Pileus Git - ~andy/linux/blob - fs/btrfs/inode.c

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