr = 0;
4603         u64 index = 0;
4604
4605         if (!new_valid_dev(rdev))
4606                 return -EINVAL;
4607
4608         /*
4609          * 2 for inode item and ref
4610          * 2 for dir items
4611          * 1 for xattr if selinux is on
4612          */
4613         trans = btrfs_start_transaction(root, 5);
4614         if (IS_ERR(trans))
4615                 return PTR_ERR(trans);
4616
4617         err = btrfs_find_free_ino(root, &objectid);
4618         if (err)
4619                 goto out_unlock;
4620
4621         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4622                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4623                                 mode, &index);
4624         if (IS_ERR(inode)) {
4625                 err = PTR_ERR(inode);
4626                 goto out_unlock;
4627         }
4628
4629         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4630         if (err) {
4631                 drop_inode = 1;
4632                 goto out_unlock;
4633         }
4634
4635         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4636         if (err)
4637                 drop_inode = 1;
4638         else {
4639                 inode->i_op = &btrfs_special_inode_operations;
4640                 init_special_inode(inode, inode->i_mode, rdev);
4641                 btrfs_update_inode(trans, root, inode);
4642         }
4643 out_unlock:
4644         nr = trans->blocks_used;
4645         btrfs_end_transaction_throttle(trans, root);
4646         btrfs_btree_balance_dirty(root, nr);
4647         if (drop_inode) {
4648                 inode_dec_link_count(inode);
4649                 iput(inode);
4650         }
4651         return err;
4652 }
4653
4654 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4655                         int mode, struct nameidata *nd)
4656 {
4657         struct btrfs_trans_handle *trans;
4658         struct btrfs_root *root = BTRFS_I(dir)->root;
4659         struct inode *inode = NULL;
4660         int drop_inode = 0;
4661         int err;
4662         unsigned long nr = 0;
4663         u64 objectid;
4664         u64 index = 0;
4665
4666         /*
4667          * 2 for inode item and ref
4668          * 2 for dir items
4669          * 1 for xattr if selinux is on
4670          */
4671         trans = btrfs_start_transaction(root, 5);
4672         if (IS_ERR(trans))
4673                 return PTR_ERR(trans);
4674
4675         err = btrfs_find_free_ino(root, &objectid);
4676         if (err)
4677                 goto out_unlock;
4678
4679         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4680                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4681                                 mode, &index);
4682         if (IS_ERR(inode)) {
4683                 err = PTR_ERR(inode);
4684                 goto out_unlock;
4685         }
4686
4687         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4688         if (err) {
4689                 drop_inode = 1;
4690                 goto out_unlock;
4691         }
4692
4693         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4694         if (err)
4695                 drop_inode = 1;
4696         else {
4697                 inode->i_mapping->a_ops = &btrfs_aops;
4698                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4699                 inode->i_fop = &btrfs_file_operations;
4700                 inode->i_op = &btrfs_file_inode_operations;
4701                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4702         }
4703 out_unlock:
4704         nr = trans->blocks_used;
4705         btrfs_end_transaction_throttle(trans, root);
4706         if (drop_inode) {
4707                 inode_dec_link_count(inode);
4708                 iput(inode);
4709         }
4710         btrfs_btree_balance_dirty(root, nr);
4711         return err;
4712 }
4713
4714 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4715                       struct dentry *dentry)
4716 {
4717         struct btrfs_trans_handle *trans;
4718         struct btrfs_root *root = BTRFS_I(dir)->root;
4719         struct inode *inode = old_dentry->d_inode;
4720         u64 index;
4721         unsigned long nr = 0;
4722         int err;
4723         int drop_inode = 0;
4724
4725         /* do not allow sys_link's with other subvols of the same device */
4726         if (root->objectid != BTRFS_I(inode)->root->objectid)
4727                 return -EXDEV;
4728
4729         if (inode->i_nlink == ~0U)
4730                 return -EMLINK;
4731
4732         err = btrfs_set_inode_index(dir, &index);
4733         if (err)
4734                 goto fail;
4735
4736         /*
4737          * 2 items for inode and inode ref
4738          * 2 items for dir items
4739          * 1 item for parent inode
4740          */
4741         trans = btrfs_start_transaction(root, 5);
4742         if (IS_ERR(trans)) {
4743                 err = PTR_ERR(trans);
4744                 goto fail;
4745         }
4746
4747         btrfs_inc_nlink(inode);
4748         inode->i_ctime = CURRENT_TIME;
4749         ihold(inode);
4750
4751         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4752
4753         if (err) {
4754                 drop_inode = 1;
4755         } else {
4756                 struct dentry *parent = dentry->d_parent;
4757                 err = btrfs_update_inode(trans, root, inode);
4758                 BUG_ON(err);
4759                 btrfs_log_new_name(trans, inode, NULL, parent);
4760         }
4761
4762         nr = trans->blocks_used;
4763         btrfs_end_transaction_throttle(trans, root);
4764 fail:
4765         if (drop_inode) {
4766                 inode_dec_link_count(inode);
4767                 iput(inode);
4768         }
4769         btrfs_btree_balance_dirty(root, nr);
4770         return err;
4771 }
4772
4773 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4774 {
4775         struct inode *inode = NULL;
4776         struct btrfs_trans_handle *trans;
4777         struct btrfs_root *root = BTRFS_I(dir)->root;
4778         int err = 0;
4779         int drop_on_err = 0;
4780         u64 objectid = 0;
4781         u64 index = 0;
4782         unsigned long nr = 1;
4783
4784         /*
4785          * 2 items for inode and ref
4786          * 2 items for dir items
4787          * 1 for xattr if selinux is on
4788          */
4789         trans = btrfs_start_transaction(root, 5);
4790         if (IS_ERR(trans))
4791                 return PTR_ERR(trans);
4792
4793         err = btrfs_find_free_ino(root, &objectid);
4794         if (err)
4795                 goto out_fail;
4796
4797         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4798                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4799                                 S_IFDIR | mode, &index);
4800         if (IS_ERR(inode)) {
4801                 err = PTR_ERR(inode);
4802                 goto out_fail;
4803         }
4804
4805         drop_on_err = 1;
4806
4807         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4808         if (err)
4809                 goto out_fail;
4810
4811         inode->i_op = &btrfs_dir_inode_operations;
4812         inode->i_fop = &btrfs_dir_file_operations;
4813
4814         btrfs_i_size_write(inode, 0);
4815         err = btrfs_update_inode(trans, root, inode);
4816         if (err)
4817                 goto out_fail;
4818
4819         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4820                              dentry->d_name.len, 0, index);
4821         if (err)
4822                 goto out_fail;
4823
4824         d_instantiate(dentry, inode);
4825         drop_on_err = 0;
4826
4827 out_fail:
4828         nr = trans->blocks_used;
4829         btrfs_end_transaction_throttle(trans, root);
4830         if (drop_on_err)
4831                 iput(inode);
4832         btrfs_btree_balance_dirty(root, nr);
4833         return err;
4834 }
4835
4836 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4837  * and an extent that you want to insert, deal with overlap and insert
4838  * the new extent into the tree.
4839  */
4840 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4841                                 struct extent_map *existing,
4842                                 struct extent_map *em,
4843                                 u64 map_start, u64 map_len)
4844 {
4845         u64 start_diff;
4846
4847         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4848         start_diff = map_start - em->start;
4849         em->start = map_start;
4850         em->len = map_len;
4851         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4852             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4853                 em->block_start += start_diff;
4854                 em->block_len -= start_diff;
4855         }
4856         return add_extent_mapping(em_tree, em);
4857 }
4858
4859 static noinline int uncompress_inline(struct btrfs_path *path,
4860                                       struct inode *inode, struct page *page,
4861                                       size_t pg_offset, u64 extent_offset,
4862                                       struct btrfs_file_extent_item *item)
4863 {
4864         int ret;
4865         struct extent_buffer *leaf = path->nodes[0];
4866         char *tmp;
4867         size_t max_size;
4868         unsigned long inline_size;
4869         unsigned long ptr;
4870         int compress_type;
4871
4872         WARN_ON(pg_offset != 0);
4873         compress_type = btrfs_file_extent_compression(leaf, item);
4874         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4875         inline_size = btrfs_file_extent_inline_item_len(leaf,
4876                                         btrfs_item_nr(leaf, path->slots[0]));
4877         tmp = kmalloc(inline_size, GFP_NOFS);
4878         if (!tmp)
4879                 return -ENOMEM;
4880         ptr = btrfs_file_extent_inline_start(item);
4881
4882         read_extent_buffer(leaf, tmp, ptr, inline_size);
4883
4884         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4885         ret = btrfs_decompress(compress_type, tmp, page,
4886                                extent_offset, inline_size, max_size);
4887         if (ret) {
4888                 char *kaddr = kmap_atomic(page, KM_USER0);
4889                 unsigned long copy_size = min_t(u64,
4890                                   PAGE_CACHE_SIZE - pg_offset,
4891                                   max_size - extent_offset);
4892                 memset(kaddr + pg_offset, 0, copy_size);
4893                 kunmap_atomic(kaddr, KM_USER0);
4894         }
4895         kfree(tmp);
4896         return 0;
4897 }
4898
4899 /*
4900  * a bit scary, this does extent mapping from logical file offset to the disk.
4901  * the ugly parts come from merging extents from the disk with the in-ram
4902  * representation.  This gets more complex because of the data=ordered code,
4903  * where the in-ram extents might be locked pending data=ordered completion.
4904  *
4905  * This also copies inline extents directly into the page.
4906  */
4907
4908 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4909                                     size_t pg_offset, u64 start, u64 len,
4910                                     int create)
4911 {
4912         int ret;
4913         int err = 0;
4914         u64 bytenr;
4915         u64 extent_start = 0;
4916         u64 extent_end = 0;
4917         u64 objectid = btrfs_ino(inode);
4918         u32 found_type;
4919         struct btrfs_path *path = NULL;
4920         struct btrfs_root *root = BTRFS_I(inode)->root;
4921         struct btrfs_file_extent_item *item;
4922         struct extent_buffer *leaf;
4923         struct btrfs_key found_key;
4924         struct extent_map *em = NULL;
4925         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4926         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4927         struct btrfs_trans_handle *trans = NULL;
4928         int compress_type;
4929
4930 again:
4931         read_lock(&em_tree->lock);
4932         em = lookup_extent_mapping(em_tree, start, len);
4933         if (em)
4934                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4935         read_unlock(&em_tree->lock);
4936
4937         if (em) {
4938                 if (em->start > start || em->start + em->len <= start)
4939                         free_extent_map(em);
4940                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4941                         free_extent_map(em);
4942                 else
4943                         goto out;
4944         }
4945         em = alloc_extent_map();
4946         if (!em) {
4947                 err = -ENOMEM;
4948                 goto out;
4949         }
4950         em->bdev = root->fs_info->fs_devices->latest_bdev;
4951         em->start = EXTENT_MAP_HOLE;
4952         em->orig_start = EXTENT_MAP_HOLE;
4953         em->len = (u64)-1;
4954         em->block_len = (u64)-1;
4955
4956         if (!path) {
4957                 path = btrfs_alloc_path();
4958                 if (!path) {
4959                         err = -ENOMEM;
4960                         goto out;
4961                 }
4962                 /*
4963                  * Chances are we'll be called again, so go ahead and do
4964                  * readahead
4965                  */
4966                 path->reada = 1;
4967         }
4968
4969         ret = btrfs_lookup_file_extent(trans, root, path,
4970                                        objectid, start, trans != NULL);
4971         if (ret < 0) {
4972                 err = ret;
4973                 goto out;
4974         }
4975
4976         if (ret != 0) {
4977                 if (path->slots[0] == 0)
4978                         goto not_found;
4979                 path->slots[0]--;
4980         }
4981
4982         leaf = path->nodes[0];
4983         item = btrfs_item_ptr(leaf, path->slots[0],
4984                               struct btrfs_file_extent_item);
4985         /* are we inside the extent that was found? */
4986         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4987         found_type = btrfs_key_type(&found_key);
4988         if (found_key.objectid != objectid ||
4989             found_type != BTRFS_EXTENT_DATA_KEY) {
4990                 goto not_found;
4991         }
4992
4993         found_type = btrfs_file_extent_type(leaf, item);
4994         extent_start = found_key.offset;
4995         compress_type = btrfs_file_extent_compression(leaf, item);
4996         if (found_type == BTRFS_FILE_EXTENT_REG ||
4997             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4998                 extent_end = extent_start +
4999                        btrfs_file_extent_num_bytes(leaf, item);
5000         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5001                 size_t size;
5002                 size = btrfs_file_extent_inline_len(leaf, item);
5003                 extent_end = (extent_start + size + root->sectorsize - 1) &
5004                         ~((u64)root->sectorsize - 1);
5005         }
5006
5007         if (start >= extent_end) {
5008                 path->slots[0]++;
5009                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5010                         ret = btrfs_next_leaf(root, path);
5011                         if (ret < 0) {
5012                                 err = ret;
5013                                 goto out;
5014                         }
5015                         if (ret > 0)
5016                                 goto not_found;
5017                         leaf = path->nodes[0];
5018                 }
5019                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5020                 if (found_key.objectid != objectid ||
5021                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5022                         goto not_found;
5023                 if (start + len <= found_key.offset)
5024                         goto not_found;
5025                 em->start = start;
5026                 em->len = found_key.offset - start;
5027                 goto not_found_em;
5028         }
5029
5030         if (found_type == BTRFS_FILE_EXTENT_REG ||
5031             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5032                 em->start = extent_start;
5033                 em->len = extent_end - extent_start;
5034                 em->orig_start = extent_start -
5035                                  btrfs_file_extent_offset(leaf, item);
5036                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5037                 if (bytenr == 0) {
5038                         em->block_start = EXTENT_MAP_HOLE;
5039                         goto insert;
5040                 }
5041                 if (compress_type != BTRFS_COMPRESS_NONE) {
5042                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5043                         em->compress_type = compress_type;
5044                         em->block_start = bytenr;
5045                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5046                                                                          item);
5047                 } else {
5048                         bytenr += btrfs_file_extent_offset(leaf, item);
5049                         em->block_start = bytenr;
5050                         em->block_len = em->len;
5051                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5052                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5053                 }
5054                 goto insert;
5055         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5056                 unsigned long ptr;
5057                 char *map;
5058                 size_t size;
5059                 size_t extent_offset;
5060                 size_t copy_size;
5061
5062                 em->block_start = EXTENT_MAP_INLINE;
5063                 if (!page || create) {
5064                         em->start = extent_start;
5065                         em->len = extent_end - extent_start;
5066                         goto out;
5067                 }
5068
5069                 size = btrfs_file_extent_inline_len(leaf, item);
5070                 extent_offset = page_offset(page) + pg_offset - extent_start;
5071                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5072                                 size - extent_offset);
5073                 em->start = extent_start + extent_offset;
5074                 em->len = (copy_size + root->sectorsize - 1) &
5075                         ~((u64)root->sectorsize - 1);
5076                 em->orig_start = EXTENT_MAP_INLINE;
5077                 if (compress_type) {
5078                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5079                         em->compress_type = compress_type;
5080                 }
5081                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5082                 if (create == 0 && !PageUptodate(page)) {
5083                         if (btrfs_file_extent_compression(leaf, item) !=
5084                             BTRFS_COMPRESS_NONE) {
5085                                 ret = uncompress_inline(path, inode, page,
5086                                                         pg_offset,
5087                                                         extent_offset, item);
5088                                 BUG_ON(ret);
5089                         } else {
5090                                 map = kmap(page);
5091                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5092                                                    copy_size);
5093                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5094                                         memset(map + pg_offset + copy_size, 0,
5095                                                PAGE_CACHE_SIZE - pg_offset -
5096                                                copy_size);
5097                                 }
5098                                 kunmap(page);
5099                         }
5100                         flush_dcache_page(page);
5101                 } else if (create && PageUptodate(page)) {
5102                         WARN_ON(1);
5103                         if (!trans) {
5104                                 kunmap(page);
5105                                 free_extent_map(em);
5106                                 em = NULL;
5107
5108                                 btrfs_release_path(path);
5109                                 trans = btrfs_join_transaction(root);
5110
5111                                 if (IS_ERR(trans))
5112                                         return ERR_CAST(trans);
5113                                 goto again;
5114                         }
5115                         map = kmap(page);
5116                         write_extent_buffer(leaf, map + pg_offset, ptr,
5117                                             copy_size);
5118                         kunmap(page);
5119                         btrfs_mark_buffer_dirty(leaf);
5120                 }
5121                 set_extent_uptodate(io_tree, em->start,
5122                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5123                 goto insert;
5124         } else {
5125                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5126                 WARN_ON(1);
5127         }
5128 not_found:
5129         em->start = start;
5130         em->len = len;
5131 not_found_em:
5132         em->block_start = EXTENT_MAP_HOLE;
5133         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5134 insert:
5135         btrfs_release_path(path);
5136         if (em->start > start || extent_map_end(em) <= start) {
5137                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5138                        "[%llu %llu]\n", (unsigned long long)em->start,
5139                        (unsigned long long)em->len,
5140                        (unsigned long long)start,
5141                        (unsigned long long)len);
5142                 err = -EIO;
5143                 goto out;
5144         }
5145
5146         err = 0;
5147         write_lock(&em_tree->lock);
5148         ret = add_extent_mapping(em_tree, em);
5149         /* it is possible that someone inserted the extent into the tree
5150          * while we had the lock dropped.  It is also possible that
5151          * an overlapping map exists in the tree
5152          */
5153         if (ret == -EEXIST) {
5154                 struct extent_map *existing;
5155
5156                 ret = 0;
5157
5158                 existing = lookup_extent_mapping(em_tree, start, len);
5159                 if (existing && (existing->start > start ||
5160                     existing->start + existing->len <= start)) {
5161                         free_extent_map(existing);
5162                         existing = NULL;
5163                 }
5164                 if (!existing) {
5165                         existing = lookup_extent_mapping(em_tree, em->start,
5166                                                          em->len);
5167                         if (existing) {
5168                                 err = merge_extent_mapping(em_tree, existing,
5169                                                            em, start,
5170                                                            root->sectorsize);
5171                                 free_extent_map(existing);
5172                                 if (err) {
5173                                         free_extent_map(em);
5174                                         em = NULL;
5175                                 }
5176                         } else {
5177                                 err = -EIO;
5178                                 free_extent_map(em);
5179                                 em = NULL;
5180                         }
5181                 } else {
5182                         free_extent_map(em);
5183                         em = existing;
5184                         err = 0;
5185                 }
5186         }
5187         write_unlock(&em_tree->lock);
5188 out:
5189
5190         trace_btrfs_get_extent(root, em);
5191
5192         if (path)
5193                 btrfs_free_path(path);
5194         if (trans) {
5195                 ret = btrfs_end_transaction(trans, root);
5196                 if (!err)
5197                         err = ret;
5198         }
5199         if (err) {
5200                 free_extent_map(em);
5201                 return ERR_PTR(err);
5202         }
5203         return em;
5204 }
5205
5206 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5207                                            size_t pg_offset, u64 start, u64 len,
5208                                            int create)
5209 {
5210         struct extent_map *em;
5211         struct extent_map *hole_em = NULL;
5212         u64 range_start = start;
5213         u64 end;
5214         u64 found;
5215         u64 found_end;
5216         int err = 0;
5217
5218         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5219         if (IS_ERR(em))
5220                 return em;
5221         if (em) {
5222                 /*
5223                  * if our em maps to a hole, there might
5224                  * actually be delalloc bytes behind it
5225                  */
5226                 if (em->block_start != EXTENT_MAP_HOLE)
5227                         return em;
5228                 else
5229                         hole_em = em;
5230         }
5231
5232         /* check to see if we've wrapped (len == -1 or similar) */
5233         end = start + len;
5234         if (end < start)
5235                 end = (u64)-1;
5236         else
5237                 end -= 1;
5238
5239         em = NULL;
5240
5241         /* ok, we didn't find anything, lets look for delalloc */
5242         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5243                                  end, len, EXTENT_DELALLOC, 1);
5244         found_end = range_start + found;
5245         if (found_end < range_start)
5246                 found_end = (u64)-1;
5247
5248         /*
5249          * we didn't find anything useful, return
5250          * the original results from get_extent()
5251          */
5252         if (range_start > end || found_end <= start) {
5253                 em = hole_em;
5254                 hole_em = NULL;
5255                 goto out;
5256         }
5257
5258         /* adjust the range_start to make sure it doesn't
5259          * go backwards from the start they passed in
5260          */
5261         range_start = max(start,range_start);
5262         found = found_end - range_start;
5263
5264         if (found > 0) {
5265                 u64 hole_start = start;
5266                 u64 hole_len = len;
5267
5268                 em = alloc_extent_map();
5269                 if (!em) {
5270                         err = -ENOMEM;
5271                         goto out;
5272                 }
5273                 /*
5274                  * when btrfs_get_extent can't find anything it
5275                  * returns one huge hole
5276                  *
5277                  * make sure what it found really fits our range, and
5278                  * adjust to make sure it is based on the start from
5279                  * the caller
5280                  */
5281                 if (hole_em) {
5282                         u64 calc_end = extent_map_end(hole_em);
5283
5284                         if (calc_end <= start || (hole_em->start > end)) {
5285                                 free_extent_map(hole_em);
5286                                 hole_em = NULL;
5287                         } else {
5288                                 hole_start = max(hole_em->start, start);
5289                                 hole_len = calc_end - hole_start;
5290                         }
5291                 }
5292                 em->bdev = NULL;
5293                 if (hole_em && range_start > hole_start) {
5294                         /* our hole starts before our delalloc, so we
5295                          * have to return just the parts of the hole
5296                          * that go until  the delalloc starts
5297                          */
5298                         em->len = min(hole_len,
5299                                       range_start - hole_start);
5300                         em->start = hole_start;
5301                         em->orig_start = hole_start;
5302                         /*
5303                          * don't adjust block start at all,
5304                          * it is fixed at EXTENT_MAP_HOLE
5305                          */
5306                         em->block_start = hole_em->block_start;
5307                         em->block_len = hole_len;
5308                 } else {
5309                         em->start = range_start;
5310                         em->len = found;
5311                         em->orig_start = range_start;
5312                         em->block_start = EXTENT_MAP_DELALLOC;
5313                         em->block_len = found;
5314                 }
5315         } else if (hole_em) {
5316                 return hole_em;
5317         }
5318 out:
5319
5320         free_extent_map(hole_em);
5321         if (err) {
5322                 free_extent_map(em);
5323                 return ERR_PTR(err);
5324         }
5325         return em;
5326 }
5327
5328 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5329                                                   struct extent_map *em,
5330                                                   u64 start, u64 len)
5331 {
5332         struct btrfs_root *root = BTRFS_I(inode)->root;
5333         struct btrfs_trans_handle *trans;
5334         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5335         struct btrfs_key ins;
5336         u64 alloc_hint;
5337         int ret;
5338         bool insert = false;
5339
5340         /*
5341          * Ok if the extent map we looked up is a hole and is for the exact
5342          * range we want, there is no reason to allocate a new one, however if
5343          * it is not right then we need to free this one and drop the cache for
5344          * our range.
5345          */
5346         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5347             em->len != len) {
5348                 free_extent_map(em);
5349                 em = NULL;
5350                 insert = true;
5351                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5352         }
5353
5354         trans = btrfs_join_transaction(root);
5355         if (IS_ERR(trans))
5356                 return ERR_CAST(trans);
5357
5358         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5359                 btrfs_add_inode_defrag(trans, inode);
5360
5361         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5362
5363         alloc_hint = get_extent_allocation_hint(inode, start, len);
5364         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5365                                    alloc_hint, (u64)-1, &ins, 1);
5366         if (ret) {
5367                 em = ERR_PTR(ret);
5368                 goto out;
5369         }
5370
5371         if (!em) {
5372                 em = alloc_extent_map();
5373                 if (!em) {
5374                         em = ERR_PTR(-ENOMEM);
5375                         goto out;
5376                 }
5377         }
5378
5379         em->start = start;
5380         em->orig_start = em->start;
5381         em->len = ins.offset;
5382
5383         em->block_start = ins.objectid;
5384         em->block_len = ins.offset;
5385         em->bdev = root->fs_info->fs_devices->latest_bdev;
5386
5387         /*
5388          * We need to do this because if we're using the original em we searched
5389          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5390          */
5391         em->flags = 0;
5392         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5393
5394         while (insert) {
5395                 write_lock(&em_tree->lock);
5396                 ret = add_extent_mapping(em_tree, em);
5397                 write_unlock(&em_tree->lock);
5398                 if (ret != -EEXIST)
5399                         break;
5400                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5401         }
5402
5403         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5404                                            ins.offset, ins.offset, 0);
5405         if (ret) {
5406                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5407                 em = ERR_PTR(ret);
5408         }
5409 out:
5410         btrfs_end_transaction(trans, root);
5411         return em;
5412 }
5413
5414 /*
5415  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5416  * block must be cow'd
5417  */
5418 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5419                                       struct inode *inode, u64 offset, u64 len)
5420 {
5421         struct btrfs_path *path;
5422         int ret;
5423         struct extent_buffer *leaf;
5424         struct btrfs_root *root = BTRFS_I(inode)->root;
5425         struct btrfs_file_extent_item *fi;
5426         struct btrfs_key key;
5427         u64 disk_bytenr;
5428         u64 backref_offset;
5429         u64 extent_end;
5430         u64 num_bytes;
5431         int slot;
5432         int found_type;
5433
5434         path = btrfs_alloc_path();
5435         if (!path)
5436                 return -ENOMEM;
5437
5438         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5439                                        offset, 0);
5440         if (ret < 0)
5441                 goto out;
5442
5443         slot = path->slots[0];
5444         if (ret == 1) {
5445                 if (slot == 0) {
5446                         /* can't find the item, must cow */
5447                         ret = 0;
5448                         goto out;
5449                 }
5450                 slot--;
5451         }
5452         ret = 0;
5453         leaf = path->nodes[0];
5454         btrfs_item_key_to_cpu(leaf, &key, slot);
5455         if (key.objectid != btrfs_ino(inode) ||
5456             key.type != BTRFS_EXTENT_DATA_KEY) {
5457                 /* not our file or wrong item type, must cow */
5458                 goto out;
5459         }
5460
5461         if (key.offset > offset) {
5462                 /* Wrong offset, must cow */
5463                 goto out;
5464         }
5465
5466         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5467         found_type = btrfs_file_extent_type(leaf, fi);
5468         if (found_type != BTRFS_FILE_EXTENT_REG &&
5469             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5470                 /* not a regular extent, must cow */
5471                 goto out;
5472         }
5473         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5474         backref_offset = btrfs_file_extent_offset(leaf, fi);
5475
5476         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5477         if (extent_end < offset + len) {
5478                 /* extent doesn't include our full range, must cow */
5479                 goto out;
5480         }
5481
5482         if (btrfs_extent_readonly(root, disk_bytenr))
5483                 goto out;
5484
5485         /*
5486          * look for other files referencing this extent, if we
5487          * find any we must cow
5488          */
5489         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5490                                   key.offset - backref_offset, disk_bytenr))
5491                 goto out;
5492
5493         /*
5494          * adjust disk_bytenr and num_bytes to cover just the bytes
5495          * in this extent we are about to write.  If there
5496          * are any csums in that range we have to cow in order
5497          * to keep the csums correct
5498          */
5499         disk_bytenr += backref_offset;
5500         disk_bytenr += offset - key.offset;
5501         num_bytes = min(offset + len, extent_end) - offset;
5502         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5503                                 goto out;
5504         /*
5505          * all of the above have passed, it is safe to overwrite this extent
5506          * without cow
5507          */
5508         ret = 1;
5509 out:
5510         btrfs_free_path(path);
5511         return ret;
5512 }
5513
5514 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5515                                    struct buffer_head *bh_result, int create)
5516 {
5517         struct extent_map *em;
5518         struct btrfs_root *root = BTRFS_I(inode)->root;
5519         u64 start = iblock << inode->i_blkbits;
5520         u64 len = bh_result->b_size;
5521         struct btrfs_trans_handle *trans;
5522
5523         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5524         if (IS_ERR(em))
5525                 return PTR_ERR(em);
5526
5527         /*
5528          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5529          * io.  INLINE is special, and we could probably kludge it in here, but
5530          * it's still buffered so for safety lets just fall back to the generic
5531          * buffered path.
5532          *
5533          * For COMPRESSED we _have_ to read the entire extent in so we can
5534          * decompress it, so there will be buffering required no matter what we
5535          * do, so go ahead and fallback to buffered.
5536          *
5537          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5538          * to buffered IO.  Don't blame me, this is the price we pay for using
5539          * the generic code.
5540          */
5541         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5542             em->block_start == EXTENT_MAP_INLINE) {
5543                 free_extent_map(em);
5544                 return -ENOTBLK;
5545         }
5546
5547         /* Just a good old fashioned hole, return */
5548         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5549                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5550                 free_extent_map(em);
5551                 /* DIO will do one hole at a time, so just unlock a sector */
5552                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5553                               start + root->sectorsize - 1, GFP_NOFS);
5554                 return 0;
5555         }
5556
5557         /*
5558          * We don't allocate a new extent in the following cases
5559          *
5560          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5561          * existing extent.
5562          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5563          * just use the extent.
5564          *
5565          */
5566         if (!create) {
5567                 len = em->len - (start - em->start);
5568                 goto map;
5569         }
5570
5571         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5572             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5573              em->block_start != EXTENT_MAP_HOLE)) {
5574                 int type;
5575                 int ret;
5576                 u64 block_start;
5577
5578                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5579                         type = BTRFS_ORDERED_PREALLOC;
5580                 else
5581                         type = BTRFS_ORDERED_NOCOW;
5582                 len = min(len, em->len - (start - em->start));
5583                 block_start = em->block_start + (start - em->start);
5584
5585                 /*
5586                  * we're not going to log anything, but we do need
5587                  * to make sure the current transaction stays open
5588                  * while we look for nocow cross refs
5589                  */
5590                 trans = btrfs_join_transaction(root);
5591                 if (IS_ERR(trans))
5592                         goto must_cow;
5593
5594                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5595                         ret = btrfs_add_ordered_extent_dio(inode, start,
5596                                            block_start, len, len, type);
5597                         btrfs_end_transaction(trans, root);
5598                         if (ret) {
5599                                 free_extent_map(em);
5600                                 return ret;
5601                         }
5602                         goto unlock;
5603                 }
5604                 btrfs_end_transaction(trans, root);
5605         }
5606 must_cow:
5607         /*
5608          * this will cow the extent, reset the len in case we changed
5609          * it above
5610          */
5611         len = bh_result->b_size;
5612         em = btrfs_new_extent_direct(inode, em, start, len);
5613         if (IS_ERR(em))
5614                 return PTR_ERR(em);
5615         len = min(len, em->len - (start - em->start));
5616 unlock:
5617         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5618                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5619                           0, NULL, GFP_NOFS);
5620 map:
5621         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5622                 inode->i_blkbits;
5623         bh_result->b_size = len;
5624         bh_result->b_bdev = em->bdev;
5625         set_buffer_mapped(bh_result);
5626         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5627                 set_buffer_new(bh_result);
5628
5629         free_extent_map(em);
5630
5631         return 0;
5632 }
5633
5634 struct btrfs_dio_private {
5635         struct inode *inode;
5636         u64 logical_offset;
5637         u64 disk_bytenr;
5638         u64 bytes;
5639         u32 *csums;
5640         void *private;
5641
5642         /* number of bios pending for this dio */
5643         atomic_t pending_bios;
5644
5645         /* IO errors */
5646         int errors;
5647
5648         struct bio *orig_bio;
5649 };
5650
5651 static void btrfs_endio_direct_read(struct bio *bio, int err)
5652 {
5653         struct btrfs_dio_private *dip = bio->bi_private;
5654         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5655         struct bio_vec *bvec = bio->bi_io_vec;
5656         struct inode *inode = dip->inode;
5657         struct btrfs_root *root = BTRFS_I(inode)->root;
5658         u64 start;
5659         u32 *private = dip->csums;
5660
5661         start = dip->logical_offset;
5662         do {
5663                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5664                         struct page *page = bvec->bv_page;
5665                         char *kaddr;
5666                         u32 csum = ~(u32)0;
5667                         unsigned long flags;
5668
5669                         local_irq_save(flags);
5670                         kaddr = kmap_atomic(page, KM_IRQ0);
5671                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5672                                                csum, bvec->bv_len);
5673                         btrfs_csum_final(csum, (char *)&csum);
5674                         kunmap_atomic(kaddr, KM_IRQ0);
5675                         local_irq_restore(flags);
5676
5677                         flush_dcache_page(bvec->bv_page);
5678                         if (csum != *private) {
5679                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5680                                       " %llu csum %u private %u\n",
5681                                       (unsigned long long)btrfs_ino(inode),
5682                                       (unsigned long long)start,
5683                                       csum, *private);
5684                                 err = -EIO;
5685                         }
5686                 }
5687
5688                 start += bvec->bv_len;
5689                 private++;
5690                 bvec++;
5691         } while (bvec <= bvec_end);
5692
5693         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5694                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5695         bio->bi_private = dip->private;
5696
5697         kfree(dip->csums);
5698         kfree(dip);
5699
5700         /* If we had a csum failure make sure to clear the uptodate flag */
5701         if (err)
5702                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5703         dio_end_io(bio, err);
5704 }
5705
5706 static void btrfs_endio_direct_write(struct bio *bio, int err)
5707 {
5708         struct btrfs_dio_private *dip = bio->bi_private;
5709         struct inode *inode = dip->inode;
5710         struct btrfs_root *root = BTRFS_I(inode)->root;
5711         struct btrfs_trans_handle *trans;
5712         struct btrfs_ordered_extent *ordered = NULL;
5713         struct extent_state *cached_state = NULL;
5714         u64 ordered_offset = dip->logical_offset;
5715         u64 ordered_bytes = dip->bytes;
5716         int ret;
5717
5718         if (err)
5719                 goto out_done;
5720 again:
5721         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5722                                                    &ordered_offset,
5723                                                    ordered_bytes);
5724         if (!ret)
5725                 goto out_test;
5726
5727         BUG_ON(!ordered);
5728
5729         trans = btrfs_join_transaction(root);
5730         if (IS_ERR(trans)) {
5731                 err = -ENOMEM;
5732                 goto out;
5733         }
5734         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5735
5736         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5737                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5738                 if (!ret)
5739                         ret = btrfs_update_inode(trans, root, inode);
5740                 err = ret;
5741                 goto out;
5742         }
5743
5744         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5745                          ordered->file_offset + ordered->len - 1, 0,
5746                          &cached_state, GFP_NOFS);
5747
5748         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5749                 ret = btrfs_mark_extent_written(trans, inode,
5750                                                 ordered->file_offset,
5751                                                 ordered->file_offset +
5752                                                 ordered->len);
5753                 if (ret) {
5754                         err = ret;
5755                         goto out_unlock;
5756                 }
5757         } else {
5758                 ret = insert_reserved_file_extent(trans, inode,
5759                                                   ordered->file_offset,
5760                                                   ordered->start,
5761                                                   ordered->disk_len,
5762                                                   ordered->len,
5763                                                   ordered->len,
5764                                                   0, 0, 0,
5765                                                   BTRFS_FILE_EXTENT_REG);
5766                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5767                                    ordered->file_offset, ordered->len);
5768                 if (ret) {
5769                         err = ret;
5770                         WARN_ON(1);
5771                         goto out_unlock;
5772                 }
5773         }
5774
5775         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5776         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5777         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5778                 btrfs_update_inode(trans, root, inode);
5779         ret = 0;
5780 out_unlock:
5781         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5782                              ordered->file_offset + ordered->len - 1,
5783                              &cached_state, GFP_NOFS);
5784 out:
5785         btrfs_delalloc_release_metadata(inode, ordered->len);
5786         btrfs_end_transaction(trans, root);
5787         ordered_offset = ordered->file_offset + ordered->len;
5788         btrfs_put_ordered_extent(ordered);
5789         btrfs_put_ordered_extent(ordered);
5790
5791 out_test:
5792         /*
5793          * our bio might span multiple ordered extents.  If we haven't
5794          * completed the accounting for the whole dio, go back and try again
5795          */
5796         if (ordered_offset < dip->logical_offset + dip->bytes) {
5797                 ordered_bytes = dip->logical_offset + dip->bytes -
5798                         ordered_offset;
5799                 goto again;
5800         }
5801 out_done:
5802         bio->bi_private = dip->private;
5803
5804         kfree(dip->csums);
5805         kfree(dip);
5806
5807         /* If we had an error make sure to clear the uptodate flag */
5808         if (err)
5809                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5810         dio_end_io(bio, err);
5811 }
5812
5813 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5814                                     struct bio *bio, int mirror_num,
5815                                     unsigned long bio_flags, u64 offset)
5816 {
5817         int ret;
5818         struct btrfs_root *root = BTRFS_I(inode)->root;
5819         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5820         BUG_ON(ret);
5821         return 0;
5822 }
5823
5824 static void btrfs_end_dio_bio(struct bio *bio, int err)
5825 {
5826         struct btrfs_dio_private *dip = bio->bi_private;
5827
5828         if (err) {
5829                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5830                       "sector %#Lx len %u err no %d\n",
5831                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5832                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5833                 dip->errors = 1;
5834
5835                 /*
5836                  * before atomic variable goto zero, we must make sure
5837                  * dip->errors is perceived to be set.
5838                  */
5839                 smp_mb__before_atomic_dec();
5840         }
5841
5842         /* if there are more bios still pending for this dio, just exit */
5843         if (!atomic_dec_and_test(&dip->pending_bios))
5844                 goto out;
5845
5846         if (dip->errors)
5847                 bio_io_error(dip->orig_bio);
5848         else {
5849                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5850                 bio_endio(dip->orig_bio, 0);
5851         }
5852 out:
5853         bio_put(bio);
5854 }
5855
5856 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5857                                        u64 first_sector, gfp_t gfp_flags)
5858 {
5859         int nr_vecs = bio_get_nr_vecs(bdev);
5860         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5861 }
5862
5863 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5864                                          int rw, u64 file_offset, int skip_sum,
5865                                          u32 *csums, int async_submit)
5866 {
5867         int write = rw & REQ_WRITE;
5868         struct btrfs_root *root = BTRFS_I(inode)->root;
5869         int ret;
5870
5871         bio_get(bio);
5872         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5873         if (ret)
5874                 goto err;
5875
5876         if (skip_sum)
5877                 goto map;
5878
5879         if (write && async_submit) {
5880                 ret = btrfs_wq_submit_bio(root->fs_info,
5881                                    inode, rw, bio, 0, 0,
5882                                    file_offset,
5883                                    __btrfs_submit_bio_start_direct_io,
5884                                    __btrfs_submit_bio_done);
5885                 goto err;
5886         } else if (write) {
5887                 /*
5888                  * If we aren't doing async submit, calculate the csum of the
5889                  * bio now.
5890                  */
5891                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5892                 if (ret)
5893                         goto err;
5894         } else if (!skip_sum) {
5895                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5896                                           file_offset, csums);
5897                 if (ret)
5898                         goto err;
5899         }
5900
5901 map:
5902         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5903 err:
5904         bio_put(bio);
5905         return ret;
5906 }
5907
5908 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5909                                     int skip_sum)
5910 {
5911         struct inode *inode = dip->inode;
5912         struct btrfs_root *root = BTRFS_I(inode)->root;
5913         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5914         struct bio *bio;
5915         struct bio *orig_bio = dip->orig_bio;
5916         struct bio_vec *bvec = orig_bio->bi_io_vec;
5917         u64 start_sector = orig_bio->bi_sector;
5918         u64 file_offset = dip->logical_offset;
5919         u64 submit_len = 0;
5920         u64 map_length;
5921         int nr_pages = 0;
5922         u32 *csums = dip->csums;
5923         int ret = 0;
5924         int async_submit = 0;
5925         int write = rw & REQ_WRITE;
5926
5927         map_length = orig_bio->bi_size;
5928         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5929                               &map_length, NULL, 0);
5930         if (ret) {
5931                 bio_put(orig_bio);
5932                 return -EIO;
5933         }
5934
5935         if (map_length >= orig_bio->bi_size) {
5936                 bio = orig_bio;
5937                 goto submit;
5938         }
5939
5940         async_submit = 1;
5941         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5942         if (!bio)
5943                 return -ENOMEM;
5944         bio->bi_private = dip;
5945         bio->bi_end_io = btrfs_end_dio_bio;
5946         atomic_inc(&dip->pending_bios);
5947
5948         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5949                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5950                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5951                                  bvec->bv_offset) < bvec->bv_len)) {
5952                         /*
5953                          * inc the count before we submit the bio so
5954                          * we know the end IO handler won't happen before
5955                          * we inc the count. Otherwise, the dip might get freed
5956                          * before we're done setting it up
5957                          */
5958                         atomic_inc(&dip->pending_bios);
5959                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5960                                                      file_offset, skip_sum,
5961                                                      csums, async_submit);
5962                         if (ret) {
5963                                 bio_put(bio);
5964                                 atomic_dec(&dip->pending_bios);
5965                                 goto out_err;
5966                         }
5967
5968                         /* Write's use the ordered csums */
5969                         if (!write && !skip_sum)
5970                                 csums = csums + nr_pages;
5971                         start_sector += submit_len >> 9;
5972                         file_offset += submit_len;
5973
5974                         submit_len = 0;
5975                         nr_pages = 0;
5976
5977                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5978                                                   start_sector, GFP_NOFS);
5979                         if (!bio)
5980                                 goto out_err;
5981                         bio->bi_private = dip;
5982                         bio->bi_end_io = btrfs_end_dio_bio;
5983
5984                         map_length = orig_bio->bi_size;
5985                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5986                                               &map_length, NULL, 0);
5987                         if (ret) {
5988                                 bio_put(bio);
5989                                 goto out_err;
5990                         }
5991                 } else {
5992                         submit_len += bvec->bv_len;
5993                         nr_pages ++;
5994                         bvec++;
5995                 }
5996         }
5997
5998 submit:
5999         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6000                                      csums, async_submit);
6001         if (!ret)
6002                 return 0;
6003
6004         bio_put(bio);
6005 out_err:
6006         dip->errors = 1;
6007         /*
6008          * before atomic variable goto zero, we must
6009          * make sure dip->errors is perceived to be set.
6010          */
6011         smp_mb__before_atomic_dec();
6012         if (atomic_dec_and_test(&dip->pending_bios))
6013                 bio_io_error(dip->orig_bio);
6014
6015         /* bio_end_io() will handle error, so we needn't return it */
6016         return 0;
6017 }
6018
6019 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6020                                 loff_t file_offset)
6021 {
6022         struct btrfs_root *root = BTRFS_I(inode)->root;
6023         struct btrfs_dio_private *dip;
6024         struct bio_vec *bvec = bio->bi_io_vec;
6025         int skip_sum;
6026         int write = rw & REQ_WRITE;
6027         int ret = 0;
6028
6029         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6030
6031         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6032         if (!dip) {
6033                 ret = -ENOMEM;
6034                 goto free_ordered;
6035         }
6036         dip->csums = NULL;
6037
6038         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6039         if (!write && !skip_sum) {
6040                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6041                 if (!dip->csums) {
6042                         kfree(dip);
6043                         ret = -ENOMEM;
6044                         goto free_ordered;
6045                 }
6046         }
6047
6048         dip->private = bio->bi_private;
6049         dip->inode = inode;
6050         dip->logical_offset = file_offset;
6051
6052         dip->bytes = 0;
6053         do {
6054                 dip->bytes += bvec->bv_len;
6055                 bvec++;
6056         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6057
6058         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6059         bio->bi_private = dip;
6060         dip->errors = 0;
6061         dip->orig_bio = bio;
6062         atomic_set(&dip->pending_bios, 0);
6063
6064         if (write)
6065                 bio->bi_end_io = btrfs_endio_direct_write;
6066         else
6067                 bio->bi_end_io = btrfs_endio_direct_read;
6068
6069         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6070         if (!ret)
6071                 return;
6072 free_ordered:
6073         /*
6074          * If this is a write, we need to clean up the reserved space and kill
6075          * the ordered extent.
6076          */
6077         if (write) {
6078                 struct btrfs_ordered_extent *ordered;
6079                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6080                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6081                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6082                         btrfs_free_reserved_extent(root, ordered->start,
6083                                                    ordered->disk_len);
6084                 btrfs_put_ordered_extent(ordered);
6085                 btrfs_put_ordered_extent(ordered);
6086         }
6087         bio_endio(bio, ret);
6088 }
6089
6090 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6091                         const struct iovec *iov, loff_t offset,
6092                         unsigned long nr_segs)
6093 {
6094         int seg;
6095         int i;
6096         size_t size;
6097         unsigned long addr;
6098         unsigned blocksize_mask = root->sectorsize - 1;
6099         ssize_t retval = -EINVAL;
6100         loff_t end = offset;
6101
6102         if (offset & blocksize_mask)
6103                 goto out;
6104
6105         /* Check the memory alignment.  Blocks cannot straddle pages */
6106         for (seg = 0; seg < nr_segs; seg++) {
6107                 addr = (unsigned long)iov[seg].iov_base;
6108                 size = iov[seg].iov_len;
6109                 end += size;
6110                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6111                         goto out;
6112
6113                 /* If this is a write we don't need to check anymore */
6114                 if (rw & WRITE)
6115                         continue;
6116
6117                 /*
6118                  * Check to make sure we don't have duplicate iov_base's in this
6119                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6120                  * when reading back.
6121                  */
6122                 for (i = seg + 1; i < nr_segs; i++) {
6123                         if (iov[seg].iov_base == iov[i].iov_base)
6124                                 goto out;
6125                 }
6126         }
6127         retval = 0;
6128 out:
6129         return retval;
6130 }
6131 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6132                         const struct iovec *iov, loff_t offset,
6133                         unsigned long nr_segs)
6134 {
6135         struct file *file = iocb->ki_filp;
6136         struct inode *inode = file->f_mapping->host;
6137         struct btrfs_ordered_extent *ordered;
6138         struct extent_state *cached_state = NULL;
6139         u64 lockstart, lockend;
6140         ssize_t ret;
6141         int writing = rw & WRITE;
6142         int write_bits = 0;
6143         size_t count = iov_length(iov, nr_segs);
6144
6145         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6146                             offset, nr_segs)) {
6147                 return 0;
6148         }
6149
6150         lockstart = offset;
6151         lockend = offset + count - 1;
6152
6153         if (writing) {
6154                 ret = btrfs_delalloc_reserve_space(inode, count);
6155                 if (ret)
6156                         goto out;
6157         }
6158
6159         while (1) {
6160                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6161                                  0, &cached_state, GFP_NOFS);
6162                 /*
6163                  * We're concerned with the entire range that we're going to be
6164                  * doing DIO to, so we need to make sure theres no ordered
6165                  * extents in this range.
6166                  */
6167                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6168                                                      lockend - lockstart + 1);
6169                 if (!ordered)
6170                         break;
6171                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6172                                      &cached_state, GFP_NOFS);
6173                 btrfs_start_ordered_extent(inode, ordered, 1);
6174                 btrfs_put_ordered_extent(ordered);
6175                 cond_resched();
6176         }
6177
6178         /*
6179          * we don't use btrfs_set_extent_delalloc because we don't want
6180          * the dirty or uptodate bits
6181          */
6182         if (writing) {
6183                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6184                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6185                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6186                                      GFP_NOFS);
6187                 if (ret) {
6188                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6189                                          lockend, EXTENT_LOCKED | write_bits,
6190                                          1, 0, &cached_state, GFP_NOFS);
6191                         goto out;
6192                 }
6193         }
6194
6195         free_extent_state(cached_state);
6196         cached_state = NULL;
6197
6198         ret = __blockdev_direct_IO(rw, iocb, inode,
6199                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6200                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6201                    btrfs_submit_direct, 0);
6202
6203         if (ret < 0 && ret != -EIOCBQUEUED) {
6204                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6205                               offset + iov_length(iov, nr_segs) - 1,
6206                               EXTENT_LOCKED | write_bits, 1, 0,
6207                               &cached_state, GFP_NOFS);
6208         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6209                 /*
6210                  * We're falling back to buffered, unlock the section we didn't
6211                  * do IO on.
6212                  */
6213                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6214                               offset + iov_length(iov, nr_segs) - 1,
6215                               EXTENT_LOCKED | write_bits, 1, 0,
6216                               &cached_state, GFP_NOFS);
6217         }
6218 out:
6219         free_extent_state(cached_state);
6220         return ret;
6221 }
6222
6223 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6224                 __u64 start, __u64 len)
6225 {
6226         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6227 }
6228
6229 int btrfs_readpage(struct file *file, struct page *page)
6230 {
6231         struct extent_io_tree *tree;
6232         tree = &BTRFS_I(page->mapping->host)->io_tree;
6233         return extent_read_full_page(tree, page, btrfs_get_extent);
6234 }
6235
6236 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6237 {
6238         struct extent_io_tree *tree;
6239
6240
6241         if (current->flags & PF_MEMALLOC) {
6242                 redirty_page_for_writepage(wbc, page);
6243                 unlock_page(page);
6244                 return 0;
6245         }
6246         tree = &BTRFS_I(page->mapping->host)->io_tree;
6247         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6248 }
6249
6250 int btrfs_writepages(struct address_space *mapping,
6251                      struct writeback_control *wbc)
6252 {
6253         struct extent_io_tree *tree;
6254
6255         tree = &BTRFS_I(mapping->host)->io_tree;
6256         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6257 }
6258
6259 static int
6260 btrfs_readpages(struct file *file, struct address_space *mapping,
6261                 struct list_head *pages, unsigned nr_pages)
6262 {
6263         struct extent_io_tree *tree;
6264         tree = &BTRFS_I(mapping->host)->io_tree;
6265         return extent_readpages(tree, mapping, pages, nr_pages,
6266                                 btrfs_get_extent);
6267 }
6268 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6269 {
6270         struct extent_io_tree *tree;
6271         struct extent_map_tree *map;
6272         int ret;
6273
6274         tree = &BTRFS_I(page->mapping->host)->io_tree;
6275         map = &BTRFS_I(page->mapping->host)->extent_tree;
6276         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6277         if (ret == 1) {
6278                 ClearPagePrivate(page);
6279                 set_page_private(page, 0);
6280                 page_cache_release(page);
6281         }
6282         return ret;
6283 }
6284
6285 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6286 {
6287         if (PageWriteback(page) || PageDirty(page))
6288                 return 0;
6289         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6290 }
6291
6292 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6293 {
6294         struct extent_io_tree *tree;
6295         struct btrfs_ordered_extent *ordered;
6296         struct extent_state *cached_state = NULL;
6297         u64 page_start = page_offset(page);
6298         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6299
6300
6301         /*
6302          * we have the page locked, so new writeback can't start,
6303          * and the dirty bit won't be cleared while we are here.
6304          *
6305          * Wait for IO on this page so that we can safely clear
6306          * the PagePrivate2 bit and do ordered accounting
6307          */
6308         wait_on_page_writeback(page);
6309
6310         tree = &BTRFS_I(page->mapping->host)->io_tree;
6311         if (offset) {
6312                 btrfs_releasepage(page, GFP_NOFS);
6313                 return;
6314         }
6315         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6316                          GFP_NOFS);
6317         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6318                                            page_offset(page));
6319         if (ordered) {
6320                 /*
6321                  * IO on this page will never be started, so we need
6322                  * to account for any ordered extents now
6323                  */
6324                 clear_extent_bit(tree, page_start, page_end,
6325                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6326                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6327                                  &cached_state, GFP_NOFS);
6328                 /*
6329                  * whoever cleared the private bit is responsible
6330                  * for the finish_ordered_io
6331                  */
6332                 if (TestClearPagePrivate2(page)) {
6333                         btrfs_finish_ordered_io(page->mapping->host,
6334                                                 page_start, page_end);
6335                 }
6336                 btrfs_put_ordered_extent(ordered);
6337                 cached_state = NULL;
6338                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6339                                  GFP_NOFS);
6340         }
6341         clear_extent_bit(tree, page_start, page_end,
6342                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6343                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6344         __btrfs_releasepage(page, GFP_NOFS);
6345
6346         ClearPageChecked(page);
6347         if (PagePrivate(page)) {
6348                 ClearPagePrivate(page);
6349                 set_page_private(page, 0);
6350                 page_cache_release(page);
6351         }
6352 }
6353
6354 /*
6355  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6356  * called from a page fault handler when a page is first dirtied. Hence we must
6357  * be careful to check for EOF conditions here. We set the page up correctly
6358  * for a written page which means we get ENOSPC checking when writing into
6359  * holes and correct delalloc and unwritten extent mapping on filesystems that
6360  * support these features.
6361  *
6362  * We are not allowed to take the i_mutex here so we have to play games to
6363  * protect against truncate races as the page could now be beyond EOF.  Because
6364  * vmtruncate() writes the inode size before removing pages, once we have the
6365  * page lock we can determine safely if the page is beyond EOF. If it is not
6366  * beyond EOF, then the page is guaranteed safe against truncation until we
6367  * unlock the page.
6368  */
6369 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6370 {
6371         struct page *page = vmf->page;
6372         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6373         struct btrfs_root *root = BTRFS_I(inode)->root;
6374         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6375         struct btrfs_ordered_extent *ordered;
6376         struct extent_state *cached_state = NULL;
6377         char *kaddr;
6378         unsigned long zero_start;
6379         loff_t size;
6380         int ret;
6381         u64 page_start;
6382         u64 page_end;
6383
6384         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6385         if (ret) {
6386                 if (ret == -ENOMEM)
6387                         ret = VM_FAULT_OOM;
6388                 else /* -ENOSPC, -EIO, etc */
6389                         ret = VM_FAULT_SIGBUS;
6390                 goto out;
6391         }
6392
6393         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6394 again:
6395         lock_page(page);
6396         size = i_size_read(inode);
6397         page_start = page_offset(page);
6398         page_end = page_start + PAGE_CACHE_SIZE - 1;
6399
6400         if ((page->mapping != inode->i_mapping) ||
6401             (page_start >= size)) {
6402                 /* page got truncated out from underneath us */
6403                 goto out_unlock;
6404         }
6405         wait_on_page_writeback(page);
6406
6407         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6408                          GFP_NOFS);
6409         set_page_extent_mapped(page);
6410
6411         /*
6412          * we can't set the delalloc bits if there are pending ordered
6413          * extents.  Drop our locks and wait for them to finish
6414          */
6415         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6416         if (ordered) {
6417                 unlock_extent_cached(io_tree, page_start, page_end,
6418                                      &cached_state, GFP_NOFS);
6419                 unlock_page(page);
6420                 btrfs_start_ordered_extent(inode, ordered, 1);
6421                 btrfs_put_ordered_extent(ordered);
6422                 goto again;
6423         }
6424
6425         /*
6426          * XXX - page_mkwrite gets called every time the page is dirtied, even
6427          * if it was already dirty, so for space accounting reasons we need to
6428          * clear any delalloc bits for the range we are fixing to save.  There
6429          * is probably a better way to do this, but for now keep consistent with
6430          * prepare_pages in the normal write path.
6431          */
6432         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6433                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6434                           0, 0, &cached_state, GFP_NOFS);
6435
6436         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6437                                         &cached_state);
6438         if (ret) {
6439                 unlock_extent_cached(io_tree, page_start, page_end,
6440                                      &cached_state, GFP_NOFS);
6441                 ret = VM_FAULT_SIGBUS;
6442                 goto out_unlock;
6443         }
6444         ret = 0;
6445
6446         /* page is wholly or partially inside EOF */
6447         if (page_start + PAGE_CACHE_SIZE > size)
6448                 zero_start = size & ~PAGE_CACHE_MASK;
6449         else
6450                 zero_start = PAGE_CACHE_SIZE;
6451
6452         if (zero_start != PAGE_CACHE_SIZE) {
6453                 kaddr = kmap(page);
6454                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6455                 flush_dcache_page(page);
6456                 kunmap(page);
6457         }
6458         ClearPageChecked(page);
6459         set_page_dirty(page);
6460         SetPageUptodate(page);
6461
6462         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6463         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6464
6465         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6466
6467 out_unlock:
6468         if (!ret)
6469                 return VM_FAULT_LOCKED;
6470         unlock_page(page);
6471         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6472 out:
6473         return ret;
6474 }
6475
6476 static int btrfs_truncate(struct inode *inode)
6477 {
6478         struct btrfs_root *root = BTRFS_I(inode)->root;
6479         struct btrfs_block_rsv *rsv;
6480         int ret;
6481         int err = 0;
6482         struct btrfs_trans_handle *trans;
6483         unsigned long nr;
6484         u64 mask = root->sectorsize - 1;
6485         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6486
6487         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6488         if (ret)
6489                 return ret;
6490
6491         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6492         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6493
6494         /*
6495          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6496          * 3 things going on here
6497          *
6498          * 1) We need to reserve space for our orphan item and the space to
6499          * delete our orphan item.  Lord knows we don't want to have a dangling
6500          * orphan item because we didn't reserve space to remove it.
6501          *
6502          * 2) We need to reserve space to update our inode.
6503          *
6504          * 3) We need to have something to cache all the space that is going to
6505          * be free'd up by the truncate operation, but also have some slack
6506          * space reserved in case it uses space during the truncate (thank you
6507          * very much snapshotting).
6508          *
6509          * And we need these to all be seperate.  The fact is we can use alot of
6510          * space doing the truncate, and we have no earthly idea how much space
6511          * we will use, so we need the truncate reservation to be seperate so it
6512          * doesn't end up using space reserved for updating the inode or
6513          * removing the orphan item.  We also need to be able to stop the
6514          * transaction and start a new one, which means we need to be able to
6515          * update the inode several times, and we have no idea of knowing how
6516          * many times that will be, so we can't just reserve 1 item for the
6517          * entirety of the opration, so that has to be done seperately as well.
6518          * Then there is the orphan item, which does indeed need to be held on
6519          * to for the whole operation, and we need nobody to touch this reserved
6520          * space except the orphan code.
6521          *
6522          * So that leaves us with
6523          *
6524          * 1) root->orphan_block_rsv - for the orphan deletion.
6525          * 2) rsv - for the truncate reservation, which we will steal from the
6526          * transaction reservation.
6527          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6528          * updating the inode.
6529          */
6530         rsv = btrfs_alloc_block_rsv(root);
6531         if (!rsv)
6532                 return -ENOMEM;
6533         rsv->size = min_size;
6534
6535         /*
6536          * 1 for the truncate slack space
6537          * 1 for the orphan item we're going to add
6538          * 1 for the orphan item deletion
6539          * 1 for updating the inode.
6540          */
6541         trans = btrfs_start_transaction(root, 4);
6542         if (IS_ERR(trans)) {
6543                 err = PTR_ERR(trans);
6544                 goto out;
6545         }
6546
6547         /* Migrate the slack space for the truncate to our reserve */
6548         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6549                                       min_size);
6550         BUG_ON(ret);
6551
6552         ret = btrfs_orphan_add(trans, inode);
6553         if (ret) {
6554                 btrfs_end_transaction(trans, root);
6555                 goto out;
6556         }
6557
6558         /*
6559          * setattr is responsible for setting the ordered_data_close flag,
6560          * but that is only tested during the last file release.  That
6561          * could happen well after the next commit, leaving a great big
6562          * window where new writes may get lost if someone chooses to write
6563          * to this file after truncating to zero
6564          *
6565          * The inode doesn't have any dirty data here, and so if we commit
6566          * this is a noop.  If someone immediately starts writing to the inode
6567          * it is very likely we'll catch some of their writes in this
6568          * transaction, and the commit will find this file on the ordered
6569          * data list with good things to send down.
6570          *
6571          * This is a best effort solution, there is still a window where
6572          * using truncate to replace the contents of the file will
6573          * end up with a zero length file after a crash.
6574          */
6575         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6576                 btrfs_add_ordered_operation(trans, root, inode);
6577
6578         while (1) {
6579                 ret = btrfs_block_rsv_check(root, rsv, min_size, 0, 1);
6580                 if (ret) {
6581                         /*
6582                          * This can only happen with the original transaction we
6583                          * started above, every other time we shouldn't have a
6584                          * transaction started yet.
6585                          */
6586                         if (ret == -EAGAIN)
6587                                 goto end_trans;
6588                         err = ret;
6589                         break;
6590                 }
6591
6592                 if (!trans) {
6593                         /* Just need the 1 for updating the inode */
6594                         trans = btrfs_start_transaction(root, 1);
6595                         if (IS_ERR(trans)) {
6596                                 err = PTR_ERR(trans);
6597                                 goto out;
6598                         }
6599                 }
6600
6601                 trans->block_rsv = rsv;
6602
6603                 ret = btrfs_truncate_inode_items(trans, root, inode,
6604                                                  inode->i_size,
6605                                                  BTRFS_EXTENT_DATA_KEY);
6606                 if (ret != -EAGAIN) {
6607                         err = ret;
6608                         break;
6609                 }
6610
6611                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6612                 ret = btrfs_update_inode(trans, root, inode);
6613                 if (ret) {
6614                         err = ret;
6615                         break;
6616                 }
6617 end_trans:
6618                 nr = trans->blocks_used;
6619                 btrfs_end_transaction(trans, root);
6620                 trans = NULL;
6621                 btrfs_btree_balance_dirty(root, nr);
6622         }
6623
6624         if (ret == 0 && inode->i_nlink > 0) {
6625                 trans->block_rsv = root->orphan_block_rsv;
6626                 ret = btrfs_orphan_del(trans, inode);
6627                 if (ret)
6628                         err = ret;
6629         } else if (ret && inode->i_nlink > 0) {
6630                 /*
6631                  * Failed to do the truncate, remove us from the in memory
6632                  * orphan list.
6633                  */
6634                 ret = btrfs_orphan_del(NULL, inode);
6635         }
6636
6637         trans->block_rsv = &root->fs_info->trans_block_rsv;
6638         ret = btrfs_update_inode(trans, root, inode);
6639         if (ret && !err)
6640                 err = ret;
6641
6642         nr = trans->blocks_used;
6643         ret = btrfs_end_transaction_throttle(trans, root);
6644         btrfs_btree_balance_dirty(root, nr);
6645
6646 out:
6647         btrfs_free_block_rsv(root, rsv);
6648
6649         if (ret && !err)
6650                 err = ret;
6651
6652         return err;
6653 }
6654
6655 /*
6656  * create a new subvolume directory/inode (helper for the ioctl).
6657  */
6658 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6659                              struct btrfs_root *new_root, u64 new_dirid)
6660 {
6661         struct inode *inode;
6662         int err;
6663         u64 index = 0;
6664
6665         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6666                                 new_dirid, S_IFDIR | 0700, &index);
6667         if (IS_ERR(inode))
6668                 return PTR_ERR(inode);
6669         inode->i_op = &btrfs_dir_inode_operations;
6670         inode->i_fop = &btrfs_dir_file_operations;
6671
6672         inode->i_nlink = 1;
6673         btrfs_i_size_write(inode, 0);
6674
6675         err = btrfs_update_inode(trans, new_root, inode);
6676         BUG_ON(err);
6677
6678         iput(inode);
6679         return 0;
6680 }
6681
6682 struct inode *btrfs_alloc_inode(struct super_block *sb)
6683 {
6684         struct btrfs_inode *ei;
6685         struct inode *inode;
6686
6687         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6688         if (!ei)
6689                 return NULL;
6690
6691         ei->root = NULL;
6692         ei->space_info = NULL;
6693         ei->generation = 0;
6694         ei->sequence = 0;
6695         ei->last_trans = 0;
6696         ei->last_sub_trans = 0;
6697         ei->logged_trans = 0;
6698         ei->delalloc_bytes = 0;
6699         ei->disk_i_size = 0;
6700         ei->flags = 0;
6701         ei->csum_bytes = 0;
6702         ei->index_cnt = (u64)-1;
6703         ei->last_unlink_trans = 0;
6704
6705         spin_lock_init(&ei->lock);
6706         ei->outstanding_extents = 0;
6707         ei->reserved_extents = 0;
6708
6709         ei->ordered_data_close = 0;
6710         ei->orphan_meta_reserved = 0;
6711         ei->dummy_inode = 0;
6712         ei->in_defrag = 0;
6713         ei->force_compress = BTRFS_COMPRESS_NONE;
6714
6715         ei->delayed_node = NULL;
6716
6717         inode = &ei->vfs_inode;
6718         extent_map_tree_init(&ei->extent_tree);
6719         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6720         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6721         mutex_init(&ei->log_mutex);
6722         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6723         INIT_LIST_HEAD(&ei->i_orphan);
6724         INIT_LIST_HEAD(&ei->delalloc_inodes);
6725         INIT_LIST_HEAD(&ei->ordered_operations);
6726         RB_CLEAR_NODE(&ei->rb_node);
6727
6728         return inode;
6729 }
6730
6731 static void btrfs_i_callback(struct rcu_head *head)
6732 {
6733         struct inode *inode = container_of(head, struct inode, i_rcu);
6734         INIT_LIST_HEAD(&inode->i_dentry);
6735         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6736 }
6737
6738 void btrfs_destroy_inode(struct inode *inode)
6739 {
6740         struct btrfs_ordered_extent *ordered;
6741         struct btrfs_root *root = BTRFS_I(inode)->root;
6742
6743         WARN_ON(!list_empty(&inode->i_dentry));
6744         WARN_ON(inode->i_data.nrpages);
6745         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6746         WARN_ON(BTRFS_I(inode)->reserved_extents);
6747         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6748         WARN_ON(BTRFS_I(inode)->csum_bytes);
6749
6750         /*
6751          * This can happen where we create an inode, but somebody else also
6752          * created the same inode and we need to destroy the one we already
6753          * created.
6754          */
6755         if (!root)
6756                 goto free;
6757
6758         /*
6759          * Make sure we're properly removed from the ordered operation
6760          * lists.
6761          */
6762         smp_mb();
6763         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6764                 spin_lock(&root->fs_info->ordered_extent_lock);
6765                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6766                 spin_unlock(&root->fs_info->ordered_extent_lock);
6767         }
6768
6769         spin_lock(&root->orphan_lock);
6770         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6771                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6772                        (unsigned long long)btrfs_ino(inode));
6773                 list_del_init(&BTRFS_I(inode)->i_orphan);
6774         }
6775         spin_unlock(&root->orphan_lock);
6776
6777         while (1) {
6778                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6779                 if (!ordered)
6780                         break;
6781                 else {
6782                         printk(KERN_ERR "btrfs found ordered "
6783                                "extent %llu %llu on inode cleanup\n",
6784                                (unsigned long long)ordered->file_offset,
6785                                (unsigned long long)ordered->len);
6786                         btrfs_remove_ordered_extent(inode, ordered);
6787                         btrfs_put_ordered_extent(ordered);
6788                         btrfs_put_ordered_extent(ordered);
6789                 }
6790         }
6791         inode_tree_del(inode);
6792         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6793 free:
6794         btrfs_remove_delayed_node(inode);
6795         call_rcu(&inode->i_rcu, btrfs_i_callback);
6796 }
6797
6798 int btrfs_drop_inode(struct inode *inode)
6799 {
6800         struct btrfs_root *root = BTRFS_I(inode)->root;
6801
6802         if (btrfs_root_refs(&root->root_item) == 0 &&
6803             !btrfs_is_free_space_inode(root, inode))
6804                 return 1;
6805         else
6806                 return generic_drop_inode(inode);
6807 }
6808
6809 static void init_once(void *foo)
6810 {
6811         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6812
6813         inode_init_once(&ei->vfs_inode);
6814 }
6815
6816 void btrfs_destroy_cachep(void)
6817 {
6818         if (btrfs_inode_cachep)
6819                 kmem_cache_destroy(btrfs_inode_cachep);
6820         if (btrfs_trans_handle_cachep)
6821                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6822         if (btrfs_transaction_cachep)
6823                 kmem_cache_destroy(btrfs_transaction_cachep);
6824         if (btrfs_path_cachep)
6825                 kmem_cache_destroy(btrfs_path_cachep);
6826         if (btrfs_free_space_cachep)
6827                 kmem_cache_destroy(btrfs_free_space_cachep);
6828 }
6829
6830 int btrfs_init_cachep(void)
6831 {
6832         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6833                         sizeof(struct btrfs_inode), 0,
6834                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6835         if (!btrfs_inode_cachep)
6836                 goto fail;
6837
6838         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6839                         sizeof(struct btrfs_trans_handle), 0,
6840                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6841         if (!btrfs_trans_handle_cachep)
6842                 goto fail;
6843
6844         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6845                         sizeof(struct btrfs_transaction), 0,
6846                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6847         if (!btrfs_transaction_cachep)
6848                 goto fail;
6849
6850         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6851                         sizeof(struct btrfs_path), 0,
6852                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6853         if (!btrfs_path_cachep)
6854                 goto fail;
6855
6856         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6857                         sizeof(struct btrfs_free_space), 0,
6858                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6859         if (!btrfs_free_space_cachep)
6860                 goto fail;
6861
6862         return 0;
6863 fail:
6864         btrfs_destroy_cachep();
6865         return -ENOMEM;
6866 }
6867
6868 static int btrfs_getattr(struct vfsmount *mnt,
6869                          struct dentry *dentry, struct kstat *stat)
6870 {
6871         struct inode *inode = dentry->d_inode;
6872         generic_fillattr(inode, stat);
6873         stat->dev = BTRFS_I(inode)->root->anon_dev;
6874         stat->blksize = PAGE_CACHE_SIZE;
6875         stat->blocks = (inode_get_bytes(inode) +
6876                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6877         return 0;
6878 }
6879
6880 /*
6881  * If a file is moved, it will inherit the cow and compression flags of the new
6882  * directory.
6883  */
6884 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6885 {
6886         struct btrfs_inode *b_dir = BTRFS_I(dir);
6887         struct btrfs_inode *b_inode = BTRFS_I(inode);
6888
6889         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6890                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6891         else
6892                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6893
6894         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6895                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6896         else
6897                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6898 }
6899
6900 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6901                            struct inode *new_dir, struct dentry *new_dentry)
6902 {
6903         struct btrfs_trans_handle *trans;
6904         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6905         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6906         struct inode *new_inode = new_dentry->d_inode;
6907         struct inode *old_inode = old_dentry->d_inode;
6908         struct timespec ctime = CURRENT_TIME;
6909         u64 index = 0;
6910         u64 root_objectid;
6911         int ret;
6912         u64 old_ino = btrfs_ino(old_inode);
6913
6914         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6915                 return -EPERM;
6916
6917         /* we only allow rename subvolume link between subvolumes */
6918         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6919                 return -EXDEV;
6920
6921         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6922             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6923                 return -ENOTEMPTY;
6924
6925         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6926             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6927                 return -ENOTEMPTY;
6928         /*
6929          * we're using rename to replace one file with another.
6930          * and the replacement file is large.  Start IO on it now so
6931          * we don't add too much work to the end of the transaction
6932          */
6933         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6934             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6935                 filemap_flush(old_inode->i_mapping);
6936
6937         /* close the racy window with snapshot create/destroy ioctl */
6938         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6939                 down_read(&root->fs_info->subvol_sem);
6940         /*
6941          * We want to reserve the absolute worst case amount of items.  So if
6942          * both inodes are subvols and we need to unlink them then that would
6943          * require 4 item modifications, but if they are both normal inodes it
6944          * would require 5 item modifications, so we'll assume their normal
6945          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6946          * should cover the worst case number of items we'll modify.
6947          */
6948         trans = btrfs_start_transaction(root, 20);
6949         if (IS_ERR(trans)) {
6950                 ret = PTR_ERR(trans);
6951                 goto out_notrans;
6952         }
6953
6954         if (dest != root)
6955                 btrfs_record_root_in_trans(trans, dest);
6956
6957         ret = btrfs_set_inode_index(new_dir, &index);
6958         if (ret)
6959                 goto out_fail;
6960
6961         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6962                 /* force full log commit if subvolume involved. */
6963                 root->fs_info->last_trans_log_full_commit = trans->transid;
6964         } else {
6965                 ret = btrfs_insert_inode_ref(trans, dest,
6966                                              new_dentry->d_name.name,
6967                                              new_dentry->d_name.len,
6968                                              old_ino,
6969                                              btrfs_ino(new_dir), index);
6970                 if (ret)
6971                         goto out_fail;
6972                 /*
6973                  * this is an ugly little race, but the rename is required
6974                  * to make sure that if we crash, the inode is either at the
6975                  * old name or the new one.  pinning the log transaction lets
6976                  * us make sure we don't allow a log commit to come in after
6977                  * we unlink the name but before we add the new name back in.
6978                  */
6979                 btrfs_pin_log_trans(root);
6980         }
6981         /*
6982          * make sure the inode gets flushed if it is replacing
6983          * something.
6984          */
6985         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6986                 btrfs_add_ordered_operation(trans, root, old_inode);
6987
6988         old_dir->i_ctime = old_dir->i_mtime = ctime;
6989         new_dir->i_ctime = new_dir->i_mtime = ctime;
6990         old_inode->i_ctime = ctime;
6991
6992         if (old_dentry->d_parent != new_dentry->d_parent)
6993                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6994
6995         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6996                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6997                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6998                                         old_dentry->d_name.name,
6999                                         old_dentry->d_name.len);
7000         } else {
7001                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7002                                         old_dentry->d_inode,
7003                                         old_dentry->d_name.name,
7004                                         old_dentry->d_name.len);
7005                 if (!ret)
7006                         ret = btrfs_update_inode(trans, root, old_inode);
7007         }
7008         BUG_ON(ret);
7009
7010         if (new_inode) {
7011                 new_inode->i_ctime = CURRENT_TIME;
7012                 if (unlikely(btrfs_ino(new_inode) ==
7013                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7014                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7015                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7016                                                 root_objectid,
7017                                                 new_dentry->d_name.name,
7018                                                 new_dentry->d_name.len);
7019                         BUG_ON(new_inode->i_nlink == 0);
7020                 } else {
7021                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7022                                                  new_dentry->d_inode,
7023                                                  new_dentry->d_name.name,
7024                                                  new_dentry->d_name.len);
7025                 }
7026                 BUG_ON(ret);
7027                 if (new_inode->i_nlink == 0) {
7028                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7029                         BUG_ON(ret);
7030                 }
7031         }
7032
7033         fixup_inode_flags(new_dir, old_inode);
7034
7035         ret = btrfs_add_link(trans, new_dir, old_inode,
7036                              new_dentry->d_name.name,
7037                              new_dentry->d_name.len, 0, index);
7038         BUG_ON(ret);
7039
7040         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7041                 struct dentry *parent = new_dentry->d_parent;
7042                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7043                 btrfs_end_log_trans(root);
7044         }
7045 out_fail:
7046         btrfs_end_transaction_throttle(trans, root);
7047 out_notrans:
7048         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7049                 up_read(&root->fs_info->subvol_sem);
7050
7051         return ret;
7052 }
7053
7054 /*
7055  * some fairly slow code that needs optimization. This walks the list
7056  * of all the inodes with pending delalloc and forces them to disk.
7057  */
7058 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7059 {
7060         struct list_head *head = &root->fs_info->delalloc_inodes;
7061         struct btrfs_inode *binode;
7062         struct inode *inode;
7063
7064         if (root->fs_info->sb->s_flags & MS_RDONLY)
7065                 return -EROFS;
7066
7067         spin_lock(&root->fs_info->delalloc_lock);
7068         while (!list_empty(head)) {
7069                 binode = list_entry(head->next, struct btrfs_inode,
7070                                     delalloc_inodes);
7071                 inode = igrab(&binode->vfs_inode);
7072                 if (!inode)
7073                         list_del_init(&binode->delalloc_inodes);
7074                 spin_unlock(&root->fs_info->delalloc_lock);
7075                 if (inode) {
7076                         filemap_flush(inode->i_mapping);
7077                         if (delay_iput)
7078                                 btrfs_add_delayed_iput(inode);
7079                         else
7080                                 iput(inode);
7081                 }
7082                 cond_resched();
7083                 spin_lock(&root->fs_info->delalloc_lock);
7084         }
7085         spin_unlock(&root->fs_info->delalloc_lock);
7086
7087         /* the filemap_flush will queue IO into the worker threads, but
7088          * we have to make sure the IO is actually started and that
7089          * ordered extents get created before we return
7090          */
7091         atomic_inc(&root->fs_info->async_submit_draining);
7092         while (atomic_read(&root->fs_info->nr_async_submits) ||
7093               atomic_read(&root->fs_info->async_delalloc_pages)) {
7094                 wait_event(root->fs_info->async_submit_wait,
7095                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7096                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7097         }
7098         atomic_dec(&root->fs_info->async_submit_draining);
7099         return 0;
7100 }
7101
7102 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7103                          const char *symname)
7104 {
7105         struct btrfs_trans_handle *trans;
7106         struct btrfs_root *root = BTRFS_I(dir)->root;
7107         struct btrfs_path *path;
7108         struct btrfs_key key;
7109         struct inode *inode = NULL;
7110         int err;
7111         int drop_inode = 0;
7112         u64 objectid;
7113         u64 index = 0 ;
7114         int name_len;
7115         int datasize;
7116         unsigned long ptr;
7117         struct btrfs_file_extent_item *ei;
7118         struct extent_buffer *leaf;
7119         unsigned long nr = 0;
7120
7121         name_len = strlen(symname) + 1;
7122         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7123                 return -ENAMETOOLONG;
7124
7125         /*
7126          * 2 items for inode item and ref
7127          * 2 items for dir items
7128          * 1 item for xattr if selinux is on
7129          */
7130         trans = btrfs_start_transaction(root, 5);
7131         if (IS_ERR(trans))
7132                 return PTR_ERR(trans);
7133
7134         err = btrfs_find_free_ino(root, &objectid);
7135         if (err)
7136                 goto out_unlock;
7137
7138         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7139                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7140                                 S_IFLNK|S_IRWXUGO, &index);
7141         if (IS_ERR(inode)) {
7142                 err = PTR_ERR(inode);
7143                 goto out_unlock;
7144         }
7145
7146         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7147         if (err) {
7148                 drop_inode = 1;
7149                 goto out_unlock;
7150         }
7151
7152         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7153         if (err)
7154                 drop_inode = 1;
7155         else {
7156                 inode->i_mapping->a_ops = &btrfs_aops;
7157                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7158                 inode->i_fop = &btrfs_file_operations;
7159                 inode->i_op = &btrfs_file_inode_operations;
7160                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7161         }
7162         if (drop_inode)
7163                 goto out_unlock;
7164
7165         path = btrfs_alloc_path();
7166         if (!path) {
7167                 err = -ENOMEM;
7168                 drop_inode = 1;
7169                 goto out_unlock;
7170         }
7171         key.objectid = btrfs_ino(inode);
7172         key.offset = 0;
7173         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7174         datasize = btrfs_file_extent_calc_inline_size(name_len);
7175         err = btrfs_insert_empty_item(trans, root, path, &key,
7176                                       datasize);
7177         if (err) {
7178                 drop_inode = 1;
7179                 btrfs_free_path(path);
7180                 goto out_unlock;
7181         }
7182         leaf = path->nodes[0];
7183         ei = btrfs_item_ptr(leaf, path->slots[0],
7184                             struct btrfs_file_extent_item);
7185         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7186         btrfs_set_file_extent_type(leaf, ei,
7187                                    BTRFS_FILE_EXTENT_INLINE);
7188         btrfs_set_file_extent_encryption(leaf, ei, 0);
7189         btrfs_set_file_extent_compression(leaf, ei, 0);
7190         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7191         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7192
7193         ptr = btrfs_file_extent_inline_start(ei);
7194         write_extent_buffer(leaf, symname, ptr, name_len);
7195         btrfs_mark_buffer_dirty(leaf);
7196         btrfs_free_path(path);
7197
7198         inode->i_op = &btrfs_symlink_inode_operations;
7199         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7200         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7201         inode_set_bytes(inode, name_len);
7202         btrfs_i_size_write(inode, name_len - 1);
7203         err = btrfs_update_inode(trans, root, inode);
7204         if (err)
7205                 drop_inode = 1;
7206
7207 out_unlock:
7208         nr = trans->blocks_used;
7209         btrfs_end_transaction_throttle(trans, root);
7210         if (drop_inode) {
7211                 inode_dec_link_count(inode);
7212                 iput(inode);
7213         }
7214         btrfs_btree_balance_dirty(root, nr);
7215         return err;
7216 }
7217
7218 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7219                                        u64 start, u64 num_bytes, u64 min_size,
7220                                        loff_t actual_len, u64 *alloc_hint,
7221                                        struct btrfs_trans_handle *trans)
7222 {
7223         struct btrfs_root *root = BTRFS_I(inode)->root;
7224         struct btrfs_key ins;
7225         u64 cur_offset = start;
7226         u64 i_size;
7227         int ret = 0;
7228         bool own_trans = true;
7229
7230         if (trans)
7231                 own_trans = false;
7232         while (num_bytes > 0) {
7233                 if (own_trans) {
7234                         trans = btrfs_start_transaction(root, 3);
7235                         if (IS_ERR(trans)) {
7236                                 ret = PTR_ERR(trans);
7237                                 break;
7238                         }
7239                 }
7240
7241                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7242                                            0, *alloc_hint, (u64)-1, &ins, 1);
7243                 if (ret) {
7244                         if (own_trans)
7245                                 btrfs_end_transaction(trans, root);
7246                         break;
7247                 }
7248
7249                 ret = insert_reserved_file_extent(trans, inode,
7250                                                   cur_offset, ins.objectid,
7251                                                   ins.offset, ins.offset,
7252                                                   ins.offset, 0, 0, 0,
7253                                                   BTRFS_FILE_EXTENT_PREALLOC);
7254                 BUG_ON(ret);
7255                 btrfs_drop_extent_cache(inode, cur_offset,
7256                                         cur_offset + ins.offset -1, 0);
7257
7258                 num_bytes -= ins.offset;
7259                 cur_offset += ins.offset;
7260                 *alloc_hint = ins.objectid + ins.offset;
7261
7262                 inode->i_ctime = CURRENT_TIME;
7263                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7264                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7265                     (actual_len > inode->i_size) &&
7266                     (cur_offset > inode->i_size)) {
7267                         if (cur_offset > actual_len)
7268                                 i_size = actual_len;
7269                         else
7270                                 i_size = cur_offset;
7271                         i_size_write(inode, i_size);
7272                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7273                 }
7274
7275                 ret = btrfs_update_inode(trans, root, inode);
7276                 BUG_ON(ret);
7277
7278                 if (own_trans)
7279                         btrfs_end_transaction(trans, root);
7280         }
7281         return ret;
7282 }
7283
7284 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7285                               u64 start, u64 num_bytes, u64 min_size,
7286                               loff_t actual_len, u64 *alloc_hint)
7287 {
7288         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7289                                            min_size, actual_len, alloc_hint,
7290                                            NULL);
7291 }
7292
7293 int btrfs_prealloc_file_range_trans(struct inode *inode,
7294                                     struct btrfs_trans_handle *trans, int mode,
7295                                     u64 start, u64 num_bytes, u64 min_size,
7296                                     loff_t actual_len, u64 *alloc_hint)
7297 {
7298         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7299                                            min_size, actual_len, alloc_hint, trans);
7300 }
7301
7302 static int btrfs_set_page_dirty(struct page *page)
7303 {
7304         return __set_page_dirty_nobuffers(page);
7305 }
7306
7307 static int btrfs_permission(struct inode *inode, int mask)
7308 {
7309         struct btrfs_root *root = BTRFS_I(inode)->root;
7310         umode_t mode = inode->i_mode;
7311
7312         if (mask & MAY_WRITE &&
7313             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7314                 if (btrfs_root_readonly(root))
7315                         return -EROFS;
7316                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7317                         return -EACCES;
7318         }
7319         return generic_permission(inode, mask);
7320 }
7321
7322 static const struct inode_operations btrfs_dir_inode_operations = {
7323         .getattr        = btrfs_getattr,
7324         .lookup         = btrfs_lookup,
7325         .create         = btrfs_create,
7326         .unlink         = btrfs_unlink,
7327         .link           = btrfs_link,
7328         .mkdir          = btrfs_mkdir,
7329         .rmdir          = btrfs_rmdir,
7330         .rename         = btrfs_rename,
7331         .symlink        = btrfs_symlink,
7332         .setattr        = btrfs_setattr,
7333         .mknod          = btrfs_mknod,
7334         .setxattr       = btrfs_setxattr,
7335         .getxattr       = btrfs_getxattr,
7336         .listxattr      = btrfs_listxattr,
7337         .removexattr    = btrfs_removexattr,
7338         .permission     = btrfs_permission,
7339         .get_acl        = btrfs_get_acl,
7340 };
7341 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7342         .lookup         = btrfs_lookup,
7343         .permission     = btrfs_permission,
7344         .get_acl        = btrfs_get_acl,
7345 };
7346
7347 static const struct file_operations btrfs_dir_file_operations = {
7348         .llseek         = generic_file_llseek,
7349         .read           = generic_read_dir,
7350         .readdir        = btrfs_real_readdir,
7351         .unlocked_ioctl = btrfs_ioctl,
7352 #ifdef CONFIG_COMPAT
7353         .compat_ioctl   = btrfs_ioctl,
7354 #endif
7355         .release        = btrfs_release_file,
7356         .fsync          = btrfs_sync_file,
7357 };
7358
7359 static struct extent_io_ops btrfs_extent_io_ops = {
7360         .fill_delalloc = run_delalloc_range,
7361         .submit_bio_hook = btrfs_submit_bio_hook,
7362         .merge_bio_hook = btrfs_merge_bio_hook,
7363         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7364         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7365         .writepage_start_hook = btrfs_writepage_start_hook,
7366         .readpage_io_failed_hook = btrfs_io_failed_hook,
7367         .set_bit_hook = btrfs_set_bit_hook,
7368         .clear_bit_hook = btrfs_clear_bit_hook,
7369         .merge_extent_hook = btrfs_merge_extent_hook,
7370         .split_extent_hook = btrfs_split_extent_hook,
7371 };
7372
7373 /*
7374  * btrfs doesn't support the bmap operation because swapfiles
7375  * use bmap to make a mapping of extents in the file.  They assume
7376  * these extents won't change over the life of the file and they
7377  * use the bmap result to do IO directly to the drive.
7378  *
7379  * the btrfs bmap call would return logical addresses that aren't
7380  * suitable for IO and they also will change frequently as COW
7381  * operations happen.  So, swapfile + btrfs == corruption.
7382  *
7383  * For now we're avoiding this by dropping bmap.
7384  */
7385 static const struct address_space_operations btrfs_aops = {
7386         .readpage       = btrfs_readpage,
7387         .writepage      = btrfs_writepage,
7388         .writepages     = btrfs_writepages,
7389         .readpages      = btrfs_readpages,
7390         .direct_IO      = btrfs_direct_IO,
7391         .invalidatepage = btrfs_invalidatepage,
7392         .releasepage    = btrfs_releasepage,
7393         .set_page_dirty = btrfs_set_page_dirty,
7394         .error_remove_page = generic_error_remove_page,
7395 };
7396
7397 static const struct address_space_operations btrfs_symlink_aops = {
7398         .readpage       = btrfs_readpage,
7399         .writepage      = btrfs_writepage,
7400         .invalidatepage = btrfs_invalidatepage,
7401         .releasepage    = btrfs_releasepage,
7402 };
7403
7404 static const struct inode_operations btrfs_file_inode_operations = {
7405         .getattr        = btrfs_getattr,
7406         .setattr        = btrfs_setattr,
7407         .setxattr       = btrfs_setxattr,
7408         .getxattr       = btrfs_getxattr,
7409         .listxattr      = btrfs_listxattr,
7410         .removexattr    = btrfs_removexattr,
7411         .permission     = btrfs_permission,
7412         .fiemap         = btrfs_fiemap,
7413         .get_acl        = btrfs_get_acl,
7414 };
7415 static const struct inode_operations btrfs_special_inode_operations = {
7416         .getattr        = btrfs_getattr,
7417         .setattr        = btrfs_setattr,
7418         .permission     = btrfs_permission,
7419         .setxattr       = btrfs_setxattr,
7420         .getxattr       = btrfs_getxattr,
7421         .listxattr      = btrfs_listxattr,
7422         .removexattr    = btrfs_removexattr,
7423         .get_acl        = btrfs_get_acl,
7424 };
7425 static const struct inode_operations btrfs_symlink_inode_operations = {
7426         .readlink       = generic_readlink,
7427         .follow_link    = page_follow_link_light,
7428         .put_link       = page_put_link,
7429         .getattr        = btrfs_getattr,
7430         .permission     = btrfs_permission,
7431         .setxattr       = btrfs_setxattr,
7432         .getxattr       = btrfs_getxattr,
7433         .listxattr      = btrfs_listxattr,
7434         .removexattr    = btrfs_removexattr,
7435         .get_acl        = btrfs_get_acl,
7436 };
7437
7438 const struct dentry_operations btrfs_dentry_operations = {
7439         .d_delete       = btrfs_dentry_delete,
7440         .d_release      = btrfs_dentry_release,
7441 };