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 static inline bool is_free_space_inode(struct btrfs_root *root,
 754                                        struct inode *inode)
 755 {
 756         if (root == root->fs_info->tree_root ||
 757             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
 758                 return true;
 759         return false;
 760 }
 761
 762 /*
 763  * when extent_io.c finds a delayed allocation range in the file,
 764  * the call backs end up in this code.  The basic idea is to
 765  * allocate extents on disk for the range, and create ordered data structs
 766  * in ram to track those extents.
 767  *
 768  * locked_page is the page that writepage had locked already.  We use
 769  * it to make sure we don't do extra locks or unlocks.
 770  *
 771  * *page_started is set to one if we unlock locked_page and do everything
 772  * required to start IO on it.  It may be clean and already done with
 773  * IO when we return.
 774  */
 775 static noinline int cow_file_range(struct inode *inode,
 776                                    struct page *locked_page,
 777                                    u64 start, u64 end, int *page_started,
 778                                    unsigned long *nr_written,
 779                                    int unlock)
 780 {
 781         struct btrfs_root *root = BTRFS_I(inode)->root;
 782         struct btrfs_trans_handle *trans;
 783         u64 alloc_hint = 0;
 784         u64 num_bytes;
 785         unsigned long ram_size;
 786         u64 disk_num_bytes;
 787         u64 cur_alloc_size;
 788         u64 blocksize = root->sectorsize;
 789         struct btrfs_key ins;
 790         struct extent_map *em;
 791         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 792         int ret = 0;
 793
 794         BUG_ON(is_free_space_inode(root, inode));
 795         trans = btrfs_join_transaction(root);
 796         BUG_ON(IS_ERR(trans));
 797         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 798
 799         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 800         num_bytes = max(blocksize,  num_bytes);
 801         disk_num_bytes = num_bytes;
 802         ret = 0;
 803
 804         /* if this is a small write inside eof, kick off defrag */
 805         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 806                 btrfs_add_inode_defrag(trans, inode);
 807
 808         if (start == 0) {
 809                 /* lets try to make an inline extent */
 810                 ret = cow_file_range_inline(trans, root, inode,
 811                                             start, end, 0, 0, NULL);
 812                 if (ret == 0) {
 813                         extent_clear_unlock_delalloc(inode,
 814                                      &BTRFS_I(inode)->io_tree,
 815                                      start, end, NULL,
 816                                      EXTENT_CLEAR_UNLOCK_PAGE |
 817                                      EXTENT_CLEAR_UNLOCK |
 818                                      EXTENT_CLEAR_DELALLOC |
 819                                      EXTENT_CLEAR_DIRTY |
 820                                      EXTENT_SET_WRITEBACK |
 821                                      EXTENT_END_WRITEBACK);
 822
 823                         *nr_written = *nr_written +
 824                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 825                         *page_started = 1;
 826                         ret = 0;
 827                         goto out;
 828                 }
 829         }
 830
 831         BUG_ON(disk_num_bytes >
 832                btrfs_super_total_bytes(&root->fs_info->super_copy));
 833
 834         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 835         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 836
 837         while (disk_num_bytes > 0) {
 838                 unsigned long op;
 839
 840                 cur_alloc_size = disk_num_bytes;
 841                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 842                                            root->sectorsize, 0, alloc_hint,
 843                                            (u64)-1, &ins, 1);
 844                 BUG_ON(ret);
 845
 846                 em = alloc_extent_map();
 847                 BUG_ON(!em);
 848                 em->start = start;
 849                 em->orig_start = em->start;
 850                 ram_size = ins.offset;
 851                 em->len = ins.offset;
 852
 853                 em->block_start = ins.objectid;
 854                 em->block_len = ins.offset;
 855                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 856                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 857
 858                 while (1) {
 859                         write_lock(&em_tree->lock);
 860                         ret = add_extent_mapping(em_tree, em);
 861                         write_unlock(&em_tree->lock);
 862                         if (ret != -EEXIST) {
 863                                 free_extent_map(em);
 864                                 break;
 865                         }
 866                         btrfs_drop_extent_cache(inode, start,
 867                                                 start + ram_size - 1, 0);
 868                 }
 869
 870                 cur_alloc_size = ins.offset;
 871                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 872                                                ram_size, cur_alloc_size, 0);
 873                 BUG_ON(ret);
 874
 875                 if (root->root_key.objectid ==
 876                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 877                         ret = btrfs_reloc_clone_csums(inode, start,
 878                                                       cur_alloc_size);
 879                         BUG_ON(ret);
 880                 }
 881
 882                 if (disk_num_bytes < cur_alloc_size)
 883                         break;
 884
 885                 /* we're not doing compressed IO, don't unlock the first
 886                  * page (which the caller expects to stay locked), don't
 887                  * clear any dirty bits and don't set any writeback bits
 888                  *
 889                  * Do set the Private2 bit so we know this page was properly
 890                  * setup for writepage
 891                  */
 892                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 893                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 894                         EXTENT_SET_PRIVATE2;
 895
 896                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 897                                              start, start + ram_size - 1,
 898                                              locked_page, op);
 899                 disk_num_bytes -= cur_alloc_size;
 900                 num_bytes -= cur_alloc_size;
 901                 alloc_hint = ins.objectid + ins.offset;
 902                 start += cur_alloc_size;
 903         }
 904 out:
 905         ret = 0;
 906         btrfs_end_transaction(trans, root);
 907
 908         return ret;
 909 }
 910
 911 /*
 912  * work queue call back to started compression on a file and pages
 913  */
 914 static noinline void async_cow_start(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         int num_added = 0;
 918         async_cow = container_of(work, struct async_cow, work);
 919
 920         compress_file_range(async_cow->inode, async_cow->locked_page,
 921                             async_cow->start, async_cow->end, async_cow,
 922                             &num_added);
 923         if (num_added == 0)
 924                 async_cow->inode = NULL;
 925 }
 926
 927 /*
 928  * work queue call back to submit previously compressed pages
 929  */
 930 static noinline void async_cow_submit(struct btrfs_work *work)
 931 {
 932         struct async_cow *async_cow;
 933         struct btrfs_root *root;
 934         unsigned long nr_pages;
 935
 936         async_cow = container_of(work, struct async_cow, work);
 937
 938         root = async_cow->root;
 939         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 940                 PAGE_CACHE_SHIFT;
 941
 942         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 943
 944         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 945             5 * 1042 * 1024 &&
 946             waitqueue_active(&root->fs_info->async_submit_wait))
 947                 wake_up(&root->fs_info->async_submit_wait);
 948
 949         if (async_cow->inode)
 950                 submit_compressed_extents(async_cow->inode, async_cow);
 951 }
 952
 953 static noinline void async_cow_free(struct btrfs_work *work)
 954 {
 955         struct async_cow *async_cow;
 956         async_cow = container_of(work, struct async_cow, work);
 957         kfree(async_cow);
 958 }
 959
 960 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 961                                 u64 start, u64 end, int *page_started,
 962                                 unsigned long *nr_written)
 963 {
 964         struct async_cow *async_cow;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         unsigned long nr_pages;
 967         u64 cur_end;
 968         int limit = 10 * 1024 * 1042;
 969
 970         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 971                          1, 0, NULL, GFP_NOFS);
 972         while (start < end) {
 973                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 974                 BUG_ON(!async_cow);
 975                 async_cow->inode = inode;
 976                 async_cow->root = root;
 977                 async_cow->locked_page = locked_page;
 978                 async_cow->start = start;
 979
 980                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 981                         cur_end = end;
 982                 else
 983                         cur_end = min(end, start + 512 * 1024 - 1);
 984
 985                 async_cow->end = cur_end;
 986                 INIT_LIST_HEAD(&async_cow->extents);
 987
 988                 async_cow->work.func = async_cow_start;
 989                 async_cow->work.ordered_func = async_cow_submit;
 990                 async_cow->work.ordered_free = async_cow_free;
 991                 async_cow->work.flags = 0;
 992
 993                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 994                         PAGE_CACHE_SHIFT;
 995                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 996
 997                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 998                                    &async_cow->work);
 999
1000                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1001                         wait_event(root->fs_info->async_submit_wait,
1002                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1003                             limit));
1004                 }
1005
1006                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1007                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1008                         wait_event(root->fs_info->async_submit_wait,
1009                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010                            0));
1011                 }
1012
1013                 *nr_written += nr_pages;
1014                 start = cur_end + 1;
1015         }
1016         *page_started = 1;
1017         return 0;
1018 }
1019
1020 static noinline int csum_exist_in_range(struct btrfs_root *root,
1021                                         u64 bytenr, u64 num_bytes)
1022 {
1023         int ret;
1024         struct btrfs_ordered_sum *sums;
1025         LIST_HEAD(list);
1026
1027         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1028                                        bytenr + num_bytes - 1, &list, 0);
1029         if (ret == 0 && list_empty(&list))
1030                 return 0;
1031
1032         while (!list_empty(&list)) {
1033                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1034                 list_del(&sums->list);
1035                 kfree(sums);
1036         }
1037         return 1;
1038 }
1039
1040 /*
1041  * when nowcow writeback call back.  This checks for snapshots or COW copies
1042  * of the extents that exist in the file, and COWs the file as required.
1043  *
1044  * If no cow copies or snapshots exist, we write directly to the existing
1045  * blocks on disk
1046  */
1047 static noinline int run_delalloc_nocow(struct inode *inode,
1048                                        struct page *locked_page,
1049                               u64 start, u64 end, int *page_started, int force,
1050                               unsigned long *nr_written)
1051 {
1052         struct btrfs_root *root = BTRFS_I(inode)->root;
1053         struct btrfs_trans_handle *trans;
1054         struct extent_buffer *leaf;
1055         struct btrfs_path *path;
1056         struct btrfs_file_extent_item *fi;
1057         struct btrfs_key found_key;
1058         u64 cow_start;
1059         u64 cur_offset;
1060         u64 extent_end;
1061         u64 extent_offset;
1062         u64 disk_bytenr;
1063         u64 num_bytes;
1064         int extent_type;
1065         int ret;
1066         int type;
1067         int nocow;
1068         int check_prev = 1;
1069         bool nolock;
1070         u64 ino = btrfs_ino(inode);
1071
1072         path = btrfs_alloc_path();
1073         BUG_ON(!path);
1074
1075         nolock = is_free_space_inode(root, inode);
1076
1077         if (nolock)
1078                 trans = btrfs_join_transaction_nolock(root);
1079         else
1080                 trans = btrfs_join_transaction(root);
1081
1082         BUG_ON(IS_ERR(trans));
1083         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1084
1085         cow_start = (u64)-1;
1086         cur_offset = start;
1087         while (1) {
1088                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089                                                cur_offset, 0);
1090                 BUG_ON(ret < 0);
1091                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1092                         leaf = path->nodes[0];
1093                         btrfs_item_key_to_cpu(leaf, &found_key,
1094                                               path->slots[0] - 1);
1095                         if (found_key.objectid == ino &&
1096                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1097                                 path->slots[0]--;
1098                 }
1099                 check_prev = 0;
1100 next_slot:
1101                 leaf = path->nodes[0];
1102                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1103                         ret = btrfs_next_leaf(root, path);
1104                         if (ret < 0)
1105                                 BUG_ON(1);
1106                         if (ret > 0)
1107                                 break;
1108                         leaf = path->nodes[0];
1109                 }
1110
1111                 nocow = 0;
1112                 disk_bytenr = 0;
1113                 num_bytes = 0;
1114                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1115
1116                 if (found_key.objectid > ino ||
1117                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1118                     found_key.offset > end)
1119                         break;
1120
1121                 if (found_key.offset > cur_offset) {
1122                         extent_end = found_key.offset;
1123                         extent_type = 0;
1124                         goto out_check;
1125                 }
1126
1127                 fi = btrfs_item_ptr(leaf, path->slots[0],
1128                                     struct btrfs_file_extent_item);
1129                 extent_type = btrfs_file_extent_type(leaf, fi);
1130
1131                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1132                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1133                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1134                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1135                         extent_end = found_key.offset +
1136                                 btrfs_file_extent_num_bytes(leaf, fi);
1137                         if (extent_end <= start) {
1138                                 path->slots[0]++;
1139                                 goto next_slot;
1140                         }
1141                         if (disk_bytenr == 0)
1142                                 goto out_check;
1143                         if (btrfs_file_extent_compression(leaf, fi) ||
1144                             btrfs_file_extent_encryption(leaf, fi) ||
1145                             btrfs_file_extent_other_encoding(leaf, fi))
1146                                 goto out_check;
1147                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1148                                 goto out_check;
1149                         if (btrfs_extent_readonly(root, disk_bytenr))
1150                                 goto out_check;
1151                         if (btrfs_cross_ref_exist(trans, root, ino,
1152                                                   found_key.offset -
1153                                                   extent_offset, disk_bytenr))
1154                                 goto out_check;
1155                         disk_bytenr += extent_offset;
1156                         disk_bytenr += cur_offset - found_key.offset;
1157                         num_bytes = min(end + 1, extent_end) - cur_offset;
1158                         /*
1159                          * force cow if csum exists in the range.
1160                          * this ensure that csum for a given extent are
1161                          * either valid or do not exist.
1162                          */
1163                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164                                 goto out_check;
1165                         nocow = 1;
1166                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1167                         extent_end = found_key.offset +
1168                                 btrfs_file_extent_inline_len(leaf, fi);
1169                         extent_end = ALIGN(extent_end, root->sectorsize);
1170                 } else {
1171                         BUG_ON(1);
1172                 }
1173 out_check:
1174                 if (extent_end <= start) {
1175                         path->slots[0]++;
1176                         goto next_slot;
1177                 }
1178                 if (!nocow) {
1179                         if (cow_start == (u64)-1)
1180                                 cow_start = cur_offset;
1181                         cur_offset = extent_end;
1182                         if (cur_offset > end)
1183                                 break;
1184                         path->slots[0]++;
1185                         goto next_slot;
1186                 }
1187
1188                 btrfs_release_path(path);
1189                 if (cow_start != (u64)-1) {
1190                         ret = cow_file_range(inode, locked_page, cow_start,
1191                                         found_key.offset - 1, page_started,
1192                                         nr_written, 1);
1193                         BUG_ON(ret);
1194                         cow_start = (u64)-1;
1195                 }
1196
1197                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1198                         struct extent_map *em;
1199                         struct extent_map_tree *em_tree;
1200                         em_tree = &BTRFS_I(inode)->extent_tree;
1201                         em = alloc_extent_map();
1202                         BUG_ON(!em);
1203                         em->start = cur_offset;
1204                         em->orig_start = em->start;
1205                         em->len = num_bytes;
1206                         em->block_len = num_bytes;
1207                         em->block_start = disk_bytenr;
1208                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1209                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1210                         while (1) {
1211                                 write_lock(&em_tree->lock);
1212                                 ret = add_extent_mapping(em_tree, em);
1213                                 write_unlock(&em_tree->lock);
1214                                 if (ret != -EEXIST) {
1215                                         free_extent_map(em);
1216                                         break;
1217                                 }
1218                                 btrfs_drop_extent_cache(inode, em->start,
1219                                                 em->start + em->len - 1, 0);
1220                         }
1221                         type = BTRFS_ORDERED_PREALLOC;
1222                 } else {
1223                         type = BTRFS_ORDERED_NOCOW;
1224                 }
1225
1226                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1227                                                num_bytes, num_bytes, type);
1228                 BUG_ON(ret);
1229
1230                 if (root->root_key.objectid ==
1231                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1232                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1233                                                       num_bytes);
1234                         BUG_ON(ret);
1235                 }
1236
1237                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1238                                 cur_offset, cur_offset + num_bytes - 1,
1239                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1240                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1241                                 EXTENT_SET_PRIVATE2);
1242                 cur_offset = extent_end;
1243                 if (cur_offset > end)
1244                         break;
1245         }
1246         btrfs_release_path(path);
1247
1248         if (cur_offset <= end && cow_start == (u64)-1)
1249                 cow_start = cur_offset;
1250         if (cow_start != (u64)-1) {
1251                 ret = cow_file_range(inode, locked_page, cow_start, end,
1252                                      page_started, nr_written, 1);
1253                 BUG_ON(ret);
1254         }
1255
1256         if (nolock) {
1257                 ret = btrfs_end_transaction_nolock(trans, root);
1258                 BUG_ON(ret);
1259         } else {
1260                 ret = btrfs_end_transaction(trans, root);
1261                 BUG_ON(ret);
1262         }
1263         btrfs_free_path(path);
1264         return 0;
1265 }
1266
1267 /*
1268  * extent_io.c call back to do delayed allocation processing
1269  */
1270 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1271                               u64 start, u64 end, int *page_started,
1272                               unsigned long *nr_written)
1273 {
1274         int ret;
1275         struct btrfs_root *root = BTRFS_I(inode)->root;
1276
1277         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1278                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1279                                          page_started, 1, nr_written);
1280         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1281                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1282                                          page_started, 0, nr_written);
1283         else if (!btrfs_test_opt(root, COMPRESS) &&
1284                  !(BTRFS_I(inode)->force_compress) &&
1285                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1286                 ret = cow_file_range(inode, locked_page, start, end,
1287                                       page_started, nr_written, 1);
1288         else
1289                 ret = cow_file_range_async(inode, locked_page, start, end,
1290                                            page_started, nr_written);
1291         return ret;
1292 }
1293
1294 static int btrfs_split_extent_hook(struct inode *inode,
1295                                    struct extent_state *orig, u64 split)
1296 {
1297         /* not delalloc, ignore it */
1298         if (!(orig->state & EXTENT_DELALLOC))
1299                 return 0;
1300
1301         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1302         return 0;
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static int btrfs_merge_extent_hook(struct inode *inode,
1312                                    struct extent_state *new,
1313                                    struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return 0;
1318
1319         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1320         return 0;
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static int btrfs_set_bit_hook(struct inode *inode,
1329                               struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC)
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 else
1345                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1346
1347                 spin_lock(&root->fs_info->delalloc_lock);
1348                 BTRFS_I(inode)->delalloc_bytes += len;
1349                 root->fs_info->delalloc_bytes += len;
1350                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1351                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1352                                       &root->fs_info->delalloc_inodes);
1353                 }
1354                 spin_unlock(&root->fs_info->delalloc_lock);
1355         }
1356         return 0;
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static int btrfs_clear_bit_hook(struct inode *inode,
1363                                 struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC)
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1378                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1379
1380                 if (*bits & EXTENT_DO_ACCOUNTING)
1381                         btrfs_delalloc_release_metadata(inode, len);
1382
1383                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1384                     && do_list)
1385                         btrfs_free_reserved_data_space(inode, len);
1386
1387                 spin_lock(&root->fs_info->delalloc_lock);
1388                 root->fs_info->delalloc_bytes -= len;
1389                 BTRFS_I(inode)->delalloc_bytes -= len;
1390
1391                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1392                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1393                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1394                 }
1395                 spin_unlock(&root->fs_info->delalloc_lock);
1396         }
1397         return 0;
1398 }
1399
1400 /*
1401  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1402  * we don't create bios that span stripes or chunks
1403  */
1404 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1405                          size_t size, struct bio *bio,
1406                          unsigned long bio_flags)
1407 {
1408         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1409         struct btrfs_mapping_tree *map_tree;
1410         u64 logical = (u64)bio->bi_sector << 9;
1411         u64 length = 0;
1412         u64 map_length;
1413         int ret;
1414
1415         if (bio_flags & EXTENT_BIO_COMPRESSED)
1416                 return 0;
1417
1418         length = bio->bi_size;
1419         map_tree = &root->fs_info->mapping_tree;
1420         map_length = length;
1421         ret = btrfs_map_block(map_tree, READ, logical,
1422                               &map_length, NULL, 0);
1423
1424         if (map_length < length + size)
1425                 return 1;
1426         return ret;
1427 }
1428
1429 /*
1430  * in order to insert checksums into the metadata in large chunks,
1431  * we wait until bio submission time.   All the pages in the bio are
1432  * checksummed and sums are attached onto the ordered extent record.
1433  *
1434  * At IO completion time the cums attached on the ordered extent record
1435  * are inserted into the btree
1436  */
1437 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1438                                     struct bio *bio, int mirror_num,
1439                                     unsigned long bio_flags,
1440                                     u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         int ret = 0;
1444
1445         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1446         BUG_ON(ret);
1447         return 0;
1448 }
1449
1450 /*
1451  * in order to insert checksums into the metadata in large chunks,
1452  * we wait until bio submission time.   All the pages in the bio are
1453  * checksummed and sums are attached onto the ordered extent record.
1454  *
1455  * At IO completion time the cums attached on the ordered extent record
1456  * are inserted into the btree
1457  */
1458 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1459                           int mirror_num, unsigned long bio_flags,
1460                           u64 bio_offset)
1461 {
1462         struct btrfs_root *root = BTRFS_I(inode)->root;
1463         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1464 }
1465
1466 /*
1467  * extent_io.c submission hook. This does the right thing for csum calculation
1468  * on write, or reading the csums from the tree before a read
1469  */
1470 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1471                           int mirror_num, unsigned long bio_flags,
1472                           u64 bio_offset)
1473 {
1474         struct btrfs_root *root = BTRFS_I(inode)->root;
1475         int ret = 0;
1476         int skip_sum;
1477
1478         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1479
1480         if (is_free_space_inode(root, inode))
1481                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1482         else
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1484         BUG_ON(ret);
1485
1486         if (!(rw & REQ_WRITE)) {
1487                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1488                         return btrfs_submit_compressed_read(inode, bio,
1489                                                     mirror_num, bio_flags);
1490                 } else if (!skip_sum) {
1491                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1492                         if (ret)
1493                                 return ret;
1494                 }
1495                 goto mapit;
1496         } else if (!skip_sum) {
1497                 /* csum items have already been cloned */
1498                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1499                         goto mapit;
1500                 /* we're doing a write, do the async checksumming */
1501                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1502                                    inode, rw, bio, mirror_num,
1503                                    bio_flags, bio_offset,
1504                                    __btrfs_submit_bio_start,
1505                                    __btrfs_submit_bio_done);
1506         }
1507
1508 mapit:
1509         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1510 }
1511
1512 /*
1513  * given a list of ordered sums record them in the inode.  This happens
1514  * at IO completion time based on sums calculated at bio submission time.
1515  */
1516 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1517                              struct inode *inode, u64 file_offset,
1518                              struct list_head *list)
1519 {
1520         struct btrfs_ordered_sum *sum;
1521
1522         list_for_each_entry(sum, list, list) {
1523                 btrfs_csum_file_blocks(trans,
1524                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1525         }
1526         return 0;
1527 }
1528
1529 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1530                               struct extent_state **cached_state)
1531 {
1532         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1533                 WARN_ON(1);
1534         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1535                                    cached_state, GFP_NOFS);
1536 }
1537
1538 /* see btrfs_writepage_start_hook for details on why this is required */
1539 struct btrfs_writepage_fixup {
1540         struct page *page;
1541         struct btrfs_work work;
1542 };
1543
1544 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1545 {
1546         struct btrfs_writepage_fixup *fixup;
1547         struct btrfs_ordered_extent *ordered;
1548         struct extent_state *cached_state = NULL;
1549         struct page *page;
1550         struct inode *inode;
1551         u64 page_start;
1552         u64 page_end;
1553
1554         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1555         page = fixup->page;
1556 again:
1557         lock_page(page);
1558         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1559                 ClearPageChecked(page);
1560                 goto out_page;
1561         }
1562
1563         inode = page->mapping->host;
1564         page_start = page_offset(page);
1565         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1566
1567         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1568                          &cached_state, GFP_NOFS);
1569
1570         /* already ordered? We're done */
1571         if (PagePrivate2(page))
1572                 goto out;
1573
1574         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1575         if (ordered) {
1576                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1577                                      page_end, &cached_state, GFP_NOFS);
1578                 unlock_page(page);
1579                 btrfs_start_ordered_extent(inode, ordered, 1);
1580                 goto again;
1581         }
1582
1583         BUG();
1584         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1585         ClearPageChecked(page);
1586 out:
1587         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1588                              &cached_state, GFP_NOFS);
1589 out_page:
1590         unlock_page(page);
1591         page_cache_release(page);
1592         kfree(fixup);
1593 }
1594
1595 /*
1596  * There are a few paths in the higher layers of the kernel that directly
1597  * set the page dirty bit without asking the filesystem if it is a
1598  * good idea.  This causes problems because we want to make sure COW
1599  * properly happens and the data=ordered rules are followed.
1600  *
1601  * In our case any range that doesn't have the ORDERED bit set
1602  * hasn't been properly setup for IO.  We kick off an async process
1603  * to fix it up.  The async helper will wait for ordered extents, set
1604  * the delalloc bit and make it safe to write the page.
1605  */
1606 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1607 {
1608         struct inode *inode = page->mapping->host;
1609         struct btrfs_writepage_fixup *fixup;
1610         struct btrfs_root *root = BTRFS_I(inode)->root;
1611
1612         /* this page is properly in the ordered list */
1613         if (TestClearPagePrivate2(page))
1614                 return 0;
1615
1616         if (PageChecked(page))
1617                 return -EAGAIN;
1618
1619         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1620         if (!fixup)
1621                 return -EAGAIN;
1622
1623         SetPageChecked(page);
1624         page_cache_get(page);
1625         fixup->work.func = btrfs_writepage_fixup_worker;
1626         fixup->page = page;
1627         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1628         return -EAGAIN;
1629 }
1630
1631 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1632                                        struct inode *inode, u64 file_pos,
1633                                        u64 disk_bytenr, u64 disk_num_bytes,
1634                                        u64 num_bytes, u64 ram_bytes,
1635                                        u8 compression, u8 encryption,
1636                                        u16 other_encoding, int extent_type)
1637 {
1638         struct btrfs_root *root = BTRFS_I(inode)->root;
1639         struct btrfs_file_extent_item *fi;
1640         struct btrfs_path *path;
1641         struct extent_buffer *leaf;
1642         struct btrfs_key ins;
1643         u64 hint;
1644         int ret;
1645
1646         path = btrfs_alloc_path();
1647         BUG_ON(!path);
1648
1649         path->leave_spinning = 1;
1650
1651         /*
1652          * we may be replacing one extent in the tree with another.
1653          * The new extent is pinned in the extent map, and we don't want
1654          * to drop it from the cache until it is completely in the btree.
1655          *
1656          * So, tell btrfs_drop_extents to leave this extent in the cache.
1657          * the caller is expected to unpin it and allow it to be merged
1658          * with the others.
1659          */
1660         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1661                                  &hint, 0);
1662         BUG_ON(ret);
1663
1664         ins.objectid = btrfs_ino(inode);
1665         ins.offset = file_pos;
1666         ins.type = BTRFS_EXTENT_DATA_KEY;
1667         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1668         BUG_ON(ret);
1669         leaf = path->nodes[0];
1670         fi = btrfs_item_ptr(leaf, path->slots[0],
1671                             struct btrfs_file_extent_item);
1672         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1673         btrfs_set_file_extent_type(leaf, fi, extent_type);
1674         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1675         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1676         btrfs_set_file_extent_offset(leaf, fi, 0);
1677         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1678         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1679         btrfs_set_file_extent_compression(leaf, fi, compression);
1680         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1681         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1682
1683         btrfs_unlock_up_safe(path, 1);
1684         btrfs_set_lock_blocking(leaf);
1685
1686         btrfs_mark_buffer_dirty(leaf);
1687
1688         inode_add_bytes(inode, num_bytes);
1689
1690         ins.objectid = disk_bytenr;
1691         ins.offset = disk_num_bytes;
1692         ins.type = BTRFS_EXTENT_ITEM_KEY;
1693         ret = btrfs_alloc_reserved_file_extent(trans, root,
1694                                         root->root_key.objectid,
1695                                         btrfs_ino(inode), file_pos, &ins);
1696         BUG_ON(ret);
1697         btrfs_free_path(path);
1698
1699         return 0;
1700 }
1701
1702 /*
1703  * helper function for btrfs_finish_ordered_io, this
1704  * just reads in some of the csum leaves to prime them into ram
1705  * before we start the transaction.  It limits the amount of btree
1706  * reads required while inside the transaction.
1707  */
1708 /* as ordered data IO finishes, this gets called so we can finish
1709  * an ordered extent if the range of bytes in the file it covers are
1710  * fully written.
1711  */
1712 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1713 {
1714         struct btrfs_root *root = BTRFS_I(inode)->root;
1715         struct btrfs_trans_handle *trans = NULL;
1716         struct btrfs_ordered_extent *ordered_extent = NULL;
1717         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1718         struct extent_state *cached_state = NULL;
1719         int compress_type = 0;
1720         int ret;
1721         bool nolock;
1722
1723         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1724                                              end - start + 1);
1725         if (!ret)
1726                 return 0;
1727         BUG_ON(!ordered_extent);
1728
1729         nolock = is_free_space_inode(root, inode);
1730
1731         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1732                 BUG_ON(!list_empty(&ordered_extent->list));
1733                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1734                 if (!ret) {
1735                         if (nolock)
1736                                 trans = btrfs_join_transaction_nolock(root);
1737                         else
1738                                 trans = btrfs_join_transaction(root);
1739                         BUG_ON(IS_ERR(trans));
1740                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741                         ret = btrfs_update_inode(trans, root, inode);
1742                         BUG_ON(ret);
1743                 }
1744                 goto out;
1745         }
1746
1747         lock_extent_bits(io_tree, ordered_extent->file_offset,
1748                          ordered_extent->file_offset + ordered_extent->len - 1,
1749                          0, &cached_state, GFP_NOFS);
1750
1751         if (nolock)
1752                 trans = btrfs_join_transaction_nolock(root);
1753         else
1754                 trans = btrfs_join_transaction(root);
1755         BUG_ON(IS_ERR(trans));
1756         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1757
1758         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1759                 compress_type = ordered_extent->compress_type;
1760         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1761                 BUG_ON(compress_type);
1762                 ret = btrfs_mark_extent_written(trans, inode,
1763                                                 ordered_extent->file_offset,
1764                                                 ordered_extent->file_offset +
1765                                                 ordered_extent->len);
1766                 BUG_ON(ret);
1767         } else {
1768                 BUG_ON(root == root->fs_info->tree_root);
1769                 ret = insert_reserved_file_extent(trans, inode,
1770                                                 ordered_extent->file_offset,
1771                                                 ordered_extent->start,
1772                                                 ordered_extent->disk_len,
1773                                                 ordered_extent->len,
1774                                                 ordered_extent->len,
1775                                                 compress_type, 0, 0,
1776                                                 BTRFS_FILE_EXTENT_REG);
1777                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1778                                    ordered_extent->file_offset,
1779                                    ordered_extent->len);
1780                 BUG_ON(ret);
1781         }
1782         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1783                              ordered_extent->file_offset +
1784                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1785
1786         add_pending_csums(trans, inode, ordered_extent->file_offset,
1787                           &ordered_extent->list);
1788
1789         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1790         if (!ret) {
1791                 ret = btrfs_update_inode(trans, root, inode);
1792                 BUG_ON(ret);
1793         }
1794         ret = 0;
1795 out:
1796         if (nolock) {
1797                 if (trans)
1798                         btrfs_end_transaction_nolock(trans, root);
1799         } else {
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801                 if (trans)
1802                         btrfs_end_transaction(trans, root);
1803         }
1804
1805         /* once for us */
1806         btrfs_put_ordered_extent(ordered_extent);
1807         /* once for the tree */
1808         btrfs_put_ordered_extent(ordered_extent);
1809
1810         return 0;
1811 }
1812
1813 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1814                                 struct extent_state *state, int uptodate)
1815 {
1816         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1817
1818         ClearPagePrivate2(page);
1819         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1820 }
1821
1822 /*
1823  * When IO fails, either with EIO or csum verification fails, we
1824  * try other mirrors that might have a good copy of the data.  This
1825  * io_failure_record is used to record state as we go through all the
1826  * mirrors.  If another mirror has good data, the page is set up to date
1827  * and things continue.  If a good mirror can't be found, the original
1828  * bio end_io callback is called to indicate things have failed.
1829  */
1830 struct io_failure_record {
1831         struct page *page;
1832         u64 start;
1833         u64 len;
1834         u64 logical;
1835         unsigned long bio_flags;
1836         int last_mirror;
1837 };
1838
1839 static int btrfs_io_failed_hook(struct bio *failed_bio,
1840                          struct page *page, u64 start, u64 end,
1841                          struct extent_state *state)
1842 {
1843         struct io_failure_record *failrec = NULL;
1844         u64 private;
1845         struct extent_map *em;
1846         struct inode *inode = page->mapping->host;
1847         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1848         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1849         struct bio *bio;
1850         int num_copies;
1851         int ret;
1852         int rw;
1853         u64 logical;
1854
1855         ret = get_state_private(failure_tree, start, &private);
1856         if (ret) {
1857                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1858                 if (!failrec)
1859                         return -ENOMEM;
1860                 failrec->start = start;
1861                 failrec->len = end - start + 1;
1862                 failrec->last_mirror = 0;
1863                 failrec->bio_flags = 0;
1864
1865                 read_lock(&em_tree->lock);
1866                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1867                 if (em->start > start || em->start + em->len < start) {
1868                         free_extent_map(em);
1869                         em = NULL;
1870                 }
1871                 read_unlock(&em_tree->lock);
1872
1873                 if (IS_ERR_OR_NULL(em)) {
1874                         kfree(failrec);
1875                         return -EIO;
1876                 }
1877                 logical = start - em->start;
1878                 logical = em->block_start + logical;
1879                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1880                         logical = em->block_start;
1881                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1882                         extent_set_compress_type(&failrec->bio_flags,
1883                                                  em->compress_type);
1884                 }
1885                 failrec->logical = logical;
1886                 free_extent_map(em);
1887                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1888                                 EXTENT_DIRTY, GFP_NOFS);
1889                 set_state_private(failure_tree, start,
1890                                  (u64)(unsigned long)failrec);
1891         } else {
1892                 failrec = (struct io_failure_record *)(unsigned long)private;
1893         }
1894         num_copies = btrfs_num_copies(
1895                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1896                               failrec->logical, failrec->len);
1897         failrec->last_mirror++;
1898         if (!state) {
1899                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1900                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1901                                                     failrec->start,
1902                                                     EXTENT_LOCKED);
1903                 if (state && state->start != failrec->start)
1904                         state = NULL;
1905                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1906         }
1907         if (!state || failrec->last_mirror > num_copies) {
1908                 set_state_private(failure_tree, failrec->start, 0);
1909                 clear_extent_bits(failure_tree, failrec->start,
1910                                   failrec->start + failrec->len - 1,
1911                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1912                 kfree(failrec);
1913                 return -EIO;
1914         }
1915         bio = bio_alloc(GFP_NOFS, 1);
1916         bio->bi_private = state;
1917         bio->bi_end_io = failed_bio->bi_end_io;
1918         bio->bi_sector = failrec->logical >> 9;
1919         bio->bi_bdev = failed_bio->bi_bdev;
1920         bio->bi_size = 0;
1921
1922         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1923         if (failed_bio->bi_rw & REQ_WRITE)
1924                 rw = WRITE;
1925         else
1926                 rw = READ;
1927
1928         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1929                                                       failrec->last_mirror,
1930                                                       failrec->bio_flags, 0);
1931         return ret;
1932 }
1933
1934 /*
1935  * each time an IO finishes, we do a fast check in the IO failure tree
1936  * to see if we need to process or clean up an io_failure_record
1937  */
1938 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1939 {
1940         u64 private;
1941         u64 private_failure;
1942         struct io_failure_record *failure;
1943         int ret;
1944
1945         private = 0;
1946         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1947                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1948                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1949                                         start, &private_failure);
1950                 if (ret == 0) {
1951                         failure = (struct io_failure_record *)(unsigned long)
1952                                    private_failure;
1953                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1954                                           failure->start, 0);
1955                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1956                                           failure->start,
1957                                           failure->start + failure->len - 1,
1958                                           EXTENT_DIRTY | EXTENT_LOCKED,
1959                                           GFP_NOFS);
1960                         kfree(failure);
1961                 }
1962         }
1963         return 0;
1964 }
1965
1966 /*
1967  * when reads are done, we need to check csums to verify the data is correct
1968  * if there's a match, we allow the bio to finish.  If not, we go through
1969  * the io_failure_record routines to find good copies
1970  */
1971 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1972                                struct extent_state *state)
1973 {
1974         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1975         struct inode *inode = page->mapping->host;
1976         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1977         char *kaddr;
1978         u64 private = ~(u32)0;
1979         int ret;
1980         struct btrfs_root *root = BTRFS_I(inode)->root;
1981         u32 csum = ~(u32)0;
1982
1983         if (PageChecked(page)) {
1984                 ClearPageChecked(page);
1985                 goto good;
1986         }
1987
1988         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1989                 goto good;
1990
1991         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1992             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1993                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1994                                   GFP_NOFS);
1995                 return 0;
1996         }
1997
1998         if (state && state->start == start) {
1999                 private = state->private;
2000                 ret = 0;
2001         } else {
2002                 ret = get_state_private(io_tree, start, &private);
2003         }
2004         kaddr = kmap_atomic(page, KM_USER0);
2005         if (ret)
2006                 goto zeroit;
2007
2008         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2009         btrfs_csum_final(csum, (char *)&csum);
2010         if (csum != private)
2011                 goto zeroit;
2012
2013         kunmap_atomic(kaddr, KM_USER0);
2014 good:
2015         /* if the io failure tree for this inode is non-empty,
2016          * check to see if we've recovered from a failed IO
2017          */
2018         btrfs_clean_io_failures(inode, start);
2019         return 0;
2020
2021 zeroit:
2022         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2023                        "private %llu\n",
2024                        (unsigned long long)btrfs_ino(page->mapping->host),
2025                        (unsigned long long)start, csum,
2026                        (unsigned long long)private);
2027         memset(kaddr + offset, 1, end - start + 1);
2028         flush_dcache_page(page);
2029         kunmap_atomic(kaddr, KM_USER0);
2030         if (private == 0)
2031                 return 0;
2032         return -EIO;
2033 }
2034
2035 struct delayed_iput {
2036         struct list_head list;
2037         struct inode *inode;
2038 };
2039
2040 void btrfs_add_delayed_iput(struct inode *inode)
2041 {
2042         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2043         struct delayed_iput *delayed;
2044
2045         if (atomic_add_unless(&inode->i_count, -1, 1))
2046                 return;
2047
2048         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2049         delayed->inode = inode;
2050
2051         spin_lock(&fs_info->delayed_iput_lock);
2052         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2053         spin_unlock(&fs_info->delayed_iput_lock);
2054 }
2055
2056 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2057 {
2058         LIST_HEAD(list);
2059         struct btrfs_fs_info *fs_info = root->fs_info;
2060         struct delayed_iput *delayed;
2061         int empty;
2062
2063         spin_lock(&fs_info->delayed_iput_lock);
2064         empty = list_empty(&fs_info->delayed_iputs);
2065         spin_unlock(&fs_info->delayed_iput_lock);
2066         if (empty)
2067                 return;
2068
2069         down_read(&root->fs_info->cleanup_work_sem);
2070         spin_lock(&fs_info->delayed_iput_lock);
2071         list_splice_init(&fs_info->delayed_iputs, &list);
2072         spin_unlock(&fs_info->delayed_iput_lock);
2073
2074         while (!list_empty(&list)) {
2075                 delayed = list_entry(list.next, struct delayed_iput, list);
2076                 list_del(&delayed->list);
2077                 iput(delayed->inode);
2078                 kfree(delayed);
2079         }
2080         up_read(&root->fs_info->cleanup_work_sem);
2081 }
2082
2083 /*
2084  * calculate extra metadata reservation when snapshotting a subvolume
2085  * contains orphan files.
2086  */
2087 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2088                                 struct btrfs_pending_snapshot *pending,
2089                                 u64 *bytes_to_reserve)
2090 {
2091         struct btrfs_root *root;
2092         struct btrfs_block_rsv *block_rsv;
2093         u64 num_bytes;
2094         int index;
2095
2096         root = pending->root;
2097         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2098                 return;
2099
2100         block_rsv = root->orphan_block_rsv;
2101
2102         /* orphan block reservation for the snapshot */
2103         num_bytes = block_rsv->size;
2104
2105         /*
2106          * after the snapshot is created, COWing tree blocks may use more
2107          * space than it frees. So we should make sure there is enough
2108          * reserved space.
2109          */
2110         index = trans->transid & 0x1;
2111         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2112                 num_bytes += block_rsv->size -
2113                              (block_rsv->reserved + block_rsv->freed[index]);
2114         }
2115
2116         *bytes_to_reserve += num_bytes;
2117 }
2118
2119 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2120                                 struct btrfs_pending_snapshot *pending)
2121 {
2122         struct btrfs_root *root = pending->root;
2123         struct btrfs_root *snap = pending->snap;
2124         struct btrfs_block_rsv *block_rsv;
2125         u64 num_bytes;
2126         int index;
2127         int ret;
2128
2129         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2130                 return;
2131
2132         /* refill source subvolume's orphan block reservation */
2133         block_rsv = root->orphan_block_rsv;
2134         index = trans->transid & 0x1;
2135         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2136                 num_bytes = block_rsv->size -
2137                             (block_rsv->reserved + block_rsv->freed[index]);
2138                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                               root->orphan_block_rsv,
2140                                               num_bytes);
2141                 BUG_ON(ret);
2142         }
2143
2144         /* setup orphan block reservation for the snapshot */
2145         block_rsv = btrfs_alloc_block_rsv(snap);
2146         BUG_ON(!block_rsv);
2147
2148         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2149         snap->orphan_block_rsv = block_rsv;
2150
2151         num_bytes = root->orphan_block_rsv->size;
2152         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2153                                       block_rsv, num_bytes);
2154         BUG_ON(ret);
2155
2156 #if 0
2157         /* insert orphan item for the snapshot */
2158         WARN_ON(!root->orphan_item_inserted);
2159         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2160                                        snap->root_key.objectid);
2161         BUG_ON(ret);
2162         snap->orphan_item_inserted = 1;
2163 #endif
2164 }
2165
2166 enum btrfs_orphan_cleanup_state {
2167         ORPHAN_CLEANUP_STARTED  = 1,
2168         ORPHAN_CLEANUP_DONE     = 2,
2169 };
2170
2171 /*
2172  * This is called in transaction commmit time. If there are no orphan
2173  * files in the subvolume, it removes orphan item and frees block_rsv
2174  * structure.
2175  */
2176 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2177                               struct btrfs_root *root)
2178 {
2179         int ret;
2180
2181         if (!list_empty(&root->orphan_list) ||
2182             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2183                 return;
2184
2185         if (root->orphan_item_inserted &&
2186             btrfs_root_refs(&root->root_item) > 0) {
2187                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2188                                             root->root_key.objectid);
2189                 BUG_ON(ret);
2190                 root->orphan_item_inserted = 0;
2191         }
2192
2193         if (root->orphan_block_rsv) {
2194                 WARN_ON(root->orphan_block_rsv->size > 0);
2195                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2196                 root->orphan_block_rsv = NULL;
2197         }
2198 }
2199
2200 /*
2201  * This creates an orphan entry for the given inode in case something goes
2202  * wrong in the middle of an unlink/truncate.
2203  *
2204  * NOTE: caller of this function should reserve 5 units of metadata for
2205  *       this function.
2206  */
2207 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2208 {
2209         struct btrfs_root *root = BTRFS_I(inode)->root;
2210         struct btrfs_block_rsv *block_rsv = NULL;
2211         int reserve = 0;
2212         int insert = 0;
2213         int ret;
2214
2215         if (!root->orphan_block_rsv) {
2216                 block_rsv = btrfs_alloc_block_rsv(root);
2217                 BUG_ON(!block_rsv);
2218         }
2219
2220         spin_lock(&root->orphan_lock);
2221         if (!root->orphan_block_rsv) {
2222                 root->orphan_block_rsv = block_rsv;
2223         } else if (block_rsv) {
2224                 btrfs_free_block_rsv(root, block_rsv);
2225                 block_rsv = NULL;
2226         }
2227
2228         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2229                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2230 #if 0
2231                 /*
2232                  * For proper ENOSPC handling, we should do orphan
2233                  * cleanup when mounting. But this introduces backward
2234                  * compatibility issue.
2235                  */
2236                 if (!xchg(&root->orphan_item_inserted, 1))
2237                         insert = 2;
2238                 else
2239                         insert = 1;
2240 #endif
2241                 insert = 1;
2242         }
2243
2244         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2245                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2246                 reserve = 1;
2247         }
2248         spin_unlock(&root->orphan_lock);
2249
2250         if (block_rsv)
2251                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2252
2253         /* grab metadata reservation from transaction handle */
2254         if (reserve) {
2255                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2256                 BUG_ON(ret);
2257         }
2258
2259         /* insert an orphan item to track this unlinked/truncated file */
2260         if (insert >= 1) {
2261                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2262                 BUG_ON(ret);
2263         }
2264
2265         /* insert an orphan item to track subvolume contains orphan files */
2266         if (insert >= 2) {
2267                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2268                                                root->root_key.objectid);
2269                 BUG_ON(ret);
2270         }
2271         return 0;
2272 }
2273
2274 /*
2275  * We have done the truncate/delete so we can go ahead and remove the orphan
2276  * item for this particular inode.
2277  */
2278 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2279 {
2280         struct btrfs_root *root = BTRFS_I(inode)->root;
2281         int delete_item = 0;
2282         int release_rsv = 0;
2283         int ret = 0;
2284
2285         spin_lock(&root->orphan_lock);
2286         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2287                 list_del_init(&BTRFS_I(inode)->i_orphan);
2288                 delete_item = 1;
2289         }
2290
2291         if (BTRFS_I(inode)->orphan_meta_reserved) {
2292                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2293                 release_rsv = 1;
2294         }
2295         spin_unlock(&root->orphan_lock);
2296
2297         if (trans && delete_item) {
2298                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2299                 BUG_ON(ret);
2300         }
2301
2302         if (release_rsv)
2303                 btrfs_orphan_release_metadata(inode);
2304
2305         return 0;
2306 }
2307
2308 /*
2309  * this cleans up any orphans that may be left on the list from the last use
2310  * of this root.
2311  */
2312 int btrfs_orphan_cleanup(struct btrfs_root *root)
2313 {
2314         struct btrfs_path *path;
2315         struct extent_buffer *leaf;
2316         struct btrfs_key key, found_key;
2317         struct btrfs_trans_handle *trans;
2318         struct inode *inode;
2319         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2320
2321         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2322                 return 0;
2323
2324         path = btrfs_alloc_path();
2325         if (!path) {
2326                 ret = -ENOMEM;
2327                 goto out;
2328         }
2329         path->reada = -1;
2330
2331         key.objectid = BTRFS_ORPHAN_OBJECTID;
2332         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2333         key.offset = (u64)-1;
2334
2335         while (1) {
2336                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2337                 if (ret < 0)
2338                         goto out;
2339
2340                 /*
2341                  * if ret == 0 means we found what we were searching for, which
2342                  * is weird, but possible, so only screw with path if we didn't
2343                  * find the key and see if we have stuff that matches
2344                  */
2345                 if (ret > 0) {
2346                         ret = 0;
2347                         if (path->slots[0] == 0)
2348                                 break;
2349                         path->slots[0]--;
2350                 }
2351
2352                 /* pull out the item */
2353                 leaf = path->nodes[0];
2354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356                 /* make sure the item matches what we want */
2357                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2358                         break;
2359                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2360                         break;
2361
2362                 /* release the path since we're done with it */
2363                 btrfs_release_path(path);
2364
2365                 /*
2366                  * this is where we are basically btrfs_lookup, without the
2367                  * crossing root thing.  we store the inode number in the
2368                  * offset of the orphan item.
2369                  */
2370                 found_key.objectid = found_key.offset;
2371                 found_key.type = BTRFS_INODE_ITEM_KEY;
2372                 found_key.offset = 0;
2373                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2374                 if (IS_ERR(inode)) {
2375                         ret = PTR_ERR(inode);
2376                         goto out;
2377                 }
2378
2379                 /*
2380                  * add this inode to the orphan list so btrfs_orphan_del does
2381                  * the proper thing when we hit it
2382                  */
2383                 spin_lock(&root->orphan_lock);
2384                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2385                 spin_unlock(&root->orphan_lock);
2386
2387                 /*
2388                  * if this is a bad inode, means we actually succeeded in
2389                  * removing the inode, but not the orphan record, which means
2390                  * we need to manually delete the orphan since iput will just
2391                  * do a destroy_inode
2392                  */
2393                 if (is_bad_inode(inode)) {
2394                         trans = btrfs_start_transaction(root, 0);
2395                         if (IS_ERR(trans)) {
2396                                 ret = PTR_ERR(trans);
2397                                 goto out;
2398                         }
2399                         btrfs_orphan_del(trans, inode);
2400                         btrfs_end_transaction(trans, root);
2401                         iput(inode);
2402                         continue;
2403                 }
2404
2405                 /* if we have links, this was a truncate, lets do that */
2406                 if (inode->i_nlink) {
2407                         if (!S_ISREG(inode->i_mode)) {
2408                                 WARN_ON(1);
2409                                 iput(inode);
2410                                 continue;
2411                         }
2412                         nr_truncate++;
2413                         ret = btrfs_truncate(inode);
2414                 } else {
2415                         nr_unlink++;
2416                 }
2417
2418                 /* this will do delete_inode and everything for us */
2419                 iput(inode);
2420                 if (ret)
2421                         goto out;
2422         }
2423         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2424
2425         if (root->orphan_block_rsv)
2426                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2427                                         (u64)-1);
2428
2429         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2430                 trans = btrfs_join_transaction(root);
2431                 if (!IS_ERR(trans))
2432                         btrfs_end_transaction(trans, root);
2433         }
2434
2435         if (nr_unlink)
2436                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2437         if (nr_truncate)
2438                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2439
2440 out:
2441         if (ret)
2442                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2443         btrfs_free_path(path);
2444         return ret;
2445 }
2446
2447 /*
2448  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2449  * don't find any xattrs, we know there can't be any acls.
2450  *
2451  * slot is the slot the inode is in, objectid is the objectid of the inode
2452  */
2453 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2454                                           int slot, u64 objectid)
2455 {
2456         u32 nritems = btrfs_header_nritems(leaf);
2457         struct btrfs_key found_key;
2458         int scanned = 0;
2459
2460         slot++;
2461         while (slot < nritems) {
2462                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2463
2464                 /* we found a different objectid, there must not be acls */
2465                 if (found_key.objectid != objectid)
2466                         return 0;
2467
2468                 /* we found an xattr, assume we've got an acl */
2469                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2470                         return 1;
2471
2472                 /*
2473                  * we found a key greater than an xattr key, there can't
2474                  * be any acls later on
2475                  */
2476                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2477                         return 0;
2478
2479                 slot++;
2480                 scanned++;
2481
2482                 /*
2483                  * it goes inode, inode backrefs, xattrs, extents,
2484                  * so if there are a ton of hard links to an inode there can
2485                  * be a lot of backrefs.  Don't waste time searching too hard,
2486                  * this is just an optimization
2487                  */
2488                 if (scanned >= 8)
2489                         break;
2490         }
2491         /* we hit the end of the leaf before we found an xattr or
2492          * something larger than an xattr.  We have to assume the inode
2493          * has acls
2494          */
2495         return 1;
2496 }
2497
2498 /*
2499  * read an inode from the btree into the in-memory inode
2500  */
2501 static void btrfs_read_locked_inode(struct inode *inode)
2502 {
2503         struct btrfs_path *path;
2504         struct extent_buffer *leaf;
2505         struct btrfs_inode_item *inode_item;
2506         struct btrfs_timespec *tspec;
2507         struct btrfs_root *root = BTRFS_I(inode)->root;
2508         struct btrfs_key location;
2509         int maybe_acls;
2510         u32 rdev;
2511         int ret;
2512         bool filled = false;
2513
2514         ret = btrfs_fill_inode(inode, &rdev);
2515         if (!ret)
2516                 filled = true;
2517
2518         path = btrfs_alloc_path();
2519         BUG_ON(!path);
2520         path->leave_spinning = 1;
2521         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2522
2523         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2524         if (ret)
2525                 goto make_bad;
2526
2527         leaf = path->nodes[0];
2528
2529         if (filled)
2530                 goto cache_acl;
2531
2532         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2533                                     struct btrfs_inode_item);
2534         if (!leaf->map_token)
2535                 map_private_extent_buffer(leaf, (unsigned long)inode_item,
2536                                           sizeof(struct btrfs_inode_item),
2537                                           &leaf->map_token, &leaf->kaddr,
2538                                           &leaf->map_start, &leaf->map_len,
2539                                           KM_USER1);
2540
2541         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2542         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2543         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2544         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2545         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2546
2547         tspec = btrfs_inode_atime(inode_item);
2548         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2549         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2550
2551         tspec = btrfs_inode_mtime(inode_item);
2552         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2553         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2554
2555         tspec = btrfs_inode_ctime(inode_item);
2556         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2557         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2558
2559         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2560         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2561         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2562         inode->i_generation = BTRFS_I(inode)->generation;
2563         inode->i_rdev = 0;
2564         rdev = btrfs_inode_rdev(leaf, inode_item);
2565
2566         BTRFS_I(inode)->index_cnt = (u64)-1;
2567         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2568 cache_acl:
2569         /*
2570          * try to precache a NULL acl entry for files that don't have
2571          * any xattrs or acls
2572          */
2573         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2574                                            btrfs_ino(inode));
2575         if (!maybe_acls)
2576                 cache_no_acl(inode);
2577
2578         if (leaf->map_token) {
2579                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2580                 leaf->map_token = NULL;
2581         }
2582
2583         btrfs_free_path(path);
2584
2585         switch (inode->i_mode & S_IFMT) {
2586         case S_IFREG:
2587                 inode->i_mapping->a_ops = &btrfs_aops;
2588                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2589                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2590                 inode->i_fop = &btrfs_file_operations;
2591                 inode->i_op = &btrfs_file_inode_operations;
2592                 break;
2593         case S_IFDIR:
2594                 inode->i_fop = &btrfs_dir_file_operations;
2595                 if (root == root->fs_info->tree_root)
2596                         inode->i_op = &btrfs_dir_ro_inode_operations;
2597                 else
2598                         inode->i_op = &btrfs_dir_inode_operations;
2599                 break;
2600         case S_IFLNK:
2601                 inode->i_op = &btrfs_symlink_inode_operations;
2602                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2603                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2604                 break;
2605         default:
2606                 inode->i_op = &btrfs_special_inode_operations;
2607                 init_special_inode(inode, inode->i_mode, rdev);
2608                 break;
2609         }
2610
2611         btrfs_update_iflags(inode);
2612         return;
2613
2614 make_bad:
2615         btrfs_free_path(path);
2616         make_bad_inode(inode);
2617 }
2618
2619 /*
2620  * given a leaf and an inode, copy the inode fields into the leaf
2621  */
2622 static void fill_inode_item(struct btrfs_trans_handle *trans,
2623                             struct extent_buffer *leaf,
2624                             struct btrfs_inode_item *item,
2625                             struct inode *inode)
2626 {
2627         if (!leaf->map_token)
2628                 map_private_extent_buffer(leaf, (unsigned long)item,
2629                                           sizeof(struct btrfs_inode_item),
2630                                           &leaf->map_token, &leaf->kaddr,
2631                                           &leaf->map_start, &leaf->map_len,
2632                                           KM_USER1);
2633
2634         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2635         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2636         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2637         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2638         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2639
2640         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2641                                inode->i_atime.tv_sec);
2642         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2643                                 inode->i_atime.tv_nsec);
2644
2645         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2646                                inode->i_mtime.tv_sec);
2647         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2648                                 inode->i_mtime.tv_nsec);
2649
2650         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2651                                inode->i_ctime.tv_sec);
2652         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2653                                 inode->i_ctime.tv_nsec);
2654
2655         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2656         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2657         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2658         btrfs_set_inode_transid(leaf, item, trans->transid);
2659         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2660         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2661         btrfs_set_inode_block_group(leaf, item, 0);
2662
2663         if (leaf->map_token) {
2664                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2665                 leaf->map_token = NULL;
2666         }
2667 }
2668
2669 /*
2670  * copy everything in the in-memory inode into the btree.
2671  */
2672 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2673                                 struct btrfs_root *root, struct inode *inode)
2674 {
2675         struct btrfs_inode_item *inode_item;
2676         struct btrfs_path *path;
2677         struct extent_buffer *leaf;
2678         int ret;
2679
2680         /*
2681          * If the inode is a free space inode, we can deadlock during commit
2682          * if we put it into the delayed code.
2683          *
2684          * The data relocation inode should also be directly updated
2685          * without delay
2686          */
2687         if (!is_free_space_inode(root, inode)
2688             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2689                 ret = btrfs_delayed_update_inode(trans, root, inode);
2690                 if (!ret)
2691                         btrfs_set_inode_last_trans(trans, inode);
2692                 return ret;
2693         }
2694
2695         path = btrfs_alloc_path();
2696         if (!path)
2697                 return -ENOMEM;
2698
2699         path->leave_spinning = 1;
2700         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2701                                  1);
2702         if (ret) {
2703                 if (ret > 0)
2704                         ret = -ENOENT;
2705                 goto failed;
2706         }
2707
2708         btrfs_unlock_up_safe(path, 1);
2709         leaf = path->nodes[0];
2710         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2711                                     struct btrfs_inode_item);
2712
2713         fill_inode_item(trans, leaf, inode_item, inode);
2714         btrfs_mark_buffer_dirty(leaf);
2715         btrfs_set_inode_last_trans(trans, inode);
2716         ret = 0;
2717 failed:
2718         btrfs_free_path(path);
2719         return ret;
2720 }
2721
2722 /*
2723  * unlink helper that gets used here in inode.c and in the tree logging
2724  * recovery code.  It remove a link in a directory with a given name, and
2725  * also drops the back refs in the inode to the directory
2726  */
2727 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2728                                 struct btrfs_root *root,
2729                                 struct inode *dir, struct inode *inode,
2730                                 const char *name, int name_len)
2731 {
2732         struct btrfs_path *path;
2733         int ret = 0;
2734         struct extent_buffer *leaf;
2735         struct btrfs_dir_item *di;
2736         struct btrfs_key key;
2737         u64 index;
2738         u64 ino = btrfs_ino(inode);
2739         u64 dir_ino = btrfs_ino(dir);
2740
2741         path = btrfs_alloc_path();
2742         if (!path) {
2743                 ret = -ENOMEM;
2744                 goto out;
2745         }
2746
2747         path->leave_spinning = 1;
2748         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2749                                     name, name_len, -1);
2750         if (IS_ERR(di)) {
2751                 ret = PTR_ERR(di);
2752                 goto err;
2753         }
2754         if (!di) {
2755                 ret = -ENOENT;
2756                 goto err;
2757         }
2758         leaf = path->nodes[0];
2759         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2760         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2761         if (ret)
2762                 goto err;
2763         btrfs_release_path(path);
2764
2765         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2766                                   dir_ino, &index);
2767         if (ret) {
2768                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2769                        "inode %llu parent %llu\n", name_len, name,
2770                        (unsigned long long)ino, (unsigned long long)dir_ino);
2771                 goto err;
2772         }
2773
2774         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2775         if (ret)
2776                 goto err;
2777
2778         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2779                                          inode, dir_ino);
2780         BUG_ON(ret != 0 && ret != -ENOENT);
2781
2782         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2783                                            dir, index);
2784         if (ret == -ENOENT)
2785                 ret = 0;
2786 err:
2787         btrfs_free_path(path);
2788         if (ret)
2789                 goto out;
2790
2791         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2792         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2793         btrfs_update_inode(trans, root, dir);
2794 out:
2795         return ret;
2796 }
2797
2798 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2799                        struct btrfs_root *root,
2800                        struct inode *dir, struct inode *inode,
2801                        const char *name, int name_len)
2802 {
2803         int ret;
2804         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2805         if (!ret) {
2806                 btrfs_drop_nlink(inode);
2807                 ret = btrfs_update_inode(trans, root, inode);
2808         }
2809         return ret;
2810 }
2811
2812
2813 /* helper to check if there is any shared block in the path */
2814 static int check_path_shared(struct btrfs_root *root,
2815                              struct btrfs_path *path)
2816 {
2817         struct extent_buffer *eb;
2818         int level;
2819         u64 refs = 1;
2820
2821         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2822                 int ret;
2823
2824                 if (!path->nodes[level])
2825                         break;
2826                 eb = path->nodes[level];
2827                 if (!btrfs_block_can_be_shared(root, eb))
2828                         continue;
2829                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2830                                                &refs, NULL);
2831                 if (refs > 1)
2832                         return 1;
2833         }
2834         return 0;
2835 }
2836
2837 /*
2838  * helper to start transaction for unlink and rmdir.
2839  *
2840  * unlink and rmdir are special in btrfs, they do not always free space.
2841  * so in enospc case, we should make sure they will free space before
2842  * allowing them to use the global metadata reservation.
2843  */
2844 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2845                                                        struct dentry *dentry)
2846 {
2847         struct btrfs_trans_handle *trans;
2848         struct btrfs_root *root = BTRFS_I(dir)->root;
2849         struct btrfs_path *path;
2850         struct btrfs_inode_ref *ref;
2851         struct btrfs_dir_item *di;
2852         struct inode *inode = dentry->d_inode;
2853         u64 index;
2854         int check_link = 1;
2855         int err = -ENOSPC;
2856         int ret;
2857         u64 ino = btrfs_ino(inode);
2858         u64 dir_ino = btrfs_ino(dir);
2859
2860         trans = btrfs_start_transaction(root, 10);
2861         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2862                 return trans;
2863
2864         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2865                 return ERR_PTR(-ENOSPC);
2866
2867         /* check if there is someone else holds reference */
2868         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2869                 return ERR_PTR(-ENOSPC);
2870
2871         if (atomic_read(&inode->i_count) > 2)
2872                 return ERR_PTR(-ENOSPC);
2873
2874         if (xchg(&root->fs_info->enospc_unlink, 1))
2875                 return ERR_PTR(-ENOSPC);
2876
2877         path = btrfs_alloc_path();
2878         if (!path) {
2879                 root->fs_info->enospc_unlink = 0;
2880                 return ERR_PTR(-ENOMEM);
2881         }
2882
2883         trans = btrfs_start_transaction(root, 0);
2884         if (IS_ERR(trans)) {
2885                 btrfs_free_path(path);
2886                 root->fs_info->enospc_unlink = 0;
2887                 return trans;
2888         }
2889
2890         path->skip_locking = 1;
2891         path->search_commit_root = 1;
2892
2893         ret = btrfs_lookup_inode(trans, root, path,
2894                                 &BTRFS_I(dir)->location, 0);
2895         if (ret < 0) {
2896                 err = ret;
2897                 goto out;
2898         }
2899         if (ret == 0) {
2900                 if (check_path_shared(root, path))
2901                         goto out;
2902         } else {
2903                 check_link = 0;
2904         }
2905         btrfs_release_path(path);
2906
2907         ret = btrfs_lookup_inode(trans, root, path,
2908                                 &BTRFS_I(inode)->location, 0);
2909         if (ret < 0) {
2910                 err = ret;
2911                 goto out;
2912         }
2913         if (ret == 0) {
2914                 if (check_path_shared(root, path))
2915                         goto out;
2916         } else {
2917                 check_link = 0;
2918         }
2919         btrfs_release_path(path);
2920
2921         if (ret == 0 && S_ISREG(inode->i_mode)) {
2922                 ret = btrfs_lookup_file_extent(trans, root, path,
2923                                                ino, (u64)-1, 0);
2924                 if (ret < 0) {
2925                         err = ret;
2926                         goto out;
2927                 }
2928                 BUG_ON(ret == 0);
2929                 if (check_path_shared(root, path))
2930                         goto out;
2931                 btrfs_release_path(path);
2932         }
2933
2934         if (!check_link) {
2935                 err = 0;
2936                 goto out;
2937         }
2938
2939         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2940                                 dentry->d_name.name, dentry->d_name.len, 0);
2941         if (IS_ERR(di)) {
2942                 err = PTR_ERR(di);
2943                 goto out;
2944         }
2945         if (di) {
2946                 if (check_path_shared(root, path))
2947                         goto out;
2948         } else {
2949                 err = 0;
2950                 goto out;
2951         }
2952         btrfs_release_path(path);
2953
2954         ref = btrfs_lookup_inode_ref(trans, root, path,
2955                                 dentry->d_name.name, dentry->d_name.len,
2956                                 ino, dir_ino, 0);
2957         if (IS_ERR(ref)) {
2958                 err = PTR_ERR(ref);
2959                 goto out;
2960         }
2961         BUG_ON(!ref);
2962         if (check_path_shared(root, path))
2963                 goto out;
2964         index = btrfs_inode_ref_index(path->nodes[0], ref);
2965         btrfs_release_path(path);
2966
2967         /*
2968          * This is a commit root search, if we can lookup inode item and other
2969          * relative items in the commit root, it means the transaction of
2970          * dir/file creation has been committed, and the dir index item that we
2971          * delay to insert has also been inserted into the commit root. So
2972          * we needn't worry about the delayed insertion of the dir index item
2973          * here.
2974          */
2975         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2976                                 dentry->d_name.name, dentry->d_name.len, 0);
2977         if (IS_ERR(di)) {
2978                 err = PTR_ERR(di);
2979                 goto out;
2980         }
2981         BUG_ON(ret == -ENOENT);
2982         if (check_path_shared(root, path))
2983                 goto out;
2984
2985         err = 0;
2986 out:
2987         btrfs_free_path(path);
2988         if (err) {
2989                 btrfs_end_transaction(trans, root);
2990                 root->fs_info->enospc_unlink = 0;
2991                 return ERR_PTR(err);
2992         }
2993
2994         trans->block_rsv = &root->fs_info->global_block_rsv;
2995         return trans;
2996 }
2997
2998 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2999                                struct btrfs_root *root)
3000 {
3001         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3002                 BUG_ON(!root->fs_info->enospc_unlink);
3003                 root->fs_info->enospc_unlink = 0;
3004         }
3005         btrfs_end_transaction_throttle(trans, root);
3006 }
3007
3008 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3009 {
3010         struct btrfs_root *root = BTRFS_I(dir)->root;
3011         struct btrfs_trans_handle *trans;
3012         struct inode *inode = dentry->d_inode;
3013         int ret;
3014         unsigned long nr = 0;
3015
3016         trans = __unlink_start_trans(dir, dentry);
3017         if (IS_ERR(trans))
3018                 return PTR_ERR(trans);
3019
3020         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3021
3022         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3023                                  dentry->d_name.name, dentry->d_name.len);
3024         BUG_ON(ret);
3025
3026         if (inode->i_nlink == 0) {
3027                 ret = btrfs_orphan_add(trans, inode);
3028                 BUG_ON(ret);
3029         }
3030
3031         nr = trans->blocks_used;
3032         __unlink_end_trans(trans, root);
3033         btrfs_btree_balance_dirty(root, nr);
3034         return ret;
3035 }
3036
3037 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3038                         struct btrfs_root *root,
3039                         struct inode *dir, u64 objectid,
3040                         const char *name, int name_len)
3041 {
3042         struct btrfs_path *path;
3043         struct extent_buffer *leaf;
3044         struct btrfs_dir_item *di;
3045         struct btrfs_key key;
3046         u64 index;
3047         int ret;
3048         u64 dir_ino = btrfs_ino(dir);
3049
3050         path = btrfs_alloc_path();
3051         if (!path)
3052                 return -ENOMEM;
3053
3054         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3055                                    name, name_len, -1);
3056         BUG_ON(IS_ERR_OR_NULL(di));
3057
3058         leaf = path->nodes[0];
3059         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3060         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3061         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3062         BUG_ON(ret);
3063         btrfs_release_path(path);
3064
3065         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3066                                  objectid, root->root_key.objectid,
3067                                  dir_ino, &index, name, name_len);
3068         if (ret < 0) {
3069                 BUG_ON(ret != -ENOENT);
3070                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3071                                                  name, name_len);
3072                 BUG_ON(IS_ERR_OR_NULL(di));
3073
3074                 leaf = path->nodes[0];
3075                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3076                 btrfs_release_path(path);
3077                 index = key.offset;
3078         }
3079         btrfs_release_path(path);
3080
3081         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3082         BUG_ON(ret);
3083
3084         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3085         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3086         ret = btrfs_update_inode(trans, root, dir);
3087         BUG_ON(ret);
3088
3089         btrfs_free_path(path);
3090         return 0;
3091 }
3092
3093 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3094 {
3095         struct inode *inode = dentry->d_inode;
3096         int err = 0;
3097         struct btrfs_root *root = BTRFS_I(dir)->root;
3098         struct btrfs_trans_handle *trans;
3099         unsigned long nr = 0;
3100
3101         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3102             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3103                 return -ENOTEMPTY;
3104
3105         trans = __unlink_start_trans(dir, dentry);
3106         if (IS_ERR(trans))
3107                 return PTR_ERR(trans);
3108
3109         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3110                 err = btrfs_unlink_subvol(trans, root, dir,
3111                                           BTRFS_I(inode)->location.objectid,
3112                                           dentry->d_name.name,
3113                                           dentry->d_name.len);
3114                 goto out;
3115         }
3116
3117         err = btrfs_orphan_add(trans, inode);
3118         if (err)
3119                 goto out;
3120
3121         /* now the directory is empty */
3122         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3123                                  dentry->d_name.name, dentry->d_name.len);
3124         if (!err)
3125                 btrfs_i_size_write(inode, 0);
3126 out:
3127         nr = trans->blocks_used;
3128         __unlink_end_trans(trans, root);
3129         btrfs_btree_balance_dirty(root, nr);
3130
3131         return err;
3132 }
3133
3134 /*
3135  * this can truncate away extent items, csum items and directory items.
3136  * It starts at a high offset and removes keys until it can't find
3137  * any higher than new_size
3138  *
3139  * csum items that cross the new i_size are truncated to the new size
3140  * as well.
3141  *
3142  * min_type is the minimum key type to truncate down to.  If set to 0, this
3143  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3144  */
3145 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3146                                struct btrfs_root *root,
3147                                struct inode *inode,
3148                                u64 new_size, u32 min_type)
3149 {
3150         struct btrfs_path *path;
3151         struct extent_buffer *leaf;
3152         struct btrfs_file_extent_item *fi;
3153         struct btrfs_key key;
3154         struct btrfs_key found_key;
3155         u64 extent_start = 0;
3156         u64 extent_num_bytes = 0;
3157         u64 extent_offset = 0;
3158         u64 item_end = 0;
3159         u64 mask = root->sectorsize - 1;
3160         u32 found_type = (u8)-1;
3161         int found_extent;
3162         int del_item;
3163         int pending_del_nr = 0;
3164         int pending_del_slot = 0;
3165         int extent_type = -1;
3166         int encoding;
3167         int ret;
3168         int err = 0;
3169         u64 ino = btrfs_ino(inode);
3170
3171         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3172
3173         if (root->ref_cows || root == root->fs_info->tree_root)
3174                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3175
3176         /*
3177          * This function is also used to drop the items in the log tree before
3178          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3179          * it is used to drop the loged items. So we shouldn't kill the delayed
3180          * items.
3181          */
3182         if (min_type == 0 && root == BTRFS_I(inode)->root)
3183                 btrfs_kill_delayed_inode_items(inode);
3184
3185         path = btrfs_alloc_path();
3186         BUG_ON(!path);
3187         path->reada = -1;
3188
3189         key.objectid = ino;
3190         key.offset = (u64)-1;
3191         key.type = (u8)-1;
3192
3193 search_again:
3194         path->leave_spinning = 1;
3195         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3196         if (ret < 0) {
3197                 err = ret;
3198                 goto out;
3199         }
3200
3201         if (ret > 0) {
3202                 /* there are no items in the tree for us to truncate, we're
3203                  * done
3204                  */
3205                 if (path->slots[0] == 0)
3206                         goto out;
3207                 path->slots[0]--;
3208         }
3209
3210         while (1) {
3211                 fi = NULL;
3212                 leaf = path->nodes[0];
3213                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3214                 found_type = btrfs_key_type(&found_key);
3215                 encoding = 0;
3216
3217                 if (found_key.objectid != ino)
3218                         break;
3219
3220                 if (found_type < min_type)
3221                         break;
3222
3223                 item_end = found_key.offset;
3224                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3225                         fi = btrfs_item_ptr(leaf, path->slots[0],
3226                                             struct btrfs_file_extent_item);
3227                         extent_type = btrfs_file_extent_type(leaf, fi);
3228                         encoding = btrfs_file_extent_compression(leaf, fi);
3229                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3230                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3231
3232                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3233                                 item_end +=
3234                                     btrfs_file_extent_num_bytes(leaf, fi);
3235                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3236                                 item_end += btrfs_file_extent_inline_len(leaf,
3237                                                                          fi);
3238                         }
3239                         item_end--;
3240                 }
3241                 if (found_type > min_type) {
3242                         del_item = 1;
3243                 } else {
3244                         if (item_end < new_size)
3245                                 break;
3246                         if (found_key.offset >= new_size)
3247                                 del_item = 1;
3248                         else
3249                                 del_item = 0;
3250                 }
3251                 found_extent = 0;
3252                 /* FIXME, shrink the extent if the ref count is only 1 */
3253                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3254                         goto delete;
3255
3256                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3257                         u64 num_dec;
3258                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3259                         if (!del_item && !encoding) {
3260                                 u64 orig_num_bytes =
3261                                         btrfs_file_extent_num_bytes(leaf, fi);
3262                                 extent_num_bytes = new_size -
3263                                         found_key.offset + root->sectorsize - 1;
3264                                 extent_num_bytes = extent_num_bytes &
3265                                         ~((u64)root->sectorsize - 1);
3266                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3267                                                          extent_num_bytes);
3268                                 num_dec = (orig_num_bytes -
3269                                            extent_num_bytes);
3270                                 if (root->ref_cows && extent_start != 0)
3271                                         inode_sub_bytes(inode, num_dec);
3272                                 btrfs_mark_buffer_dirty(leaf);
3273                         } else {
3274                                 extent_num_bytes =
3275                                         btrfs_file_extent_disk_num_bytes(leaf,
3276                                                                          fi);
3277                                 extent_offset = found_key.offset -
3278                                         btrfs_file_extent_offset(leaf, fi);
3279
3280                                 /* FIXME blocksize != 4096 */
3281                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3282                                 if (extent_start != 0) {
3283                                         found_extent = 1;
3284                                         if (root->ref_cows)
3285                                                 inode_sub_bytes(inode, num_dec);
3286                                 }
3287                         }
3288                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3289                         /*
3290                          * we can't truncate inline items that have had
3291                          * special encodings
3292                          */
3293                         if (!del_item &&
3294                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3295                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3296                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3297                                 u32 size = new_size - found_key.offset;
3298
3299                                 if (root->ref_cows) {
3300                                         inode_sub_bytes(inode, item_end + 1 -
3301                                                         new_size);
3302                                 }
3303                                 size =
3304                                     btrfs_file_extent_calc_inline_size(size);
3305                                 ret = btrfs_truncate_item(trans, root, path,
3306                                                           size, 1);
3307                         } else if (root->ref_cows) {
3308                                 inode_sub_bytes(inode, item_end + 1 -
3309                                                 found_key.offset);
3310                         }
3311                 }
3312 delete:
3313                 if (del_item) {
3314                         if (!pending_del_nr) {
3315                                 /* no pending yet, add ourselves */
3316                                 pending_del_slot = path->slots[0];
3317                                 pending_del_nr = 1;
3318                         } else if (pending_del_nr &&
3319                                    path->slots[0] + 1 == pending_del_slot) {
3320                                 /* hop on the pending chunk */
3321                                 pending_del_nr++;
3322                                 pending_del_slot = path->slots[0];
3323                         } else {
3324                                 BUG();
3325                         }
3326                 } else {
3327                         break;
3328                 }
3329                 if (found_extent && (root->ref_cows ||
3330                                      root == root->fs_info->tree_root)) {
3331                         btrfs_set_path_blocking(path);
3332                         ret = btrfs_free_extent(trans, root, extent_start,
3333                                                 extent_num_bytes, 0,
3334                                                 btrfs_header_owner(leaf),
3335                                                 ino, extent_offset);
3336                         BUG_ON(ret);
3337                 }
3338
3339                 if (found_type == BTRFS_INODE_ITEM_KEY)
3340                         break;
3341
3342                 if (path->slots[0] == 0 ||
3343                     path->slots[0] != pending_del_slot) {
3344                         if (root->ref_cows &&
3345                             BTRFS_I(inode)->location.objectid !=
3346                                                 BTRFS_FREE_INO_OBJECTID) {
3347                                 err = -EAGAIN;
3348                                 goto out;
3349                         }
3350                         if (pending_del_nr) {
3351                                 ret = btrfs_del_items(trans, root, path,
3352                                                 pending_del_slot,
3353                                                 pending_del_nr);
3354                                 BUG_ON(ret);
3355                                 pending_del_nr = 0;
3356                         }
3357                         btrfs_release_path(path);
3358                         goto search_again;
3359                 } else {
3360                         path->slots[0]--;
3361                 }
3362         }
3363 out:
3364         if (pending_del_nr) {
3365                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3366                                       pending_del_nr);
3367                 BUG_ON(ret);
3368         }
3369         btrfs_free_path(path);
3370         return err;
3371 }
3372
3373 /*
3374  * taken from block_truncate_page, but does cow as it zeros out
3375  * any bytes left in the last page in the file.
3376  */
3377 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3378 {
3379         struct inode *inode = mapping->host;
3380         struct btrfs_root *root = BTRFS_I(inode)->root;
3381         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3382         struct btrfs_ordered_extent *ordered;
3383         struct extent_state *cached_state = NULL;
3384         char *kaddr;
3385         u32 blocksize = root->sectorsize;
3386         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3387         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3388         struct page *page;
3389         int ret = 0;
3390         u64 page_start;
3391         u64 page_end;
3392
3393         if ((offset & (blocksize - 1)) == 0)
3394                 goto out;
3395         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3396         if (ret)
3397                 goto out;
3398
3399         ret = -ENOMEM;
3400 again:
3401         page = grab_cache_page(mapping, index);
3402         if (!page) {
3403                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3404                 goto out;
3405         }
3406
3407         page_start = page_offset(page);
3408         page_end = page_start + PAGE_CACHE_SIZE - 1;
3409
3410         if (!PageUptodate(page)) {
3411                 ret = btrfs_readpage(NULL, page);
3412                 lock_page(page);
3413                 if (page->mapping != mapping) {
3414                         unlock_page(page);
3415                         page_cache_release(page);
3416                         goto again;
3417                 }
3418                 if (!PageUptodate(page)) {
3419                         ret = -EIO;
3420                         goto out_unlock;
3421                 }
3422         }
3423         wait_on_page_writeback(page);
3424
3425         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3426                          GFP_NOFS);
3427         set_page_extent_mapped(page);
3428
3429         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3430         if (ordered) {
3431                 unlock_extent_cached(io_tree, page_start, page_end,
3432                                      &cached_state, GFP_NOFS);
3433                 unlock_page(page);
3434                 page_cache_release(page);
3435                 btrfs_start_ordered_extent(inode, ordered, 1);
3436                 btrfs_put_ordered_extent(ordered);
3437                 goto again;
3438         }
3439
3440         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3441                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3442                           0, 0, &cached_state, GFP_NOFS);
3443
3444         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3445                                         &cached_state);
3446         if (ret) {
3447                 unlock_extent_cached(io_tree, page_start, page_end,
3448                                      &cached_state, GFP_NOFS);
3449                 goto out_unlock;
3450         }
3451
3452         ret = 0;
3453         if (offset != PAGE_CACHE_SIZE) {
3454                 kaddr = kmap(page);
3455                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3456                 flush_dcache_page(page);
3457                 kunmap(page);
3458         }
3459         ClearPageChecked(page);
3460         set_page_dirty(page);
3461         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3462                              GFP_NOFS);
3463
3464 out_unlock:
3465         if (ret)
3466                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3467         unlock_page(page);
3468         page_cache_release(page);
3469 out:
3470         return ret;
3471 }
3472
3473 /*
3474  * This function puts in dummy file extents for the area we're creating a hole
3475  * for.  So if we are truncating this file to a larger size we need to insert
3476  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3477  * the range between oldsize and size
3478  */
3479 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3480 {
3481         struct btrfs_trans_handle *trans;
3482         struct btrfs_root *root = BTRFS_I(inode)->root;
3483         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3484         struct extent_map *em = NULL;
3485         struct extent_state *cached_state = NULL;
3486         u64 mask = root->sectorsize - 1;
3487         u64 hole_start = (oldsize + mask) & ~mask;
3488         u64 block_end = (size + mask) & ~mask;
3489         u64 last_byte;
3490         u64 cur_offset;
3491         u64 hole_size;
3492         int err = 0;
3493
3494         if (size <= hole_start)
3495                 return 0;
3496
3497         while (1) {
3498                 struct btrfs_ordered_extent *ordered;
3499                 btrfs_wait_ordered_range(inode, hole_start,
3500                                          block_end - hole_start);
3501                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3502                                  &cached_state, GFP_NOFS);
3503                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3504                 if (!ordered)
3505                         break;
3506                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3507                                      &cached_state, GFP_NOFS);
3508                 btrfs_put_ordered_extent(ordered);
3509         }
3510
3511         cur_offset = hole_start;
3512         while (1) {
3513                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3514                                 block_end - cur_offset, 0);
3515                 BUG_ON(IS_ERR_OR_NULL(em));
3516                 last_byte = min(extent_map_end(em), block_end);
3517                 last_byte = (last_byte + mask) & ~mask;
3518                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3519                         u64 hint_byte = 0;
3520                         hole_size = last_byte - cur_offset;
3521
3522                         trans = btrfs_start_transaction(root, 2);
3523                         if (IS_ERR(trans)) {
3524                                 err = PTR_ERR(trans);
3525                                 break;
3526                         }
3527
3528                         err = btrfs_drop_extents(trans, inode, cur_offset,
3529                                                  cur_offset + hole_size,
3530                                                  &hint_byte, 1);
3531                         if (err)
3532                                 break;
3533
3534                         err = btrfs_insert_file_extent(trans, root,
3535                                         btrfs_ino(inode), cur_offset, 0,
3536                                         0, hole_size, 0, hole_size,
3537                                         0, 0, 0);
3538                         if (err)
3539                                 break;
3540
3541                         btrfs_drop_extent_cache(inode, hole_start,
3542                                         last_byte - 1, 0);
3543
3544                         btrfs_end_transaction(trans, root);
3545                 }
3546                 free_extent_map(em);
3547                 em = NULL;
3548                 cur_offset = last_byte;
3549                 if (cur_offset >= block_end)
3550                         break;
3551         }
3552
3553         free_extent_map(em);
3554         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3555                              GFP_NOFS);
3556         return err;
3557 }
3558
3559 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3560 {
3561         loff_t oldsize = i_size_read(inode);
3562         int ret;
3563
3564         if (newsize == oldsize)
3565                 return 0;
3566
3567         if (newsize > oldsize) {
3568                 i_size_write(inode, newsize);
3569                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3570                 truncate_pagecache(inode, oldsize, newsize);
3571                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3572                 if (ret) {
3573                         btrfs_setsize(inode, oldsize);
3574                         return ret;
3575                 }
3576
3577                 mark_inode_dirty(inode);
3578         } else {
3579
3580                 /*
3581                  * We're truncating a file that used to have good data down to
3582                  * zero. Make sure it gets into the ordered flush list so that
3583                  * any new writes get down to disk quickly.
3584                  */
3585                 if (newsize == 0)
3586                         BTRFS_I(inode)->ordered_data_close = 1;
3587
3588                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3589                 truncate_setsize(inode, newsize);
3590                 ret = btrfs_truncate(inode);
3591         }
3592
3593         return ret;
3594 }
3595
3596 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3597 {
3598         struct inode *inode = dentry->d_inode;
3599         struct btrfs_root *root = BTRFS_I(inode)->root;
3600         int err;
3601
3602         if (btrfs_root_readonly(root))
3603                 return -EROFS;
3604
3605         err = inode_change_ok(inode, attr);
3606         if (err)
3607                 return err;
3608
3609         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3610                 err = btrfs_setsize(inode, attr->ia_size);
3611                 if (err)
3612                         return err;
3613         }
3614
3615         if (attr->ia_valid) {
3616                 setattr_copy(inode, attr);
3617                 mark_inode_dirty(inode);
3618
3619                 if (attr->ia_valid & ATTR_MODE)
3620                         err = btrfs_acl_chmod(inode);
3621         }
3622
3623         return err;
3624 }
3625
3626 void btrfs_evict_inode(struct inode *inode)
3627 {
3628         struct btrfs_trans_handle *trans;
3629         struct btrfs_root *root = BTRFS_I(inode)->root;
3630         unsigned long nr;
3631         int ret;
3632
3633         trace_btrfs_inode_evict(inode);
3634
3635         truncate_inode_pages(&inode->i_data, 0);
3636         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3637                                is_free_space_inode(root, inode)))
3638                 goto no_delete;
3639
3640         if (is_bad_inode(inode)) {
3641                 btrfs_orphan_del(NULL, inode);
3642                 goto no_delete;
3643         }
3644         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3645         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3646
3647         if (root->fs_info->log_root_recovering) {
3648                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3649                 goto no_delete;
3650         }
3651
3652         if (inode->i_nlink > 0) {
3653                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3654                 goto no_delete;
3655         }
3656
3657         btrfs_i_size_write(inode, 0);
3658
3659         while (1) {
3660                 trans = btrfs_join_transaction(root);
3661                 BUG_ON(IS_ERR(trans));
3662                 trans->block_rsv = root->orphan_block_rsv;
3663
3664                 ret = btrfs_block_rsv_check(trans, root,
3665                                             root->orphan_block_rsv, 0, 5);
3666                 if (ret) {
3667                         BUG_ON(ret != -EAGAIN);
3668                         ret = btrfs_commit_transaction(trans, root);
3669                         BUG_ON(ret);
3670                         continue;
3671                 }
3672
3673                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3674                 if (ret != -EAGAIN)
3675                         break;
3676
3677                 nr = trans->blocks_used;
3678                 btrfs_end_transaction(trans, root);
3679                 trans = NULL;
3680                 btrfs_btree_balance_dirty(root, nr);
3681
3682         }
3683
3684         if (ret == 0) {
3685                 ret = btrfs_orphan_del(trans, inode);
3686                 BUG_ON(ret);
3687         }
3688
3689         if (!(root == root->fs_info->tree_root ||
3690               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3691                 btrfs_return_ino(root, btrfs_ino(inode));
3692
3693         nr = trans->blocks_used;
3694         btrfs_end_transaction(trans, root);
3695         btrfs_btree_balance_dirty(root, nr);
3696 no_delete:
3697         end_writeback(inode);
3698         return;
3699 }
3700
3701 /*
3702  * this returns the key found in the dir entry in the location pointer.
3703  * If no dir entries were found, location->objectid is 0.
3704  */
3705 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3706                                struct btrfs_key *location)
3707 {
3708         const char *name = dentry->d_name.name;
3709         int namelen = dentry->d_name.len;
3710         struct btrfs_dir_item *di;
3711         struct btrfs_path *path;
3712         struct btrfs_root *root = BTRFS_I(dir)->root;
3713         int ret = 0;
3714
3715         path = btrfs_alloc_path();
3716         BUG_ON(!path);
3717
3718         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3719                                     namelen, 0);
3720         if (IS_ERR(di))
3721                 ret = PTR_ERR(di);
3722
3723         if (IS_ERR_OR_NULL(di))
3724                 goto out_err;
3725
3726         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3727 out:
3728         btrfs_free_path(path);
3729         return ret;
3730 out_err:
3731         location->objectid = 0;
3732         goto out;
3733 }
3734
3735 /*
3736  * when we hit a tree root in a directory, the btrfs part of the inode
3737  * needs to be changed to reflect the root directory of the tree root.  This
3738  * is kind of like crossing a mount point.
3739  */
3740 static int fixup_tree_root_location(struct btrfs_root *root,
3741                                     struct inode *dir,
3742                                     struct dentry *dentry,
3743                                     struct btrfs_key *location,
3744                                     struct btrfs_root **sub_root)
3745 {
3746         struct btrfs_path *path;
3747         struct btrfs_root *new_root;
3748         struct btrfs_root_ref *ref;
3749         struct extent_buffer *leaf;
3750         int ret;
3751         int err = 0;
3752
3753         path = btrfs_alloc_path();
3754         if (!path) {
3755                 err = -ENOMEM;
3756                 goto out;
3757         }
3758
3759         err = -ENOENT;
3760         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3761                                   BTRFS_I(dir)->root->root_key.objectid,
3762                                   location->objectid);
3763         if (ret) {
3764                 if (ret < 0)
3765                         err = ret;
3766                 goto out;
3767         }
3768
3769         leaf = path->nodes[0];
3770         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3771         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3772             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3773                 goto out;
3774
3775         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3776                                    (unsigned long)(ref + 1),
3777                                    dentry->d_name.len);
3778         if (ret)
3779                 goto out;
3780
3781         btrfs_release_path(path);
3782
3783         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3784         if (IS_ERR(new_root)) {
3785                 err = PTR_ERR(new_root);
3786                 goto out;
3787         }
3788
3789         if (btrfs_root_refs(&new_root->root_item) == 0) {
3790                 err = -ENOENT;
3791                 goto out;
3792         }
3793
3794         *sub_root = new_root;
3795         location->objectid = btrfs_root_dirid(&new_root->root_item);
3796         location->type = BTRFS_INODE_ITEM_KEY;
3797         location->offset = 0;
3798         err = 0;
3799 out:
3800         btrfs_free_path(path);
3801         return err;
3802 }
3803
3804 static void inode_tree_add(struct inode *inode)
3805 {
3806         struct btrfs_root *root = BTRFS_I(inode)->root;
3807         struct btrfs_inode *entry;
3808         struct rb_node **p;
3809         struct rb_node *parent;
3810         u64 ino = btrfs_ino(inode);
3811 again:
3812         p = &root->inode_tree.rb_node;
3813         parent = NULL;
3814
3815         if (inode_unhashed(inode))
3816                 return;
3817
3818         spin_lock(&root->inode_lock);
3819         while (*p) {
3820                 parent = *p;
3821                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3822
3823                 if (ino < btrfs_ino(&entry->vfs_inode))
3824                         p = &parent->rb_left;
3825                 else if (ino > btrfs_ino(&entry->vfs_inode))
3826                         p = &parent->rb_right;
3827                 else {
3828                         WARN_ON(!(entry->vfs_inode.i_state &
3829                                   (I_WILL_FREE | I_FREEING)));
3830                         rb_erase(parent, &root->inode_tree);
3831                         RB_CLEAR_NODE(parent);
3832                         spin_unlock(&root->inode_lock);
3833                         goto again;
3834                 }
3835         }
3836         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3837         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3838         spin_unlock(&root->inode_lock);
3839 }
3840
3841 static void inode_tree_del(struct inode *inode)
3842 {
3843         struct btrfs_root *root = BTRFS_I(inode)->root;
3844         int empty = 0;
3845
3846         spin_lock(&root->inode_lock);
3847         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3848                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3849                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3850                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3851         }
3852         spin_unlock(&root->inode_lock);
3853
3854         /*
3855          * Free space cache has inodes in the tree root, but the tree root has a
3856          * root_refs of 0, so this could end up dropping the tree root as a
3857          * snapshot, so we need the extra !root->fs_info->tree_root check to
3858          * make sure we don't drop it.
3859          */
3860         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3861             root != root->fs_info->tree_root) {
3862                 synchronize_srcu(&root->fs_info->subvol_srcu);
3863                 spin_lock(&root->inode_lock);
3864                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3865                 spin_unlock(&root->inode_lock);
3866                 if (empty)
3867                         btrfs_add_dead_root(root);
3868         }
3869 }
3870
3871 int btrfs_invalidate_inodes(struct btrfs_root *root)
3872 {
3873         struct rb_node *node;
3874         struct rb_node *prev;
3875         struct btrfs_inode *entry;
3876         struct inode *inode;
3877         u64 objectid = 0;
3878
3879         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3880
3881         spin_lock(&root->inode_lock);
3882 again:
3883         node = root->inode_tree.rb_node;
3884         prev = NULL;
3885         while (node) {
3886                 prev = node;
3887                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3888
3889                 if (objectid < btrfs_ino(&entry->vfs_inode))
3890                         node = node->rb_left;
3891                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3892                         node = node->rb_right;
3893                 else
3894                         break;
3895         }
3896         if (!node) {
3897                 while (prev) {
3898                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3899                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3900                                 node = prev;
3901                                 break;
3902                         }
3903                         prev = rb_next(prev);
3904                 }
3905         }
3906         while (node) {
3907                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3908                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3909                 inode = igrab(&entry->vfs_inode);
3910                 if (inode) {
3911                         spin_unlock(&root->inode_lock);
3912                         if (atomic_read(&inode->i_count) > 1)
3913                                 d_prune_aliases(inode);
3914                         /*
3915                          * btrfs_drop_inode will have it removed from
3916                          * the inode cache when its usage count
3917                          * hits zero.
3918                          */
3919                         iput(inode);
3920                         cond_resched();
3921                         spin_lock(&root->inode_lock);
3922                         goto again;
3923                 }
3924
3925                 if (cond_resched_lock(&root->inode_lock))
3926                         goto again;
3927
3928                 node = rb_next(node);
3929         }
3930         spin_unlock(&root->inode_lock);
3931         return 0;
3932 }
3933
3934 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3935 {
3936         struct btrfs_iget_args *args = p;
3937         inode->i_ino = args->ino;
3938         BTRFS_I(inode)->root = args->root;
3939         btrfs_set_inode_space_info(args->root, inode);
3940         return 0;
3941 }
3942
3943 static int btrfs_find_actor(struct inode *inode, void *opaque)
3944 {
3945         struct btrfs_iget_args *args = opaque;
3946         return args->ino == btrfs_ino(inode) &&
3947                 args->root == BTRFS_I(inode)->root;
3948 }
3949
3950 static struct inode *btrfs_iget_locked(struct super_block *s,
3951                                        u64 objectid,
3952                                        struct btrfs_root *root)
3953 {
3954         struct inode *inode;
3955         struct btrfs_iget_args args;
3956         args.ino = objectid;
3957         args.root = root;
3958
3959         inode = iget5_locked(s, objectid, btrfs_find_actor,
3960                              btrfs_init_locked_inode,
3961                              (void *)&args);
3962         return inode;
3963 }
3964
3965 /* Get an inode object given its location and corresponding root.
3966  * Returns in *is_new if the inode was read from disk
3967  */
3968 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3969                          struct btrfs_root *root, int *new)
3970 {
3971         struct inode *inode;
3972
3973         inode = btrfs_iget_locked(s, location->objectid, root);
3974         if (!inode)
3975                 return ERR_PTR(-ENOMEM);
3976
3977         if (inode->i_state & I_NEW) {
3978                 BTRFS_I(inode)->root = root;
3979                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3980                 btrfs_read_locked_inode(inode);
3981                 inode_tree_add(inode);
3982                 unlock_new_inode(inode);
3983                 if (new)
3984                         *new = 1;
3985         }
3986
3987         return inode;
3988 }
3989
3990 static struct inode *new_simple_dir(struct super_block *s,
3991                                     struct btrfs_key *key,
3992                                     struct btrfs_root *root)
3993 {
3994         struct inode *inode = new_inode(s);
3995
3996         if (!inode)
3997                 return ERR_PTR(-ENOMEM);
3998
3999         BTRFS_I(inode)->root = root;
4000         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4001         BTRFS_I(inode)->dummy_inode = 1;
4002
4003         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4004         inode->i_op = &simple_dir_inode_operations;
4005         inode->i_fop = &simple_dir_operations;
4006         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4007         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4008
4009         return inode;
4010 }
4011
4012 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4013 {
4014         struct inode *inode;
4015         struct btrfs_root *root = BTRFS_I(dir)->root;
4016         struct btrfs_root *sub_root = root;
4017         struct btrfs_key location;
4018         int index;
4019         int ret;
4020
4021         if (dentry->d_name.len > BTRFS_NAME_LEN)
4022                 return ERR_PTR(-ENAMETOOLONG);
4023
4024         ret = btrfs_inode_by_name(dir, dentry, &location);
4025
4026         if (ret < 0)
4027                 return ERR_PTR(ret);
4028
4029         if (location.objectid == 0)
4030                 return NULL;
4031
4032         if (location.type == BTRFS_INODE_ITEM_KEY) {
4033                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4034                 return inode;
4035         }
4036
4037         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4038
4039         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4040         ret = fixup_tree_root_location(root, dir, dentry,
4041                                        &location, &sub_root);
4042         if (ret < 0) {
4043                 if (ret != -ENOENT)
4044                         inode = ERR_PTR(ret);
4045                 else
4046                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4047         } else {
4048                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4049         }
4050         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4051
4052         if (!IS_ERR(inode) && root != sub_root) {
4053                 down_read(&root->fs_info->cleanup_work_sem);
4054                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4055                         ret = btrfs_orphan_cleanup(sub_root);
4056                 up_read(&root->fs_info->cleanup_work_sem);
4057                 if (ret)
4058                         inode = ERR_PTR(ret);
4059         }
4060
4061         return inode;
4062 }
4063
4064 static int btrfs_dentry_delete(const struct dentry *dentry)
4065 {
4066         struct btrfs_root *root;
4067
4068         if (!dentry->d_inode && !IS_ROOT(dentry))
4069                 dentry = dentry->d_parent;
4070
4071         if (dentry->d_inode) {
4072                 root = BTRFS_I(dentry->d_inode)->root;
4073                 if (btrfs_root_refs(&root->root_item) == 0)
4074                         return 1;
4075         }
4076         return 0;
4077 }
4078
4079 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4080                                    struct nameidata *nd)
4081 {
4082         return d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4083 }
4084
4085 unsigned char btrfs_filetype_table[] = {
4086         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4087 };
4088
4089 static int btrfs_real_readdir(struct file *filp, void *dirent,
4090                               filldir_t filldir)
4091 {
4092         struct inode *inode = filp->f_dentry->d_inode;
4093         struct btrfs_root *root = BTRFS_I(inode)->root;
4094         struct btrfs_item *item;
4095         struct btrfs_dir_item *di;
4096         struct btrfs_key key;
4097         struct btrfs_key found_key;
4098         struct btrfs_path *path;
4099         struct list_head ins_list;
4100         struct list_head del_list;
4101         int ret;
4102         struct extent_buffer *leaf;
4103         int slot;
4104         unsigned char d_type;
4105         int over = 0;
4106         u32 di_cur;
4107         u32 di_total;
4108         u32 di_len;
4109         int key_type = BTRFS_DIR_INDEX_KEY;
4110         char tmp_name[32];
4111         char *name_ptr;
4112         int name_len;
4113         int is_curr = 0;        /* filp->f_pos points to the current index? */
4114
4115         /* FIXME, use a real flag for deciding about the key type */
4116         if (root->fs_info->tree_root == root)
4117                 key_type = BTRFS_DIR_ITEM_KEY;
4118
4119         /* special case for "." */
4120         if (filp->f_pos == 0) {
4121                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4122                 if (over)
4123                         return 0;
4124                 filp->f_pos = 1;
4125         }
4126         /* special case for .., just use the back ref */
4127         if (filp->f_pos == 1) {
4128                 u64 pino = parent_ino(filp->f_path.dentry);
4129                 over = filldir(dirent, "..", 2,
4130                                2, pino, DT_DIR);
4131                 if (over)
4132                         return 0;
4133                 filp->f_pos = 2;
4134         }
4135         path = btrfs_alloc_path();
4136         if (!path)
4137                 return -ENOMEM;
4138
4139         path->reada = 1;
4140
4141         if (key_type == BTRFS_DIR_INDEX_KEY) {
4142                 INIT_LIST_HEAD(&ins_list);
4143                 INIT_LIST_HEAD(&del_list);
4144                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4145         }
4146
4147         btrfs_set_key_type(&key, key_type);
4148         key.offset = filp->f_pos;
4149         key.objectid = btrfs_ino(inode);
4150
4151         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4152         if (ret < 0)
4153                 goto err;
4154
4155         while (1) {
4156                 leaf = path->nodes[0];
4157                 slot = path->slots[0];
4158                 if (slot >= btrfs_header_nritems(leaf)) {
4159                         ret = btrfs_next_leaf(root, path);
4160                         if (ret < 0)
4161                                 goto err;
4162                         else if (ret > 0)
4163                                 break;
4164                         continue;
4165                 }
4166
4167                 item = btrfs_item_nr(leaf, slot);
4168                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4169
4170                 if (found_key.objectid != key.objectid)
4171                         break;
4172                 if (btrfs_key_type(&found_key) != key_type)
4173                         break;
4174                 if (found_key.offset < filp->f_pos)
4175                         goto next;
4176                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4177                     btrfs_should_delete_dir_index(&del_list,
4178                                                   found_key.offset))
4179                         goto next;
4180
4181                 filp->f_pos = found_key.offset;
4182                 is_curr = 1;
4183
4184                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4185                 di_cur = 0;
4186                 di_total = btrfs_item_size(leaf, item);
4187
4188                 while (di_cur < di_total) {
4189                         struct btrfs_key location;
4190
4191                         if (verify_dir_item(root, leaf, di))
4192                                 break;
4193
4194                         name_len = btrfs_dir_name_len(leaf, di);
4195                         if (name_len <= sizeof(tmp_name)) {
4196                                 name_ptr = tmp_name;
4197                         } else {
4198                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4199                                 if (!name_ptr) {
4200                                         ret = -ENOMEM;
4201                                         goto err;
4202                                 }
4203                         }
4204                         read_extent_buffer(leaf, name_ptr,
4205                                            (unsigned long)(di + 1), name_len);
4206
4207                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4208                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4209
4210                         /* is this a reference to our own snapshot? If so
4211                          * skip it
4212                          */
4213                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4214                             location.objectid == root->root_key.objectid) {
4215                                 over = 0;
4216                                 goto skip;
4217                         }
4218                         over = filldir(dirent, name_ptr, name_len,
4219                                        found_key.offset, location.objectid,
4220                                        d_type);
4221
4222 skip:
4223                         if (name_ptr != tmp_name)
4224                                 kfree(name_ptr);
4225
4226                         if (over)
4227                                 goto nopos;
4228                         di_len = btrfs_dir_name_len(leaf, di) +
4229                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4230                         di_cur += di_len;
4231                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4232                 }
4233 next:
4234                 path->slots[0]++;
4235         }
4236
4237         if (key_type == BTRFS_DIR_INDEX_KEY) {
4238                 if (is_curr)
4239                         filp->f_pos++;
4240                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4241                                                       &ins_list);
4242                 if (ret)
4243                         goto nopos;
4244         }
4245
4246         /* Reached end of directory/root. Bump pos past the last item. */
4247         if (key_type == BTRFS_DIR_INDEX_KEY)
4248                 /*
4249                  * 32-bit glibc will use getdents64, but then strtol -
4250                  * so the last number we can serve is this.
4251                  */
4252                 filp->f_pos = 0x7fffffff;
4253         else
4254                 filp->f_pos++;
4255 nopos:
4256         ret = 0;
4257 err:
4258         if (key_type == BTRFS_DIR_INDEX_KEY)
4259                 btrfs_put_delayed_items(&ins_list, &del_list);
4260         btrfs_free_path(path);
4261         return ret;
4262 }
4263
4264 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4265 {
4266         struct btrfs_root *root = BTRFS_I(inode)->root;
4267         struct btrfs_trans_handle *trans;
4268         int ret = 0;
4269         bool nolock = false;
4270
4271         if (BTRFS_I(inode)->dummy_inode)
4272                 return 0;
4273
4274         if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
4275                 nolock = true;
4276
4277         if (wbc->sync_mode == WB_SYNC_ALL) {
4278                 if (nolock)
4279                         trans = btrfs_join_transaction_nolock(root);
4280                 else
4281                         trans = btrfs_join_transaction(root);
4282                 if (IS_ERR(trans))
4283                         return PTR_ERR(trans);
4284                 if (nolock)
4285                         ret = btrfs_end_transaction_nolock(trans, root);
4286                 else
4287                         ret = btrfs_commit_transaction(trans, root);
4288         }
4289         return ret;
4290 }
4291
4292 /*
4293  * This is somewhat expensive, updating the tree every time the
4294  * inode changes.  But, it is most likely to find the inode in cache.
4295  * FIXME, needs more benchmarking...there are no reasons other than performance
4296  * to keep or drop this code.
4297  */
4298 void btrfs_dirty_inode(struct inode *inode, int flags)
4299 {
4300         struct btrfs_root *root = BTRFS_I(inode)->root;
4301         struct btrfs_trans_handle *trans;
4302         int ret;
4303
4304         if (BTRFS_I(inode)->dummy_inode)
4305                 return;
4306
4307         trans = btrfs_join_transaction(root);
4308         BUG_ON(IS_ERR(trans));
4309
4310         ret = btrfs_update_inode(trans, root, inode);
4311         if (ret && ret == -ENOSPC) {
4312                 /* whoops, lets try again with the full transaction */
4313                 btrfs_end_transaction(trans, root);
4314                 trans = btrfs_start_transaction(root, 1);
4315                 if (IS_ERR(trans)) {
4316                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4317                                        "dirty  inode %llu error %ld\n",
4318                                        (unsigned long long)btrfs_ino(inode),
4319                                        PTR_ERR(trans));
4320                         return;
4321                 }
4322
4323                 ret = btrfs_update_inode(trans, root, inode);
4324                 if (ret) {
4325                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4326                                        "dirty  inode %llu error %d\n",
4327                                        (unsigned long long)btrfs_ino(inode),
4328                                        ret);
4329                 }
4330         }
4331         btrfs_end_transaction(trans, root);
4332         if (BTRFS_I(inode)->delayed_node)
4333                 btrfs_balance_delayed_items(root);
4334 }
4335
4336 /*
4337  * find the highest existing sequence number in a directory
4338  * and then set the in-memory index_cnt variable to reflect
4339  * free sequence numbers
4340  */
4341 static int btrfs_set_inode_index_count(struct inode *inode)
4342 {
4343         struct btrfs_root *root = BTRFS_I(inode)->root;
4344         struct btrfs_key key, found_key;
4345         struct btrfs_path *path;
4346         struct extent_buffer *leaf;
4347         int ret;
4348
4349         key.objectid = btrfs_ino(inode);
4350         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4351         key.offset = (u64)-1;
4352
4353         path = btrfs_alloc_path();
4354         if (!path)
4355                 return -ENOMEM;
4356
4357         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4358         if (ret < 0)
4359                 goto out;
4360         /* FIXME: we should be able to handle this */
4361         if (ret == 0)
4362                 goto out;
4363         ret = 0;
4364
4365         /*
4366          * MAGIC NUMBER EXPLANATION:
4367          * since we search a directory based on f_pos we have to start at 2
4368          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4369          * else has to start at 2
4370          */
4371         if (path->slots[0] == 0) {
4372                 BTRFS_I(inode)->index_cnt = 2;
4373                 goto out;
4374         }
4375
4376         path->slots[0]--;
4377
4378         leaf = path->nodes[0];
4379         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4380
4381         if (found_key.objectid != btrfs_ino(inode) ||
4382             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4383                 BTRFS_I(inode)->index_cnt = 2;
4384                 goto out;
4385         }
4386
4387         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4388 out:
4389         btrfs_free_path(path);
4390         return ret;
4391 }
4392
4393 /*
4394  * helper to find a free sequence number in a given directory.  This current
4395  * code is very simple, later versions will do smarter things in the btree
4396  */
4397 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4398 {
4399         int ret = 0;
4400
4401         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4402                 ret = btrfs_inode_delayed_dir_index_count(dir);
4403                 if (ret) {
4404                         ret = btrfs_set_inode_index_count(dir);
4405                         if (ret)
4406                                 return ret;
4407                 }
4408         }
4409
4410         *index = BTRFS_I(dir)->index_cnt;
4411         BTRFS_I(dir)->index_cnt++;
4412
4413         return ret;
4414 }
4415
4416 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4417                                      struct btrfs_root *root,
4418                                      struct inode *dir,
4419                                      const char *name, int name_len,
4420                                      u64 ref_objectid, u64 objectid, int mode,
4421                                      u64 *index)
4422 {
4423         struct inode *inode;
4424         struct btrfs_inode_item *inode_item;
4425         struct btrfs_key *location;
4426         struct btrfs_path *path;
4427         struct btrfs_inode_ref *ref;
4428         struct btrfs_key key[2];
4429         u32 sizes[2];
4430         unsigned long ptr;
4431         int ret;
4432         int owner;
4433
4434         path = btrfs_alloc_path();
4435         BUG_ON(!path);
4436
4437         inode = new_inode(root->fs_info->sb);
4438         if (!inode) {
4439                 btrfs_free_path(path);
4440                 return ERR_PTR(-ENOMEM);
4441         }
4442
4443         /*
4444          * we have to initialize this early, so we can reclaim the inode
4445          * number if we fail afterwards in this function.
4446          */
4447         inode->i_ino = objectid;
4448
4449         if (dir) {
4450                 trace_btrfs_inode_request(dir);
4451
4452                 ret = btrfs_set_inode_index(dir, index);
4453                 if (ret) {
4454                         btrfs_free_path(path);
4455                         iput(inode);
4456                         return ERR_PTR(ret);
4457                 }
4458         }
4459         /*
4460          * index_cnt is ignored for everything but a dir,
4461          * btrfs_get_inode_index_count has an explanation for the magic
4462          * number
4463          */
4464         BTRFS_I(inode)->index_cnt = 2;
4465         BTRFS_I(inode)->root = root;
4466         BTRFS_I(inode)->generation = trans->transid;
4467         inode->i_generation = BTRFS_I(inode)->generation;
4468         btrfs_set_inode_space_info(root, inode);
4469
4470         if (mode & S_IFDIR)
4471                 owner = 0;
4472         else
4473                 owner = 1;
4474
4475         key[0].objectid = objectid;
4476         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4477         key[0].offset = 0;
4478
4479         key[1].objectid = objectid;
4480         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4481         key[1].offset = ref_objectid;
4482
4483         sizes[0] = sizeof(struct btrfs_inode_item);
4484         sizes[1] = name_len + sizeof(*ref);
4485
4486         path->leave_spinning = 1;
4487         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4488         if (ret != 0)
4489                 goto fail;
4490
4491         inode_init_owner(inode, dir, mode);
4492         inode_set_bytes(inode, 0);
4493         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4494         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4495                                   struct btrfs_inode_item);
4496         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4497
4498         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4499                              struct btrfs_inode_ref);
4500         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4501         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4502         ptr = (unsigned long)(ref + 1);
4503         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4504
4505         btrfs_mark_buffer_dirty(path->nodes[0]);
4506         btrfs_free_path(path);
4507
4508         location = &BTRFS_I(inode)->location;
4509         location->objectid = objectid;
4510         location->offset = 0;
4511         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4512
4513         btrfs_inherit_iflags(inode, dir);
4514
4515         if ((mode & S_IFREG)) {
4516                 if (btrfs_test_opt(root, NODATASUM))
4517                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4518                 if (btrfs_test_opt(root, NODATACOW) ||
4519                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4520                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4521         }
4522
4523         insert_inode_hash(inode);
4524         inode_tree_add(inode);
4525
4526         trace_btrfs_inode_new(inode);
4527         btrfs_set_inode_last_trans(trans, inode);
4528
4529         return inode;
4530 fail:
4531         if (dir)
4532                 BTRFS_I(dir)->index_cnt--;
4533         btrfs_free_path(path);
4534         iput(inode);
4535         return ERR_PTR(ret);
4536 }
4537
4538 static inline u8 btrfs_inode_type(struct inode *inode)
4539 {
4540         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4541 }
4542
4543 /*
4544  * utility function to add 'inode' into 'parent_inode' with
4545  * a give name and a given sequence number.
4546  * if 'add_backref' is true, also insert a backref from the
4547  * inode to the parent directory.
4548  */
4549 int btrfs_add_link(struct btrfs_trans_handle *trans,
4550                    struct inode *parent_inode, struct inode *inode,
4551                    const char *name, int name_len, int add_backref, u64 index)
4552 {
4553         int ret = 0;
4554         struct btrfs_key key;
4555         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4556         u64 ino = btrfs_ino(inode);
4557         u64 parent_ino = btrfs_ino(parent_inode);
4558
4559         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4560                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4561         } else {
4562                 key.objectid = ino;
4563                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4564                 key.offset = 0;
4565         }
4566
4567         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4568                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4569                                          key.objectid, root->root_key.objectid,
4570                                          parent_ino, index, name, name_len);
4571         } else if (add_backref) {
4572                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4573                                              parent_ino, index);
4574         }
4575
4576         if (ret == 0) {
4577                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4578                                             parent_inode, &key,
4579                                             btrfs_inode_type(inode), index);
4580                 BUG_ON(ret);
4581
4582                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4583                                    name_len * 2);
4584                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4585                 ret = btrfs_update_inode(trans, root, parent_inode);
4586         }
4587         return ret;
4588 }
4589
4590 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4591                             struct inode *dir, struct dentry *dentry,
4592                             struct inode *inode, int backref, u64 index)
4593 {
4594         int err = btrfs_add_link(trans, dir, inode,
4595                                  dentry->d_name.name, dentry->d_name.len,
4596                                  backref, index);
4597         if (!err) {
4598                 d_instantiate(dentry, inode);
4599                 return 0;
4600         }
4601         if (err > 0)
4602                 err = -EEXIST;
4603         return err;
4604 }
4605
4606 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4607                         int mode, dev_t rdev)
4608 {
4609         struct btrfs_trans_handle *trans;
4610         struct btrfs_root *root = BTRFS_I(dir)->root;
4611         struct inode *inode = NULL;
4612         int err;
4613         int drop_inode = 0;
4614         u64 objectid;
4615         unsigned long nr = 0;
4616         u64 index = 0;
4617
4618         if (!new_valid_dev(rdev))
4619                 return -EINVAL;
4620
4621         /*
4622          * 2 for inode item and ref
4623          * 2 for dir items
4624          * 1 for xattr if selinux is on
4625          */
4626         trans = btrfs_start_transaction(root, 5);
4627         if (IS_ERR(trans))
4628                 return PTR_ERR(trans);
4629
4630         err = btrfs_find_free_ino(root, &objectid);
4631         if (err)
4632                 goto out_unlock;
4633
4634         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4635                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4636                                 mode, &index);
4637         if (IS_ERR(inode)) {
4638                 err = PTR_ERR(inode);
4639                 goto out_unlock;
4640         }
4641
4642         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4643         if (err) {
4644                 drop_inode = 1;
4645                 goto out_unlock;
4646         }
4647
4648         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4649         if (err)
4650                 drop_inode = 1;
4651         else {
4652                 inode->i_op = &btrfs_special_inode_operations;
4653                 init_special_inode(inode, inode->i_mode, rdev);
4654                 btrfs_update_inode(trans, root, inode);
4655         }
4656 out_unlock:
4657         nr = trans->blocks_used;
4658         btrfs_end_transaction_throttle(trans, root);
4659         btrfs_btree_balance_dirty(root, nr);
4660         if (drop_inode) {
4661                 inode_dec_link_count(inode);
4662                 iput(inode);
4663         }
4664         return err;
4665 }
4666
4667 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4668                         int mode, struct nameidata *nd)
4669 {
4670         struct btrfs_trans_handle *trans;
4671         struct btrfs_root *root = BTRFS_I(dir)->root;
4672         struct inode *inode = NULL;
4673         int drop_inode = 0;
4674         int err;
4675         unsigned long nr = 0;
4676         u64 objectid;
4677         u64 index = 0;
4678
4679         /*
4680          * 2 for inode item and ref
4681          * 2 for dir items
4682          * 1 for xattr if selinux is on
4683          */
4684         trans = btrfs_start_transaction(root, 5);
4685         if (IS_ERR(trans))
4686                 return PTR_ERR(trans);
4687
4688         err = btrfs_find_free_ino(root, &objectid);
4689         if (err)
4690                 goto out_unlock;
4691
4692         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4693                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4694                                 mode, &index);
4695         if (IS_ERR(inode)) {
4696                 err = PTR_ERR(inode);
4697                 goto out_unlock;
4698         }
4699
4700         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4701         if (err) {
4702                 drop_inode = 1;
4703                 goto out_unlock;
4704         }
4705
4706         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4707         if (err)
4708                 drop_inode = 1;
4709         else {
4710                 inode->i_mapping->a_ops = &btrfs_aops;
4711                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4712                 inode->i_fop = &btrfs_file_operations;
4713                 inode->i_op = &btrfs_file_inode_operations;
4714                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4715         }
4716 out_unlock:
4717         nr = trans->blocks_used;
4718         btrfs_end_transaction_throttle(trans, root);
4719         if (drop_inode) {
4720                 inode_dec_link_count(inode);
4721                 iput(inode);
4722         }
4723         btrfs_btree_balance_dirty(root, nr);
4724         return err;
4725 }
4726
4727 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4728                       struct dentry *dentry)
4729 {
4730         struct btrfs_trans_handle *trans;
4731         struct btrfs_root *root = BTRFS_I(dir)->root;
4732         struct inode *inode = old_dentry->d_inode;
4733         u64 index;
4734         unsigned long nr = 0;
4735         int err;
4736         int drop_inode = 0;
4737
4738         /* do not allow sys_link's with other subvols of the same device */
4739         if (root->objectid != BTRFS_I(inode)->root->objectid)
4740                 return -EXDEV;
4741
4742         if (inode->i_nlink == ~0U)
4743                 return -EMLINK;
4744
4745         err = btrfs_set_inode_index(dir, &index);
4746         if (err)
4747                 goto fail;
4748
4749         /*
4750          * 2 items for inode and inode ref
4751          * 2 items for dir items
4752          * 1 item for parent inode
4753          */
4754         trans = btrfs_start_transaction(root, 5);
4755         if (IS_ERR(trans)) {
4756                 err = PTR_ERR(trans);
4757                 goto fail;
4758         }
4759
4760         btrfs_inc_nlink(inode);
4761         inode->i_ctime = CURRENT_TIME;
4762         ihold(inode);
4763
4764         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4765
4766         if (err) {
4767                 drop_inode = 1;
4768         } else {
4769                 struct dentry *parent = dentry->d_parent;
4770                 err = btrfs_update_inode(trans, root, inode);
4771                 BUG_ON(err);
4772                 btrfs_log_new_name(trans, inode, NULL, parent);
4773         }
4774
4775         nr = trans->blocks_used;
4776         btrfs_end_transaction_throttle(trans, root);
4777 fail:
4778         if (drop_inode) {
4779                 inode_dec_link_count(inode);
4780                 iput(inode);
4781         }
4782         btrfs_btree_balance_dirty(root, nr);
4783         return err;
4784 }
4785
4786 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4787 {
4788         struct inode *inode = NULL;
4789         struct btrfs_trans_handle *trans;
4790         struct btrfs_root *root = BTRFS_I(dir)->root;
4791         int err = 0;
4792         int drop_on_err = 0;
4793         u64 objectid = 0;
4794         u64 index = 0;
4795         unsigned long nr = 1;
4796
4797         /*
4798          * 2 items for inode and ref
4799          * 2 items for dir items
4800          * 1 for xattr if selinux is on
4801          */
4802         trans = btrfs_start_transaction(root, 5);
4803         if (IS_ERR(trans))
4804                 return PTR_ERR(trans);
4805
4806         err = btrfs_find_free_ino(root, &objectid);
4807         if (err)
4808                 goto out_fail;
4809
4810         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4811                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4812                                 S_IFDIR | mode, &index);
4813         if (IS_ERR(inode)) {
4814                 err = PTR_ERR(inode);
4815                 goto out_fail;
4816         }
4817
4818         drop_on_err = 1;
4819
4820         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4821         if (err)
4822                 goto out_fail;
4823
4824         inode->i_op = &btrfs_dir_inode_operations;
4825         inode->i_fop = &btrfs_dir_file_operations;
4826
4827         btrfs_i_size_write(inode, 0);
4828         err = btrfs_update_inode(trans, root, inode);
4829         if (err)
4830                 goto out_fail;
4831
4832         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4833                              dentry->d_name.len, 0, index);
4834         if (err)
4835                 goto out_fail;
4836
4837         d_instantiate(dentry, inode);
4838         drop_on_err = 0;
4839
4840 out_fail:
4841         nr = trans->blocks_used;
4842         btrfs_end_transaction_throttle(trans, root);
4843         if (drop_on_err)
4844                 iput(inode);
4845         btrfs_btree_balance_dirty(root, nr);
4846         return err;
4847 }
4848
4849 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4850  * and an extent that you want to insert, deal with overlap and insert
4851  * the new extent into the tree.
4852  */
4853 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4854                                 struct extent_map *existing,
4855                                 struct extent_map *em,
4856                                 u64 map_start, u64 map_len)
4857 {
4858         u64 start_diff;
4859
4860         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4861         start_diff = map_start - em->start;
4862         em->start = map_start;
4863         em->len = map_len;
4864         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4865             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4866                 em->block_start += start_diff;
4867                 em->block_len -= start_diff;
4868         }
4869         return add_extent_mapping(em_tree, em);
4870 }
4871
4872 static noinline int uncompress_inline(struct btrfs_path *path,
4873                                       struct inode *inode, struct page *page,
4874                                       size_t pg_offset, u64 extent_offset,
4875                                       struct btrfs_file_extent_item *item)
4876 {
4877         int ret;
4878         struct extent_buffer *leaf = path->nodes[0];
4879         char *tmp;
4880         size_t max_size;
4881         unsigned long inline_size;
4882         unsigned long ptr;
4883         int compress_type;
4884
4885         WARN_ON(pg_offset != 0);
4886         compress_type = btrfs_file_extent_compression(leaf, item);
4887         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4888         inline_size = btrfs_file_extent_inline_item_len(leaf,
4889                                         btrfs_item_nr(leaf, path->slots[0]));
4890         tmp = kmalloc(inline_size, GFP_NOFS);
4891         if (!tmp)
4892                 return -ENOMEM;
4893         ptr = btrfs_file_extent_inline_start(item);
4894
4895         read_extent_buffer(leaf, tmp, ptr, inline_size);
4896
4897         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4898         ret = btrfs_decompress(compress_type, tmp, page,
4899                                extent_offset, inline_size, max_size);
4900         if (ret) {
4901                 char *kaddr = kmap_atomic(page, KM_USER0);
4902                 unsigned long copy_size = min_t(u64,
4903                                   PAGE_CACHE_SIZE - pg_offset,
4904                                   max_size - extent_offset);
4905                 memset(kaddr + pg_offset, 0, copy_size);
4906                 kunmap_atomic(kaddr, KM_USER0);
4907         }
4908         kfree(tmp);
4909         return 0;
4910 }
4911
4912 /*
4913  * a bit scary, this does extent mapping from logical file offset to the disk.
4914  * the ugly parts come from merging extents from the disk with the in-ram
4915  * representation.  This gets more complex because of the data=ordered code,
4916  * where the in-ram extents might be locked pending data=ordered completion.
4917  *
4918  * This also copies inline extents directly into the page.
4919  */
4920
4921 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4922                                     size_t pg_offset, u64 start, u64 len,
4923                                     int create)
4924 {
4925         int ret;
4926         int err = 0;
4927         u64 bytenr;
4928         u64 extent_start = 0;
4929         u64 extent_end = 0;
4930         u64 objectid = btrfs_ino(inode);
4931         u32 found_type;
4932         struct btrfs_path *path = NULL;
4933         struct btrfs_root *root = BTRFS_I(inode)->root;
4934         struct btrfs_file_extent_item *item;
4935         struct extent_buffer *leaf;
4936         struct btrfs_key found_key;
4937         struct extent_map *em = NULL;
4938         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4939         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4940         struct btrfs_trans_handle *trans = NULL;
4941         int compress_type;
4942
4943 again:
4944         read_lock(&em_tree->lock);
4945         em = lookup_extent_mapping(em_tree, start, len);
4946         if (em)
4947                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4948         read_unlock(&em_tree->lock);
4949
4950         if (em) {
4951                 if (em->start > start || em->start + em->len <= start)
4952                         free_extent_map(em);
4953                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4954                         free_extent_map(em);
4955                 else
4956                         goto out;
4957         }
4958         em = alloc_extent_map();
4959         if (!em) {
4960                 err = -ENOMEM;
4961                 goto out;
4962         }
4963         em->bdev = root->fs_info->fs_devices->latest_bdev;
4964         em->start = EXTENT_MAP_HOLE;
4965         em->orig_start = EXTENT_MAP_HOLE;
4966         em->len = (u64)-1;
4967         em->block_len = (u64)-1;
4968
4969         if (!path) {
4970                 path = btrfs_alloc_path();
4971                 if (!path) {
4972                         err = -ENOMEM;
4973                         goto out;
4974                 }
4975                 /*
4976                  * Chances are we'll be called again, so go ahead and do
4977                  * readahead
4978                  */
4979                 path->reada = 1;
4980         }
4981
4982         ret = btrfs_lookup_file_extent(trans, root, path,
4983                                        objectid, start, trans != NULL);
4984         if (ret < 0) {
4985                 err = ret;
4986                 goto out;
4987         }
4988
4989         if (ret != 0) {
4990                 if (path->slots[0] == 0)
4991                         goto not_found;
4992                 path->slots[0]--;
4993         }
4994
4995         leaf = path->nodes[0];
4996         item = btrfs_item_ptr(leaf, path->slots[0],
4997                               struct btrfs_file_extent_item);
4998         /* are we inside the extent that was found? */
4999         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5000         found_type = btrfs_key_type(&found_key);
5001         if (found_key.objectid != objectid ||
5002             found_type != BTRFS_EXTENT_DATA_KEY) {
5003                 goto not_found;
5004         }
5005
5006         found_type = btrfs_file_extent_type(leaf, item);
5007         extent_start = found_key.offset;
5008         compress_type = btrfs_file_extent_compression(leaf, item);
5009         if (found_type == BTRFS_FILE_EXTENT_REG ||
5010             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5011                 extent_end = extent_start +
5012                        btrfs_file_extent_num_bytes(leaf, item);
5013         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5014                 size_t size;
5015                 size = btrfs_file_extent_inline_len(leaf, item);
5016                 extent_end = (extent_start + size + root->sectorsize - 1) &
5017                         ~((u64)root->sectorsize - 1);
5018         }
5019
5020         if (start >= extent_end) {
5021                 path->slots[0]++;
5022                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5023                         ret = btrfs_next_leaf(root, path);
5024                         if (ret < 0) {
5025                                 err = ret;
5026                                 goto out;
5027                         }
5028                         if (ret > 0)
5029                                 goto not_found;
5030                         leaf = path->nodes[0];
5031                 }
5032                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5033                 if (found_key.objectid != objectid ||
5034                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5035                         goto not_found;
5036                 if (start + len <= found_key.offset)
5037                         goto not_found;
5038                 em->start = start;
5039                 em->len = found_key.offset - start;
5040                 goto not_found_em;
5041         }
5042
5043         if (found_type == BTRFS_FILE_EXTENT_REG ||
5044             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5045                 em->start = extent_start;
5046                 em->len = extent_end - extent_start;
5047                 em->orig_start = extent_start -
5048                                  btrfs_file_extent_offset(leaf, item);
5049                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5050                 if (bytenr == 0) {
5051                         em->block_start = EXTENT_MAP_HOLE;
5052                         goto insert;
5053                 }
5054                 if (compress_type != BTRFS_COMPRESS_NONE) {
5055                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5056                         em->compress_type = compress_type;
5057                         em->block_start = bytenr;
5058                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5059                                                                          item);
5060                 } else {
5061                         bytenr += btrfs_file_extent_offset(leaf, item);
5062                         em->block_start = bytenr;
5063                         em->block_len = em->len;
5064                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5065                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5066                 }
5067                 goto insert;
5068         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5069                 unsigned long ptr;
5070                 char *map;
5071                 size_t size;
5072                 size_t extent_offset;
5073                 size_t copy_size;
5074
5075                 em->block_start = EXTENT_MAP_INLINE;
5076                 if (!page || create) {
5077                         em->start = extent_start;
5078                         em->len = extent_end - extent_start;
5079                         goto out;
5080                 }
5081
5082                 size = btrfs_file_extent_inline_len(leaf, item);
5083                 extent_offset = page_offset(page) + pg_offset - extent_start;
5084                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5085                                 size - extent_offset);
5086                 em->start = extent_start + extent_offset;
5087                 em->len = (copy_size + root->sectorsize - 1) &
5088                         ~((u64)root->sectorsize - 1);
5089                 em->orig_start = EXTENT_MAP_INLINE;
5090                 if (compress_type) {
5091                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5092                         em->compress_type = compress_type;
5093                 }
5094                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5095                 if (create == 0 && !PageUptodate(page)) {
5096                         if (btrfs_file_extent_compression(leaf, item) !=
5097                             BTRFS_COMPRESS_NONE) {
5098                                 ret = uncompress_inline(path, inode, page,
5099                                                         pg_offset,
5100                                                         extent_offset, item);
5101                                 BUG_ON(ret);
5102                         } else {
5103                                 map = kmap(page);
5104                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5105                                                    copy_size);
5106                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5107                                         memset(map + pg_offset + copy_size, 0,
5108                                                PAGE_CACHE_SIZE - pg_offset -
5109                                                copy_size);
5110                                 }
5111                                 kunmap(page);
5112                         }
5113                         flush_dcache_page(page);
5114                 } else if (create && PageUptodate(page)) {
5115                         WARN_ON(1);
5116                         if (!trans) {
5117                                 kunmap(page);
5118                                 free_extent_map(em);
5119                                 em = NULL;
5120
5121                                 btrfs_release_path(path);
5122                                 trans = btrfs_join_transaction(root);
5123
5124                                 if (IS_ERR(trans))
5125                                         return ERR_CAST(trans);
5126                                 goto again;
5127                         }
5128                         map = kmap(page);
5129                         write_extent_buffer(leaf, map + pg_offset, ptr,
5130                                             copy_size);
5131                         kunmap(page);
5132                         btrfs_mark_buffer_dirty(leaf);
5133                 }
5134                 set_extent_uptodate(io_tree, em->start,
5135                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5136                 goto insert;
5137         } else {
5138                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5139                 WARN_ON(1);
5140         }
5141 not_found:
5142         em->start = start;
5143         em->len = len;
5144 not_found_em:
5145         em->block_start = EXTENT_MAP_HOLE;
5146         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5147 insert:
5148         btrfs_release_path(path);
5149         if (em->start > start || extent_map_end(em) <= start) {
5150                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5151                        "[%llu %llu]\n", (unsigned long long)em->start,
5152                        (unsigned long long)em->len,
5153                        (unsigned long long)start,
5154                        (unsigned long long)len);
5155                 err = -EIO;
5156                 goto out;
5157         }
5158
5159         err = 0;
5160         write_lock(&em_tree->lock);
5161         ret = add_extent_mapping(em_tree, em);
5162         /* it is possible that someone inserted the extent into the tree
5163          * while we had the lock dropped.  It is also possible that
5164          * an overlapping map exists in the tree
5165          */
5166         if (ret == -EEXIST) {
5167                 struct extent_map *existing;
5168
5169                 ret = 0;
5170
5171                 existing = lookup_extent_mapping(em_tree, start, len);
5172                 if (existing && (existing->start > start ||
5173                     existing->start + existing->len <= start)) {
5174                         free_extent_map(existing);
5175                         existing = NULL;
5176                 }
5177                 if (!existing) {
5178                         existing = lookup_extent_mapping(em_tree, em->start,
5179                                                          em->len);
5180                         if (existing) {
5181                                 err = merge_extent_mapping(em_tree, existing,
5182                                                            em, start,
5183                                                            root->sectorsize);
5184                                 free_extent_map(existing);
5185                                 if (err) {
5186                                         free_extent_map(em);
5187                                         em = NULL;
5188                                 }
5189                         } else {
5190                                 err = -EIO;
5191                                 free_extent_map(em);
5192                                 em = NULL;
5193                         }
5194                 } else {
5195                         free_extent_map(em);
5196                         em = existing;
5197                         err = 0;
5198                 }
5199         }
5200         write_unlock(&em_tree->lock);
5201 out:
5202
5203         trace_btrfs_get_extent(root, em);
5204
5205         if (path)
5206                 btrfs_free_path(path);
5207         if (trans) {
5208                 ret = btrfs_end_transaction(trans, root);
5209                 if (!err)
5210                         err = ret;
5211         }
5212         if (err) {
5213                 free_extent_map(em);
5214                 return ERR_PTR(err);
5215         }
5216         return em;
5217 }
5218
5219 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5220                                            size_t pg_offset, u64 start, u64 len,
5221                                            int create)
5222 {
5223         struct extent_map *em;
5224         struct extent_map *hole_em = NULL;
5225         u64 range_start = start;
5226         u64 end;
5227         u64 found;
5228         u64 found_end;
5229         int err = 0;
5230
5231         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5232         if (IS_ERR(em))
5233                 return em;
5234         if (em) {
5235                 /*
5236                  * if our em maps to a hole, there might
5237                  * actually be delalloc bytes behind it
5238                  */
5239                 if (em->block_start != EXTENT_MAP_HOLE)
5240                         return em;
5241                 else
5242                         hole_em = em;
5243         }
5244
5245         /* check to see if we've wrapped (len == -1 or similar) */
5246         end = start + len;
5247         if (end < start)
5248                 end = (u64)-1;
5249         else
5250                 end -= 1;
5251
5252         em = NULL;
5253
5254         /* ok, we didn't find anything, lets look for delalloc */
5255         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5256                                  end, len, EXTENT_DELALLOC, 1);
5257         found_end = range_start + found;
5258         if (found_end < range_start)
5259                 found_end = (u64)-1;
5260
5261         /*
5262          * we didn't find anything useful, return
5263          * the original results from get_extent()
5264          */
5265         if (range_start > end || found_end <= start) {
5266                 em = hole_em;
5267                 hole_em = NULL;
5268                 goto out;
5269         }
5270
5271         /* adjust the range_start to make sure it doesn't
5272          * go backwards from the start they passed in
5273          */
5274         range_start = max(start,range_start);
5275         found = found_end - range_start;
5276
5277         if (found > 0) {
5278                 u64 hole_start = start;
5279                 u64 hole_len = len;
5280
5281                 em = alloc_extent_map();
5282                 if (!em) {
5283                         err = -ENOMEM;
5284                         goto out;
5285                 }
5286                 /*
5287                  * when btrfs_get_extent can't find anything it
5288                  * returns one huge hole
5289                  *
5290                  * make sure what it found really fits our range, and
5291                  * adjust to make sure it is based on the start from
5292                  * the caller
5293                  */
5294                 if (hole_em) {
5295                         u64 calc_end = extent_map_end(hole_em);
5296
5297                         if (calc_end <= start || (hole_em->start > end)) {
5298                                 free_extent_map(hole_em);
5299                                 hole_em = NULL;
5300                         } else {
5301                                 hole_start = max(hole_em->start, start);
5302                                 hole_len = calc_end - hole_start;
5303                         }
5304                 }
5305                 em->bdev = NULL;
5306                 if (hole_em && range_start > hole_start) {
5307                         /* our hole starts before our delalloc, so we
5308                          * have to return just the parts of the hole
5309                          * that go until  the delalloc starts
5310                          */
5311                         em->len = min(hole_len,
5312                                       range_start - hole_start);
5313                         em->start = hole_start;
5314                         em->orig_start = hole_start;
5315                         /*
5316                          * don't adjust block start at all,
5317                          * it is fixed at EXTENT_MAP_HOLE
5318                          */
5319                         em->block_start = hole_em->block_start;
5320                         em->block_len = hole_len;
5321                 } else {
5322                         em->start = range_start;
5323                         em->len = found;
5324                         em->orig_start = range_start;
5325                         em->block_start = EXTENT_MAP_DELALLOC;
5326                         em->block_len = found;
5327                 }
5328         } else if (hole_em) {
5329                 return hole_em;
5330         }
5331 out:
5332
5333         free_extent_map(hole_em);
5334         if (err) {
5335                 free_extent_map(em);
5336                 return ERR_PTR(err);
5337         }
5338         return em;
5339 }
5340
5341 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5342                                                   struct extent_map *em,
5343                                                   u64 start, u64 len)
5344 {
5345         struct btrfs_root *root = BTRFS_I(inode)->root;
5346         struct btrfs_trans_handle *trans;
5347         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5348         struct btrfs_key ins;
5349         u64 alloc_hint;
5350         int ret;
5351         bool insert = false;
5352
5353         /*
5354          * Ok if the extent map we looked up is a hole and is for the exact
5355          * range we want, there is no reason to allocate a new one, however if
5356          * it is not right then we need to free this one and drop the cache for
5357          * our range.
5358          */
5359         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5360             em->len != len) {
5361                 free_extent_map(em);
5362                 em = NULL;
5363                 insert = true;
5364                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5365         }
5366
5367         trans = btrfs_join_transaction(root);
5368         if (IS_ERR(trans))
5369                 return ERR_CAST(trans);
5370
5371         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5372                 btrfs_add_inode_defrag(trans, inode);
5373
5374         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5375
5376         alloc_hint = get_extent_allocation_hint(inode, start, len);
5377         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5378                                    alloc_hint, (u64)-1, &ins, 1);
5379         if (ret) {
5380                 em = ERR_PTR(ret);
5381                 goto out;
5382         }
5383
5384         if (!em) {
5385                 em = alloc_extent_map();
5386                 if (!em) {
5387                         em = ERR_PTR(-ENOMEM);
5388                         goto out;
5389                 }
5390         }
5391
5392         em->start = start;
5393         em->orig_start = em->start;
5394         em->len = ins.offset;
5395
5396         em->block_start = ins.objectid;
5397         em->block_len = ins.offset;
5398         em->bdev = root->fs_info->fs_devices->latest_bdev;
5399
5400         /*
5401          * We need to do this because if we're using the original em we searched
5402          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5403          */
5404         em->flags = 0;
5405         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5406
5407         while (insert) {
5408                 write_lock(&em_tree->lock);
5409                 ret = add_extent_mapping(em_tree, em);
5410                 write_unlock(&em_tree->lock);
5411                 if (ret != -EEXIST)
5412                         break;
5413                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5414         }
5415
5416         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5417                                            ins.offset, ins.offset, 0);
5418         if (ret) {
5419                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5420                 em = ERR_PTR(ret);
5421         }
5422 out:
5423         btrfs_end_transaction(trans, root);
5424         return em;
5425 }
5426
5427 /*
5428  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5429  * block must be cow'd
5430  */
5431 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5432                                       struct inode *inode, u64 offset, u64 len)
5433 {
5434         struct btrfs_path *path;
5435         int ret;
5436         struct extent_buffer *leaf;
5437         struct btrfs_root *root = BTRFS_I(inode)->root;
5438         struct btrfs_file_extent_item *fi;
5439         struct btrfs_key key;
5440         u64 disk_bytenr;
5441         u64 backref_offset;
5442         u64 extent_end;
5443         u64 num_bytes;
5444         int slot;
5445         int found_type;
5446
5447         path = btrfs_alloc_path();
5448         if (!path)
5449                 return -ENOMEM;
5450
5451         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5452                                        offset, 0);
5453         if (ret < 0)
5454                 goto out;
5455
5456         slot = path->slots[0];
5457         if (ret == 1) {
5458                 if (slot == 0) {
5459                         /* can't find the item, must cow */
5460                         ret = 0;
5461                         goto out;
5462                 }
5463                 slot--;
5464         }
5465         ret = 0;
5466         leaf = path->nodes[0];
5467         btrfs_item_key_to_cpu(leaf, &key, slot);
5468         if (key.objectid != btrfs_ino(inode) ||
5469             key.type != BTRFS_EXTENT_DATA_KEY) {
5470                 /* not our file or wrong item type, must cow */
5471                 goto out;
5472         }
5473
5474         if (key.offset > offset) {
5475                 /* Wrong offset, must cow */
5476                 goto out;
5477         }
5478
5479         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5480         found_type = btrfs_file_extent_type(leaf, fi);
5481         if (found_type != BTRFS_FILE_EXTENT_REG &&
5482             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5483                 /* not a regular extent, must cow */
5484                 goto out;
5485         }
5486         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5487         backref_offset = btrfs_file_extent_offset(leaf, fi);
5488
5489         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5490         if (extent_end < offset + len) {
5491                 /* extent doesn't include our full range, must cow */
5492                 goto out;
5493         }
5494
5495         if (btrfs_extent_readonly(root, disk_bytenr))
5496                 goto out;
5497
5498         /*
5499          * look for other files referencing this extent, if we
5500          * find any we must cow
5501          */
5502         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5503                                   key.offset - backref_offset, disk_bytenr))
5504                 goto out;
5505
5506         /*
5507          * adjust disk_bytenr and num_bytes to cover just the bytes
5508          * in this extent we are about to write.  If there
5509          * are any csums in that range we have to cow in order
5510          * to keep the csums correct
5511          */
5512         disk_bytenr += backref_offset;
5513         disk_bytenr += offset - key.offset;
5514         num_bytes = min(offset + len, extent_end) - offset;
5515         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5516                                 goto out;
5517         /*
5518          * all of the above have passed, it is safe to overwrite this extent
5519          * without cow
5520          */
5521         ret = 1;
5522 out:
5523         btrfs_free_path(path);
5524         return ret;
5525 }
5526
5527 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5528                                    struct buffer_head *bh_result, int create)
5529 {
5530         struct extent_map *em;
5531         struct btrfs_root *root = BTRFS_I(inode)->root;
5532         u64 start = iblock << inode->i_blkbits;
5533         u64 len = bh_result->b_size;
5534         struct btrfs_trans_handle *trans;
5535
5536         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5537         if (IS_ERR(em))
5538                 return PTR_ERR(em);
5539
5540         /*
5541          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5542          * io.  INLINE is special, and we could probably kludge it in here, but
5543          * it's still buffered so for safety lets just fall back to the generic
5544          * buffered path.
5545          *
5546          * For COMPRESSED we _have_ to read the entire extent in so we can
5547          * decompress it, so there will be buffering required no matter what we
5548          * do, so go ahead and fallback to buffered.
5549          *
5550          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5551          * to buffered IO.  Don't blame me, this is the price we pay for using
5552          * the generic code.
5553          */
5554         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5555             em->block_start == EXTENT_MAP_INLINE) {
5556                 free_extent_map(em);
5557                 return -ENOTBLK;
5558         }
5559
5560         /* Just a good old fashioned hole, return */
5561         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5562                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5563                 free_extent_map(em);
5564                 /* DIO will do one hole at a time, so just unlock a sector */
5565                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5566                               start + root->sectorsize - 1, GFP_NOFS);
5567                 return 0;
5568         }
5569
5570         /*
5571          * We don't allocate a new extent in the following cases
5572          *
5573          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5574          * existing extent.
5575          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5576          * just use the extent.
5577          *
5578          */
5579         if (!create) {
5580                 len = em->len - (start - em->start);
5581                 goto map;
5582         }
5583
5584         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5585             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5586              em->block_start != EXTENT_MAP_HOLE)) {
5587                 int type;
5588                 int ret;
5589                 u64 block_start;
5590
5591                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5592                         type = BTRFS_ORDERED_PREALLOC;
5593                 else
5594                         type = BTRFS_ORDERED_NOCOW;
5595                 len = min(len, em->len - (start - em->start));
5596                 block_start = em->block_start + (start - em->start);
5597
5598                 /*
5599                  * we're not going to log anything, but we do need
5600                  * to make sure the current transaction stays open
5601                  * while we look for nocow cross refs
5602                  */
5603                 trans = btrfs_join_transaction(root);
5604                 if (IS_ERR(trans))
5605                         goto must_cow;
5606
5607                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5608                         ret = btrfs_add_ordered_extent_dio(inode, start,
5609                                            block_start, len, len, type);
5610                         btrfs_end_transaction(trans, root);
5611                         if (ret) {
5612                                 free_extent_map(em);
5613                                 return ret;
5614                         }
5615                         goto unlock;
5616                 }
5617                 btrfs_end_transaction(trans, root);
5618         }
5619 must_cow:
5620         /*
5621          * this will cow the extent, reset the len in case we changed
5622          * it above
5623          */
5624         len = bh_result->b_size;
5625         em = btrfs_new_extent_direct(inode, em, start, len);
5626         if (IS_ERR(em))
5627                 return PTR_ERR(em);
5628         len = min(len, em->len - (start - em->start));
5629 unlock:
5630         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5631                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5632                           0, NULL, GFP_NOFS);
5633 map:
5634         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5635                 inode->i_blkbits;
5636         bh_result->b_size = len;
5637         bh_result->b_bdev = em->bdev;
5638         set_buffer_mapped(bh_result);
5639         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5640                 set_buffer_new(bh_result);
5641
5642         free_extent_map(em);
5643
5644         return 0;
5645 }
5646
5647 struct btrfs_dio_private {
5648         struct inode *inode;
5649         u64 logical_offset;
5650         u64 disk_bytenr;
5651         u64 bytes;
5652         u32 *csums;
5653         void *private;
5654
5655         /* number of bios pending for this dio */
5656         atomic_t pending_bios;
5657
5658         /* IO errors */
5659         int errors;
5660
5661         struct bio *orig_bio;
5662 };
5663
5664 static void btrfs_endio_direct_read(struct bio *bio, int err)
5665 {
5666         struct btrfs_dio_private *dip = bio->bi_private;
5667         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5668         struct bio_vec *bvec = bio->bi_io_vec;
5669         struct inode *inode = dip->inode;
5670         struct btrfs_root *root = BTRFS_I(inode)->root;
5671         u64 start;
5672         u32 *private = dip->csums;
5673
5674         start = dip->logical_offset;
5675         do {
5676                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5677                         struct page *page = bvec->bv_page;
5678                         char *kaddr;
5679                         u32 csum = ~(u32)0;
5680                         unsigned long flags;
5681
5682                         local_irq_save(flags);
5683                         kaddr = kmap_atomic(page, KM_IRQ0);
5684                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5685                                                csum, bvec->bv_len);
5686                         btrfs_csum_final(csum, (char *)&csum);
5687                         kunmap_atomic(kaddr, KM_IRQ0);
5688                         local_irq_restore(flags);
5689
5690                         flush_dcache_page(bvec->bv_page);
5691                         if (csum != *private) {
5692                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5693                                       " %llu csum %u private %u\n",
5694                                       (unsigned long long)btrfs_ino(inode),
5695                                       (unsigned long long)start,
5696                                       csum, *private);
5697                                 err = -EIO;
5698                         }
5699                 }
5700
5701                 start += bvec->bv_len;
5702                 private++;
5703                 bvec++;
5704         } while (bvec <= bvec_end);
5705
5706         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5707                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5708         bio->bi_private = dip->private;
5709
5710         kfree(dip->csums);
5711         kfree(dip);
5712
5713         /* If we had a csum failure make sure to clear the uptodate flag */
5714         if (err)
5715                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5716         dio_end_io(bio, err);
5717 }
5718
5719 static void btrfs_endio_direct_write(struct bio *bio, int err)
5720 {
5721         struct btrfs_dio_private *dip = bio->bi_private;
5722         struct inode *inode = dip->inode;
5723         struct btrfs_root *root = BTRFS_I(inode)->root;
5724         struct btrfs_trans_handle *trans;
5725         struct btrfs_ordered_extent *ordered = NULL;
5726         struct extent_state *cached_state = NULL;
5727         u64 ordered_offset = dip->logical_offset;
5728         u64 ordered_bytes = dip->bytes;
5729         int ret;
5730
5731         if (err)
5732                 goto out_done;
5733 again:
5734         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5735                                                    &ordered_offset,
5736                                                    ordered_bytes);
5737         if (!ret)
5738                 goto out_test;
5739
5740         BUG_ON(!ordered);
5741
5742         trans = btrfs_join_transaction(root);
5743         if (IS_ERR(trans)) {
5744                 err = -ENOMEM;
5745                 goto out;
5746         }
5747         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5748
5749         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5750                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5751                 if (!ret)
5752                         ret = btrfs_update_inode(trans, root, inode);
5753                 err = ret;
5754                 goto out;
5755         }
5756
5757         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5758                          ordered->file_offset + ordered->len - 1, 0,
5759                          &cached_state, GFP_NOFS);
5760
5761         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5762                 ret = btrfs_mark_extent_written(trans, inode,
5763                                                 ordered->file_offset,
5764                                                 ordered->file_offset +
5765                                                 ordered->len);
5766                 if (ret) {
5767                         err = ret;
5768                         goto out_unlock;
5769                 }
5770         } else {
5771                 ret = insert_reserved_file_extent(trans, inode,
5772                                                   ordered->file_offset,
5773                                                   ordered->start,
5774                                                   ordered->disk_len,
5775                                                   ordered->len,
5776                                                   ordered->len,
5777                                                   0, 0, 0,
5778                                                   BTRFS_FILE_EXTENT_REG);
5779                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5780                                    ordered->file_offset, ordered->len);
5781                 if (ret) {
5782                         err = ret;
5783                         WARN_ON(1);
5784                         goto out_unlock;
5785                 }
5786         }
5787
5788         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5789         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5790         if (!ret)
5791                 btrfs_update_inode(trans, root, inode);
5792         ret = 0;
5793 out_unlock:
5794         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5795                              ordered->file_offset + ordered->len - 1,
5796                              &cached_state, GFP_NOFS);
5797 out:
5798         btrfs_delalloc_release_metadata(inode, ordered->len);
5799         btrfs_end_transaction(trans, root);
5800         ordered_offset = ordered->file_offset + ordered->len;
5801         btrfs_put_ordered_extent(ordered);
5802         btrfs_put_ordered_extent(ordered);
5803
5804 out_test:
5805         /*
5806          * our bio might span multiple ordered extents.  If we haven't
5807          * completed the accounting for the whole dio, go back and try again
5808          */
5809         if (ordered_offset < dip->logical_offset + dip->bytes) {
5810                 ordered_bytes = dip->logical_offset + dip->bytes -
5811                         ordered_offset;
5812                 goto again;
5813         }
5814 out_done:
5815         bio->bi_private = dip->private;
5816
5817         kfree(dip->csums);
5818         kfree(dip);
5819
5820         /* If we had an error make sure to clear the uptodate flag */
5821         if (err)
5822                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5823         dio_end_io(bio, err);
5824 }
5825
5826 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5827                                     struct bio *bio, int mirror_num,
5828                                     unsigned long bio_flags, u64 offset)
5829 {
5830         int ret;
5831         struct btrfs_root *root = BTRFS_I(inode)->root;
5832         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5833         BUG_ON(ret);
5834         return 0;
5835 }
5836
5837 static void btrfs_end_dio_bio(struct bio *bio, int err)
5838 {
5839         struct btrfs_dio_private *dip = bio->bi_private;
5840
5841         if (err) {
5842                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5843                       "sector %#Lx len %u err no %d\n",
5844                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5845                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5846                 dip->errors = 1;
5847
5848                 /*
5849                  * before atomic variable goto zero, we must make sure
5850                  * dip->errors is perceived to be set.
5851                  */
5852                 smp_mb__before_atomic_dec();
5853         }
5854
5855         /* if there are more bios still pending for this dio, just exit */
5856         if (!atomic_dec_and_test(&dip->pending_bios))
5857                 goto out;
5858
5859         if (dip->errors)
5860                 bio_io_error(dip->orig_bio);
5861         else {
5862                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5863                 bio_endio(dip->orig_bio, 0);
5864         }
5865 out:
5866         bio_put(bio);
5867 }
5868
5869 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5870                                        u64 first_sector, gfp_t gfp_flags)
5871 {
5872         int nr_vecs = bio_get_nr_vecs(bdev);
5873         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5874 }
5875
5876 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5877                                          int rw, u64 file_offset, int skip_sum,
5878                                          u32 *csums, int async_submit)
5879 {
5880         int write = rw & REQ_WRITE;
5881         struct btrfs_root *root = BTRFS_I(inode)->root;
5882         int ret;
5883
5884         bio_get(bio);
5885         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5886         if (ret)
5887                 goto err;
5888
5889         if (skip_sum)
5890                 goto map;
5891
5892         if (write && async_submit) {
5893                 ret = btrfs_wq_submit_bio(root->fs_info,
5894                                    inode, rw, bio, 0, 0,
5895                                    file_offset,
5896                                    __btrfs_submit_bio_start_direct_io,
5897                                    __btrfs_submit_bio_done);
5898                 goto err;
5899         } else if (write) {
5900                 /*
5901                  * If we aren't doing async submit, calculate the csum of the
5902                  * bio now.
5903                  */
5904                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5905                 if (ret)
5906                         goto err;
5907         } else if (!skip_sum) {
5908                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5909                                           file_offset, csums);
5910                 if (ret)
5911                         goto err;
5912         }
5913
5914 map:
5915         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5916 err:
5917         bio_put(bio);
5918         return ret;
5919 }
5920
5921 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5922                                     int skip_sum)
5923 {
5924         struct inode *inode = dip->inode;
5925         struct btrfs_root *root = BTRFS_I(inode)->root;
5926         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5927         struct bio *bio;
5928         struct bio *orig_bio = dip->orig_bio;
5929         struct bio_vec *bvec = orig_bio->bi_io_vec;
5930         u64 start_sector = orig_bio->bi_sector;
5931         u64 file_offset = dip->logical_offset;
5932         u64 submit_len = 0;
5933         u64 map_length;
5934         int nr_pages = 0;
5935         u32 *csums = dip->csums;
5936         int ret = 0;
5937         int async_submit = 0;
5938         int write = rw & REQ_WRITE;
5939
5940         map_length = orig_bio->bi_size;
5941         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5942                               &map_length, NULL, 0);
5943         if (ret) {
5944                 bio_put(orig_bio);
5945                 return -EIO;
5946         }
5947
5948         if (map_length >= orig_bio->bi_size) {
5949                 bio = orig_bio;
5950                 goto submit;
5951         }
5952
5953         async_submit = 1;
5954         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5955         if (!bio)
5956                 return -ENOMEM;
5957         bio->bi_private = dip;
5958         bio->bi_end_io = btrfs_end_dio_bio;
5959         atomic_inc(&dip->pending_bios);
5960
5961         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5962                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5963                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5964                                  bvec->bv_offset) < bvec->bv_len)) {
5965                         /*
5966                          * inc the count before we submit the bio so
5967                          * we know the end IO handler won't happen before
5968                          * we inc the count. Otherwise, the dip might get freed
5969                          * before we're done setting it up
5970                          */
5971                         atomic_inc(&dip->pending_bios);
5972                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5973                                                      file_offset, skip_sum,
5974                                                      csums, async_submit);
5975                         if (ret) {
5976                                 bio_put(bio);
5977                                 atomic_dec(&dip->pending_bios);
5978                                 goto out_err;
5979                         }
5980
5981                         /* Write's use the ordered csums */
5982                         if (!write && !skip_sum)
5983                                 csums = csums + nr_pages;
5984                         start_sector += submit_len >> 9;
5985                         file_offset += submit_len;
5986
5987                         submit_len = 0;
5988                         nr_pages = 0;
5989
5990                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5991                                                   start_sector, GFP_NOFS);
5992                         if (!bio)
5993                                 goto out_err;
5994                         bio->bi_private = dip;
5995                         bio->bi_end_io = btrfs_end_dio_bio;
5996
5997                         map_length = orig_bio->bi_size;
5998                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5999                                               &map_length, NULL, 0);
6000                         if (ret) {
6001                                 bio_put(bio);
6002                                 goto out_err;
6003                         }
6004                 } else {
6005                         submit_len += bvec->bv_len;
6006                         nr_pages ++;
6007                         bvec++;
6008                 }
6009         }
6010
6011 submit:
6012         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6013                                      csums, async_submit);
6014         if (!ret)
6015                 return 0;
6016
6017         bio_put(bio);
6018 out_err:
6019         dip->errors = 1;
6020         /*
6021          * before atomic variable goto zero, we must
6022          * make sure dip->errors is perceived to be set.
6023          */
6024         smp_mb__before_atomic_dec();
6025         if (atomic_dec_and_test(&dip->pending_bios))
6026                 bio_io_error(dip->orig_bio);
6027
6028         /* bio_end_io() will handle error, so we needn't return it */
6029         return 0;
6030 }
6031
6032 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6033                                 loff_t file_offset)
6034 {
6035         struct btrfs_root *root = BTRFS_I(inode)->root;
6036         struct btrfs_dio_private *dip;
6037         struct bio_vec *bvec = bio->bi_io_vec;
6038         int skip_sum;
6039         int write = rw & REQ_WRITE;
6040         int ret = 0;
6041
6042         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6043
6044         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6045         if (!dip) {
6046                 ret = -ENOMEM;
6047                 goto free_ordered;
6048         }
6049         dip->csums = NULL;
6050
6051         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6052         if (!write && !skip_sum) {
6053                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6054                 if (!dip->csums) {
6055                         kfree(dip);
6056                         ret = -ENOMEM;
6057                         goto free_ordered;
6058                 }
6059         }
6060
6061         dip->private = bio->bi_private;
6062         dip->inode = inode;
6063         dip->logical_offset = file_offset;
6064
6065         dip->bytes = 0;
6066         do {
6067                 dip->bytes += bvec->bv_len;
6068                 bvec++;
6069         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6070
6071         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6072         bio->bi_private = dip;
6073         dip->errors = 0;
6074         dip->orig_bio = bio;
6075         atomic_set(&dip->pending_bios, 0);
6076
6077         if (write)
6078                 bio->bi_end_io = btrfs_endio_direct_write;
6079         else
6080                 bio->bi_end_io = btrfs_endio_direct_read;
6081
6082         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6083         if (!ret)
6084                 return;
6085 free_ordered:
6086         /*
6087          * If this is a write, we need to clean up the reserved space and kill
6088          * the ordered extent.
6089          */
6090         if (write) {
6091                 struct btrfs_ordered_extent *ordered;
6092                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6093                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6094                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6095                         btrfs_free_reserved_extent(root, ordered->start,
6096                                                    ordered->disk_len);
6097                 btrfs_put_ordered_extent(ordered);
6098                 btrfs_put_ordered_extent(ordered);
6099         }
6100         bio_endio(bio, ret);
6101 }
6102
6103 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6104                         const struct iovec *iov, loff_t offset,
6105                         unsigned long nr_segs)
6106 {
6107         int seg;
6108         int i;
6109         size_t size;
6110         unsigned long addr;
6111         unsigned blocksize_mask = root->sectorsize - 1;
6112         ssize_t retval = -EINVAL;
6113         loff_t end = offset;
6114
6115         if (offset & blocksize_mask)
6116                 goto out;
6117
6118         /* Check the memory alignment.  Blocks cannot straddle pages */
6119         for (seg = 0; seg < nr_segs; seg++) {
6120                 addr = (unsigned long)iov[seg].iov_base;
6121                 size = iov[seg].iov_len;
6122                 end += size;
6123                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6124                         goto out;
6125
6126                 /* If this is a write we don't need to check anymore */
6127                 if (rw & WRITE)
6128                         continue;
6129
6130                 /*
6131                  * Check to make sure we don't have duplicate iov_base's in this
6132                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6133                  * when reading back.
6134                  */
6135                 for (i = seg + 1; i < nr_segs; i++) {
6136                         if (iov[seg].iov_base == iov[i].iov_base)
6137                                 goto out;
6138                 }
6139         }
6140         retval = 0;
6141 out:
6142         return retval;
6143 }
6144 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6145                         const struct iovec *iov, loff_t offset,
6146                         unsigned long nr_segs)
6147 {
6148         struct file *file = iocb->ki_filp;
6149         struct inode *inode = file->f_mapping->host;
6150         struct btrfs_ordered_extent *ordered;
6151         struct extent_state *cached_state = NULL;
6152         u64 lockstart, lockend;
6153         ssize_t ret;
6154         int writing = rw & WRITE;
6155         int write_bits = 0;
6156         size_t count = iov_length(iov, nr_segs);
6157
6158         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6159                             offset, nr_segs)) {
6160                 return 0;
6161         }
6162
6163         lockstart = offset;
6164         lockend = offset + count - 1;
6165
6166         if (writing) {
6167                 ret = btrfs_delalloc_reserve_space(inode, count);
6168                 if (ret)
6169                         goto out;
6170         }
6171
6172         while (1) {
6173                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6174                                  0, &cached_state, GFP_NOFS);
6175                 /*
6176                  * We're concerned with the entire range that we're going to be
6177                  * doing DIO to, so we need to make sure theres no ordered
6178                  * extents in this range.
6179                  */
6180                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6181                                                      lockend - lockstart + 1);
6182                 if (!ordered)
6183                         break;
6184                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6185                                      &cached_state, GFP_NOFS);
6186                 btrfs_start_ordered_extent(inode, ordered, 1);
6187                 btrfs_put_ordered_extent(ordered);
6188                 cond_resched();
6189         }
6190
6191         /*
6192          * we don't use btrfs_set_extent_delalloc because we don't want
6193          * the dirty or uptodate bits
6194          */
6195         if (writing) {
6196                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6197                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6198                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6199                                      GFP_NOFS);
6200                 if (ret) {
6201                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6202                                          lockend, EXTENT_LOCKED | write_bits,
6203                                          1, 0, &cached_state, GFP_NOFS);
6204                         goto out;
6205                 }
6206         }
6207
6208         free_extent_state(cached_state);
6209         cached_state = NULL;
6210
6211         ret = __blockdev_direct_IO(rw, iocb, inode,
6212                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6213                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6214                    btrfs_submit_direct, 0);
6215
6216         if (ret < 0 && ret != -EIOCBQUEUED) {
6217                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6218                               offset + iov_length(iov, nr_segs) - 1,
6219                               EXTENT_LOCKED | write_bits, 1, 0,
6220                               &cached_state, GFP_NOFS);
6221         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6222                 /*
6223                  * We're falling back to buffered, unlock the section we didn't
6224                  * do IO on.
6225                  */
6226                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6227                               offset + iov_length(iov, nr_segs) - 1,
6228                               EXTENT_LOCKED | write_bits, 1, 0,
6229                               &cached_state, GFP_NOFS);
6230         }
6231 out:
6232         free_extent_state(cached_state);
6233         return ret;
6234 }
6235
6236 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6237                 __u64 start, __u64 len)
6238 {
6239         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6240 }
6241
6242 int btrfs_readpage(struct file *file, struct page *page)
6243 {
6244         struct extent_io_tree *tree;
6245         tree = &BTRFS_I(page->mapping->host)->io_tree;
6246         return extent_read_full_page(tree, page, btrfs_get_extent);
6247 }
6248
6249 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6250 {
6251         struct extent_io_tree *tree;
6252
6253
6254         if (current->flags & PF_MEMALLOC) {
6255                 redirty_page_for_writepage(wbc, page);
6256                 unlock_page(page);
6257                 return 0;
6258         }
6259         tree = &BTRFS_I(page->mapping->host)->io_tree;
6260         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6261 }
6262
6263 int btrfs_writepages(struct address_space *mapping,
6264                      struct writeback_control *wbc)
6265 {
6266         struct extent_io_tree *tree;
6267
6268         tree = &BTRFS_I(mapping->host)->io_tree;
6269         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6270 }
6271
6272 static int
6273 btrfs_readpages(struct file *file, struct address_space *mapping,
6274                 struct list_head *pages, unsigned nr_pages)
6275 {
6276         struct extent_io_tree *tree;
6277         tree = &BTRFS_I(mapping->host)->io_tree;
6278         return extent_readpages(tree, mapping, pages, nr_pages,
6279                                 btrfs_get_extent);
6280 }
6281 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6282 {
6283         struct extent_io_tree *tree;
6284         struct extent_map_tree *map;
6285         int ret;
6286
6287         tree = &BTRFS_I(page->mapping->host)->io_tree;
6288         map = &BTRFS_I(page->mapping->host)->extent_tree;
6289         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6290         if (ret == 1) {
6291                 ClearPagePrivate(page);
6292                 set_page_private(page, 0);
6293                 page_cache_release(page);
6294         }
6295         return ret;
6296 }
6297
6298 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6299 {
6300         if (PageWriteback(page) || PageDirty(page))
6301                 return 0;
6302         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6303 }
6304
6305 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6306 {
6307         struct extent_io_tree *tree;
6308         struct btrfs_ordered_extent *ordered;
6309         struct extent_state *cached_state = NULL;
6310         u64 page_start = page_offset(page);
6311         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6312
6313
6314         /*
6315          * we have the page locked, so new writeback can't start,
6316          * and the dirty bit won't be cleared while we are here.
6317          *
6318          * Wait for IO on this page so that we can safely clear
6319          * the PagePrivate2 bit and do ordered accounting
6320          */
6321         wait_on_page_writeback(page);
6322
6323         tree = &BTRFS_I(page->mapping->host)->io_tree;
6324         if (offset) {
6325                 btrfs_releasepage(page, GFP_NOFS);
6326                 return;
6327         }
6328         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6329                          GFP_NOFS);
6330         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6331                                            page_offset(page));
6332         if (ordered) {
6333                 /*
6334                  * IO on this page will never be started, so we need
6335                  * to account for any ordered extents now
6336                  */
6337                 clear_extent_bit(tree, page_start, page_end,
6338                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6339                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6340                                  &cached_state, GFP_NOFS);
6341                 /*
6342                  * whoever cleared the private bit is responsible
6343                  * for the finish_ordered_io
6344                  */
6345                 if (TestClearPagePrivate2(page)) {
6346                         btrfs_finish_ordered_io(page->mapping->host,
6347                                                 page_start, page_end);
6348                 }
6349                 btrfs_put_ordered_extent(ordered);
6350                 cached_state = NULL;
6351                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6352                                  GFP_NOFS);
6353         }
6354         clear_extent_bit(tree, page_start, page_end,
6355                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6356                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6357         __btrfs_releasepage(page, GFP_NOFS);
6358
6359         ClearPageChecked(page);
6360         if (PagePrivate(page)) {
6361                 ClearPagePrivate(page);
6362                 set_page_private(page, 0);
6363                 page_cache_release(page);
6364         }
6365 }
6366
6367 /*
6368  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6369  * called from a page fault handler when a page is first dirtied. Hence we must
6370  * be careful to check for EOF conditions here. We set the page up correctly
6371  * for a written page which means we get ENOSPC checking when writing into
6372  * holes and correct delalloc and unwritten extent mapping on filesystems that
6373  * support these features.
6374  *
6375  * We are not allowed to take the i_mutex here so we have to play games to
6376  * protect against truncate races as the page could now be beyond EOF.  Because
6377  * vmtruncate() writes the inode size before removing pages, once we have the
6378  * page lock we can determine safely if the page is beyond EOF. If it is not
6379  * beyond EOF, then the page is guaranteed safe against truncation until we
6380  * unlock the page.
6381  */
6382 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6383 {
6384         struct page *page = vmf->page;
6385         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6386         struct btrfs_root *root = BTRFS_I(inode)->root;
6387         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6388         struct btrfs_ordered_extent *ordered;
6389         struct extent_state *cached_state = NULL;
6390         char *kaddr;
6391         unsigned long zero_start;
6392         loff_t size;
6393         int ret;
6394         u64 page_start;
6395         u64 page_end;
6396
6397         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6398         if (ret) {
6399                 if (ret == -ENOMEM)
6400                         ret = VM_FAULT_OOM;
6401                 else /* -ENOSPC, -EIO, etc */
6402                         ret = VM_FAULT_SIGBUS;
6403                 goto out;
6404         }
6405
6406         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6407 again:
6408         lock_page(page);
6409         size = i_size_read(inode);
6410         page_start = page_offset(page);
6411         page_end = page_start + PAGE_CACHE_SIZE - 1;
6412
6413         if ((page->mapping != inode->i_mapping) ||
6414             (page_start >= size)) {
6415                 /* page got truncated out from underneath us */
6416                 goto out_unlock;
6417         }
6418         wait_on_page_writeback(page);
6419
6420         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6421                          GFP_NOFS);
6422         set_page_extent_mapped(page);
6423
6424         /*
6425          * we can't set the delalloc bits if there are pending ordered
6426          * extents.  Drop our locks and wait for them to finish
6427          */
6428         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6429         if (ordered) {
6430                 unlock_extent_cached(io_tree, page_start, page_end,
6431                                      &cached_state, GFP_NOFS);
6432                 unlock_page(page);
6433                 btrfs_start_ordered_extent(inode, ordered, 1);
6434                 btrfs_put_ordered_extent(ordered);
6435                 goto again;
6436         }
6437
6438         /*
6439          * XXX - page_mkwrite gets called every time the page is dirtied, even
6440          * if it was already dirty, so for space accounting reasons we need to
6441          * clear any delalloc bits for the range we are fixing to save.  There
6442          * is probably a better way to do this, but for now keep consistent with
6443          * prepare_pages in the normal write path.
6444          */
6445         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6446                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6447                           0, 0, &cached_state, GFP_NOFS);
6448
6449         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6450                                         &cached_state);
6451         if (ret) {
6452                 unlock_extent_cached(io_tree, page_start, page_end,
6453                                      &cached_state, GFP_NOFS);
6454                 ret = VM_FAULT_SIGBUS;
6455                 goto out_unlock;
6456         }
6457         ret = 0;
6458
6459         /* page is wholly or partially inside EOF */
6460         if (page_start + PAGE_CACHE_SIZE > size)
6461                 zero_start = size & ~PAGE_CACHE_MASK;
6462         else
6463                 zero_start = PAGE_CACHE_SIZE;
6464
6465         if (zero_start != PAGE_CACHE_SIZE) {
6466                 kaddr = kmap(page);
6467                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6468                 flush_dcache_page(page);
6469                 kunmap(page);
6470         }
6471         ClearPageChecked(page);
6472         set_page_dirty(page);
6473         SetPageUptodate(page);
6474
6475         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6476         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6477
6478         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6479
6480 out_unlock:
6481         if (!ret)
6482                 return VM_FAULT_LOCKED;
6483         unlock_page(page);
6484         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6485 out:
6486         return ret;
6487 }
6488
6489 static int btrfs_truncate(struct inode *inode)
6490 {
6491         struct btrfs_root *root = BTRFS_I(inode)->root;
6492         struct btrfs_block_rsv *rsv;
6493         int ret;
6494         int err = 0;
6495         struct btrfs_trans_handle *trans;
6496         unsigned long nr;
6497         u64 mask = root->sectorsize - 1;
6498
6499         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6500         if (ret)
6501                 return ret;
6502
6503         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6504         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6505
6506         /*
6507          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6508          * 3 things going on here
6509          *
6510          * 1) We need to reserve space for our orphan item and the space to
6511          * delete our orphan item.  Lord knows we don't want to have a dangling
6512          * orphan item because we didn't reserve space to remove it.
6513          *
6514          * 2) We need to reserve space to update our inode.
6515          *
6516          * 3) We need to have something to cache all the space that is going to
6517          * be free'd up by the truncate operation, but also have some slack
6518          * space reserved in case it uses space during the truncate (thank you
6519          * very much snapshotting).
6520          *
6521          * And we need these to all be seperate.  The fact is we can use alot of
6522          * space doing the truncate, and we have no earthly idea how much space
6523          * we will use, so we need the truncate reservation to be seperate so it
6524          * doesn't end up using space reserved for updating the inode or
6525          * removing the orphan item.  We also need to be able to stop the
6526          * transaction and start a new one, which means we need to be able to
6527          * update the inode several times, and we have no idea of knowing how
6528          * many times that will be, so we can't just reserve 1 item for the
6529          * entirety of the opration, so that has to be done seperately as well.
6530          * Then there is the orphan item, which does indeed need to be held on
6531          * to for the whole operation, and we need nobody to touch this reserved
6532          * space except the orphan code.
6533          *
6534          * So that leaves us with
6535          *
6536          * 1) root->orphan_block_rsv - for the orphan deletion.
6537          * 2) rsv - for the truncate reservation, which we will steal from the
6538          * transaction reservation.
6539          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6540          * updating the inode.
6541          */
6542         rsv = btrfs_alloc_block_rsv(root);
6543         if (!rsv)
6544                 return -ENOMEM;
6545         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6546
6547         trans = btrfs_start_transaction(root, 4);
6548         if (IS_ERR(trans)) {
6549                 err = PTR_ERR(trans);
6550                 goto out;
6551         }
6552
6553         /*
6554          * Reserve space for the truncate process.  Truncate should be adding
6555          * space, but if there are snapshots it may end up using space.
6556          */
6557         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6558         BUG_ON(ret);
6559
6560         ret = btrfs_orphan_add(trans, inode);
6561         if (ret) {
6562                 btrfs_end_transaction(trans, root);
6563                 goto out;
6564         }
6565
6566         nr = trans->blocks_used;
6567         btrfs_end_transaction(trans, root);
6568         btrfs_btree_balance_dirty(root, nr);
6569
6570         /*
6571          * Ok so we've already migrated our bytes over for the truncate, so here
6572          * just reserve the one slot we need for updating the inode.
6573          */
6574         trans = btrfs_start_transaction(root, 1);
6575         if (IS_ERR(trans)) {
6576                 err = PTR_ERR(trans);
6577                 goto out;
6578         }
6579         trans->block_rsv = rsv;
6580
6581         /*
6582          * setattr is responsible for setting the ordered_data_close flag,
6583          * but that is only tested during the last file release.  That
6584          * could happen well after the next commit, leaving a great big
6585          * window where new writes may get lost if someone chooses to write
6586          * to this file after truncating to zero
6587          *
6588          * The inode doesn't have any dirty data here, and so if we commit
6589          * this is a noop.  If someone immediately starts writing to the inode
6590          * it is very likely we'll catch some of their writes in this
6591          * transaction, and the commit will find this file on the ordered
6592          * data list with good things to send down.
6593          *
6594          * This is a best effort solution, there is still a window where
6595          * using truncate to replace the contents of the file will
6596          * end up with a zero length file after a crash.
6597          */
6598         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6599                 btrfs_add_ordered_operation(trans, root, inode);
6600
6601         while (1) {
6602                 if (!trans) {
6603                         trans = btrfs_start_transaction(root, 3);
6604                         if (IS_ERR(trans)) {
6605                                 err = PTR_ERR(trans);
6606                                 goto out;
6607                         }
6608
6609                         ret = btrfs_truncate_reserve_metadata(trans, root,
6610                                                               rsv);
6611                         BUG_ON(ret);
6612
6613                         trans->block_rsv = rsv;
6614                 }
6615
6616                 ret = btrfs_truncate_inode_items(trans, root, inode,
6617                                                  inode->i_size,
6618                                                  BTRFS_EXTENT_DATA_KEY);
6619                 if (ret != -EAGAIN) {
6620                         err = ret;
6621                         break;
6622                 }
6623
6624                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6625                 ret = btrfs_update_inode(trans, root, inode);
6626                 if (ret) {
6627                         err = ret;
6628                         break;
6629                 }
6630
6631                 nr = trans->blocks_used;
6632                 btrfs_end_transaction(trans, root);
6633                 trans = NULL;
6634                 btrfs_btree_balance_dirty(root, nr);
6635         }
6636
6637         if (ret == 0 && inode->i_nlink > 0) {
6638                 trans->block_rsv = root->orphan_block_rsv;
6639                 ret = btrfs_orphan_del(trans, inode);
6640                 if (ret)
6641                         err = ret;
6642         } else if (ret && inode->i_nlink > 0) {
6643                 /*
6644                  * Failed to do the truncate, remove us from the in memory
6645                  * orphan list.
6646                  */
6647                 ret = btrfs_orphan_del(NULL, inode);
6648         }
6649
6650         trans->block_rsv = &root->fs_info->trans_block_rsv;
6651         ret = btrfs_update_inode(trans, root, inode);
6652         if (ret && !err)
6653                 err = ret;
6654
6655         nr = trans->blocks_used;
6656         ret = btrfs_end_transaction_throttle(trans, root);
6657         btrfs_btree_balance_dirty(root, nr);
6658
6659 out:
6660         btrfs_free_block_rsv(root, rsv);
6661
6662         if (ret && !err)
6663                 err = ret;
6664
6665         return err;
6666 }
6667
6668 /*
6669  * create a new subvolume directory/inode (helper for the ioctl).
6670  */
6671 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6672                              struct btrfs_root *new_root, u64 new_dirid)
6673 {
6674         struct inode *inode;
6675         int err;
6676         u64 index = 0;
6677
6678         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6679                                 new_dirid, S_IFDIR | 0700, &index);
6680         if (IS_ERR(inode))
6681                 return PTR_ERR(inode);
6682         inode->i_op = &btrfs_dir_inode_operations;
6683         inode->i_fop = &btrfs_dir_file_operations;
6684
6685         inode->i_nlink = 1;
6686         btrfs_i_size_write(inode, 0);
6687
6688         err = btrfs_update_inode(trans, new_root, inode);
6689         BUG_ON(err);
6690
6691         iput(inode);
6692         return 0;
6693 }
6694
6695 /* helper function for file defrag and space balancing.  This
6696  * forces readahead on a given range of bytes in an inode
6697  */
6698 unsigned long btrfs_force_ra(struct address_space *mapping,
6699                               struct file_ra_state *ra, struct file *file,
6700                               pgoff_t offset, pgoff_t last_index)
6701 {
6702         pgoff_t req_size = last_index - offset + 1;
6703
6704         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6705         return offset + req_size;
6706 }
6707
6708 struct inode *btrfs_alloc_inode(struct super_block *sb)
6709 {
6710         struct btrfs_inode *ei;
6711         struct inode *inode;
6712
6713         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6714         if (!ei)
6715                 return NULL;
6716
6717         ei->root = NULL;
6718         ei->space_info = NULL;
6719         ei->generation = 0;
6720         ei->sequence = 0;
6721         ei->last_trans = 0;
6722         ei->last_sub_trans = 0;
6723         ei->logged_trans = 0;
6724         ei->delalloc_bytes = 0;
6725         ei->reserved_bytes = 0;
6726         ei->disk_i_size = 0;
6727         ei->flags = 0;
6728         ei->index_cnt = (u64)-1;
6729         ei->last_unlink_trans = 0;
6730
6731         atomic_set(&ei->outstanding_extents, 0);
6732         atomic_set(&ei->reserved_extents, 0);
6733
6734         ei->ordered_data_close = 0;
6735         ei->orphan_meta_reserved = 0;
6736         ei->dummy_inode = 0;
6737         ei->in_defrag = 0;
6738         ei->force_compress = BTRFS_COMPRESS_NONE;
6739
6740         ei->delayed_node = NULL;
6741
6742         inode = &ei->vfs_inode;
6743         extent_map_tree_init(&ei->extent_tree);
6744         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6745         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6746         mutex_init(&ei->log_mutex);
6747         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6748         INIT_LIST_HEAD(&ei->i_orphan);
6749         INIT_LIST_HEAD(&ei->delalloc_inodes);
6750         INIT_LIST_HEAD(&ei->ordered_operations);
6751         RB_CLEAR_NODE(&ei->rb_node);
6752
6753         return inode;
6754 }
6755
6756 static void btrfs_i_callback(struct rcu_head *head)
6757 {
6758         struct inode *inode = container_of(head, struct inode, i_rcu);
6759         INIT_LIST_HEAD(&inode->i_dentry);
6760         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6761 }
6762
6763 void btrfs_destroy_inode(struct inode *inode)
6764 {
6765         struct btrfs_ordered_extent *ordered;
6766         struct btrfs_root *root = BTRFS_I(inode)->root;
6767
6768         WARN_ON(!list_empty(&inode->i_dentry));
6769         WARN_ON(inode->i_data.nrpages);
6770         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6771         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6772
6773         /*
6774          * This can happen where we create an inode, but somebody else also
6775          * created the same inode and we need to destroy the one we already
6776          * created.
6777          */
6778         if (!root)
6779                 goto free;
6780
6781         /*
6782          * Make sure we're properly removed from the ordered operation
6783          * lists.
6784          */
6785         smp_mb();
6786         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6787                 spin_lock(&root->fs_info->ordered_extent_lock);
6788                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6789                 spin_unlock(&root->fs_info->ordered_extent_lock);
6790         }
6791
6792         spin_lock(&root->orphan_lock);
6793         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6794                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6795                        (unsigned long long)btrfs_ino(inode));
6796                 list_del_init(&BTRFS_I(inode)->i_orphan);
6797         }
6798         spin_unlock(&root->orphan_lock);
6799
6800         while (1) {
6801                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6802                 if (!ordered)
6803                         break;
6804                 else {
6805                         printk(KERN_ERR "btrfs found ordered "
6806                                "extent %llu %llu on inode cleanup\n",
6807                                (unsigned long long)ordered->file_offset,
6808                                (unsigned long long)ordered->len);
6809                         btrfs_remove_ordered_extent(inode, ordered);
6810                         btrfs_put_ordered_extent(ordered);
6811                         btrfs_put_ordered_extent(ordered);
6812                 }
6813         }
6814         inode_tree_del(inode);
6815         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6816 free:
6817         btrfs_remove_delayed_node(inode);
6818         call_rcu(&inode->i_rcu, btrfs_i_callback);
6819 }
6820
6821 int btrfs_drop_inode(struct inode *inode)
6822 {
6823         struct btrfs_root *root = BTRFS_I(inode)->root;
6824
6825         if (btrfs_root_refs(&root->root_item) == 0 &&
6826             !is_free_space_inode(root, inode))
6827                 return 1;
6828         else
6829                 return generic_drop_inode(inode);
6830 }
6831
6832 static void init_once(void *foo)
6833 {
6834         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6835
6836         inode_init_once(&ei->vfs_inode);
6837 }
6838
6839 void btrfs_destroy_cachep(void)
6840 {
6841         if (btrfs_inode_cachep)
6842                 kmem_cache_destroy(btrfs_inode_cachep);
6843         if (btrfs_trans_handle_cachep)
6844                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6845         if (btrfs_transaction_cachep)
6846                 kmem_cache_destroy(btrfs_transaction_cachep);
6847         if (btrfs_path_cachep)
6848                 kmem_cache_destroy(btrfs_path_cachep);
6849         if (btrfs_free_space_cachep)
6850                 kmem_cache_destroy(btrfs_free_space_cachep);
6851 }
6852
6853 int btrfs_init_cachep(void)
6854 {
6855         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6856                         sizeof(struct btrfs_inode), 0,
6857                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6858         if (!btrfs_inode_cachep)
6859                 goto fail;
6860
6861         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6862                         sizeof(struct btrfs_trans_handle), 0,
6863                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6864         if (!btrfs_trans_handle_cachep)
6865                 goto fail;
6866
6867         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6868                         sizeof(struct btrfs_transaction), 0,
6869                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6870         if (!btrfs_transaction_cachep)
6871                 goto fail;
6872
6873         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6874                         sizeof(struct btrfs_path), 0,
6875                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6876         if (!btrfs_path_cachep)
6877                 goto fail;
6878
6879         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6880                         sizeof(struct btrfs_free_space), 0,
6881                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6882         if (!btrfs_free_space_cachep)
6883                 goto fail;
6884
6885         return 0;
6886 fail:
6887         btrfs_destroy_cachep();
6888         return -ENOMEM;
6889 }
6890
6891 static int btrfs_getattr(struct vfsmount *mnt,
6892                          struct dentry *dentry, struct kstat *stat)
6893 {
6894         struct inode *inode = dentry->d_inode;
6895         generic_fillattr(inode, stat);
6896         stat->dev = BTRFS_I(inode)->root->anon_dev;
6897         stat->blksize = PAGE_CACHE_SIZE;
6898         stat->blocks = (inode_get_bytes(inode) +
6899                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6900         return 0;
6901 }
6902
6903 /*
6904  * If a file is moved, it will inherit the cow and compression flags of the new
6905  * directory.
6906  */
6907 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6908 {
6909         struct btrfs_inode *b_dir = BTRFS_I(dir);
6910         struct btrfs_inode *b_inode = BTRFS_I(inode);
6911
6912         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6913                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6914         else
6915                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6916
6917         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6918                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6919         else
6920                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6921 }
6922
6923 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6924                            struct inode *new_dir, struct dentry *new_dentry)
6925 {
6926         struct btrfs_trans_handle *trans;
6927         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6928         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6929         struct inode *new_inode = new_dentry->d_inode;
6930         struct inode *old_inode = old_dentry->d_inode;
6931         struct timespec ctime = CURRENT_TIME;
6932         u64 index = 0;
6933         u64 root_objectid;
6934         int ret;
6935         u64 old_ino = btrfs_ino(old_inode);
6936
6937         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6938                 return -EPERM;
6939
6940         /* we only allow rename subvolume link between subvolumes */
6941         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6942                 return -EXDEV;
6943
6944         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6945             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6946                 return -ENOTEMPTY;
6947
6948         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6949             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6950                 return -ENOTEMPTY;
6951         /*
6952          * we're using rename to replace one file with another.
6953          * and the replacement file is large.  Start IO on it now so
6954          * we don't add too much work to the end of the transaction
6955          */
6956         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6957             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6958                 filemap_flush(old_inode->i_mapping);
6959
6960         /* close the racy window with snapshot create/destroy ioctl */
6961         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6962                 down_read(&root->fs_info->subvol_sem);
6963         /*
6964          * We want to reserve the absolute worst case amount of items.  So if
6965          * both inodes are subvols and we need to unlink them then that would
6966          * require 4 item modifications, but if they are both normal inodes it
6967          * would require 5 item modifications, so we'll assume their normal
6968          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6969          * should cover the worst case number of items we'll modify.
6970          */
6971         trans = btrfs_start_transaction(root, 20);
6972         if (IS_ERR(trans)) {
6973                 ret = PTR_ERR(trans);
6974                 goto out_notrans;
6975         }
6976
6977         if (dest != root)
6978                 btrfs_record_root_in_trans(trans, dest);
6979
6980         ret = btrfs_set_inode_index(new_dir, &index);
6981         if (ret)
6982                 goto out_fail;
6983
6984         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6985                 /* force full log commit if subvolume involved. */
6986                 root->fs_info->last_trans_log_full_commit = trans->transid;
6987         } else {
6988                 ret = btrfs_insert_inode_ref(trans, dest,
6989                                              new_dentry->d_name.name,
6990                                              new_dentry->d_name.len,
6991                                              old_ino,
6992                                              btrfs_ino(new_dir), index);
6993                 if (ret)
6994                         goto out_fail;
6995                 /*
6996                  * this is an ugly little race, but the rename is required
6997                  * to make sure that if we crash, the inode is either at the
6998                  * old name or the new one.  pinning the log transaction lets
6999                  * us make sure we don't allow a log commit to come in after
7000                  * we unlink the name but before we add the new name back in.
7001                  */
7002                 btrfs_pin_log_trans(root);
7003         }
7004         /*
7005          * make sure the inode gets flushed if it is replacing
7006          * something.
7007          */
7008         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7009                 btrfs_add_ordered_operation(trans, root, old_inode);
7010
7011         old_dir->i_ctime = old_dir->i_mtime = ctime;
7012         new_dir->i_ctime = new_dir->i_mtime = ctime;
7013         old_inode->i_ctime = ctime;
7014
7015         if (old_dentry->d_parent != new_dentry->d_parent)
7016                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7017
7018         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7019                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7020                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7021                                         old_dentry->d_name.name,
7022                                         old_dentry->d_name.len);
7023         } else {
7024                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7025                                         old_dentry->d_inode,
7026                                         old_dentry->d_name.name,
7027                                         old_dentry->d_name.len);
7028                 if (!ret)
7029                         ret = btrfs_update_inode(trans, root, old_inode);
7030         }
7031         BUG_ON(ret);
7032
7033         if (new_inode) {
7034                 new_inode->i_ctime = CURRENT_TIME;
7035                 if (unlikely(btrfs_ino(new_inode) ==
7036                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7037                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7038                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7039                                                 root_objectid,
7040                                                 new_dentry->d_name.name,
7041                                                 new_dentry->d_name.len);
7042                         BUG_ON(new_inode->i_nlink == 0);
7043                 } else {
7044                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7045                                                  new_dentry->d_inode,
7046                                                  new_dentry->d_name.name,
7047                                                  new_dentry->d_name.len);
7048                 }
7049                 BUG_ON(ret);
7050                 if (new_inode->i_nlink == 0) {
7051                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7052                         BUG_ON(ret);
7053                 }
7054         }
7055
7056         fixup_inode_flags(new_dir, old_inode);
7057
7058         ret = btrfs_add_link(trans, new_dir, old_inode,
7059                              new_dentry->d_name.name,
7060                              new_dentry->d_name.len, 0, index);
7061         BUG_ON(ret);
7062
7063         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7064                 struct dentry *parent = new_dentry->d_parent;
7065                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7066                 btrfs_end_log_trans(root);
7067         }
7068 out_fail:
7069         btrfs_end_transaction_throttle(trans, root);
7070 out_notrans:
7071         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7072                 up_read(&root->fs_info->subvol_sem);
7073
7074         return ret;
7075 }
7076
7077 /*
7078  * some fairly slow code that needs optimization. This walks the list
7079  * of all the inodes with pending delalloc and forces them to disk.
7080  */
7081 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7082 {
7083         struct list_head *head = &root->fs_info->delalloc_inodes;
7084         struct btrfs_inode *binode;
7085         struct inode *inode;
7086
7087         if (root->fs_info->sb->s_flags & MS_RDONLY)
7088                 return -EROFS;
7089
7090         spin_lock(&root->fs_info->delalloc_lock);
7091         while (!list_empty(head)) {
7092                 binode = list_entry(head->next, struct btrfs_inode,
7093                                     delalloc_inodes);
7094                 inode = igrab(&binode->vfs_inode);
7095                 if (!inode)
7096                         list_del_init(&binode->delalloc_inodes);
7097                 spin_unlock(&root->fs_info->delalloc_lock);
7098                 if (inode) {
7099                         filemap_flush(inode->i_mapping);
7100                         if (delay_iput)
7101                                 btrfs_add_delayed_iput(inode);
7102                         else
7103                                 iput(inode);
7104                 }
7105                 cond_resched();
7106                 spin_lock(&root->fs_info->delalloc_lock);
7107         }
7108         spin_unlock(&root->fs_info->delalloc_lock);
7109
7110         /* the filemap_flush will queue IO into the worker threads, but
7111          * we have to make sure the IO is actually started and that
7112          * ordered extents get created before we return
7113          */
7114         atomic_inc(&root->fs_info->async_submit_draining);
7115         while (atomic_read(&root->fs_info->nr_async_submits) ||
7116               atomic_read(&root->fs_info->async_delalloc_pages)) {
7117                 wait_event(root->fs_info->async_submit_wait,
7118                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7119                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7120         }
7121         atomic_dec(&root->fs_info->async_submit_draining);
7122         return 0;
7123 }
7124
7125 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7126                          const char *symname)
7127 {
7128         struct btrfs_trans_handle *trans;
7129         struct btrfs_root *root = BTRFS_I(dir)->root;
7130         struct btrfs_path *path;
7131         struct btrfs_key key;
7132         struct inode *inode = NULL;
7133         int err;
7134         int drop_inode = 0;
7135         u64 objectid;
7136         u64 index = 0 ;
7137         int name_len;
7138         int datasize;
7139         unsigned long ptr;
7140         struct btrfs_file_extent_item *ei;
7141         struct extent_buffer *leaf;
7142         unsigned long nr = 0;
7143
7144         name_len = strlen(symname) + 1;
7145         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7146                 return -ENAMETOOLONG;
7147
7148         /*
7149          * 2 items for inode item and ref
7150          * 2 items for dir items
7151          * 1 item for xattr if selinux is on
7152          */
7153         trans = btrfs_start_transaction(root, 5);
7154         if (IS_ERR(trans))
7155                 return PTR_ERR(trans);
7156
7157         err = btrfs_find_free_ino(root, &objectid);
7158         if (err)
7159                 goto out_unlock;
7160
7161         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7162                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7163                                 S_IFLNK|S_IRWXUGO, &index);
7164         if (IS_ERR(inode)) {
7165                 err = PTR_ERR(inode);
7166                 goto out_unlock;
7167         }
7168
7169         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7170         if (err) {
7171                 drop_inode = 1;
7172                 goto out_unlock;
7173         }
7174
7175         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7176         if (err)
7177                 drop_inode = 1;
7178         else {
7179                 inode->i_mapping->a_ops = &btrfs_aops;
7180                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7181                 inode->i_fop = &btrfs_file_operations;
7182                 inode->i_op = &btrfs_file_inode_operations;
7183                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7184         }
7185         if (drop_inode)
7186                 goto out_unlock;
7187
7188         path = btrfs_alloc_path();
7189         BUG_ON(!path);
7190         key.objectid = btrfs_ino(inode);
7191         key.offset = 0;
7192         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7193         datasize = btrfs_file_extent_calc_inline_size(name_len);
7194         err = btrfs_insert_empty_item(trans, root, path, &key,
7195                                       datasize);
7196         if (err) {
7197                 drop_inode = 1;
7198                 btrfs_free_path(path);
7199                 goto out_unlock;
7200         }
7201         leaf = path->nodes[0];
7202         ei = btrfs_item_ptr(leaf, path->slots[0],
7203                             struct btrfs_file_extent_item);
7204         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7205         btrfs_set_file_extent_type(leaf, ei,
7206                                    BTRFS_FILE_EXTENT_INLINE);
7207         btrfs_set_file_extent_encryption(leaf, ei, 0);
7208         btrfs_set_file_extent_compression(leaf, ei, 0);
7209         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7210         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7211
7212         ptr = btrfs_file_extent_inline_start(ei);
7213         write_extent_buffer(leaf, symname, ptr, name_len);
7214         btrfs_mark_buffer_dirty(leaf);
7215         btrfs_free_path(path);
7216
7217         inode->i_op = &btrfs_symlink_inode_operations;
7218         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7219         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7220         inode_set_bytes(inode, name_len);
7221         btrfs_i_size_write(inode, name_len - 1);
7222         err = btrfs_update_inode(trans, root, inode);
7223         if (err)
7224                 drop_inode = 1;
7225
7226 out_unlock:
7227         nr = trans->blocks_used;
7228         btrfs_end_transaction_throttle(trans, root);
7229         if (drop_inode) {
7230                 inode_dec_link_count(inode);
7231                 iput(inode);
7232         }
7233         btrfs_btree_balance_dirty(root, nr);
7234         return err;
7235 }
7236
7237 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7238                                        u64 start, u64 num_bytes, u64 min_size,
7239                                        loff_t actual_len, u64 *alloc_hint,
7240                                        struct btrfs_trans_handle *trans)
7241 {
7242         struct btrfs_root *root = BTRFS_I(inode)->root;
7243         struct btrfs_key ins;
7244         u64 cur_offset = start;
7245         u64 i_size;
7246         int ret = 0;
7247         bool own_trans = true;
7248
7249         if (trans)
7250                 own_trans = false;
7251         while (num_bytes > 0) {
7252                 if (own_trans) {
7253                         trans = btrfs_start_transaction(root, 3);
7254                         if (IS_ERR(trans)) {
7255                                 ret = PTR_ERR(trans);
7256                                 break;
7257                         }
7258                 }
7259
7260                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7261                                            0, *alloc_hint, (u64)-1, &ins, 1);
7262                 if (ret) {
7263                         if (own_trans)
7264                                 btrfs_end_transaction(trans, root);
7265                         break;
7266                 }
7267
7268                 ret = insert_reserved_file_extent(trans, inode,
7269                                                   cur_offset, ins.objectid,
7270                                                   ins.offset, ins.offset,
7271                                                   ins.offset, 0, 0, 0,
7272                                                   BTRFS_FILE_EXTENT_PREALLOC);
7273                 BUG_ON(ret);
7274                 btrfs_drop_extent_cache(inode, cur_offset,
7275                                         cur_offset + ins.offset -1, 0);
7276
7277                 num_bytes -= ins.offset;
7278                 cur_offset += ins.offset;
7279                 *alloc_hint = ins.objectid + ins.offset;
7280
7281                 inode->i_ctime = CURRENT_TIME;
7282                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7283                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7284                     (actual_len > inode->i_size) &&
7285                     (cur_offset > inode->i_size)) {
7286                         if (cur_offset > actual_len)
7287                                 i_size = actual_len;
7288                         else
7289                                 i_size = cur_offset;
7290                         i_size_write(inode, i_size);
7291                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7292                 }
7293
7294                 ret = btrfs_update_inode(trans, root, inode);
7295                 BUG_ON(ret);
7296
7297                 if (own_trans)
7298                         btrfs_end_transaction(trans, root);
7299         }
7300         return ret;
7301 }
7302
7303 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7304                               u64 start, u64 num_bytes, u64 min_size,
7305                               loff_t actual_len, u64 *alloc_hint)
7306 {
7307         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7308                                            min_size, actual_len, alloc_hint,
7309                                            NULL);
7310 }
7311
7312 int btrfs_prealloc_file_range_trans(struct inode *inode,
7313                                     struct btrfs_trans_handle *trans, int mode,
7314                                     u64 start, u64 num_bytes, u64 min_size,
7315                                     loff_t actual_len, u64 *alloc_hint)
7316 {
7317         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7318                                            min_size, actual_len, alloc_hint, trans);
7319 }
7320
7321 static int btrfs_set_page_dirty(struct page *page)
7322 {
7323         return __set_page_dirty_nobuffers(page);
7324 }
7325
7326 static int btrfs_permission(struct inode *inode, int mask)
7327 {
7328         struct btrfs_root *root = BTRFS_I(inode)->root;
7329
7330         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7331                 return -EROFS;
7332         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7333                 return -EACCES;
7334         return generic_permission(inode, mask);
7335 }
7336
7337 static const struct inode_operations btrfs_dir_inode_operations = {
7338         .getattr        = btrfs_getattr,
7339         .lookup         = btrfs_lookup,
7340         .create         = btrfs_create,
7341         .unlink         = btrfs_unlink,
7342         .link           = btrfs_link,
7343         .mkdir          = btrfs_mkdir,
7344         .rmdir          = btrfs_rmdir,
7345         .rename         = btrfs_rename,
7346         .symlink        = btrfs_symlink,
7347         .setattr        = btrfs_setattr,
7348         .mknod          = btrfs_mknod,
7349         .setxattr       = btrfs_setxattr,
7350         .getxattr       = btrfs_getxattr,
7351         .listxattr      = btrfs_listxattr,
7352         .removexattr    = btrfs_removexattr,
7353         .permission     = btrfs_permission,
7354         .check_acl      = btrfs_check_acl,
7355 };
7356 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7357         .lookup         = btrfs_lookup,
7358         .permission     = btrfs_permission,
7359         .check_acl      = btrfs_check_acl,
7360 };
7361
7362 static const struct file_operations btrfs_dir_file_operations = {
7363         .llseek         = generic_file_llseek,
7364         .read           = generic_read_dir,
7365         .readdir        = btrfs_real_readdir,
7366         .unlocked_ioctl = btrfs_ioctl,
7367 #ifdef CONFIG_COMPAT
7368         .compat_ioctl   = btrfs_ioctl,
7369 #endif
7370         .release        = btrfs_release_file,
7371         .fsync          = btrfs_sync_file,
7372 };
7373
7374 static struct extent_io_ops btrfs_extent_io_ops = {
7375         .fill_delalloc = run_delalloc_range,
7376         .submit_bio_hook = btrfs_submit_bio_hook,
7377         .merge_bio_hook = btrfs_merge_bio_hook,
7378         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7379         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7380         .writepage_start_hook = btrfs_writepage_start_hook,
7381         .readpage_io_failed_hook = btrfs_io_failed_hook,
7382         .set_bit_hook = btrfs_set_bit_hook,
7383         .clear_bit_hook = btrfs_clear_bit_hook,
7384         .merge_extent_hook = btrfs_merge_extent_hook,
7385         .split_extent_hook = btrfs_split_extent_hook,
7386 };
7387
7388 /*
7389  * btrfs doesn't support the bmap operation because swapfiles
7390  * use bmap to make a mapping of extents in the file.  They assume
7391  * these extents won't change over the life of the file and they
7392  * use the bmap result to do IO directly to the drive.
7393  *
7394  * the btrfs bmap call would return logical addresses that aren't
7395  * suitable for IO and they also will change frequently as COW
7396  * operations happen.  So, swapfile + btrfs == corruption.
7397  *
7398  * For now we're avoiding this by dropping bmap.
7399  */
7400 static const struct address_space_operations btrfs_aops = {
7401         .readpage       = btrfs_readpage,
7402         .writepage      = btrfs_writepage,
7403         .writepages     = btrfs_writepages,
7404         .readpages      = btrfs_readpages,
7405         .direct_IO      = btrfs_direct_IO,
7406         .invalidatepage = btrfs_invalidatepage,
7407         .releasepage    = btrfs_releasepage,
7408         .set_page_dirty = btrfs_set_page_dirty,
7409         .error_remove_page = generic_error_remove_page,
7410 };
7411
7412 static const struct address_space_operations btrfs_symlink_aops = {
7413         .readpage       = btrfs_readpage,
7414         .writepage      = btrfs_writepage,
7415         .invalidatepage = btrfs_invalidatepage,
7416         .releasepage    = btrfs_releasepage,
7417 };
7418
7419 static const struct inode_operations btrfs_file_inode_operations = {
7420         .getattr        = btrfs_getattr,
7421         .setattr        = btrfs_setattr,
7422         .setxattr       = btrfs_setxattr,
7423         .getxattr       = btrfs_getxattr,
7424         .listxattr      = btrfs_listxattr,
7425         .removexattr    = btrfs_removexattr,
7426         .permission     = btrfs_permission,
7427         .fiemap         = btrfs_fiemap,
7428         .check_acl      = btrfs_check_acl,
7429 };
7430 static const struct inode_operations btrfs_special_inode_operations = {
7431         .getattr        = btrfs_getattr,
7432         .setattr        = btrfs_setattr,
7433         .permission     = btrfs_permission,
7434         .setxattr       = btrfs_setxattr,
7435         .getxattr       = btrfs_getxattr,
7436         .listxattr      = btrfs_listxattr,
7437         .removexattr    = btrfs_removexattr,
7438         .check_acl      = btrfs_check_acl,
7439 };
7440 static const struct inode_operations btrfs_symlink_inode_operations = {
7441         .readlink       = generic_readlink,
7442         .follow_link    = page_follow_link_light,
7443         .put_link       = page_put_link,
7444         .getattr        = btrfs_getattr,
7445         .permission     = btrfs_permission,
7446         .setxattr       = btrfs_setxattr,
7447         .getxattr       = btrfs_getxattr,
7448         .listxattr      = btrfs_listxattr,
7449         .removexattr    = btrfs_removexattr,
7450         .check_acl      = btrfs_check_acl,
7451 };
7452
7453 const struct dentry_operations btrfs_dentry_operations = {
7454         .d_delete       = btrfs_dentry_delete,
7455 };