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/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62
  63 struct btrfs_iget_args {
  64         struct btrfs_key *location;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end >= PAGE_CACHE_SIZE ||
 253             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 /*
 352  * we create compressed extents in two phases.  The first
 353  * phase compresses a range of pages that have already been
 354  * locked (both pages and state bits are locked).
 355  *
 356  * This is done inside an ordered work queue, and the compression
 357  * is spread across many cpus.  The actual IO submission is step
 358  * two, and the ordered work queue takes care of making sure that
 359  * happens in the same order things were put onto the queue by
 360  * writepages and friends.
 361  *
 362  * If this code finds it can't get good compression, it puts an
 363  * entry onto the work queue to write the uncompressed bytes.  This
 364  * makes sure that both compressed inodes and uncompressed inodes
 365  * are written in the same order that the flusher thread sent them
 366  * down.
 367  */
 368 static noinline int compress_file_range(struct inode *inode,
 369                                         struct page *locked_page,
 370                                         u64 start, u64 end,
 371                                         struct async_cow *async_cow,
 372                                         int *num_added)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375         u64 num_bytes;
 376         u64 blocksize = root->sectorsize;
 377         u64 actual_end;
 378         u64 isize = i_size_read(inode);
 379         int ret = 0;
 380         struct page **pages = NULL;
 381         unsigned long nr_pages;
 382         unsigned long nr_pages_ret = 0;
 383         unsigned long total_compressed = 0;
 384         unsigned long total_in = 0;
 385         unsigned long max_compressed = 128 * 1024;
 386         unsigned long max_uncompressed = 128 * 1024;
 387         int i;
 388         int will_compress;
 389         int compress_type = root->fs_info->compress_type;
 390         int redirty = 0;
 391
 392         /* if this is a small write inside eof, kick off a defrag */
 393         if ((end - start + 1) < 16 * 1024 &&
 394             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 395                 btrfs_add_inode_defrag(NULL, inode);
 396
 397         /*
 398          * skip compression for a small file range(<=blocksize) that
 399          * isn't an inline extent, since it dosen't save disk space at all.
 400          */
 401         if ((end - start + 1) <= blocksize &&
 402             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 403                 goto cleanup_and_bail_uncompressed;
 404
 405         actual_end = min_t(u64, isize, end + 1);
 406 again:
 407         will_compress = 0;
 408         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 409         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 410
 411         /*
 412          * we don't want to send crud past the end of i_size through
 413          * compression, that's just a waste of CPU time.  So, if the
 414          * end of the file is before the start of our current
 415          * requested range of bytes, we bail out to the uncompressed
 416          * cleanup code that can deal with all of this.
 417          *
 418          * It isn't really the fastest way to fix things, but this is a
 419          * very uncommon corner.
 420          */
 421         if (actual_end <= start)
 422                 goto cleanup_and_bail_uncompressed;
 423
 424         total_compressed = actual_end - start;
 425
 426         /* we want to make sure that amount of ram required to uncompress
 427          * an extent is reasonable, so we limit the total size in ram
 428          * of a compressed extent to 128k.  This is a crucial number
 429          * because it also controls how easily we can spread reads across
 430          * cpus for decompression.
 431          *
 432          * We also want to make sure the amount of IO required to do
 433          * a random read is reasonably small, so we limit the size of
 434          * a compressed extent to 128k.
 435          */
 436         total_compressed = min(total_compressed, max_uncompressed);
 437         num_bytes = ALIGN(end - start + 1, blocksize);
 438         num_bytes = max(blocksize,  num_bytes);
 439         total_in = 0;
 440         ret = 0;
 441
 442         /*
 443          * we do compression for mount -o compress and when the
 444          * inode has not been flagged as nocompress.  This flag can
 445          * change at any time if we discover bad compression ratios.
 446          */
 447         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 448             (btrfs_test_opt(root, COMPRESS) ||
 449              (BTRFS_I(inode)->force_compress) ||
 450              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 451                 WARN_ON(pages);
 452                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 453                 if (!pages) {
 454                         /* just bail out to the uncompressed code */
 455                         goto cont;
 456                 }
 457
 458                 if (BTRFS_I(inode)->force_compress)
 459                         compress_type = BTRFS_I(inode)->force_compress;
 460
 461                 /*
 462                  * we need to call clear_page_dirty_for_io on each
 463                  * page in the range.  Otherwise applications with the file
 464                  * mmap'd can wander in and change the page contents while
 465                  * we are compressing them.
 466                  *
 467                  * If the compression fails for any reason, we set the pages
 468                  * dirty again later on.
 469                  */
 470                 extent_range_clear_dirty_for_io(inode, start, end);
 471                 redirty = 1;
 472                 ret = btrfs_compress_pages(compress_type,
 473                                            inode->i_mapping, start,
 474                                            total_compressed, pages,
 475                                            nr_pages, &nr_pages_ret,
 476                                            &total_in,
 477                                            &total_compressed,
 478                                            max_compressed);
 479
 480                 if (!ret) {
 481                         unsigned long offset = total_compressed &
 482                                 (PAGE_CACHE_SIZE - 1);
 483                         struct page *page = pages[nr_pages_ret - 1];
 484                         char *kaddr;
 485
 486                         /* zero the tail end of the last page, we might be
 487                          * sending it down to disk
 488                          */
 489                         if (offset) {
 490                                 kaddr = kmap_atomic(page);
 491                                 memset(kaddr + offset, 0,
 492                                        PAGE_CACHE_SIZE - offset);
 493                                 kunmap_atomic(kaddr);
 494                         }
 495                         will_compress = 1;
 496                 }
 497         }
 498 cont:
 499         if (start == 0) {
 500                 /* lets try to make an inline extent */
 501                 if (ret || total_in < (actual_end - start)) {
 502                         /* we didn't compress the entire range, try
 503                          * to make an uncompressed inline extent.
 504                          */
 505                         ret = cow_file_range_inline(root, inode, start, end,
 506                                                     0, 0, NULL);
 507                 } else {
 508                         /* try making a compressed inline extent */
 509                         ret = cow_file_range_inline(root, inode, start, end,
 510                                                     total_compressed,
 511                                                     compress_type, pages);
 512                 }
 513                 if (ret <= 0) {
 514                         unsigned long clear_flags = EXTENT_DELALLOC |
 515                                 EXTENT_DEFRAG;
 516                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 517
 518                         /*
 519                          * inline extent creation worked or returned error,
 520                          * we don't need to create any more async work items.
 521                          * Unlock and free up our temp pages.
 522                          */
 523                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 524                                                      clear_flags, PAGE_UNLOCK |
 525                                                      PAGE_CLEAR_DIRTY |
 526                                                      PAGE_SET_WRITEBACK |
 527                                                      PAGE_END_WRITEBACK);
 528                         goto free_pages_out;
 529                 }
 530         }
 531
 532         if (will_compress) {
 533                 /*
 534                  * we aren't doing an inline extent round the compressed size
 535                  * up to a block size boundary so the allocator does sane
 536                  * things
 537                  */
 538                 total_compressed = ALIGN(total_compressed, blocksize);
 539
 540                 /*
 541                  * one last check to make sure the compression is really a
 542                  * win, compare the page count read with the blocks on disk
 543                  */
 544                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 545                 if (total_compressed >= total_in) {
 546                         will_compress = 0;
 547                 } else {
 548                         num_bytes = total_in;
 549                 }
 550         }
 551         if (!will_compress && pages) {
 552                 /*
 553                  * the compression code ran but failed to make things smaller,
 554                  * free any pages it allocated and our page pointer array
 555                  */
 556                 for (i = 0; i < nr_pages_ret; i++) {
 557                         WARN_ON(pages[i]->mapping);
 558                         page_cache_release(pages[i]);
 559                 }
 560                 kfree(pages);
 561                 pages = NULL;
 562                 total_compressed = 0;
 563                 nr_pages_ret = 0;
 564
 565                 /* flag the file so we don't compress in the future */
 566                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 567                     !(BTRFS_I(inode)->force_compress)) {
 568                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 569                 }
 570         }
 571         if (will_compress) {
 572                 *num_added += 1;
 573
 574                 /* the async work queues will take care of doing actual
 575                  * allocation on disk for these compressed pages,
 576                  * and will submit them to the elevator.
 577                  */
 578                 add_async_extent(async_cow, start, num_bytes,
 579                                  total_compressed, pages, nr_pages_ret,
 580                                  compress_type);
 581
 582                 if (start + num_bytes < end) {
 583                         start += num_bytes;
 584                         pages = NULL;
 585                         cond_resched();
 586                         goto again;
 587                 }
 588         } else {
 589 cleanup_and_bail_uncompressed:
 590                 /*
 591                  * No compression, but we still need to write the pages in
 592                  * the file we've been given so far.  redirty the locked
 593                  * page if it corresponds to our extent and set things up
 594                  * for the async work queue to run cow_file_range to do
 595                  * the normal delalloc dance
 596                  */
 597                 if (page_offset(locked_page) >= start &&
 598                     page_offset(locked_page) <= end) {
 599                         __set_page_dirty_nobuffers(locked_page);
 600                         /* unlocked later on in the async handlers */
 601                 }
 602                 if (redirty)
 603                         extent_range_redirty_for_io(inode, start, end);
 604                 add_async_extent(async_cow, start, end - start + 1,
 605                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 606                 *num_added += 1;
 607         }
 608
 609 out:
 610         return ret;
 611
 612 free_pages_out:
 613         for (i = 0; i < nr_pages_ret; i++) {
 614                 WARN_ON(pages[i]->mapping);
 615                 page_cache_release(pages[i]);
 616         }
 617         kfree(pages);
 618
 619         goto out;
 620 }
 621
 622 /*
 623  * phase two of compressed writeback.  This is the ordered portion
 624  * of the code, which only gets called in the order the work was
 625  * queued.  We walk all the async extents created by compress_file_range
 626  * and send them down to the disk.
 627  */
 628 static noinline int submit_compressed_extents(struct inode *inode,
 629                                               struct async_cow *async_cow)
 630 {
 631         struct async_extent *async_extent;
 632         u64 alloc_hint = 0;
 633         struct btrfs_key ins;
 634         struct extent_map *em;
 635         struct btrfs_root *root = BTRFS_I(inode)->root;
 636         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 637         struct extent_io_tree *io_tree;
 638         int ret = 0;
 639
 640         if (list_empty(&async_cow->extents))
 641                 return 0;
 642
 643 again:
 644         while (!list_empty(&async_cow->extents)) {
 645                 async_extent = list_entry(async_cow->extents.next,
 646                                           struct async_extent, list);
 647                 list_del(&async_extent->list);
 648
 649                 io_tree = &BTRFS_I(inode)->io_tree;
 650
 651 retry:
 652                 /* did the compression code fall back to uncompressed IO? */
 653                 if (!async_extent->pages) {
 654                         int page_started = 0;
 655                         unsigned long nr_written = 0;
 656
 657                         lock_extent(io_tree, async_extent->start,
 658                                          async_extent->start +
 659                                          async_extent->ram_size - 1);
 660
 661                         /* allocate blocks */
 662                         ret = cow_file_range(inode, async_cow->locked_page,
 663                                              async_extent->start,
 664                                              async_extent->start +
 665                                              async_extent->ram_size - 1,
 666                                              &page_started, &nr_written, 0);
 667
 668                         /* JDM XXX */
 669
 670                         /*
 671                          * if page_started, cow_file_range inserted an
 672                          * inline extent and took care of all the unlocking
 673                          * and IO for us.  Otherwise, we need to submit
 674                          * all those pages down to the drive.
 675                          */
 676                         if (!page_started && !ret)
 677                                 extent_write_locked_range(io_tree,
 678                                                   inode, async_extent->start,
 679                                                   async_extent->start +
 680                                                   async_extent->ram_size - 1,
 681                                                   btrfs_get_extent,
 682                                                   WB_SYNC_ALL);
 683                         else if (ret)
 684                                 unlock_page(async_cow->locked_page);
 685                         kfree(async_extent);
 686                         cond_resched();
 687                         continue;
 688                 }
 689
 690                 lock_extent(io_tree, async_extent->start,
 691                             async_extent->start + async_extent->ram_size - 1);
 692
 693                 ret = btrfs_reserve_extent(root,
 694                                            async_extent->compressed_size,
 695                                            async_extent->compressed_size,
 696                                            0, alloc_hint, &ins, 1);
 697                 if (ret) {
 698                         int i;
 699
 700                         for (i = 0; i < async_extent->nr_pages; i++) {
 701                                 WARN_ON(async_extent->pages[i]->mapping);
 702                                 page_cache_release(async_extent->pages[i]);
 703                         }
 704                         kfree(async_extent->pages);
 705                         async_extent->nr_pages = 0;
 706                         async_extent->pages = NULL;
 707
 708                         if (ret == -ENOSPC) {
 709                                 unlock_extent(io_tree, async_extent->start,
 710                                               async_extent->start +
 711                                               async_extent->ram_size - 1);
 712                                 goto retry;
 713                         }
 714                         goto out_free;
 715                 }
 716
 717                 /*
 718                  * here we're doing allocation and writeback of the
 719                  * compressed pages
 720                  */
 721                 btrfs_drop_extent_cache(inode, async_extent->start,
 722                                         async_extent->start +
 723                                         async_extent->ram_size - 1, 0);
 724
 725                 em = alloc_extent_map();
 726                 if (!em) {
 727                         ret = -ENOMEM;
 728                         goto out_free_reserve;
 729                 }
 730                 em->start = async_extent->start;
 731                 em->len = async_extent->ram_size;
 732                 em->orig_start = em->start;
 733                 em->mod_start = em->start;
 734                 em->mod_len = em->len;
 735
 736                 em->block_start = ins.objectid;
 737                 em->block_len = ins.offset;
 738                 em->orig_block_len = ins.offset;
 739                 em->ram_bytes = async_extent->ram_size;
 740                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 741                 em->compress_type = async_extent->compress_type;
 742                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 743                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 744                 em->generation = -1;
 745
 746                 while (1) {
 747                         write_lock(&em_tree->lock);
 748                         ret = add_extent_mapping(em_tree, em, 1);
 749                         write_unlock(&em_tree->lock);
 750                         if (ret != -EEXIST) {
 751                                 free_extent_map(em);
 752                                 break;
 753                         }
 754                         btrfs_drop_extent_cache(inode, async_extent->start,
 755                                                 async_extent->start +
 756                                                 async_extent->ram_size - 1, 0);
 757                 }
 758
 759                 if (ret)
 760                         goto out_free_reserve;
 761
 762                 ret = btrfs_add_ordered_extent_compress(inode,
 763                                                 async_extent->start,
 764                                                 ins.objectid,
 765                                                 async_extent->ram_size,
 766                                                 ins.offset,
 767                                                 BTRFS_ORDERED_COMPRESSED,
 768                                                 async_extent->compress_type);
 769                 if (ret)
 770                         goto out_free_reserve;
 771
 772                 /*
 773                  * clear dirty, set writeback and unlock the pages.
 774                  */
 775                 extent_clear_unlock_delalloc(inode, async_extent->start,
 776                                 async_extent->start +
 777                                 async_extent->ram_size - 1,
 778                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 779                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 780                                 PAGE_SET_WRITEBACK);
 781                 ret = btrfs_submit_compressed_write(inode,
 782                                     async_extent->start,
 783                                     async_extent->ram_size,
 784                                     ins.objectid,
 785                                     ins.offset, async_extent->pages,
 786                                     async_extent->nr_pages);
 787                 alloc_hint = ins.objectid + ins.offset;
 788                 kfree(async_extent);
 789                 if (ret)
 790                         goto out;
 791                 cond_resched();
 792         }
 793         ret = 0;
 794 out:
 795         return ret;
 796 out_free_reserve:
 797         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 798 out_free:
 799         extent_clear_unlock_delalloc(inode, async_extent->start,
 800                                      async_extent->start +
 801                                      async_extent->ram_size - 1,
 802                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 803                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 804                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 805                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 806         kfree(async_extent);
 807         goto again;
 808 }
 809
 810 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 811                                       u64 num_bytes)
 812 {
 813         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 814         struct extent_map *em;
 815         u64 alloc_hint = 0;
 816
 817         read_lock(&em_tree->lock);
 818         em = search_extent_mapping(em_tree, start, num_bytes);
 819         if (em) {
 820                 /*
 821                  * if block start isn't an actual block number then find the
 822                  * first block in this inode and use that as a hint.  If that
 823                  * block is also bogus then just don't worry about it.
 824                  */
 825                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 826                         free_extent_map(em);
 827                         em = search_extent_mapping(em_tree, 0, 0);
 828                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 829                                 alloc_hint = em->block_start;
 830                         if (em)
 831                                 free_extent_map(em);
 832                 } else {
 833                         alloc_hint = em->block_start;
 834                         free_extent_map(em);
 835                 }
 836         }
 837         read_unlock(&em_tree->lock);
 838
 839         return alloc_hint;
 840 }
 841
 842 /*
 843  * when extent_io.c finds a delayed allocation range in the file,
 844  * the call backs end up in this code.  The basic idea is to
 845  * allocate extents on disk for the range, and create ordered data structs
 846  * in ram to track those extents.
 847  *
 848  * locked_page is the page that writepage had locked already.  We use
 849  * it to make sure we don't do extra locks or unlocks.
 850  *
 851  * *page_started is set to one if we unlock locked_page and do everything
 852  * required to start IO on it.  It may be clean and already done with
 853  * IO when we return.
 854  */
 855 static noinline int cow_file_range(struct inode *inode,
 856                                    struct page *locked_page,
 857                                    u64 start, u64 end, int *page_started,
 858                                    unsigned long *nr_written,
 859                                    int unlock)
 860 {
 861         struct btrfs_root *root = BTRFS_I(inode)->root;
 862         u64 alloc_hint = 0;
 863         u64 num_bytes;
 864         unsigned long ram_size;
 865         u64 disk_num_bytes;
 866         u64 cur_alloc_size;
 867         u64 blocksize = root->sectorsize;
 868         struct btrfs_key ins;
 869         struct extent_map *em;
 870         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 871         int ret = 0;
 872
 873         if (btrfs_is_free_space_inode(inode)) {
 874                 WARN_ON_ONCE(1);
 875                 ret = -EINVAL;
 876                 goto out_unlock;
 877         }
 878
 879         num_bytes = ALIGN(end - start + 1, blocksize);
 880         num_bytes = max(blocksize,  num_bytes);
 881         disk_num_bytes = num_bytes;
 882
 883         /* if this is a small write inside eof, kick off defrag */
 884         if (num_bytes < 64 * 1024 &&
 885             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 886                 btrfs_add_inode_defrag(NULL, inode);
 887
 888         if (start == 0) {
 889                 /* lets try to make an inline extent */
 890                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 891                                             NULL);
 892                 if (ret == 0) {
 893                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 894                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 895                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 896                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 897                                      PAGE_END_WRITEBACK);
 898
 899                         *nr_written = *nr_written +
 900                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 901                         *page_started = 1;
 902                         goto out;
 903                 } else if (ret < 0) {
 904                         goto out_unlock;
 905                 }
 906         }
 907
 908         BUG_ON(disk_num_bytes >
 909                btrfs_super_total_bytes(root->fs_info->super_copy));
 910
 911         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 912         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 913
 914         while (disk_num_bytes > 0) {
 915                 unsigned long op;
 916
 917                 cur_alloc_size = disk_num_bytes;
 918                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 919                                            root->sectorsize, 0, alloc_hint,
 920                                            &ins, 1);
 921                 if (ret < 0)
 922                         goto out_unlock;
 923
 924                 em = alloc_extent_map();
 925                 if (!em) {
 926                         ret = -ENOMEM;
 927                         goto out_reserve;
 928                 }
 929                 em->start = start;
 930                 em->orig_start = em->start;
 931                 ram_size = ins.offset;
 932                 em->len = ins.offset;
 933                 em->mod_start = em->start;
 934                 em->mod_len = em->len;
 935
 936                 em->block_start = ins.objectid;
 937                 em->block_len = ins.offset;
 938                 em->orig_block_len = ins.offset;
 939                 em->ram_bytes = ram_size;
 940                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 941                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 942                 em->generation = -1;
 943
 944                 while (1) {
 945                         write_lock(&em_tree->lock);
 946                         ret = add_extent_mapping(em_tree, em, 1);
 947                         write_unlock(&em_tree->lock);
 948                         if (ret != -EEXIST) {
 949                                 free_extent_map(em);
 950                                 break;
 951                         }
 952                         btrfs_drop_extent_cache(inode, start,
 953                                                 start + ram_size - 1, 0);
 954                 }
 955                 if (ret)
 956                         goto out_reserve;
 957
 958                 cur_alloc_size = ins.offset;
 959                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 960                                                ram_size, cur_alloc_size, 0);
 961                 if (ret)
 962                         goto out_reserve;
 963
 964                 if (root->root_key.objectid ==
 965                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 966                         ret = btrfs_reloc_clone_csums(inode, start,
 967                                                       cur_alloc_size);
 968                         if (ret)
 969                                 goto out_reserve;
 970                 }
 971
 972                 if (disk_num_bytes < cur_alloc_size)
 973                         break;
 974
 975                 /* we're not doing compressed IO, don't unlock the first
 976                  * page (which the caller expects to stay locked), don't
 977                  * clear any dirty bits and don't set any writeback bits
 978                  *
 979                  * Do set the Private2 bit so we know this page was properly
 980                  * setup for writepage
 981                  */
 982                 op = unlock ? PAGE_UNLOCK : 0;
 983                 op |= PAGE_SET_PRIVATE2;
 984
 985                 extent_clear_unlock_delalloc(inode, start,
 986                                              start + ram_size - 1, locked_page,
 987                                              EXTENT_LOCKED | EXTENT_DELALLOC,
 988                                              op);
 989                 disk_num_bytes -= cur_alloc_size;
 990                 num_bytes -= cur_alloc_size;
 991                 alloc_hint = ins.objectid + ins.offset;
 992                 start += cur_alloc_size;
 993         }
 994 out:
 995         return ret;
 996
 997 out_reserve:
 998         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 999 out_unlock:
1000         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1001                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1002                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1003                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1004                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1005         goto out;
1006 }
1007
1008 /*
1009  * work queue call back to started compression on a file and pages
1010  */
1011 static noinline void async_cow_start(struct btrfs_work *work)
1012 {
1013         struct async_cow *async_cow;
1014         int num_added = 0;
1015         async_cow = container_of(work, struct async_cow, work);
1016
1017         compress_file_range(async_cow->inode, async_cow->locked_page,
1018                             async_cow->start, async_cow->end, async_cow,
1019                             &num_added);
1020         if (num_added == 0) {
1021                 btrfs_add_delayed_iput(async_cow->inode);
1022                 async_cow->inode = NULL;
1023         }
1024 }
1025
1026 /*
1027  * work queue call back to submit previously compressed pages
1028  */
1029 static noinline void async_cow_submit(struct btrfs_work *work)
1030 {
1031         struct async_cow *async_cow;
1032         struct btrfs_root *root;
1033         unsigned long nr_pages;
1034
1035         async_cow = container_of(work, struct async_cow, work);
1036
1037         root = async_cow->root;
1038         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1039                 PAGE_CACHE_SHIFT;
1040
1041         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1042             5 * 1024 * 1024 &&
1043             waitqueue_active(&root->fs_info->async_submit_wait))
1044                 wake_up(&root->fs_info->async_submit_wait);
1045
1046         if (async_cow->inode)
1047                 submit_compressed_extents(async_cow->inode, async_cow);
1048 }
1049
1050 static noinline void async_cow_free(struct btrfs_work *work)
1051 {
1052         struct async_cow *async_cow;
1053         async_cow = container_of(work, struct async_cow, work);
1054         if (async_cow->inode)
1055                 btrfs_add_delayed_iput(async_cow->inode);
1056         kfree(async_cow);
1057 }
1058
1059 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1060                                 u64 start, u64 end, int *page_started,
1061                                 unsigned long *nr_written)
1062 {
1063         struct async_cow *async_cow;
1064         struct btrfs_root *root = BTRFS_I(inode)->root;
1065         unsigned long nr_pages;
1066         u64 cur_end;
1067         int limit = 10 * 1024 * 1024;
1068
1069         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1070                          1, 0, NULL, GFP_NOFS);
1071         while (start < end) {
1072                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1073                 BUG_ON(!async_cow); /* -ENOMEM */
1074                 async_cow->inode = igrab(inode);
1075                 async_cow->root = root;
1076                 async_cow->locked_page = locked_page;
1077                 async_cow->start = start;
1078
1079                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1080                         cur_end = end;
1081                 else
1082                         cur_end = min(end, start + 512 * 1024 - 1);
1083
1084                 async_cow->end = cur_end;
1085                 INIT_LIST_HEAD(&async_cow->extents);
1086
1087                 btrfs_init_work(&async_cow->work, async_cow_start,
1088                                 async_cow_submit, async_cow_free);
1089
1090                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1091                         PAGE_CACHE_SHIFT;
1092                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1093
1094                 btrfs_queue_work(root->fs_info->delalloc_workers,
1095                                  &async_cow->work);
1096
1097                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1098                         wait_event(root->fs_info->async_submit_wait,
1099                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1100                             limit));
1101                 }
1102
1103                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1104                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1105                         wait_event(root->fs_info->async_submit_wait,
1106                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1107                            0));
1108                 }
1109
1110                 *nr_written += nr_pages;
1111                 start = cur_end + 1;
1112         }
1113         *page_started = 1;
1114         return 0;
1115 }
1116
1117 static noinline int csum_exist_in_range(struct btrfs_root *root,
1118                                         u64 bytenr, u64 num_bytes)
1119 {
1120         int ret;
1121         struct btrfs_ordered_sum *sums;
1122         LIST_HEAD(list);
1123
1124         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1125                                        bytenr + num_bytes - 1, &list, 0);
1126         if (ret == 0 && list_empty(&list))
1127                 return 0;
1128
1129         while (!list_empty(&list)) {
1130                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1131                 list_del(&sums->list);
1132                 kfree(sums);
1133         }
1134         return 1;
1135 }
1136
1137 /*
1138  * when nowcow writeback call back.  This checks for snapshots or COW copies
1139  * of the extents that exist in the file, and COWs the file as required.
1140  *
1141  * If no cow copies or snapshots exist, we write directly to the existing
1142  * blocks on disk
1143  */
1144 static noinline int run_delalloc_nocow(struct inode *inode,
1145                                        struct page *locked_page,
1146                               u64 start, u64 end, int *page_started, int force,
1147                               unsigned long *nr_written)
1148 {
1149         struct btrfs_root *root = BTRFS_I(inode)->root;
1150         struct btrfs_trans_handle *trans;
1151         struct extent_buffer *leaf;
1152         struct btrfs_path *path;
1153         struct btrfs_file_extent_item *fi;
1154         struct btrfs_key found_key;
1155         u64 cow_start;
1156         u64 cur_offset;
1157         u64 extent_end;
1158         u64 extent_offset;
1159         u64 disk_bytenr;
1160         u64 num_bytes;
1161         u64 disk_num_bytes;
1162         u64 ram_bytes;
1163         int extent_type;
1164         int ret, err;
1165         int type;
1166         int nocow;
1167         int check_prev = 1;
1168         bool nolock;
1169         u64 ino = btrfs_ino(inode);
1170
1171         path = btrfs_alloc_path();
1172         if (!path) {
1173                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1174                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1175                                              EXTENT_DO_ACCOUNTING |
1176                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1177                                              PAGE_CLEAR_DIRTY |
1178                                              PAGE_SET_WRITEBACK |
1179                                              PAGE_END_WRITEBACK);
1180                 return -ENOMEM;
1181         }
1182
1183         nolock = btrfs_is_free_space_inode(inode);
1184
1185         if (nolock)
1186                 trans = btrfs_join_transaction_nolock(root);
1187         else
1188                 trans = btrfs_join_transaction(root);
1189
1190         if (IS_ERR(trans)) {
1191                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1192                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1193                                              EXTENT_DO_ACCOUNTING |
1194                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1195                                              PAGE_CLEAR_DIRTY |
1196                                              PAGE_SET_WRITEBACK |
1197                                              PAGE_END_WRITEBACK);
1198                 btrfs_free_path(path);
1199                 return PTR_ERR(trans);
1200         }
1201
1202         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1203
1204         cow_start = (u64)-1;
1205         cur_offset = start;
1206         while (1) {
1207                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1208                                                cur_offset, 0);
1209                 if (ret < 0)
1210                         goto error;
1211                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1212                         leaf = path->nodes[0];
1213                         btrfs_item_key_to_cpu(leaf, &found_key,
1214                                               path->slots[0] - 1);
1215                         if (found_key.objectid == ino &&
1216                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1217                                 path->slots[0]--;
1218                 }
1219                 check_prev = 0;
1220 next_slot:
1221                 leaf = path->nodes[0];
1222                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1223                         ret = btrfs_next_leaf(root, path);
1224                         if (ret < 0)
1225                                 goto error;
1226                         if (ret > 0)
1227                                 break;
1228                         leaf = path->nodes[0];
1229                 }
1230
1231                 nocow = 0;
1232                 disk_bytenr = 0;
1233                 num_bytes = 0;
1234                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1235
1236                 if (found_key.objectid > ino ||
1237                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1238                     found_key.offset > end)
1239                         break;
1240
1241                 if (found_key.offset > cur_offset) {
1242                         extent_end = found_key.offset;
1243                         extent_type = 0;
1244                         goto out_check;
1245                 }
1246
1247                 fi = btrfs_item_ptr(leaf, path->slots[0],
1248                                     struct btrfs_file_extent_item);
1249                 extent_type = btrfs_file_extent_type(leaf, fi);
1250
1251                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1252                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1253                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1254                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1255                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1256                         extent_end = found_key.offset +
1257                                 btrfs_file_extent_num_bytes(leaf, fi);
1258                         disk_num_bytes =
1259                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1260                         if (extent_end <= start) {
1261                                 path->slots[0]++;
1262                                 goto next_slot;
1263                         }
1264                         if (disk_bytenr == 0)
1265                                 goto out_check;
1266                         if (btrfs_file_extent_compression(leaf, fi) ||
1267                             btrfs_file_extent_encryption(leaf, fi) ||
1268                             btrfs_file_extent_other_encoding(leaf, fi))
1269                                 goto out_check;
1270                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1271                                 goto out_check;
1272                         if (btrfs_extent_readonly(root, disk_bytenr))
1273                                 goto out_check;
1274                         if (btrfs_cross_ref_exist(trans, root, ino,
1275                                                   found_key.offset -
1276                                                   extent_offset, disk_bytenr))
1277                                 goto out_check;
1278                         disk_bytenr += extent_offset;
1279                         disk_bytenr += cur_offset - found_key.offset;
1280                         num_bytes = min(end + 1, extent_end) - cur_offset;
1281                         /*
1282                          * if there are pending snapshots for this root,
1283                          * we fall into common COW way.
1284                          */
1285                         if (!nolock) {
1286                                 err = btrfs_start_nocow_write(root);
1287                                 if (!err)
1288                                         goto out_check;
1289                         }
1290                         /*
1291                          * force cow if csum exists in the range.
1292                          * this ensure that csum for a given extent are
1293                          * either valid or do not exist.
1294                          */
1295                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1296                                 goto out_check;
1297                         nocow = 1;
1298                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299                         extent_end = found_key.offset +
1300                                 btrfs_file_extent_inline_len(leaf,
1301                                                      path->slots[0], fi);
1302                         extent_end = ALIGN(extent_end, root->sectorsize);
1303                 } else {
1304                         BUG_ON(1);
1305                 }
1306 out_check:
1307                 if (extent_end <= start) {
1308                         path->slots[0]++;
1309                         if (!nolock && nocow)
1310                                 btrfs_end_nocow_write(root);
1311                         goto next_slot;
1312                 }
1313                 if (!nocow) {
1314                         if (cow_start == (u64)-1)
1315                                 cow_start = cur_offset;
1316                         cur_offset = extent_end;
1317                         if (cur_offset > end)
1318                                 break;
1319                         path->slots[0]++;
1320                         goto next_slot;
1321                 }
1322
1323                 btrfs_release_path(path);
1324                 if (cow_start != (u64)-1) {
1325                         ret = cow_file_range(inode, locked_page,
1326                                              cow_start, found_key.offset - 1,
1327                                              page_started, nr_written, 1);
1328                         if (ret) {
1329                                 if (!nolock && nocow)
1330                                         btrfs_end_nocow_write(root);
1331                                 goto error;
1332                         }
1333                         cow_start = (u64)-1;
1334                 }
1335
1336                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337                         struct extent_map *em;
1338                         struct extent_map_tree *em_tree;
1339                         em_tree = &BTRFS_I(inode)->extent_tree;
1340                         em = alloc_extent_map();
1341                         BUG_ON(!em); /* -ENOMEM */
1342                         em->start = cur_offset;
1343                         em->orig_start = found_key.offset - extent_offset;
1344                         em->len = num_bytes;
1345                         em->block_len = num_bytes;
1346                         em->block_start = disk_bytenr;
1347                         em->orig_block_len = disk_num_bytes;
1348                         em->ram_bytes = ram_bytes;
1349                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1350                         em->mod_start = em->start;
1351                         em->mod_len = em->len;
1352                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1353                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1354                         em->generation = -1;
1355                         while (1) {
1356                                 write_lock(&em_tree->lock);
1357                                 ret = add_extent_mapping(em_tree, em, 1);
1358                                 write_unlock(&em_tree->lock);
1359                                 if (ret != -EEXIST) {
1360                                         free_extent_map(em);
1361                                         break;
1362                                 }
1363                                 btrfs_drop_extent_cache(inode, em->start,
1364                                                 em->start + em->len - 1, 0);
1365                         }
1366                         type = BTRFS_ORDERED_PREALLOC;
1367                 } else {
1368                         type = BTRFS_ORDERED_NOCOW;
1369                 }
1370
1371                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1372                                                num_bytes, num_bytes, type);
1373                 BUG_ON(ret); /* -ENOMEM */
1374
1375                 if (root->root_key.objectid ==
1376                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1377                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1378                                                       num_bytes);
1379                         if (ret) {
1380                                 if (!nolock && nocow)
1381                                         btrfs_end_nocow_write(root);
1382                                 goto error;
1383                         }
1384                 }
1385
1386                 extent_clear_unlock_delalloc(inode, cur_offset,
1387                                              cur_offset + num_bytes - 1,
1388                                              locked_page, EXTENT_LOCKED |
1389                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1390                                              PAGE_SET_PRIVATE2);
1391                 if (!nolock && nocow)
1392                         btrfs_end_nocow_write(root);
1393                 cur_offset = extent_end;
1394                 if (cur_offset > end)
1395                         break;
1396         }
1397         btrfs_release_path(path);
1398
1399         if (cur_offset <= end && cow_start == (u64)-1) {
1400                 cow_start = cur_offset;
1401                 cur_offset = end;
1402         }
1403
1404         if (cow_start != (u64)-1) {
1405                 ret = cow_file_range(inode, locked_page, cow_start, end,
1406                                      page_started, nr_written, 1);
1407                 if (ret)
1408                         goto error;
1409         }
1410
1411 error:
1412         err = btrfs_end_transaction(trans, root);
1413         if (!ret)
1414                 ret = err;
1415
1416         if (ret && cur_offset < end)
1417                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1418                                              locked_page, EXTENT_LOCKED |
1419                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1420                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1421                                              PAGE_CLEAR_DIRTY |
1422                                              PAGE_SET_WRITEBACK |
1423                                              PAGE_END_WRITEBACK);
1424         btrfs_free_path(path);
1425         return ret;
1426 }
1427
1428 /*
1429  * extent_io.c call back to do delayed allocation processing
1430  */
1431 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1432                               u64 start, u64 end, int *page_started,
1433                               unsigned long *nr_written)
1434 {
1435         int ret;
1436         struct btrfs_root *root = BTRFS_I(inode)->root;
1437
1438         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1439                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1440                                          page_started, 1, nr_written);
1441         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1442                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1443                                          page_started, 0, nr_written);
1444         } else if (!btrfs_test_opt(root, COMPRESS) &&
1445                    !(BTRFS_I(inode)->force_compress) &&
1446                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1447                 ret = cow_file_range(inode, locked_page, start, end,
1448                                       page_started, nr_written, 1);
1449         } else {
1450                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1451                         &BTRFS_I(inode)->runtime_flags);
1452                 ret = cow_file_range_async(inode, locked_page, start, end,
1453                                            page_started, nr_written);
1454         }
1455         return ret;
1456 }
1457
1458 static void btrfs_split_extent_hook(struct inode *inode,
1459                                     struct extent_state *orig, u64 split)
1460 {
1461         /* not delalloc, ignore it */
1462         if (!(orig->state & EXTENT_DELALLOC))
1463                 return;
1464
1465         spin_lock(&BTRFS_I(inode)->lock);
1466         BTRFS_I(inode)->outstanding_extents++;
1467         spin_unlock(&BTRFS_I(inode)->lock);
1468 }
1469
1470 /*
1471  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1472  * extents so we can keep track of new extents that are just merged onto old
1473  * extents, such as when we are doing sequential writes, so we can properly
1474  * account for the metadata space we'll need.
1475  */
1476 static void btrfs_merge_extent_hook(struct inode *inode,
1477                                     struct extent_state *new,
1478                                     struct extent_state *other)
1479 {
1480         /* not delalloc, ignore it */
1481         if (!(other->state & EXTENT_DELALLOC))
1482                 return;
1483
1484         spin_lock(&BTRFS_I(inode)->lock);
1485         BTRFS_I(inode)->outstanding_extents--;
1486         spin_unlock(&BTRFS_I(inode)->lock);
1487 }
1488
1489 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1490                                       struct inode *inode)
1491 {
1492         spin_lock(&root->delalloc_lock);
1493         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1494                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1495                               &root->delalloc_inodes);
1496                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1497                         &BTRFS_I(inode)->runtime_flags);
1498                 root->nr_delalloc_inodes++;
1499                 if (root->nr_delalloc_inodes == 1) {
1500                         spin_lock(&root->fs_info->delalloc_root_lock);
1501                         BUG_ON(!list_empty(&root->delalloc_root));
1502                         list_add_tail(&root->delalloc_root,
1503                                       &root->fs_info->delalloc_roots);
1504                         spin_unlock(&root->fs_info->delalloc_root_lock);
1505                 }
1506         }
1507         spin_unlock(&root->delalloc_lock);
1508 }
1509
1510 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1511                                      struct inode *inode)
1512 {
1513         spin_lock(&root->delalloc_lock);
1514         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1516                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1517                           &BTRFS_I(inode)->runtime_flags);
1518                 root->nr_delalloc_inodes--;
1519                 if (!root->nr_delalloc_inodes) {
1520                         spin_lock(&root->fs_info->delalloc_root_lock);
1521                         BUG_ON(list_empty(&root->delalloc_root));
1522                         list_del_init(&root->delalloc_root);
1523                         spin_unlock(&root->fs_info->delalloc_root_lock);
1524                 }
1525         }
1526         spin_unlock(&root->delalloc_lock);
1527 }
1528
1529 /*
1530  * extent_io.c set_bit_hook, used to track delayed allocation
1531  * bytes in this file, and to maintain the list of inodes that
1532  * have pending delalloc work to be done.
1533  */
1534 static void btrfs_set_bit_hook(struct inode *inode,
1535                                struct extent_state *state, unsigned long *bits)
1536 {
1537
1538         /*
1539          * set_bit and clear bit hooks normally require _irqsave/restore
1540          * but in this case, we are only testing for the DELALLOC
1541          * bit, which is only set or cleared with irqs on
1542          */
1543         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1544                 struct btrfs_root *root = BTRFS_I(inode)->root;
1545                 u64 len = state->end + 1 - state->start;
1546                 bool do_list = !btrfs_is_free_space_inode(inode);
1547
1548                 if (*bits & EXTENT_FIRST_DELALLOC) {
1549                         *bits &= ~EXTENT_FIRST_DELALLOC;
1550                 } else {
1551                         spin_lock(&BTRFS_I(inode)->lock);
1552                         BTRFS_I(inode)->outstanding_extents++;
1553                         spin_unlock(&BTRFS_I(inode)->lock);
1554                 }
1555
1556                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1557                                      root->fs_info->delalloc_batch);
1558                 spin_lock(&BTRFS_I(inode)->lock);
1559                 BTRFS_I(inode)->delalloc_bytes += len;
1560                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1561                                          &BTRFS_I(inode)->runtime_flags))
1562                         btrfs_add_delalloc_inodes(root, inode);
1563                 spin_unlock(&BTRFS_I(inode)->lock);
1564         }
1565 }
1566
1567 /*
1568  * extent_io.c clear_bit_hook, see set_bit_hook for why
1569  */
1570 static void btrfs_clear_bit_hook(struct inode *inode,
1571                                  struct extent_state *state,
1572                                  unsigned long *bits)
1573 {
1574         /*
1575          * set_bit and clear bit hooks normally require _irqsave/restore
1576          * but in this case, we are only testing for the DELALLOC
1577          * bit, which is only set or cleared with irqs on
1578          */
1579         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1580                 struct btrfs_root *root = BTRFS_I(inode)->root;
1581                 u64 len = state->end + 1 - state->start;
1582                 bool do_list = !btrfs_is_free_space_inode(inode);
1583
1584                 if (*bits & EXTENT_FIRST_DELALLOC) {
1585                         *bits &= ~EXTENT_FIRST_DELALLOC;
1586                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1587                         spin_lock(&BTRFS_I(inode)->lock);
1588                         BTRFS_I(inode)->outstanding_extents--;
1589                         spin_unlock(&BTRFS_I(inode)->lock);
1590                 }
1591
1592                 /*
1593                  * We don't reserve metadata space for space cache inodes so we
1594                  * don't need to call dellalloc_release_metadata if there is an
1595                  * error.
1596                  */
1597                 if (*bits & EXTENT_DO_ACCOUNTING &&
1598                     root != root->fs_info->tree_root)
1599                         btrfs_delalloc_release_metadata(inode, len);
1600
1601                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1602                     && do_list && !(state->state & EXTENT_NORESERVE))
1603                         btrfs_free_reserved_data_space(inode, len);
1604
1605                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1606                                      root->fs_info->delalloc_batch);
1607                 spin_lock(&BTRFS_I(inode)->lock);
1608                 BTRFS_I(inode)->delalloc_bytes -= len;
1609                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1610                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1611                              &BTRFS_I(inode)->runtime_flags))
1612                         btrfs_del_delalloc_inode(root, inode);
1613                 spin_unlock(&BTRFS_I(inode)->lock);
1614         }
1615 }
1616
1617 /*
1618  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1619  * we don't create bios that span stripes or chunks
1620  */
1621 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1622                          size_t size, struct bio *bio,
1623                          unsigned long bio_flags)
1624 {
1625         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1626         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1627         u64 length = 0;
1628         u64 map_length;
1629         int ret;
1630
1631         if (bio_flags & EXTENT_BIO_COMPRESSED)
1632                 return 0;
1633
1634         length = bio->bi_iter.bi_size;
1635         map_length = length;
1636         ret = btrfs_map_block(root->fs_info, rw, logical,
1637                               &map_length, NULL, 0);
1638         /* Will always return 0 with map_multi == NULL */
1639         BUG_ON(ret < 0);
1640         if (map_length < length + size)
1641                 return 1;
1642         return 0;
1643 }
1644
1645 /*
1646  * in order to insert checksums into the metadata in large chunks,
1647  * we wait until bio submission time.   All the pages in the bio are
1648  * checksummed and sums are attached onto the ordered extent record.
1649  *
1650  * At IO completion time the cums attached on the ordered extent record
1651  * are inserted into the btree
1652  */
1653 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1654                                     struct bio *bio, int mirror_num,
1655                                     unsigned long bio_flags,
1656                                     u64 bio_offset)
1657 {
1658         struct btrfs_root *root = BTRFS_I(inode)->root;
1659         int ret = 0;
1660
1661         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1662         BUG_ON(ret); /* -ENOMEM */
1663         return 0;
1664 }
1665
1666 /*
1667  * in order to insert checksums into the metadata in large chunks,
1668  * we wait until bio submission time.   All the pages in the bio are
1669  * checksummed and sums are attached onto the ordered extent record.
1670  *
1671  * At IO completion time the cums attached on the ordered extent record
1672  * are inserted into the btree
1673  */
1674 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1675                           int mirror_num, unsigned long bio_flags,
1676                           u64 bio_offset)
1677 {
1678         struct btrfs_root *root = BTRFS_I(inode)->root;
1679         int ret;
1680
1681         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1682         if (ret)
1683                 bio_endio(bio, ret);
1684         return ret;
1685 }
1686
1687 /*
1688  * extent_io.c submission hook. This does the right thing for csum calculation
1689  * on write, or reading the csums from the tree before a read
1690  */
1691 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1692                           int mirror_num, unsigned long bio_flags,
1693                           u64 bio_offset)
1694 {
1695         struct btrfs_root *root = BTRFS_I(inode)->root;
1696         int ret = 0;
1697         int skip_sum;
1698         int metadata = 0;
1699         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1700
1701         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1702
1703         if (btrfs_is_free_space_inode(inode))
1704                 metadata = 2;
1705
1706         if (!(rw & REQ_WRITE)) {
1707                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1708                 if (ret)
1709                         goto out;
1710
1711                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1712                         ret = btrfs_submit_compressed_read(inode, bio,
1713                                                            mirror_num,
1714                                                            bio_flags);
1715                         goto out;
1716                 } else if (!skip_sum) {
1717                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1718                         if (ret)
1719                                 goto out;
1720                 }
1721                 goto mapit;
1722         } else if (async && !skip_sum) {
1723                 /* csum items have already been cloned */
1724                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1725                         goto mapit;
1726                 /* we're doing a write, do the async checksumming */
1727                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1728                                    inode, rw, bio, mirror_num,
1729                                    bio_flags, bio_offset,
1730                                    __btrfs_submit_bio_start,
1731                                    __btrfs_submit_bio_done);
1732                 goto out;
1733         } else if (!skip_sum) {
1734                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1735                 if (ret)
1736                         goto out;
1737         }
1738
1739 mapit:
1740         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1741
1742 out:
1743         if (ret < 0)
1744                 bio_endio(bio, ret);
1745         return ret;
1746 }
1747
1748 /*
1749  * given a list of ordered sums record them in the inode.  This happens
1750  * at IO completion time based on sums calculated at bio submission time.
1751  */
1752 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1753                              struct inode *inode, u64 file_offset,
1754                              struct list_head *list)
1755 {
1756         struct btrfs_ordered_sum *sum;
1757
1758         list_for_each_entry(sum, list, list) {
1759                 trans->adding_csums = 1;
1760                 btrfs_csum_file_blocks(trans,
1761                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1762                 trans->adding_csums = 0;
1763         }
1764         return 0;
1765 }
1766
1767 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1768                               struct extent_state **cached_state)
1769 {
1770         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1771         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1772                                    cached_state, GFP_NOFS);
1773 }
1774
1775 /* see btrfs_writepage_start_hook for details on why this is required */
1776 struct btrfs_writepage_fixup {
1777         struct page *page;
1778         struct btrfs_work work;
1779 };
1780
1781 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1782 {
1783         struct btrfs_writepage_fixup *fixup;
1784         struct btrfs_ordered_extent *ordered;
1785         struct extent_state *cached_state = NULL;
1786         struct page *page;
1787         struct inode *inode;
1788         u64 page_start;
1789         u64 page_end;
1790         int ret;
1791
1792         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1793         page = fixup->page;
1794 again:
1795         lock_page(page);
1796         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1797                 ClearPageChecked(page);
1798                 goto out_page;
1799         }
1800
1801         inode = page->mapping->host;
1802         page_start = page_offset(page);
1803         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1804
1805         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1806                          &cached_state);
1807
1808         /* already ordered? We're done */
1809         if (PagePrivate2(page))
1810                 goto out;
1811
1812         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1813         if (ordered) {
1814                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1815                                      page_end, &cached_state, GFP_NOFS);
1816                 unlock_page(page);
1817                 btrfs_start_ordered_extent(inode, ordered, 1);
1818                 btrfs_put_ordered_extent(ordered);
1819                 goto again;
1820         }
1821
1822         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1823         if (ret) {
1824                 mapping_set_error(page->mapping, ret);
1825                 end_extent_writepage(page, ret, page_start, page_end);
1826                 ClearPageChecked(page);
1827                 goto out;
1828          }
1829
1830         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1831         ClearPageChecked(page);
1832         set_page_dirty(page);
1833 out:
1834         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1835                              &cached_state, GFP_NOFS);
1836 out_page:
1837         unlock_page(page);
1838         page_cache_release(page);
1839         kfree(fixup);
1840 }
1841
1842 /*
1843  * There are a few paths in the higher layers of the kernel that directly
1844  * set the page dirty bit without asking the filesystem if it is a
1845  * good idea.  This causes problems because we want to make sure COW
1846  * properly happens and the data=ordered rules are followed.
1847  *
1848  * In our case any range that doesn't have the ORDERED bit set
1849  * hasn't been properly setup for IO.  We kick off an async process
1850  * to fix it up.  The async helper will wait for ordered extents, set
1851  * the delalloc bit and make it safe to write the page.
1852  */
1853 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1854 {
1855         struct inode *inode = page->mapping->host;
1856         struct btrfs_writepage_fixup *fixup;
1857         struct btrfs_root *root = BTRFS_I(inode)->root;
1858
1859         /* this page is properly in the ordered list */
1860         if (TestClearPagePrivate2(page))
1861                 return 0;
1862
1863         if (PageChecked(page))
1864                 return -EAGAIN;
1865
1866         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1867         if (!fixup)
1868                 return -EAGAIN;
1869
1870         SetPageChecked(page);
1871         page_cache_get(page);
1872         btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1873         fixup->page = page;
1874         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1875         return -EBUSY;
1876 }
1877
1878 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1879                                        struct inode *inode, u64 file_pos,
1880                                        u64 disk_bytenr, u64 disk_num_bytes,
1881                                        u64 num_bytes, u64 ram_bytes,
1882                                        u8 compression, u8 encryption,
1883                                        u16 other_encoding, int extent_type)
1884 {
1885         struct btrfs_root *root = BTRFS_I(inode)->root;
1886         struct btrfs_file_extent_item *fi;
1887         struct btrfs_path *path;
1888         struct extent_buffer *leaf;
1889         struct btrfs_key ins;
1890         int extent_inserted = 0;
1891         int ret;
1892
1893         path = btrfs_alloc_path();
1894         if (!path)
1895                 return -ENOMEM;
1896
1897         /*
1898          * we may be replacing one extent in the tree with another.
1899          * The new extent is pinned in the extent map, and we don't want
1900          * to drop it from the cache until it is completely in the btree.
1901          *
1902          * So, tell btrfs_drop_extents to leave this extent in the cache.
1903          * the caller is expected to unpin it and allow it to be merged
1904          * with the others.
1905          */
1906         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1907                                    file_pos + num_bytes, NULL, 0,
1908                                    1, sizeof(*fi), &extent_inserted);
1909         if (ret)
1910                 goto out;
1911
1912         if (!extent_inserted) {
1913                 ins.objectid = btrfs_ino(inode);
1914                 ins.offset = file_pos;
1915                 ins.type = BTRFS_EXTENT_DATA_KEY;
1916
1917                 path->leave_spinning = 1;
1918                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1919                                               sizeof(*fi));
1920                 if (ret)
1921                         goto out;
1922         }
1923         leaf = path->nodes[0];
1924         fi = btrfs_item_ptr(leaf, path->slots[0],
1925                             struct btrfs_file_extent_item);
1926         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1927         btrfs_set_file_extent_type(leaf, fi, extent_type);
1928         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1929         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1930         btrfs_set_file_extent_offset(leaf, fi, 0);
1931         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1932         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1933         btrfs_set_file_extent_compression(leaf, fi, compression);
1934         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1935         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1936
1937         btrfs_mark_buffer_dirty(leaf);
1938         btrfs_release_path(path);
1939
1940         inode_add_bytes(inode, num_bytes);
1941
1942         ins.objectid = disk_bytenr;
1943         ins.offset = disk_num_bytes;
1944         ins.type = BTRFS_EXTENT_ITEM_KEY;
1945         ret = btrfs_alloc_reserved_file_extent(trans, root,
1946                                         root->root_key.objectid,
1947                                         btrfs_ino(inode), file_pos, &ins);
1948 out:
1949         btrfs_free_path(path);
1950
1951         return ret;
1952 }
1953
1954 /* snapshot-aware defrag */
1955 struct sa_defrag_extent_backref {
1956         struct rb_node node;
1957         struct old_sa_defrag_extent *old;
1958         u64 root_id;
1959         u64 inum;
1960         u64 file_pos;
1961         u64 extent_offset;
1962         u64 num_bytes;
1963         u64 generation;
1964 };
1965
1966 struct old_sa_defrag_extent {
1967         struct list_head list;
1968         struct new_sa_defrag_extent *new;
1969
1970         u64 extent_offset;
1971         u64 bytenr;
1972         u64 offset;
1973         u64 len;
1974         int count;
1975 };
1976
1977 struct new_sa_defrag_extent {
1978         struct rb_root root;
1979         struct list_head head;
1980         struct btrfs_path *path;
1981         struct inode *inode;
1982         u64 file_pos;
1983         u64 len;
1984         u64 bytenr;
1985         u64 disk_len;
1986         u8 compress_type;
1987 };
1988
1989 static int backref_comp(struct sa_defrag_extent_backref *b1,
1990                         struct sa_defrag_extent_backref *b2)
1991 {
1992         if (b1->root_id < b2->root_id)
1993                 return -1;
1994         else if (b1->root_id > b2->root_id)
1995                 return 1;
1996
1997         if (b1->inum < b2->inum)
1998                 return -1;
1999         else if (b1->inum > b2->inum)
2000                 return 1;
2001
2002         if (b1->file_pos < b2->file_pos)
2003                 return -1;
2004         else if (b1->file_pos > b2->file_pos)
2005                 return 1;
2006
2007         /*
2008          * [------------------------------] ===> (a range of space)
2009          *     |<--->|   |<---->| =============> (fs/file tree A)
2010          * |<---------------------------->| ===> (fs/file tree B)
2011          *
2012          * A range of space can refer to two file extents in one tree while
2013          * refer to only one file extent in another tree.
2014          *
2015          * So we may process a disk offset more than one time(two extents in A)
2016          * and locate at the same extent(one extent in B), then insert two same
2017          * backrefs(both refer to the extent in B).
2018          */
2019         return 0;
2020 }
2021
2022 static void backref_insert(struct rb_root *root,
2023                            struct sa_defrag_extent_backref *backref)
2024 {
2025         struct rb_node **p = &root->rb_node;
2026         struct rb_node *parent = NULL;
2027         struct sa_defrag_extent_backref *entry;
2028         int ret;
2029
2030         while (*p) {
2031                 parent = *p;
2032                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2033
2034                 ret = backref_comp(backref, entry);
2035                 if (ret < 0)
2036                         p = &(*p)->rb_left;
2037                 else
2038                         p = &(*p)->rb_right;
2039         }
2040
2041         rb_link_node(&backref->node, parent, p);
2042         rb_insert_color(&backref->node, root);
2043 }
2044
2045 /*
2046  * Note the backref might has changed, and in this case we just return 0.
2047  */
2048 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2049                                        void *ctx)
2050 {
2051         struct btrfs_file_extent_item *extent;
2052         struct btrfs_fs_info *fs_info;
2053         struct old_sa_defrag_extent *old = ctx;
2054         struct new_sa_defrag_extent *new = old->new;
2055         struct btrfs_path *path = new->path;
2056         struct btrfs_key key;
2057         struct btrfs_root *root;
2058         struct sa_defrag_extent_backref *backref;
2059         struct extent_buffer *leaf;
2060         struct inode *inode = new->inode;
2061         int slot;
2062         int ret;
2063         u64 extent_offset;
2064         u64 num_bytes;
2065
2066         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2067             inum == btrfs_ino(inode))
2068                 return 0;
2069
2070         key.objectid = root_id;
2071         key.type = BTRFS_ROOT_ITEM_KEY;
2072         key.offset = (u64)-1;
2073
2074         fs_info = BTRFS_I(inode)->root->fs_info;
2075         root = btrfs_read_fs_root_no_name(fs_info, &key);
2076         if (IS_ERR(root)) {
2077                 if (PTR_ERR(root) == -ENOENT)
2078                         return 0;
2079                 WARN_ON(1);
2080                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2081                          inum, offset, root_id);
2082                 return PTR_ERR(root);
2083         }
2084
2085         key.objectid = inum;
2086         key.type = BTRFS_EXTENT_DATA_KEY;
2087         if (offset > (u64)-1 << 32)
2088                 key.offset = 0;
2089         else
2090                 key.offset = offset;
2091
2092         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2093         if (WARN_ON(ret < 0))
2094                 return ret;
2095         ret = 0;
2096
2097         while (1) {
2098                 cond_resched();
2099
2100                 leaf = path->nodes[0];
2101                 slot = path->slots[0];
2102
2103                 if (slot >= btrfs_header_nritems(leaf)) {
2104                         ret = btrfs_next_leaf(root, path);
2105                         if (ret < 0) {
2106                                 goto out;
2107                         } else if (ret > 0) {
2108                                 ret = 0;
2109                                 goto out;
2110                         }
2111                         continue;
2112                 }
2113
2114                 path->slots[0]++;
2115
2116                 btrfs_item_key_to_cpu(leaf, &key, slot);
2117
2118                 if (key.objectid > inum)
2119                         goto out;
2120
2121                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2122                         continue;
2123
2124                 extent = btrfs_item_ptr(leaf, slot,
2125                                         struct btrfs_file_extent_item);
2126
2127                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2128                         continue;
2129
2130                 /*
2131                  * 'offset' refers to the exact key.offset,
2132                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2133                  * (key.offset - extent_offset).
2134                  */
2135                 if (key.offset != offset)
2136                         continue;
2137
2138                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2140
2141                 if (extent_offset >= old->extent_offset + old->offset +
2142                     old->len || extent_offset + num_bytes <=
2143                     old->extent_offset + old->offset)
2144                         continue;
2145                 break;
2146         }
2147
2148         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2149         if (!backref) {
2150                 ret = -ENOENT;
2151                 goto out;
2152         }
2153
2154         backref->root_id = root_id;
2155         backref->inum = inum;
2156         backref->file_pos = offset;
2157         backref->num_bytes = num_bytes;
2158         backref->extent_offset = extent_offset;
2159         backref->generation = btrfs_file_extent_generation(leaf, extent);
2160         backref->old = old;
2161         backref_insert(&new->root, backref);
2162         old->count++;
2163 out:
2164         btrfs_release_path(path);
2165         WARN_ON(ret);
2166         return ret;
2167 }
2168
2169 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2170                                    struct new_sa_defrag_extent *new)
2171 {
2172         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2173         struct old_sa_defrag_extent *old, *tmp;
2174         int ret;
2175
2176         new->path = path;
2177
2178         list_for_each_entry_safe(old, tmp, &new->head, list) {
2179                 ret = iterate_inodes_from_logical(old->bytenr +
2180                                                   old->extent_offset, fs_info,
2181                                                   path, record_one_backref,
2182                                                   old);
2183                 if (ret < 0 && ret != -ENOENT)
2184                         return false;
2185
2186                 /* no backref to be processed for this extent */
2187                 if (!old->count) {
2188                         list_del(&old->list);
2189                         kfree(old);
2190                 }
2191         }
2192
2193         if (list_empty(&new->head))
2194                 return false;
2195
2196         return true;
2197 }
2198
2199 static int relink_is_mergable(struct extent_buffer *leaf,
2200                               struct btrfs_file_extent_item *fi,
2201                               struct new_sa_defrag_extent *new)
2202 {
2203         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2204                 return 0;
2205
2206         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2207                 return 0;
2208
2209         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2210                 return 0;
2211
2212         if (btrfs_file_extent_encryption(leaf, fi) ||
2213             btrfs_file_extent_other_encoding(leaf, fi))
2214                 return 0;
2215
2216         return 1;
2217 }
2218
2219 /*
2220  * Note the backref might has changed, and in this case we just return 0.
2221  */
2222 static noinline int relink_extent_backref(struct btrfs_path *path,
2223                                  struct sa_defrag_extent_backref *prev,
2224                                  struct sa_defrag_extent_backref *backref)
2225 {
2226         struct btrfs_file_extent_item *extent;
2227         struct btrfs_file_extent_item *item;
2228         struct btrfs_ordered_extent *ordered;
2229         struct btrfs_trans_handle *trans;
2230         struct btrfs_fs_info *fs_info;
2231         struct btrfs_root *root;
2232         struct btrfs_key key;
2233         struct extent_buffer *leaf;
2234         struct old_sa_defrag_extent *old = backref->old;
2235         struct new_sa_defrag_extent *new = old->new;
2236         struct inode *src_inode = new->inode;
2237         struct inode *inode;
2238         struct extent_state *cached = NULL;
2239         int ret = 0;
2240         u64 start;
2241         u64 len;
2242         u64 lock_start;
2243         u64 lock_end;
2244         bool merge = false;
2245         int index;
2246
2247         if (prev && prev->root_id == backref->root_id &&
2248             prev->inum == backref->inum &&
2249             prev->file_pos + prev->num_bytes == backref->file_pos)
2250                 merge = true;
2251
2252         /* step 1: get root */
2253         key.objectid = backref->root_id;
2254         key.type = BTRFS_ROOT_ITEM_KEY;
2255         key.offset = (u64)-1;
2256
2257         fs_info = BTRFS_I(src_inode)->root->fs_info;
2258         index = srcu_read_lock(&fs_info->subvol_srcu);
2259
2260         root = btrfs_read_fs_root_no_name(fs_info, &key);
2261         if (IS_ERR(root)) {
2262                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2263                 if (PTR_ERR(root) == -ENOENT)
2264                         return 0;
2265                 return PTR_ERR(root);
2266         }
2267
2268         if (btrfs_root_readonly(root)) {
2269                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2270                 return 0;
2271         }
2272
2273         /* step 2: get inode */
2274         key.objectid = backref->inum;
2275         key.type = BTRFS_INODE_ITEM_KEY;
2276         key.offset = 0;
2277
2278         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2279         if (IS_ERR(inode)) {
2280                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2281                 return 0;
2282         }
2283
2284         srcu_read_unlock(&fs_info->subvol_srcu, index);
2285
2286         /* step 3: relink backref */
2287         lock_start = backref->file_pos;
2288         lock_end = backref->file_pos + backref->num_bytes - 1;
2289         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2290                          0, &cached);
2291
2292         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2293         if (ordered) {
2294                 btrfs_put_ordered_extent(ordered);
2295                 goto out_unlock;
2296         }
2297
2298         trans = btrfs_join_transaction(root);
2299         if (IS_ERR(trans)) {
2300                 ret = PTR_ERR(trans);
2301                 goto out_unlock;
2302         }
2303
2304         key.objectid = backref->inum;
2305         key.type = BTRFS_EXTENT_DATA_KEY;
2306         key.offset = backref->file_pos;
2307
2308         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2309         if (ret < 0) {
2310                 goto out_free_path;
2311         } else if (ret > 0) {
2312                 ret = 0;
2313                 goto out_free_path;
2314         }
2315
2316         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2317                                 struct btrfs_file_extent_item);
2318
2319         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2320             backref->generation)
2321                 goto out_free_path;
2322
2323         btrfs_release_path(path);
2324
2325         start = backref->file_pos;
2326         if (backref->extent_offset < old->extent_offset + old->offset)
2327                 start += old->extent_offset + old->offset -
2328                          backref->extent_offset;
2329
2330         len = min(backref->extent_offset + backref->num_bytes,
2331                   old->extent_offset + old->offset + old->len);
2332         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2333
2334         ret = btrfs_drop_extents(trans, root, inode, start,
2335                                  start + len, 1);
2336         if (ret)
2337                 goto out_free_path;
2338 again:
2339         key.objectid = btrfs_ino(inode);
2340         key.type = BTRFS_EXTENT_DATA_KEY;
2341         key.offset = start;
2342
2343         path->leave_spinning = 1;
2344         if (merge) {
2345                 struct btrfs_file_extent_item *fi;
2346                 u64 extent_len;
2347                 struct btrfs_key found_key;
2348
2349                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2350                 if (ret < 0)
2351                         goto out_free_path;
2352
2353                 path->slots[0]--;
2354                 leaf = path->nodes[0];
2355                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2356
2357                 fi = btrfs_item_ptr(leaf, path->slots[0],
2358                                     struct btrfs_file_extent_item);
2359                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2360
2361                 if (extent_len + found_key.offset == start &&
2362                     relink_is_mergable(leaf, fi, new)) {
2363                         btrfs_set_file_extent_num_bytes(leaf, fi,
2364                                                         extent_len + len);
2365                         btrfs_mark_buffer_dirty(leaf);
2366                         inode_add_bytes(inode, len);
2367
2368                         ret = 1;
2369                         goto out_free_path;
2370                 } else {
2371                         merge = false;
2372                         btrfs_release_path(path);
2373                         goto again;
2374                 }
2375         }
2376
2377         ret = btrfs_insert_empty_item(trans, root, path, &key,
2378                                         sizeof(*extent));
2379         if (ret) {
2380                 btrfs_abort_transaction(trans, root, ret);
2381                 goto out_free_path;
2382         }
2383
2384         leaf = path->nodes[0];
2385         item = btrfs_item_ptr(leaf, path->slots[0],
2386                                 struct btrfs_file_extent_item);
2387         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2388         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2389         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2390         btrfs_set_file_extent_num_bytes(leaf, item, len);
2391         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2392         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2393         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2394         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2395         btrfs_set_file_extent_encryption(leaf, item, 0);
2396         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2397
2398         btrfs_mark_buffer_dirty(leaf);
2399         inode_add_bytes(inode, len);
2400         btrfs_release_path(path);
2401
2402         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2403                         new->disk_len, 0,
2404                         backref->root_id, backref->inum,
2405                         new->file_pos, 0);      /* start - extent_offset */
2406         if (ret) {
2407                 btrfs_abort_transaction(trans, root, ret);
2408                 goto out_free_path;
2409         }
2410
2411         ret = 1;
2412 out_free_path:
2413         btrfs_release_path(path);
2414         path->leave_spinning = 0;
2415         btrfs_end_transaction(trans, root);
2416 out_unlock:
2417         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2418                              &cached, GFP_NOFS);
2419         iput(inode);
2420         return ret;
2421 }
2422
2423 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2424 {
2425         struct old_sa_defrag_extent *old, *tmp;
2426
2427         if (!new)
2428                 return;
2429
2430         list_for_each_entry_safe(old, tmp, &new->head, list) {
2431                 list_del(&old->list);
2432                 kfree(old);
2433         }
2434         kfree(new);
2435 }
2436
2437 static void relink_file_extents(struct new_sa_defrag_extent *new)
2438 {
2439         struct btrfs_path *path;
2440         struct sa_defrag_extent_backref *backref;
2441         struct sa_defrag_extent_backref *prev = NULL;
2442         struct inode *inode;
2443         struct btrfs_root *root;
2444         struct rb_node *node;
2445         int ret;
2446
2447         inode = new->inode;
2448         root = BTRFS_I(inode)->root;
2449
2450         path = btrfs_alloc_path();
2451         if (!path)
2452                 return;
2453
2454         if (!record_extent_backrefs(path, new)) {
2455                 btrfs_free_path(path);
2456                 goto out;
2457         }
2458         btrfs_release_path(path);
2459
2460         while (1) {
2461                 node = rb_first(&new->root);
2462                 if (!node)
2463                         break;
2464                 rb_erase(node, &new->root);
2465
2466                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2467
2468                 ret = relink_extent_backref(path, prev, backref);
2469                 WARN_ON(ret < 0);
2470
2471                 kfree(prev);
2472
2473                 if (ret == 1)
2474                         prev = backref;
2475                 else
2476                         prev = NULL;
2477                 cond_resched();
2478         }
2479         kfree(prev);
2480
2481         btrfs_free_path(path);
2482 out:
2483         free_sa_defrag_extent(new);
2484
2485         atomic_dec(&root->fs_info->defrag_running);
2486         wake_up(&root->fs_info->transaction_wait);
2487 }
2488
2489 static struct new_sa_defrag_extent *
2490 record_old_file_extents(struct inode *inode,
2491                         struct btrfs_ordered_extent *ordered)
2492 {
2493         struct btrfs_root *root = BTRFS_I(inode)->root;
2494         struct btrfs_path *path;
2495         struct btrfs_key key;
2496         struct old_sa_defrag_extent *old;
2497         struct new_sa_defrag_extent *new;
2498         int ret;
2499
2500         new = kmalloc(sizeof(*new), GFP_NOFS);
2501         if (!new)
2502                 return NULL;
2503
2504         new->inode = inode;
2505         new->file_pos = ordered->file_offset;
2506         new->len = ordered->len;
2507         new->bytenr = ordered->start;
2508         new->disk_len = ordered->disk_len;
2509         new->compress_type = ordered->compress_type;
2510         new->root = RB_ROOT;
2511         INIT_LIST_HEAD(&new->head);
2512
2513         path = btrfs_alloc_path();
2514         if (!path)
2515                 goto out_kfree;
2516
2517         key.objectid = btrfs_ino(inode);
2518         key.type = BTRFS_EXTENT_DATA_KEY;
2519         key.offset = new->file_pos;
2520
2521         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2522         if (ret < 0)
2523                 goto out_free_path;
2524         if (ret > 0 && path->slots[0] > 0)
2525                 path->slots[0]--;
2526
2527         /* find out all the old extents for the file range */
2528         while (1) {
2529                 struct btrfs_file_extent_item *extent;
2530                 struct extent_buffer *l;
2531                 int slot;
2532                 u64 num_bytes;
2533                 u64 offset;
2534                 u64 end;
2535                 u64 disk_bytenr;
2536                 u64 extent_offset;
2537
2538                 l = path->nodes[0];
2539                 slot = path->slots[0];
2540
2541                 if (slot >= btrfs_header_nritems(l)) {
2542                         ret = btrfs_next_leaf(root, path);
2543                         if (ret < 0)
2544                                 goto out_free_path;
2545                         else if (ret > 0)
2546                                 break;
2547                         continue;
2548                 }
2549
2550                 btrfs_item_key_to_cpu(l, &key, slot);
2551
2552                 if (key.objectid != btrfs_ino(inode))
2553                         break;
2554                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2555                         break;
2556                 if (key.offset >= new->file_pos + new->len)
2557                         break;
2558
2559                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2560
2561                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2562                 if (key.offset + num_bytes < new->file_pos)
2563                         goto next;
2564
2565                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2566                 if (!disk_bytenr)
2567                         goto next;
2568
2569                 extent_offset = btrfs_file_extent_offset(l, extent);
2570
2571                 old = kmalloc(sizeof(*old), GFP_NOFS);
2572                 if (!old)
2573                         goto out_free_path;
2574
2575                 offset = max(new->file_pos, key.offset);
2576                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2577
2578                 old->bytenr = disk_bytenr;
2579                 old->extent_offset = extent_offset;
2580                 old->offset = offset - key.offset;
2581                 old->len = end - offset;
2582                 old->new = new;
2583                 old->count = 0;
2584                 list_add_tail(&old->list, &new->head);
2585 next:
2586                 path->slots[0]++;
2587                 cond_resched();
2588         }
2589
2590         btrfs_free_path(path);
2591         atomic_inc(&root->fs_info->defrag_running);
2592
2593         return new;
2594
2595 out_free_path:
2596         btrfs_free_path(path);
2597 out_kfree:
2598         free_sa_defrag_extent(new);
2599         return NULL;
2600 }
2601
2602 /* as ordered data IO finishes, this gets called so we can finish
2603  * an ordered extent if the range of bytes in the file it covers are
2604  * fully written.
2605  */
2606 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2607 {
2608         struct inode *inode = ordered_extent->inode;
2609         struct btrfs_root *root = BTRFS_I(inode)->root;
2610         struct btrfs_trans_handle *trans = NULL;
2611         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2612         struct extent_state *cached_state = NULL;
2613         struct new_sa_defrag_extent *new = NULL;
2614         int compress_type = 0;
2615         int ret = 0;
2616         u64 logical_len = ordered_extent->len;
2617         bool nolock;
2618         bool truncated = false;
2619
2620         nolock = btrfs_is_free_space_inode(inode);
2621
2622         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2623                 ret = -EIO;
2624                 goto out;
2625         }
2626
2627         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2628                 truncated = true;
2629                 logical_len = ordered_extent->truncated_len;
2630                 /* Truncated the entire extent, don't bother adding */
2631                 if (!logical_len)
2632                         goto out;
2633         }
2634
2635         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2636                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2637                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2638                 if (nolock)
2639                         trans = btrfs_join_transaction_nolock(root);
2640                 else
2641                         trans = btrfs_join_transaction(root);
2642                 if (IS_ERR(trans)) {
2643                         ret = PTR_ERR(trans);
2644                         trans = NULL;
2645                         goto out;
2646                 }
2647                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2648                 ret = btrfs_update_inode_fallback(trans, root, inode);
2649                 if (ret) /* -ENOMEM or corruption */
2650                         btrfs_abort_transaction(trans, root, ret);
2651                 goto out;
2652         }
2653
2654         lock_extent_bits(io_tree, ordered_extent->file_offset,
2655                          ordered_extent->file_offset + ordered_extent->len - 1,
2656                          0, &cached_state);
2657
2658         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2659                         ordered_extent->file_offset + ordered_extent->len - 1,
2660                         EXTENT_DEFRAG, 1, cached_state);
2661         if (ret) {
2662                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2663                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2664                         /* the inode is shared */
2665                         new = record_old_file_extents(inode, ordered_extent);
2666
2667                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2668                         ordered_extent->file_offset + ordered_extent->len - 1,
2669                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2670         }
2671
2672         if (nolock)
2673                 trans = btrfs_join_transaction_nolock(root);
2674         else
2675                 trans = btrfs_join_transaction(root);
2676         if (IS_ERR(trans)) {
2677                 ret = PTR_ERR(trans);
2678                 trans = NULL;
2679                 goto out_unlock;
2680         }
2681
2682         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2683
2684         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2685                 compress_type = ordered_extent->compress_type;
2686         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2687                 BUG_ON(compress_type);
2688                 ret = btrfs_mark_extent_written(trans, inode,
2689                                                 ordered_extent->file_offset,
2690                                                 ordered_extent->file_offset +
2691                                                 logical_len);
2692         } else {
2693                 BUG_ON(root == root->fs_info->tree_root);
2694                 ret = insert_reserved_file_extent(trans, inode,
2695                                                 ordered_extent->file_offset,
2696                                                 ordered_extent->start,
2697                                                 ordered_extent->disk_len,
2698                                                 logical_len, logical_len,
2699                                                 compress_type, 0, 0,
2700                                                 BTRFS_FILE_EXTENT_REG);
2701         }
2702         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2703                            ordered_extent->file_offset, ordered_extent->len,
2704                            trans->transid);
2705         if (ret < 0) {
2706                 btrfs_abort_transaction(trans, root, ret);
2707                 goto out_unlock;
2708         }
2709
2710         add_pending_csums(trans, inode, ordered_extent->file_offset,
2711                           &ordered_extent->list);
2712
2713         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2714         ret = btrfs_update_inode_fallback(trans, root, inode);
2715         if (ret) { /* -ENOMEM or corruption */
2716                 btrfs_abort_transaction(trans, root, ret);
2717                 goto out_unlock;
2718         }
2719         ret = 0;
2720 out_unlock:
2721         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2722                              ordered_extent->file_offset +
2723                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2724 out:
2725         if (root != root->fs_info->tree_root)
2726                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2727         if (trans)
2728                 btrfs_end_transaction(trans, root);
2729
2730         if (ret || truncated) {
2731                 u64 start, end;
2732
2733                 if (truncated)
2734                         start = ordered_extent->file_offset + logical_len;
2735                 else
2736                         start = ordered_extent->file_offset;
2737                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2738                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2739
2740                 /* Drop the cache for the part of the extent we didn't write. */
2741                 btrfs_drop_extent_cache(inode, start, end, 0);
2742
2743                 /*
2744                  * If the ordered extent had an IOERR or something else went
2745                  * wrong we need to return the space for this ordered extent
2746                  * back to the allocator.  We only free the extent in the
2747                  * truncated case if we didn't write out the extent at all.
2748                  */
2749                 if ((ret || !logical_len) &&
2750                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2751                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2752                         btrfs_free_reserved_extent(root, ordered_extent->start,
2753                                                    ordered_extent->disk_len);
2754         }
2755
2756
2757         /*
2758          * This needs to be done to make sure anybody waiting knows we are done
2759          * updating everything for this ordered extent.
2760          */
2761         btrfs_remove_ordered_extent(inode, ordered_extent);
2762
2763         /* for snapshot-aware defrag */
2764         if (new) {
2765                 if (ret) {
2766                         free_sa_defrag_extent(new);
2767                         atomic_dec(&root->fs_info->defrag_running);
2768                 } else {
2769                         relink_file_extents(new);
2770                 }
2771         }
2772
2773         /* once for us */
2774         btrfs_put_ordered_extent(ordered_extent);
2775         /* once for the tree */
2776         btrfs_put_ordered_extent(ordered_extent);
2777
2778         return ret;
2779 }
2780
2781 static void finish_ordered_fn(struct btrfs_work *work)
2782 {
2783         struct btrfs_ordered_extent *ordered_extent;
2784         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2785         btrfs_finish_ordered_io(ordered_extent);
2786 }
2787
2788 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2789                                 struct extent_state *state, int uptodate)
2790 {
2791         struct inode *inode = page->mapping->host;
2792         struct btrfs_root *root = BTRFS_I(inode)->root;
2793         struct btrfs_ordered_extent *ordered_extent = NULL;
2794         struct btrfs_workqueue *workers;
2795
2796         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2797
2798         ClearPagePrivate2(page);
2799         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2800                                             end - start + 1, uptodate))
2801                 return 0;
2802
2803         btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2804
2805         if (btrfs_is_free_space_inode(inode))
2806                 workers = root->fs_info->endio_freespace_worker;
2807         else
2808                 workers = root->fs_info->endio_write_workers;
2809         btrfs_queue_work(workers, &ordered_extent->work);
2810
2811         return 0;
2812 }
2813
2814 /*
2815  * when reads are done, we need to check csums to verify the data is correct
2816  * if there's a match, we allow the bio to finish.  If not, the code in
2817  * extent_io.c will try to find good copies for us.
2818  */
2819 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2820                                       u64 phy_offset, struct page *page,
2821                                       u64 start, u64 end, int mirror)
2822 {
2823         size_t offset = start - page_offset(page);
2824         struct inode *inode = page->mapping->host;
2825         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2826         char *kaddr;
2827         struct btrfs_root *root = BTRFS_I(inode)->root;
2828         u32 csum_expected;
2829         u32 csum = ~(u32)0;
2830         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2831                                       DEFAULT_RATELIMIT_BURST);
2832
2833         if (PageChecked(page)) {
2834                 ClearPageChecked(page);
2835                 goto good;
2836         }
2837
2838         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2839                 goto good;
2840
2841         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2842             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2843                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2844                                   GFP_NOFS);
2845                 return 0;
2846         }
2847
2848         phy_offset >>= inode->i_sb->s_blocksize_bits;
2849         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2850
2851         kaddr = kmap_atomic(page);
2852         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2853         btrfs_csum_final(csum, (char *)&csum);
2854         if (csum != csum_expected)
2855                 goto zeroit;
2856
2857         kunmap_atomic(kaddr);
2858 good:
2859         return 0;
2860
2861 zeroit:
2862         if (__ratelimit(&_rs))
2863                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2864                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2865         memset(kaddr + offset, 1, end - start + 1);
2866         flush_dcache_page(page);
2867         kunmap_atomic(kaddr);
2868         if (csum_expected == 0)
2869                 return 0;
2870         return -EIO;
2871 }
2872
2873 struct delayed_iput {
2874         struct list_head list;
2875         struct inode *inode;
2876 };
2877
2878 /* JDM: If this is fs-wide, why can't we add a pointer to
2879  * btrfs_inode instead and avoid the allocation? */
2880 void btrfs_add_delayed_iput(struct inode *inode)
2881 {
2882         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2883         struct delayed_iput *delayed;
2884
2885         if (atomic_add_unless(&inode->i_count, -1, 1))
2886                 return;
2887
2888         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2889         delayed->inode = inode;
2890
2891         spin_lock(&fs_info->delayed_iput_lock);
2892         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2893         spin_unlock(&fs_info->delayed_iput_lock);
2894 }
2895
2896 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2897 {
2898         LIST_HEAD(list);
2899         struct btrfs_fs_info *fs_info = root->fs_info;
2900         struct delayed_iput *delayed;
2901         int empty;
2902
2903         spin_lock(&fs_info->delayed_iput_lock);
2904         empty = list_empty(&fs_info->delayed_iputs);
2905         spin_unlock(&fs_info->delayed_iput_lock);
2906         if (empty)
2907                 return;
2908
2909         spin_lock(&fs_info->delayed_iput_lock);
2910         list_splice_init(&fs_info->delayed_iputs, &list);
2911         spin_unlock(&fs_info->delayed_iput_lock);
2912
2913         while (!list_empty(&list)) {
2914                 delayed = list_entry(list.next, struct delayed_iput, list);
2915                 list_del(&delayed->list);
2916                 iput(delayed->inode);
2917                 kfree(delayed);
2918         }
2919 }
2920
2921 /*
2922  * This is called in transaction commit time. If there are no orphan
2923  * files in the subvolume, it removes orphan item and frees block_rsv
2924  * structure.
2925  */
2926 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2927                               struct btrfs_root *root)
2928 {
2929         struct btrfs_block_rsv *block_rsv;
2930         int ret;
2931
2932         if (atomic_read(&root->orphan_inodes) ||
2933             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2934                 return;
2935
2936         spin_lock(&root->orphan_lock);
2937         if (atomic_read(&root->orphan_inodes)) {
2938                 spin_unlock(&root->orphan_lock);
2939                 return;
2940         }
2941
2942         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2943                 spin_unlock(&root->orphan_lock);
2944                 return;
2945         }
2946
2947         block_rsv = root->orphan_block_rsv;
2948         root->orphan_block_rsv = NULL;
2949         spin_unlock(&root->orphan_lock);
2950
2951         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
2952             btrfs_root_refs(&root->root_item) > 0) {
2953                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2954                                             root->root_key.objectid);
2955                 if (ret)
2956                         btrfs_abort_transaction(trans, root, ret);
2957                 else
2958                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
2959                                   &root->state);
2960         }
2961
2962         if (block_rsv) {
2963                 WARN_ON(block_rsv->size > 0);
2964                 btrfs_free_block_rsv(root, block_rsv);
2965         }
2966 }
2967
2968 /*
2969  * This creates an orphan entry for the given inode in case something goes
2970  * wrong in the middle of an unlink/truncate.
2971  *
2972  * NOTE: caller of this function should reserve 5 units of metadata for
2973  *       this function.
2974  */
2975 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2976 {
2977         struct btrfs_root *root = BTRFS_I(inode)->root;
2978         struct btrfs_block_rsv *block_rsv = NULL;
2979         int reserve = 0;
2980         int insert = 0;
2981         int ret;
2982
2983         if (!root->orphan_block_rsv) {
2984                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2985                 if (!block_rsv)
2986                         return -ENOMEM;
2987         }
2988
2989         spin_lock(&root->orphan_lock);
2990         if (!root->orphan_block_rsv) {
2991                 root->orphan_block_rsv = block_rsv;
2992         } else if (block_rsv) {
2993                 btrfs_free_block_rsv(root, block_rsv);
2994                 block_rsv = NULL;
2995         }
2996
2997         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2998                               &BTRFS_I(inode)->runtime_flags)) {
2999 #if 0
3000                 /*
3001                  * For proper ENOSPC handling, we should do orphan
3002                  * cleanup when mounting. But this introduces backward
3003                  * compatibility issue.
3004                  */
3005                 if (!xchg(&root->orphan_item_inserted, 1))
3006                         insert = 2;
3007                 else
3008                         insert = 1;
3009 #endif
3010                 insert = 1;
3011                 atomic_inc(&root->orphan_inodes);
3012         }
3013
3014         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3015                               &BTRFS_I(inode)->runtime_flags))
3016                 reserve = 1;
3017         spin_unlock(&root->orphan_lock);
3018
3019         /* grab metadata reservation from transaction handle */
3020         if (reserve) {
3021                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3022                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3023         }
3024
3025         /* insert an orphan item to track this unlinked/truncated file */
3026         if (insert >= 1) {
3027                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3028                 if (ret) {
3029                         atomic_dec(&root->orphan_inodes);
3030                         if (reserve) {
3031                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3032                                           &BTRFS_I(inode)->runtime_flags);
3033                                 btrfs_orphan_release_metadata(inode);
3034                         }
3035                         if (ret != -EEXIST) {
3036                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3037                                           &BTRFS_I(inode)->runtime_flags);
3038                                 btrfs_abort_transaction(trans, root, ret);
3039                                 return ret;
3040                         }
3041                 }
3042                 ret = 0;
3043         }
3044
3045         /* insert an orphan item to track subvolume contains orphan files */
3046         if (insert >= 2) {
3047                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3048                                                root->root_key.objectid);
3049                 if (ret && ret != -EEXIST) {
3050                         btrfs_abort_transaction(trans, root, ret);
3051                         return ret;
3052                 }
3053         }
3054         return 0;
3055 }
3056
3057 /*
3058  * We have done the truncate/delete so we can go ahead and remove the orphan
3059  * item for this particular inode.
3060  */
3061 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3062                             struct inode *inode)
3063 {
3064         struct btrfs_root *root = BTRFS_I(inode)->root;
3065         int delete_item = 0;
3066         int release_rsv = 0;
3067         int ret = 0;
3068
3069         spin_lock(&root->orphan_lock);
3070         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3071                                &BTRFS_I(inode)->runtime_flags))
3072                 delete_item = 1;
3073
3074         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3075                                &BTRFS_I(inode)->runtime_flags))
3076                 release_rsv = 1;
3077         spin_unlock(&root->orphan_lock);
3078
3079         if (delete_item) {
3080                 atomic_dec(&root->orphan_inodes);
3081                 if (trans)
3082                         ret = btrfs_del_orphan_item(trans, root,
3083                                                     btrfs_ino(inode));
3084         }
3085
3086         if (release_rsv)
3087                 btrfs_orphan_release_metadata(inode);
3088
3089         return ret;
3090 }
3091
3092 /*
3093  * this cleans up any orphans that may be left on the list from the last use
3094  * of this root.
3095  */
3096 int btrfs_orphan_cleanup(struct btrfs_root *root)
3097 {
3098         struct btrfs_path *path;
3099         struct extent_buffer *leaf;
3100         struct btrfs_key key, found_key;
3101         struct btrfs_trans_handle *trans;
3102         struct inode *inode;
3103         u64 last_objectid = 0;
3104         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3105
3106         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3107                 return 0;
3108
3109         path = btrfs_alloc_path();
3110         if (!path) {
3111                 ret = -ENOMEM;
3112                 goto out;
3113         }
3114         path->reada = -1;
3115
3116         key.objectid = BTRFS_ORPHAN_OBJECTID;
3117         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3118         key.offset = (u64)-1;
3119
3120         while (1) {
3121                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3122                 if (ret < 0)
3123                         goto out;
3124
3125                 /*
3126                  * if ret == 0 means we found what we were searching for, which
3127                  * is weird, but possible, so only screw with path if we didn't
3128                  * find the key and see if we have stuff that matches
3129                  */
3130                 if (ret > 0) {
3131                         ret = 0;
3132                         if (path->slots[0] == 0)
3133                                 break;
3134                         path->slots[0]--;
3135                 }
3136
3137                 /* pull out the item */
3138                 leaf = path->nodes[0];
3139                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3140
3141                 /* make sure the item matches what we want */
3142                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3143                         break;
3144                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3145                         break;
3146
3147                 /* release the path since we're done with it */
3148                 btrfs_release_path(path);
3149
3150                 /*
3151                  * this is where we are basically btrfs_lookup, without the
3152                  * crossing root thing.  we store the inode number in the
3153                  * offset of the orphan item.
3154                  */
3155
3156                 if (found_key.offset == last_objectid) {
3157                         btrfs_err(root->fs_info,
3158                                 "Error removing orphan entry, stopping orphan cleanup");
3159                         ret = -EINVAL;
3160                         goto out;
3161                 }
3162
3163                 last_objectid = found_key.offset;
3164
3165                 found_key.objectid = found_key.offset;
3166                 found_key.type = BTRFS_INODE_ITEM_KEY;
3167                 found_key.offset = 0;
3168                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3169                 ret = PTR_ERR_OR_ZERO(inode);
3170                 if (ret && ret != -ESTALE)
3171                         goto out;
3172
3173                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3174                         struct btrfs_root *dead_root;
3175                         struct btrfs_fs_info *fs_info = root->fs_info;
3176                         int is_dead_root = 0;
3177
3178                         /*
3179                          * this is an orphan in the tree root. Currently these
3180                          * could come from 2 sources:
3181                          *  a) a snapshot deletion in progress
3182                          *  b) a free space cache inode
3183                          * We need to distinguish those two, as the snapshot
3184                          * orphan must not get deleted.
3185                          * find_dead_roots already ran before us, so if this
3186                          * is a snapshot deletion, we should find the root
3187                          * in the dead_roots list
3188                          */
3189                         spin_lock(&fs_info->trans_lock);
3190                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3191                                             root_list) {
3192                                 if (dead_root->root_key.objectid ==
3193                                     found_key.objectid) {
3194                                         is_dead_root = 1;
3195                                         break;
3196                                 }
3197                         }
3198                         spin_unlock(&fs_info->trans_lock);
3199                         if (is_dead_root) {
3200                                 /* prevent this orphan from being found again */
3201                                 key.offset = found_key.objectid - 1;
3202                                 continue;
3203                         }
3204                 }
3205                 /*
3206                  * Inode is already gone but the orphan item is still there,
3207                  * kill the orphan item.
3208                  */
3209                 if (ret == -ESTALE) {
3210                         trans = btrfs_start_transaction(root, 1);
3211                         if (IS_ERR(trans)) {
3212                                 ret = PTR_ERR(trans);
3213                                 goto out;
3214                         }
3215                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3216                                 found_key.objectid);
3217                         ret = btrfs_del_orphan_item(trans, root,
3218                                                     found_key.objectid);
3219                         btrfs_end_transaction(trans, root);
3220                         if (ret)
3221                                 goto out;
3222                         continue;
3223                 }
3224
3225                 /*
3226                  * add this inode to the orphan list so btrfs_orphan_del does
3227                  * the proper thing when we hit it
3228                  */
3229                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3230                         &BTRFS_I(inode)->runtime_flags);
3231                 atomic_inc(&root->orphan_inodes);
3232
3233                 /* if we have links, this was a truncate, lets do that */
3234                 if (inode->i_nlink) {
3235                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3236                                 iput(inode);
3237                                 continue;
3238                         }
3239                         nr_truncate++;
3240
3241                         /* 1 for the orphan item deletion. */
3242                         trans = btrfs_start_transaction(root, 1);
3243                         if (IS_ERR(trans)) {
3244                                 iput(inode);
3245                                 ret = PTR_ERR(trans);
3246                                 goto out;
3247                         }
3248                         ret = btrfs_orphan_add(trans, inode);
3249                         btrfs_end_transaction(trans, root);
3250                         if (ret) {
3251                                 iput(inode);
3252                                 goto out;
3253                         }
3254
3255                         ret = btrfs_truncate(inode);
3256                         if (ret)
3257                                 btrfs_orphan_del(NULL, inode);
3258                 } else {
3259                         nr_unlink++;
3260                 }
3261
3262                 /* this will do delete_inode and everything for us */
3263                 iput(inode);
3264                 if (ret)
3265                         goto out;
3266         }
3267         /* release the path since we're done with it */
3268         btrfs_release_path(path);
3269
3270         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3271
3272         if (root->orphan_block_rsv)
3273                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3274                                         (u64)-1);
3275
3276         if (root->orphan_block_rsv ||
3277             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3278                 trans = btrfs_join_transaction(root);
3279                 if (!IS_ERR(trans))
3280                         btrfs_end_transaction(trans, root);
3281         }
3282
3283         if (nr_unlink)
3284                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3285         if (nr_truncate)
3286                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3287
3288 out:
3289         if (ret)
3290                 btrfs_crit(root->fs_info,
3291                         "could not do orphan cleanup %d", ret);
3292         btrfs_free_path(path);
3293         return ret;
3294 }
3295
3296 /*
3297  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3298  * don't find any xattrs, we know there can't be any acls.
3299  *
3300  * slot is the slot the inode is in, objectid is the objectid of the inode
3301  */
3302 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3303                                           int slot, u64 objectid,
3304                                           int *first_xattr_slot)
3305 {
3306         u32 nritems = btrfs_header_nritems(leaf);
3307         struct btrfs_key found_key;
3308         static u64 xattr_access = 0;
3309         static u64 xattr_default = 0;
3310         int scanned = 0;
3311
3312         if (!xattr_access) {
3313                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3314                                         strlen(POSIX_ACL_XATTR_ACCESS));
3315                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3316                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3317         }
3318
3319         slot++;
3320         *first_xattr_slot = -1;
3321         while (slot < nritems) {
3322                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3323
3324                 /* we found a different objectid, there must not be acls */
3325                 if (found_key.objectid != objectid)
3326                         return 0;
3327
3328                 /* we found an xattr, assume we've got an acl */
3329                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3330                         if (*first_xattr_slot == -1)
3331                                 *first_xattr_slot = slot;
3332                         if (found_key.offset == xattr_access ||
3333                             found_key.offset == xattr_default)
3334                                 return 1;
3335                 }
3336
3337                 /*
3338                  * we found a key greater than an xattr key, there can't
3339                  * be any acls later on
3340                  */
3341                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3342                         return 0;
3343
3344                 slot++;
3345                 scanned++;
3346
3347                 /*
3348                  * it goes inode, inode backrefs, xattrs, extents,
3349                  * so if there are a ton of hard links to an inode there can
3350                  * be a lot of backrefs.  Don't waste time searching too hard,
3351                  * this is just an optimization
3352                  */
3353                 if (scanned >= 8)
3354                         break;
3355         }
3356         /* we hit the end of the leaf before we found an xattr or
3357          * something larger than an xattr.  We have to assume the inode
3358          * has acls
3359          */
3360         if (*first_xattr_slot == -1)
3361                 *first_xattr_slot = slot;
3362         return 1;
3363 }
3364
3365 /*
3366  * read an inode from the btree into the in-memory inode
3367  */
3368 static void btrfs_read_locked_inode(struct inode *inode)
3369 {
3370         struct btrfs_path *path;
3371         struct extent_buffer *leaf;
3372         struct btrfs_inode_item *inode_item;
3373         struct btrfs_timespec *tspec;
3374         struct btrfs_root *root = BTRFS_I(inode)->root;
3375         struct btrfs_key location;
3376         unsigned long ptr;
3377         int maybe_acls;
3378         u32 rdev;
3379         int ret;
3380         bool filled = false;
3381         int first_xattr_slot;
3382
3383         ret = btrfs_fill_inode(inode, &rdev);
3384         if (!ret)
3385                 filled = true;
3386
3387         path = btrfs_alloc_path();
3388         if (!path)
3389                 goto make_bad;
3390
3391         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3392
3393         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3394         if (ret)
3395                 goto make_bad;
3396
3397         leaf = path->nodes[0];
3398
3399         if (filled)
3400                 goto cache_index;
3401
3402         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3403                                     struct btrfs_inode_item);
3404         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3405         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3406         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3407         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3408         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3409
3410         tspec = btrfs_inode_atime(inode_item);
3411         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3412         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3413
3414         tspec = btrfs_inode_mtime(inode_item);
3415         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3416         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3417
3418         tspec = btrfs_inode_ctime(inode_item);
3419         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3420         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3421
3422         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3423         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3424         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3425
3426         /*
3427          * If we were modified in the current generation and evicted from memory
3428          * and then re-read we need to do a full sync since we don't have any
3429          * idea about which extents were modified before we were evicted from
3430          * cache.
3431          */
3432         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3433                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3434                         &BTRFS_I(inode)->runtime_flags);
3435
3436         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3437         inode->i_generation = BTRFS_I(inode)->generation;
3438         inode->i_rdev = 0;
3439         rdev = btrfs_inode_rdev(leaf, inode_item);
3440
3441         BTRFS_I(inode)->index_cnt = (u64)-1;
3442         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3443
3444 cache_index:
3445         path->slots[0]++;
3446         if (inode->i_nlink != 1 ||
3447             path->slots[0] >= btrfs_header_nritems(leaf))
3448                 goto cache_acl;
3449
3450         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3451         if (location.objectid != btrfs_ino(inode))
3452                 goto cache_acl;
3453
3454         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3455         if (location.type == BTRFS_INODE_REF_KEY) {
3456                 struct btrfs_inode_ref *ref;
3457
3458                 ref = (struct btrfs_inode_ref *)ptr;
3459                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3460         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3461                 struct btrfs_inode_extref *extref;
3462
3463                 extref = (struct btrfs_inode_extref *)ptr;
3464                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3465                                                                      extref);
3466         }
3467 cache_acl:
3468         /*
3469          * try to precache a NULL acl entry for files that don't have
3470          * any xattrs or acls
3471          */
3472         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3473                                            btrfs_ino(inode), &first_xattr_slot);
3474         if (first_xattr_slot != -1) {
3475                 path->slots[0] = first_xattr_slot;
3476                 ret = btrfs_load_inode_props(inode, path);
3477                 if (ret)
3478                         btrfs_err(root->fs_info,
3479                                   "error loading props for ino %llu (root %llu): %d",
3480                                   btrfs_ino(inode),
3481                                   root->root_key.objectid, ret);
3482         }
3483         btrfs_free_path(path);
3484
3485         if (!maybe_acls)
3486                 cache_no_acl(inode);
3487
3488         switch (inode->i_mode & S_IFMT) {
3489         case S_IFREG:
3490                 inode->i_mapping->a_ops = &btrfs_aops;
3491                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3492                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3493                 inode->i_fop = &btrfs_file_operations;
3494                 inode->i_op = &btrfs_file_inode_operations;
3495                 break;
3496         case S_IFDIR:
3497                 inode->i_fop = &btrfs_dir_file_operations;
3498                 if (root == root->fs_info->tree_root)
3499                         inode->i_op = &btrfs_dir_ro_inode_operations;
3500                 else
3501                         inode->i_op = &btrfs_dir_inode_operations;
3502                 break;
3503         case S_IFLNK:
3504                 inode->i_op = &btrfs_symlink_inode_operations;
3505                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3506                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3507                 break;
3508         default:
3509                 inode->i_op = &btrfs_special_inode_operations;
3510                 init_special_inode(inode, inode->i_mode, rdev);
3511                 break;
3512         }
3513
3514         btrfs_update_iflags(inode);
3515         return;
3516
3517 make_bad:
3518         btrfs_free_path(path);
3519         make_bad_inode(inode);
3520 }
3521
3522 /*
3523  * given a leaf and an inode, copy the inode fields into the leaf
3524  */
3525 static void fill_inode_item(struct btrfs_trans_handle *trans,
3526                             struct extent_buffer *leaf,
3527                             struct btrfs_inode_item *item,
3528                             struct inode *inode)
3529 {
3530         struct btrfs_map_token token;
3531
3532         btrfs_init_map_token(&token);
3533
3534         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3535         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3536         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3537                                    &token);
3538         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3539         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3540
3541         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3542                                      inode->i_atime.tv_sec, &token);
3543         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3544                                       inode->i_atime.tv_nsec, &token);
3545
3546         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3547                                      inode->i_mtime.tv_sec, &token);
3548         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3549                                       inode->i_mtime.tv_nsec, &token);
3550
3551         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3552                                      inode->i_ctime.tv_sec, &token);
3553         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3554                                       inode->i_ctime.tv_nsec, &token);
3555
3556         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3557                                      &token);
3558         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3559                                          &token);
3560         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3561         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3562         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3563         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3564         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3565 }
3566
3567 /*
3568  * copy everything in the in-memory inode into the btree.
3569  */
3570 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3571                                 struct btrfs_root *root, struct inode *inode)
3572 {
3573         struct btrfs_inode_item *inode_item;
3574         struct btrfs_path *path;
3575         struct extent_buffer *leaf;
3576         int ret;
3577
3578         path = btrfs_alloc_path();
3579         if (!path)
3580                 return -ENOMEM;
3581
3582         path->leave_spinning = 1;
3583         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3584                                  1);
3585         if (ret) {
3586                 if (ret > 0)
3587                         ret = -ENOENT;
3588                 goto failed;
3589         }
3590
3591         leaf = path->nodes[0];
3592         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3593                                     struct btrfs_inode_item);
3594
3595         fill_inode_item(trans, leaf, inode_item, inode);
3596         btrfs_mark_buffer_dirty(leaf);
3597         btrfs_set_inode_last_trans(trans, inode);
3598         ret = 0;
3599 failed:
3600         btrfs_free_path(path);
3601         return ret;
3602 }
3603
3604 /*
3605  * copy everything in the in-memory inode into the btree.
3606  */
3607 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3608                                 struct btrfs_root *root, struct inode *inode)
3609 {
3610         int ret;
3611
3612         /*
3613          * If the inode is a free space inode, we can deadlock during commit
3614          * if we put it into the delayed code.
3615          *
3616          * The data relocation inode should also be directly updated
3617          * without delay
3618          */
3619         if (!btrfs_is_free_space_inode(inode)
3620             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3621                 btrfs_update_root_times(trans, root);
3622
3623                 ret = btrfs_delayed_update_inode(trans, root, inode);
3624                 if (!ret)
3625                         btrfs_set_inode_last_trans(trans, inode);
3626                 return ret;
3627         }
3628
3629         return btrfs_update_inode_item(trans, root, inode);
3630 }
3631
3632 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3633                                          struct btrfs_root *root,
3634                                          struct inode *inode)
3635 {
3636         int ret;
3637
3638         ret = btrfs_update_inode(trans, root, inode);
3639         if (ret == -ENOSPC)
3640                 return btrfs_update_inode_item(trans, root, inode);
3641         return ret;
3642 }
3643
3644 /*
3645  * unlink helper that gets used here in inode.c and in the tree logging
3646  * recovery code.  It remove a link in a directory with a given name, and
3647  * also drops the back refs in the inode to the directory
3648  */
3649 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3650                                 struct btrfs_root *root,
3651                                 struct inode *dir, struct inode *inode,
3652                                 const char *name, int name_len)
3653 {
3654         struct btrfs_path *path;
3655         int ret = 0;
3656         struct extent_buffer *leaf;
3657         struct btrfs_dir_item *di;
3658         struct btrfs_key key;
3659         u64 index;
3660         u64 ino = btrfs_ino(inode);
3661         u64 dir_ino = btrfs_ino(dir);
3662
3663         path = btrfs_alloc_path();
3664         if (!path) {
3665                 ret = -ENOMEM;
3666                 goto out;
3667         }
3668
3669         path->leave_spinning = 1;
3670         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3671                                     name, name_len, -1);
3672         if (IS_ERR(di)) {
3673                 ret = PTR_ERR(di);
3674                 goto err;
3675         }
3676         if (!di) {
3677                 ret = -ENOENT;
3678                 goto err;
3679         }
3680         leaf = path->nodes[0];
3681         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3682         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3683         if (ret)
3684                 goto err;
3685         btrfs_release_path(path);
3686
3687         /*
3688          * If we don't have dir index, we have to get it by looking up
3689          * the inode ref, since we get the inode ref, remove it directly,
3690          * it is unnecessary to do delayed deletion.
3691          *
3692          * But if we have dir index, needn't search inode ref to get it.
3693          * Since the inode ref is close to the inode item, it is better
3694          * that we delay to delete it, and just do this deletion when
3695          * we update the inode item.
3696          */
3697         if (BTRFS_I(inode)->dir_index) {
3698                 ret = btrfs_delayed_delete_inode_ref(inode);
3699                 if (!ret) {
3700                         index = BTRFS_I(inode)->dir_index;
3701                         goto skip_backref;
3702                 }
3703         }
3704
3705         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3706                                   dir_ino, &index);
3707         if (ret) {
3708                 btrfs_info(root->fs_info,
3709                         "failed to delete reference to %.*s, inode %llu parent %llu",
3710                         name_len, name, ino, dir_ino);
3711                 btrfs_abort_transaction(trans, root, ret);
3712                 goto err;
3713         }
3714 skip_backref:
3715         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3716         if (ret) {
3717                 btrfs_abort_transaction(trans, root, ret);
3718                 goto err;
3719         }
3720
3721         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3722                                          inode, dir_ino);
3723         if (ret != 0 && ret != -ENOENT) {
3724                 btrfs_abort_transaction(trans, root, ret);
3725                 goto err;
3726         }
3727
3728         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3729                                            dir, index);
3730         if (ret == -ENOENT)
3731                 ret = 0;
3732         else if (ret)
3733                 btrfs_abort_transaction(trans, root, ret);
3734 err:
3735         btrfs_free_path(path);
3736         if (ret)
3737                 goto out;
3738
3739         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3740         inode_inc_iversion(inode);
3741         inode_inc_iversion(dir);
3742         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3743         ret = btrfs_update_inode(trans, root, dir);
3744 out:
3745         return ret;
3746 }
3747
3748 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3749                        struct btrfs_root *root,
3750                        struct inode *dir, struct inode *inode,
3751                        const char *name, int name_len)
3752 {
3753         int ret;
3754         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3755         if (!ret) {
3756                 drop_nlink(inode);
3757                 ret = btrfs_update_inode(trans, root, inode);
3758         }
3759         return ret;
3760 }
3761
3762 /*
3763  * helper to start transaction for unlink and rmdir.
3764  *
3765  * unlink and rmdir are special in btrfs, they do not always free space, so
3766  * if we cannot make our reservations the normal way try and see if there is
3767  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3768  * allow the unlink to occur.
3769  */
3770 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3771 {
3772         struct btrfs_trans_handle *trans;
3773         struct btrfs_root *root = BTRFS_I(dir)->root;
3774         int ret;
3775
3776         /*
3777          * 1 for the possible orphan item
3778          * 1 for the dir item
3779          * 1 for the dir index
3780          * 1 for the inode ref
3781          * 1 for the inode
3782          */
3783         trans = btrfs_start_transaction(root, 5);
3784         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3785                 return trans;
3786
3787         if (PTR_ERR(trans) == -ENOSPC) {
3788                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3789
3790                 trans = btrfs_start_transaction(root, 0);
3791                 if (IS_ERR(trans))
3792                         return trans;
3793                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3794                                                &root->fs_info->trans_block_rsv,
3795                                                num_bytes, 5);
3796                 if (ret) {
3797                         btrfs_end_transaction(trans, root);
3798                         return ERR_PTR(ret);
3799                 }
3800                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3801                 trans->bytes_reserved = num_bytes;
3802         }
3803         return trans;
3804 }
3805
3806 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3807 {
3808         struct btrfs_root *root = BTRFS_I(dir)->root;
3809         struct btrfs_trans_handle *trans;
3810         struct inode *inode = dentry->d_inode;
3811         int ret;
3812
3813         trans = __unlink_start_trans(dir);
3814         if (IS_ERR(trans))
3815                 return PTR_ERR(trans);
3816
3817         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3818
3819         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3820                                  dentry->d_name.name, dentry->d_name.len);
3821         if (ret)
3822                 goto out;
3823
3824         if (inode->i_nlink == 0) {
3825                 ret = btrfs_orphan_add(trans, inode);
3826                 if (ret)
3827                         goto out;
3828         }
3829
3830 out:
3831         btrfs_end_transaction(trans, root);
3832         btrfs_btree_balance_dirty(root);
3833         return ret;
3834 }
3835
3836 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3837                         struct btrfs_root *root,
3838                         struct inode *dir, u64 objectid,
3839                         const char *name, int name_len)
3840 {
3841         struct btrfs_path *path;
3842         struct extent_buffer *leaf;
3843         struct btrfs_dir_item *di;
3844         struct btrfs_key key;
3845         u64 index;
3846         int ret;
3847         u64 dir_ino = btrfs_ino(dir);
3848
3849         path = btrfs_alloc_path();
3850         if (!path)
3851                 return -ENOMEM;
3852
3853         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3854                                    name, name_len, -1);
3855         if (IS_ERR_OR_NULL(di)) {
3856                 if (!di)
3857                         ret = -ENOENT;
3858                 else
3859                         ret = PTR_ERR(di);
3860                 goto out;
3861         }
3862
3863         leaf = path->nodes[0];
3864         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3865         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3866         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3867         if (ret) {
3868                 btrfs_abort_transaction(trans, root, ret);
3869                 goto out;
3870         }
3871         btrfs_release_path(path);
3872
3873         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3874                                  objectid, root->root_key.objectid,
3875                                  dir_ino, &index, name, name_len);
3876         if (ret < 0) {
3877                 if (ret != -ENOENT) {
3878                         btrfs_abort_transaction(trans, root, ret);
3879                         goto out;
3880                 }
3881                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3882                                                  name, name_len);
3883                 if (IS_ERR_OR_NULL(di)) {
3884                         if (!di)
3885                                 ret = -ENOENT;
3886                         else
3887                                 ret = PTR_ERR(di);
3888                         btrfs_abort_transaction(trans, root, ret);
3889                         goto out;
3890                 }
3891
3892                 leaf = path->nodes[0];
3893                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3894                 btrfs_release_path(path);
3895                 index = key.offset;
3896         }
3897         btrfs_release_path(path);
3898
3899         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3900         if (ret) {
3901                 btrfs_abort_transaction(trans, root, ret);
3902                 goto out;
3903         }
3904
3905         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3906         inode_inc_iversion(dir);
3907         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3908         ret = btrfs_update_inode_fallback(trans, root, dir);
3909         if (ret)
3910                 btrfs_abort_transaction(trans, root, ret);
3911 out:
3912         btrfs_free_path(path);
3913         return ret;
3914 }
3915
3916 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3917 {
3918         struct inode *inode = dentry->d_inode;
3919         int err = 0;
3920         struct btrfs_root *root = BTRFS_I(dir)->root;
3921         struct btrfs_trans_handle *trans;
3922
3923         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3924                 return -ENOTEMPTY;
3925         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3926                 return -EPERM;
3927
3928         trans = __unlink_start_trans(dir);
3929         if (IS_ERR(trans))
3930                 return PTR_ERR(trans);
3931
3932         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3933                 err = btrfs_unlink_subvol(trans, root, dir,
3934                                           BTRFS_I(inode)->location.objectid,
3935                                           dentry->d_name.name,
3936                                           dentry->d_name.len);
3937                 goto out;
3938         }
3939
3940         err = btrfs_orphan_add(trans, inode);
3941         if (err)
3942                 goto out;
3943
3944         /* now the directory is empty */
3945         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3946                                  dentry->d_name.name, dentry->d_name.len);
3947         if (!err)
3948                 btrfs_i_size_write(inode, 0);
3949 out:
3950         btrfs_end_transaction(trans, root);
3951         btrfs_btree_balance_dirty(root);
3952
3953         return err;
3954 }
3955
3956 /*
3957  * this can truncate away extent items, csum items and directory items.
3958  * It starts at a high offset and removes keys until it can't find
3959  * any higher than new_size
3960  *
3961  * csum items that cross the new i_size are truncated to the new size
3962  * as well.
3963  *
3964  * min_type is the minimum key type to truncate down to.  If set to 0, this
3965  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3966  */
3967 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3968                                struct btrfs_root *root,
3969                                struct inode *inode,
3970                                u64 new_size, u32 min_type)
3971 {
3972         struct btrfs_path *path;
3973         struct extent_buffer *leaf;
3974         struct btrfs_file_extent_item *fi;
3975         struct btrfs_key key;
3976         struct btrfs_key found_key;
3977         u64 extent_start = 0;
3978         u64 extent_num_bytes = 0;
3979         u64 extent_offset = 0;
3980         u64 item_end = 0;
3981         u64 last_size = (u64)-1;
3982         u32 found_type = (u8)-1;
3983         int found_extent;
3984         int del_item;
3985         int pending_del_nr = 0;
3986         int pending_del_slot = 0;
3987         int extent_type = -1;
3988         int ret;
3989         int err = 0;
3990         u64 ino = btrfs_ino(inode);
3991
3992         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3993
3994         path = btrfs_alloc_path();
3995         if (!path)
3996                 return -ENOMEM;
3997         path->reada = -1;
3998
3999         /*
4000          * We want to drop from the next block forward in case this new size is
4001          * not block aligned since we will be keeping the last block of the
4002          * extent just the way it is.
4003          */
4004         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4005             root == root->fs_info->tree_root)
4006                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4007                                         root->sectorsize), (u64)-1, 0);
4008
4009         /*
4010          * This function is also used to drop the items in the log tree before
4011          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4012          * it is used to drop the loged items. So we shouldn't kill the delayed
4013          * items.
4014          */
4015         if (min_type == 0 && root == BTRFS_I(inode)->root)
4016                 btrfs_kill_delayed_inode_items(inode);
4017
4018         key.objectid = ino;
4019         key.offset = (u64)-1;
4020         key.type = (u8)-1;
4021
4022 search_again:
4023         path->leave_spinning = 1;
4024         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4025         if (ret < 0) {
4026                 err = ret;
4027                 goto out;
4028         }
4029
4030         if (ret > 0) {
4031                 /* there are no items in the tree for us to truncate, we're
4032                  * done
4033                  */
4034                 if (path->slots[0] == 0)
4035                         goto out;
4036                 path->slots[0]--;
4037         }
4038
4039         while (1) {
4040                 fi = NULL;
4041                 leaf = path->nodes[0];
4042                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4043                 found_type = btrfs_key_type(&found_key);
4044
4045                 if (found_key.objectid != ino)
4046                         break;
4047
4048                 if (found_type < min_type)
4049                         break;
4050
4051                 item_end = found_key.offset;
4052                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4053                         fi = btrfs_item_ptr(leaf, path->slots[0],
4054                                             struct btrfs_file_extent_item);
4055                         extent_type = btrfs_file_extent_type(leaf, fi);
4056                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4057                                 item_end +=
4058                                     btrfs_file_extent_num_bytes(leaf, fi);
4059                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4060                                 item_end += btrfs_file_extent_inline_len(leaf,
4061                                                          path->slots[0], fi);
4062                         }
4063                         item_end--;
4064                 }
4065                 if (found_type > min_type) {
4066                         del_item = 1;
4067                 } else {
4068                         if (item_end < new_size)
4069                                 break;
4070                         if (found_key.offset >= new_size)
4071                                 del_item = 1;
4072                         else
4073                                 del_item = 0;
4074                 }
4075                 found_extent = 0;
4076                 /* FIXME, shrink the extent if the ref count is only 1 */
4077                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4078                         goto delete;
4079
4080                 if (del_item)
4081                         last_size = found_key.offset;
4082                 else
4083                         last_size = new_size;
4084
4085                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4086                         u64 num_dec;
4087                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4088                         if (!del_item) {
4089                                 u64 orig_num_bytes =
4090                                         btrfs_file_extent_num_bytes(leaf, fi);
4091                                 extent_num_bytes = ALIGN(new_size -
4092                                                 found_key.offset,
4093                                                 root->sectorsize);
4094                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4095                                                          extent_num_bytes);
4096                                 num_dec = (orig_num_bytes -
4097                                            extent_num_bytes);
4098                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4099                                              &root->state) &&
4100                                     extent_start != 0)
4101                                         inode_sub_bytes(inode, num_dec);
4102                                 btrfs_mark_buffer_dirty(leaf);
4103                         } else {
4104                                 extent_num_bytes =
4105                                         btrfs_file_extent_disk_num_bytes(leaf,
4106                                                                          fi);
4107                                 extent_offset = found_key.offset -
4108                                         btrfs_file_extent_offset(leaf, fi);
4109
4110                                 /* FIXME blocksize != 4096 */
4111                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4112                                 if (extent_start != 0) {
4113                                         found_extent = 1;
4114                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4115                                                      &root->state))
4116                                                 inode_sub_bytes(inode, num_dec);
4117                                 }
4118                         }
4119                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4120                         /*
4121                          * we can't truncate inline items that have had
4122                          * special encodings
4123                          */
4124                         if (!del_item &&
4125                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4126                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4127                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4128                                 u32 size = new_size - found_key.offset;
4129
4130                                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4131                                         inode_sub_bytes(inode, item_end + 1 -
4132                                                         new_size);
4133
4134                                 /*
4135                                  * update the ram bytes to properly reflect
4136                                  * the new size of our item
4137                                  */
4138                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4139                                 size =
4140                                     btrfs_file_extent_calc_inline_size(size);
4141                                 btrfs_truncate_item(root, path, size, 1);
4142                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4143                                             &root->state)) {
4144                                 inode_sub_bytes(inode, item_end + 1 -
4145                                                 found_key.offset);
4146                         }
4147                 }
4148 delete:
4149                 if (del_item) {
4150                         if (!pending_del_nr) {
4151                                 /* no pending yet, add ourselves */
4152                                 pending_del_slot = path->slots[0];
4153                                 pending_del_nr = 1;
4154                         } else if (pending_del_nr &&
4155                                    path->slots[0] + 1 == pending_del_slot) {
4156                                 /* hop on the pending chunk */
4157                                 pending_del_nr++;
4158                                 pending_del_slot = path->slots[0];
4159                         } else {
4160                                 BUG();
4161                         }
4162                 } else {
4163                         break;
4164                 }
4165                 if (found_extent &&
4166                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4167                      root == root->fs_info->tree_root)) {
4168                         btrfs_set_path_blocking(path);
4169                         ret = btrfs_free_extent(trans, root, extent_start,
4170                                                 extent_num_bytes, 0,
4171                                                 btrfs_header_owner(leaf),
4172                                                 ino, extent_offset, 0);
4173                         BUG_ON(ret);
4174                 }
4175
4176                 if (found_type == BTRFS_INODE_ITEM_KEY)
4177                         break;
4178
4179                 if (path->slots[0] == 0 ||
4180                     path->slots[0] != pending_del_slot) {
4181                         if (pending_del_nr) {
4182                                 ret = btrfs_del_items(trans, root, path,
4183                                                 pending_del_slot,
4184                                                 pending_del_nr);
4185                                 if (ret) {
4186                                         btrfs_abort_transaction(trans,
4187                                                                 root, ret);
4188                                         goto error;
4189                                 }
4190                                 pending_del_nr = 0;
4191                         }
4192                         btrfs_release_path(path);
4193                         goto search_again;
4194                 } else {
4195                         path->slots[0]--;
4196                 }
4197         }
4198 out:
4199         if (pending_del_nr) {
4200                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4201                                       pending_del_nr);
4202                 if (ret)
4203                         btrfs_abort_transaction(trans, root, ret);
4204         }
4205 error:
4206         if (last_size != (u64)-1)
4207                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4208         btrfs_free_path(path);
4209         return err;
4210 }
4211
4212 /*
4213  * btrfs_truncate_page - read, zero a chunk and write a page
4214  * @inode - inode that we're zeroing
4215  * @from - the offset to start zeroing
4216  * @len - the length to zero, 0 to zero the entire range respective to the
4217  *      offset
4218  * @front - zero up to the offset instead of from the offset on
4219  *
4220  * This will find the page for the "from" offset and cow the page and zero the
4221  * part we want to zero.  This is used with truncate and hole punching.
4222  */
4223 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4224                         int front)
4225 {
4226         struct address_space *mapping = inode->i_mapping;
4227         struct btrfs_root *root = BTRFS_I(inode)->root;
4228         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4229         struct btrfs_ordered_extent *ordered;
4230         struct extent_state *cached_state = NULL;
4231         char *kaddr;
4232         u32 blocksize = root->sectorsize;
4233         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4234         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4235         struct page *page;
4236         gfp_t mask = btrfs_alloc_write_mask(mapping);
4237         int ret = 0;
4238         u64 page_start;
4239         u64 page_end;
4240
4241         if ((offset & (blocksize - 1)) == 0 &&
4242             (!len || ((len & (blocksize - 1)) == 0)))
4243                 goto out;
4244         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4245         if (ret)
4246                 goto out;
4247
4248 again:
4249         page = find_or_create_page(mapping, index, mask);
4250         if (!page) {
4251                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4252                 ret = -ENOMEM;
4253                 goto out;
4254         }
4255
4256         page_start = page_offset(page);
4257         page_end = page_start + PAGE_CACHE_SIZE - 1;
4258
4259         if (!PageUptodate(page)) {
4260                 ret = btrfs_readpage(NULL, page);
4261                 lock_page(page);
4262                 if (page->mapping != mapping) {
4263                         unlock_page(page);
4264                         page_cache_release(page);
4265                         goto again;
4266                 }
4267                 if (!PageUptodate(page)) {
4268                         ret = -EIO;
4269                         goto out_unlock;
4270                 }
4271         }
4272         wait_on_page_writeback(page);
4273
4274         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4275         set_page_extent_mapped(page);
4276
4277         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4278         if (ordered) {
4279                 unlock_extent_cached(io_tree, page_start, page_end,
4280                                      &cached_state, GFP_NOFS);
4281                 unlock_page(page);
4282                 page_cache_release(page);
4283                 btrfs_start_ordered_extent(inode, ordered, 1);
4284                 btrfs_put_ordered_extent(ordered);
4285                 goto again;
4286         }
4287
4288         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4289                           EXTENT_DIRTY | EXTENT_DELALLOC |
4290                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4291                           0, 0, &cached_state, GFP_NOFS);
4292
4293         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4294                                         &cached_state);
4295         if (ret) {
4296                 unlock_extent_cached(io_tree, page_start, page_end,
4297                                      &cached_state, GFP_NOFS);
4298                 goto out_unlock;
4299         }
4300
4301         if (offset != PAGE_CACHE_SIZE) {
4302                 if (!len)
4303                         len = PAGE_CACHE_SIZE - offset;
4304                 kaddr = kmap(page);
4305                 if (front)
4306                         memset(kaddr, 0, offset);
4307                 else
4308                         memset(kaddr + offset, 0, len);
4309                 flush_dcache_page(page);
4310                 kunmap(page);
4311         }
4312         ClearPageChecked(page);
4313         set_page_dirty(page);
4314         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4315                              GFP_NOFS);
4316
4317 out_unlock:
4318         if (ret)
4319                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4320         unlock_page(page);
4321         page_cache_release(page);
4322 out:
4323         return ret;
4324 }
4325
4326 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4327                              u64 offset, u64 len)
4328 {
4329         struct btrfs_trans_handle *trans;
4330         int ret;
4331
4332         /*
4333          * Still need to make sure the inode looks like it's been updated so
4334          * that any holes get logged if we fsync.
4335          */
4336         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4337                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4338                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4339                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4340                 return 0;
4341         }
4342
4343         /*
4344          * 1 - for the one we're dropping
4345          * 1 - for the one we're adding
4346          * 1 - for updating the inode.
4347          */
4348         trans = btrfs_start_transaction(root, 3);
4349         if (IS_ERR(trans))
4350                 return PTR_ERR(trans);
4351
4352         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4353         if (ret) {
4354                 btrfs_abort_transaction(trans, root, ret);
4355                 btrfs_end_transaction(trans, root);
4356                 return ret;
4357         }
4358
4359         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4360                                        0, 0, len, 0, len, 0, 0, 0);
4361         if (ret)
4362                 btrfs_abort_transaction(trans, root, ret);
4363         else
4364                 btrfs_update_inode(trans, root, inode);
4365         btrfs_end_transaction(trans, root);
4366         return ret;
4367 }
4368
4369 /*
4370  * This function puts in dummy file extents for the area we're creating a hole
4371  * for.  So if we are truncating this file to a larger size we need to insert
4372  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4373  * the range between oldsize and size
4374  */
4375 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4376 {
4377         struct btrfs_root *root = BTRFS_I(inode)->root;
4378         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4379         struct extent_map *em = NULL;
4380         struct extent_state *cached_state = NULL;
4381         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4382         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4383         u64 block_end = ALIGN(size, root->sectorsize);
4384         u64 last_byte;
4385         u64 cur_offset;
4386         u64 hole_size;
4387         int err = 0;
4388
4389         /*
4390          * If our size started in the middle of a page we need to zero out the
4391          * rest of the page before we expand the i_size, otherwise we could
4392          * expose stale data.
4393          */
4394         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4395         if (err)
4396                 return err;
4397
4398         if (size <= hole_start)
4399                 return 0;
4400
4401         while (1) {
4402                 struct btrfs_ordered_extent *ordered;
4403
4404                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4405                                  &cached_state);
4406                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4407                                                      block_end - hole_start);
4408                 if (!ordered)
4409                         break;
4410                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4411                                      &cached_state, GFP_NOFS);
4412                 btrfs_start_ordered_extent(inode, ordered, 1);
4413                 btrfs_put_ordered_extent(ordered);
4414         }
4415
4416         cur_offset = hole_start;
4417         while (1) {
4418                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4419                                 block_end - cur_offset, 0);
4420                 if (IS_ERR(em)) {
4421                         err = PTR_ERR(em);
4422                         em = NULL;
4423                         break;
4424                 }
4425                 last_byte = min(extent_map_end(em), block_end);
4426                 last_byte = ALIGN(last_byte , root->sectorsize);
4427                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4428                         struct extent_map *hole_em;
4429                         hole_size = last_byte - cur_offset;
4430
4431                         err = maybe_insert_hole(root, inode, cur_offset,
4432                                                 hole_size);
4433                         if (err)
4434                                 break;
4435                         btrfs_drop_extent_cache(inode, cur_offset,
4436                                                 cur_offset + hole_size - 1, 0);
4437                         hole_em = alloc_extent_map();
4438                         if (!hole_em) {
4439                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4440                                         &BTRFS_I(inode)->runtime_flags);
4441                                 goto next;
4442                         }
4443                         hole_em->start = cur_offset;
4444                         hole_em->len = hole_size;
4445                         hole_em->orig_start = cur_offset;
4446
4447                         hole_em->block_start = EXTENT_MAP_HOLE;
4448                         hole_em->block_len = 0;
4449                         hole_em->orig_block_len = 0;
4450                         hole_em->ram_bytes = hole_size;
4451                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4452                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4453                         hole_em->generation = root->fs_info->generation;
4454
4455                         while (1) {
4456                                 write_lock(&em_tree->lock);
4457                                 err = add_extent_mapping(em_tree, hole_em, 1);
4458                                 write_unlock(&em_tree->lock);
4459                                 if (err != -EEXIST)
4460                                         break;
4461                                 btrfs_drop_extent_cache(inode, cur_offset,
4462                                                         cur_offset +
4463                                                         hole_size - 1, 0);
4464                         }
4465                         free_extent_map(hole_em);
4466                 }
4467 next:
4468                 free_extent_map(em);
4469                 em = NULL;
4470                 cur_offset = last_byte;
4471                 if (cur_offset >= block_end)
4472                         break;
4473         }
4474         free_extent_map(em);
4475         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4476                              GFP_NOFS);
4477         return err;
4478 }
4479
4480 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4481 {
4482         struct btrfs_root *root = BTRFS_I(inode)->root;
4483         struct btrfs_trans_handle *trans;
4484         loff_t oldsize = i_size_read(inode);
4485         loff_t newsize = attr->ia_size;
4486         int mask = attr->ia_valid;
4487         int ret;
4488
4489         /*
4490          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4491          * special case where we need to update the times despite not having
4492          * these flags set.  For all other operations the VFS set these flags
4493          * explicitly if it wants a timestamp update.
4494          */
4495         if (newsize != oldsize) {
4496                 inode_inc_iversion(inode);
4497                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4498                         inode->i_ctime = inode->i_mtime =
4499                                 current_fs_time(inode->i_sb);
4500         }
4501
4502         if (newsize > oldsize) {
4503                 truncate_pagecache(inode, newsize);
4504                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4505                 if (ret)
4506                         return ret;
4507
4508                 trans = btrfs_start_transaction(root, 1);
4509                 if (IS_ERR(trans))
4510                         return PTR_ERR(trans);
4511
4512                 i_size_write(inode, newsize);
4513                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4514                 ret = btrfs_update_inode(trans, root, inode);
4515                 btrfs_end_transaction(trans, root);
4516         } else {
4517
4518                 /*
4519                  * We're truncating a file that used to have good data down to
4520                  * zero. Make sure it gets into the ordered flush list so that
4521                  * any new writes get down to disk quickly.
4522                  */
4523                 if (newsize == 0)
4524                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4525                                 &BTRFS_I(inode)->runtime_flags);
4526
4527                 /*
4528                  * 1 for the orphan item we're going to add
4529                  * 1 for the orphan item deletion.
4530                  */
4531                 trans = btrfs_start_transaction(root, 2);
4532                 if (IS_ERR(trans))
4533                         return PTR_ERR(trans);
4534
4535                 /*
4536                  * We need to do this in case we fail at _any_ point during the
4537                  * actual truncate.  Once we do the truncate_setsize we could
4538                  * invalidate pages which forces any outstanding ordered io to
4539                  * be instantly completed which will give us extents that need
4540                  * to be truncated.  If we fail to get an orphan inode down we
4541                  * could have left over extents that were never meant to live,
4542                  * so we need to garuntee from this point on that everything
4543                  * will be consistent.
4544                  */
4545                 ret = btrfs_orphan_add(trans, inode);
4546                 btrfs_end_transaction(trans, root);
4547                 if (ret)
4548                         return ret;
4549
4550                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4551                 truncate_setsize(inode, newsize);
4552
4553                 /* Disable nonlocked read DIO to avoid the end less truncate */
4554                 btrfs_inode_block_unlocked_dio(inode);
4555                 inode_dio_wait(inode);
4556                 btrfs_inode_resume_unlocked_dio(inode);
4557
4558                 ret = btrfs_truncate(inode);
4559                 if (ret && inode->i_nlink) {
4560                         int err;
4561
4562                         /*
4563                          * failed to truncate, disk_i_size is only adjusted down
4564                          * as we remove extents, so it should represent the true
4565                          * size of the inode, so reset the in memory size and
4566                          * delete our orphan entry.
4567                          */
4568                         trans = btrfs_join_transaction(root);
4569                         if (IS_ERR(trans)) {
4570                                 btrfs_orphan_del(NULL, inode);
4571                                 return ret;
4572                         }
4573                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4574                         err = btrfs_orphan_del(trans, inode);
4575                         if (err)
4576                                 btrfs_abort_transaction(trans, root, err);
4577                         btrfs_end_transaction(trans, root);
4578                 }
4579         }
4580
4581         return ret;
4582 }
4583
4584 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4585 {
4586         struct inode *inode = dentry->d_inode;
4587         struct btrfs_root *root = BTRFS_I(inode)->root;
4588         int err;
4589
4590         if (btrfs_root_readonly(root))
4591                 return -EROFS;
4592
4593         err = inode_change_ok(inode, attr);
4594         if (err)
4595                 return err;
4596
4597         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4598                 err = btrfs_setsize(inode, attr);
4599                 if (err)
4600                         return err;
4601         }
4602
4603         if (attr->ia_valid) {
4604                 setattr_copy(inode, attr);
4605                 inode_inc_iversion(inode);
4606                 err = btrfs_dirty_inode(inode);
4607
4608                 if (!err && attr->ia_valid & ATTR_MODE)
4609                         err = posix_acl_chmod(inode, inode->i_mode);
4610         }
4611
4612         return err;
4613 }
4614
4615 /*
4616  * While truncating the inode pages during eviction, we get the VFS calling
4617  * btrfs_invalidatepage() against each page of the inode. This is slow because
4618  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4619  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4620  * extent_state structures over and over, wasting lots of time.
4621  *
4622  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4623  * those expensive operations on a per page basis and do only the ordered io
4624  * finishing, while we release here the extent_map and extent_state structures,
4625  * without the excessive merging and splitting.
4626  */
4627 static void evict_inode_truncate_pages(struct inode *inode)
4628 {
4629         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4630         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4631         struct rb_node *node;
4632
4633         ASSERT(inode->i_state & I_FREEING);
4634         truncate_inode_pages_final(&inode->i_data);
4635
4636         write_lock(&map_tree->lock);
4637         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4638                 struct extent_map *em;
4639
4640                 node = rb_first(&map_tree->map);
4641                 em = rb_entry(node, struct extent_map, rb_node);
4642                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4643                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4644                 remove_extent_mapping(map_tree, em);
4645                 free_extent_map(em);
4646         }
4647         write_unlock(&map_tree->lock);
4648
4649         spin_lock(&io_tree->lock);
4650         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4651                 struct extent_state *state;
4652                 struct extent_state *cached_state = NULL;
4653
4654                 node = rb_first(&io_tree->state);
4655                 state = rb_entry(node, struct extent_state, rb_node);
4656                 atomic_inc(&state->refs);
4657                 spin_unlock(&io_tree->lock);
4658
4659                 lock_extent_bits(io_tree, state->start, state->end,
4660                                  0, &cached_state);
4661                 clear_extent_bit(io_tree, state->start, state->end,
4662                                  EXTENT_LOCKED | EXTENT_DIRTY |
4663                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4664                                  EXTENT_DEFRAG, 1, 1,
4665                                  &cached_state, GFP_NOFS);
4666                 free_extent_state(state);
4667
4668                 spin_lock(&io_tree->lock);
4669         }
4670         spin_unlock(&io_tree->lock);
4671 }
4672
4673 void btrfs_evict_inode(struct inode *inode)
4674 {
4675         struct btrfs_trans_handle *trans;
4676         struct btrfs_root *root = BTRFS_I(inode)->root;
4677         struct btrfs_block_rsv *rsv, *global_rsv;
4678         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4679         int ret;
4680
4681         trace_btrfs_inode_evict(inode);
4682
4683         evict_inode_truncate_pages(inode);
4684
4685         if (inode->i_nlink &&
4686             ((btrfs_root_refs(&root->root_item) != 0 &&
4687               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4688              btrfs_is_free_space_inode(inode)))
4689                 goto no_delete;
4690
4691         if (is_bad_inode(inode)) {
4692                 btrfs_orphan_del(NULL, inode);
4693                 goto no_delete;
4694         }
4695         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4696         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4697
4698         if (root->fs_info->log_root_recovering) {
4699                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4700                                  &BTRFS_I(inode)->runtime_flags));
4701                 goto no_delete;
4702         }
4703
4704         if (inode->i_nlink > 0) {
4705                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4706                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4707                 goto no_delete;
4708         }
4709
4710         ret = btrfs_commit_inode_delayed_inode(inode);
4711         if (ret) {
4712                 btrfs_orphan_del(NULL, inode);
4713                 goto no_delete;
4714         }
4715
4716         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4717         if (!rsv) {
4718                 btrfs_orphan_del(NULL, inode);
4719                 goto no_delete;
4720         }
4721         rsv->size = min_size;
4722         rsv->failfast = 1;
4723         global_rsv = &root->fs_info->global_block_rsv;
4724
4725         btrfs_i_size_write(inode, 0);
4726
4727         /*
4728          * This is a bit simpler than btrfs_truncate since we've already
4729          * reserved our space for our orphan item in the unlink, so we just
4730          * need to reserve some slack space in case we add bytes and update
4731          * inode item when doing the truncate.
4732          */
4733         while (1) {
4734                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4735                                              BTRFS_RESERVE_FLUSH_LIMIT);
4736
4737                 /*
4738                  * Try and steal from the global reserve since we will
4739                  * likely not use this space anyway, we want to try as
4740                  * hard as possible to get this to work.
4741                  */
4742                 if (ret)
4743                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4744
4745                 if (ret) {
4746                         btrfs_warn(root->fs_info,
4747                                 "Could not get space for a delete, will truncate on mount %d",
4748                                 ret);
4749                         btrfs_orphan_del(NULL, inode);
4750                         btrfs_free_block_rsv(root, rsv);
4751                         goto no_delete;
4752                 }
4753
4754                 trans = btrfs_join_transaction(root);
4755                 if (IS_ERR(trans)) {
4756                         btrfs_orphan_del(NULL, inode);
4757                         btrfs_free_block_rsv(root, rsv);
4758                         goto no_delete;
4759                 }
4760
4761                 trans->block_rsv = rsv;
4762
4763                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4764                 if (ret != -ENOSPC)
4765                         break;
4766
4767                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4768                 btrfs_end_transaction(trans, root);
4769                 trans = NULL;
4770                 btrfs_btree_balance_dirty(root);
4771         }
4772
4773         btrfs_free_block_rsv(root, rsv);
4774
4775         /*
4776          * Errors here aren't a big deal, it just means we leave orphan items
4777          * in the tree.  They will be cleaned up on the next mount.
4778          */
4779         if (ret == 0) {
4780                 trans->block_rsv = root->orphan_block_rsv;
4781                 btrfs_orphan_del(trans, inode);
4782         } else {
4783                 btrfs_orphan_del(NULL, inode);
4784         }
4785
4786         trans->block_rsv = &root->fs_info->trans_block_rsv;
4787         if (!(root == root->fs_info->tree_root ||
4788               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4789                 btrfs_return_ino(root, btrfs_ino(inode));
4790
4791         btrfs_end_transaction(trans, root);
4792         btrfs_btree_balance_dirty(root);
4793 no_delete:
4794         btrfs_remove_delayed_node(inode);
4795         clear_inode(inode);
4796         return;
4797 }
4798
4799 /*
4800  * this returns the key found in the dir entry in the location pointer.
4801  * If no dir entries were found, location->objectid is 0.
4802  */
4803 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4804                                struct btrfs_key *location)
4805 {
4806         const char *name = dentry->d_name.name;
4807         int namelen = dentry->d_name.len;
4808         struct btrfs_dir_item *di;
4809         struct btrfs_path *path;
4810         struct btrfs_root *root = BTRFS_I(dir)->root;
4811         int ret = 0;
4812
4813         path = btrfs_alloc_path();
4814         if (!path)
4815                 return -ENOMEM;
4816
4817         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4818                                     namelen, 0);
4819         if (IS_ERR(di))
4820                 ret = PTR_ERR(di);
4821
4822         if (IS_ERR_OR_NULL(di))
4823                 goto out_err;
4824
4825         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4826 out:
4827         btrfs_free_path(path);
4828         return ret;
4829 out_err:
4830         location->objectid = 0;
4831         goto out;
4832 }
4833
4834 /*
4835  * when we hit a tree root in a directory, the btrfs part of the inode
4836  * needs to be changed to reflect the root directory of the tree root.  This
4837  * is kind of like crossing a mount point.
4838  */
4839 static int fixup_tree_root_location(struct btrfs_root *root,
4840                                     struct inode *dir,
4841                                     struct dentry *dentry,
4842                                     struct btrfs_key *location,
4843                                     struct btrfs_root **sub_root)
4844 {
4845         struct btrfs_path *path;
4846         struct btrfs_root *new_root;
4847         struct btrfs_root_ref *ref;
4848         struct extent_buffer *leaf;
4849         int ret;
4850         int err = 0;
4851
4852         path = btrfs_alloc_path();
4853         if (!path) {
4854                 err = -ENOMEM;
4855                 goto out;
4856         }
4857
4858         err = -ENOENT;
4859         ret = btrfs_find_item(root->fs_info->tree_root, path,
4860                                 BTRFS_I(dir)->root->root_key.objectid,
4861                                 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4862         if (ret) {
4863                 if (ret < 0)
4864                         err = ret;
4865                 goto out;
4866         }
4867
4868         leaf = path->nodes[0];
4869         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4870         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4871             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4872                 goto out;
4873
4874         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4875                                    (unsigned long)(ref + 1),
4876                                    dentry->d_name.len);
4877         if (ret)
4878                 goto out;
4879
4880         btrfs_release_path(path);
4881
4882         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4883         if (IS_ERR(new_root)) {
4884                 err = PTR_ERR(new_root);
4885                 goto out;
4886         }
4887
4888         *sub_root = new_root;
4889         location->objectid = btrfs_root_dirid(&new_root->root_item);
4890         location->type = BTRFS_INODE_ITEM_KEY;
4891         location->offset = 0;
4892         err = 0;
4893 out:
4894         btrfs_free_path(path);
4895         return err;
4896 }
4897
4898 static void inode_tree_add(struct inode *inode)
4899 {
4900         struct btrfs_root *root = BTRFS_I(inode)->root;
4901         struct btrfs_inode *entry;
4902         struct rb_node **p;
4903         struct rb_node *parent;
4904         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4905         u64 ino = btrfs_ino(inode);
4906
4907         if (inode_unhashed(inode))
4908                 return;
4909         parent = NULL;
4910         spin_lock(&root->inode_lock);
4911         p = &root->inode_tree.rb_node;
4912         while (*p) {
4913                 parent = *p;
4914                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4915
4916                 if (ino < btrfs_ino(&entry->vfs_inode))
4917                         p = &parent->rb_left;
4918                 else if (ino > btrfs_ino(&entry->vfs_inode))
4919                         p = &parent->rb_right;
4920                 else {
4921                         WARN_ON(!(entry->vfs_inode.i_state &
4922                                   (I_WILL_FREE | I_FREEING)));
4923                         rb_replace_node(parent, new, &root->inode_tree);
4924                         RB_CLEAR_NODE(parent);
4925                         spin_unlock(&root->inode_lock);
4926                         return;
4927                 }
4928         }
4929         rb_link_node(new, parent, p);
4930         rb_insert_color(new, &root->inode_tree);
4931         spin_unlock(&root->inode_lock);
4932 }
4933
4934 static void inode_tree_del(struct inode *inode)
4935 {
4936         struct btrfs_root *root = BTRFS_I(inode)->root;
4937         int empty = 0;
4938
4939         spin_lock(&root->inode_lock);
4940         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4941                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4942                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4943                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4944         }
4945         spin_unlock(&root->inode_lock);
4946
4947         if (empty && btrfs_root_refs(&root->root_item) == 0) {
4948                 synchronize_srcu(&root->fs_info->subvol_srcu);
4949                 spin_lock(&root->inode_lock);
4950                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4951                 spin_unlock(&root->inode_lock);
4952                 if (empty)
4953                         btrfs_add_dead_root(root);
4954         }
4955 }
4956
4957 void btrfs_invalidate_inodes(struct btrfs_root *root)
4958 {
4959         struct rb_node *node;
4960         struct rb_node *prev;
4961         struct btrfs_inode *entry;
4962         struct inode *inode;
4963         u64 objectid = 0;
4964
4965         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
4966                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4967
4968         spin_lock(&root->inode_lock);
4969 again:
4970         node = root->inode_tree.rb_node;
4971         prev = NULL;
4972         while (node) {
4973                 prev = node;
4974                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4975
4976                 if (objectid < btrfs_ino(&entry->vfs_inode))
4977                         node = node->rb_left;
4978                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4979                         node = node->rb_right;
4980                 else
4981                         break;
4982         }
4983         if (!node) {
4984                 while (prev) {
4985                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4986                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4987                                 node = prev;
4988                                 break;
4989                         }
4990                         prev = rb_next(prev);
4991                 }
4992         }
4993         while (node) {
4994                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4995                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4996                 inode = igrab(&entry->vfs_inode);
4997                 if (inode) {
4998                         spin_unlock(&root->inode_lock);
4999                         if (atomic_read(&inode->i_count) > 1)
5000                                 d_prune_aliases(inode);
5001                         /*
5002                          * btrfs_drop_inode will have it removed from
5003                          * the inode cache when its usage count
5004                          * hits zero.
5005                          */
5006                         iput(inode);
5007                         cond_resched();
5008                         spin_lock(&root->inode_lock);
5009                         goto again;
5010                 }
5011
5012                 if (cond_resched_lock(&root->inode_lock))
5013                         goto again;
5014
5015                 node = rb_next(node);
5016         }
5017         spin_unlock(&root->inode_lock);
5018 }
5019
5020 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5021 {
5022         struct btrfs_iget_args *args = p;
5023         inode->i_ino = args->location->objectid;
5024         memcpy(&BTRFS_I(inode)->location, args->location,
5025                sizeof(*args->location));
5026         BTRFS_I(inode)->root = args->root;
5027         return 0;
5028 }
5029
5030 static int btrfs_find_actor(struct inode *inode, void *opaque)
5031 {
5032         struct btrfs_iget_args *args = opaque;
5033         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5034                 args->root == BTRFS_I(inode)->root;
5035 }
5036
5037 static struct inode *btrfs_iget_locked(struct super_block *s,
5038                                        struct btrfs_key *location,
5039                                        struct btrfs_root *root)
5040 {
5041         struct inode *inode;
5042         struct btrfs_iget_args args;
5043         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5044
5045         args.location = location;
5046         args.root = root;
5047
5048         inode = iget5_locked(s, hashval, btrfs_find_actor,
5049                              btrfs_init_locked_inode,
5050                              (void *)&args);
5051         return inode;
5052 }
5053
5054 /* Get an inode object given its location and corresponding root.
5055  * Returns in *is_new if the inode was read from disk
5056  */
5057 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5058                          struct btrfs_root *root, int *new)
5059 {
5060         struct inode *inode;
5061
5062         inode = btrfs_iget_locked(s, location, root);
5063         if (!inode)
5064                 return ERR_PTR(-ENOMEM);
5065
5066         if (inode->i_state & I_NEW) {
5067                 btrfs_read_locked_inode(inode);
5068                 if (!is_bad_inode(inode)) {
5069                         inode_tree_add(inode);
5070                         unlock_new_inode(inode);
5071                         if (new)
5072                                 *new = 1;
5073                 } else {
5074                         unlock_new_inode(inode);
5075                         iput(inode);
5076                         inode = ERR_PTR(-ESTALE);
5077                 }
5078         }
5079
5080         return inode;
5081 }
5082
5083 static struct inode *new_simple_dir(struct super_block *s,
5084                                     struct btrfs_key *key,
5085                                     struct btrfs_root *root)
5086 {
5087         struct inode *inode = new_inode(s);
5088
5089         if (!inode)
5090                 return ERR_PTR(-ENOMEM);
5091
5092         BTRFS_I(inode)->root = root;
5093         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5094         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5095
5096         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5097         inode->i_op = &btrfs_dir_ro_inode_operations;
5098         inode->i_fop = &simple_dir_operations;
5099         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5100         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5101
5102         return inode;
5103 }
5104
5105 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5106 {
5107         struct inode *inode;
5108         struct btrfs_root *root = BTRFS_I(dir)->root;
5109         struct btrfs_root *sub_root = root;
5110         struct btrfs_key location;
5111         int index;
5112         int ret = 0;
5113
5114         if (dentry->d_name.len > BTRFS_NAME_LEN)
5115                 return ERR_PTR(-ENAMETOOLONG);
5116
5117         ret = btrfs_inode_by_name(dir, dentry, &location);
5118         if (ret < 0)
5119                 return ERR_PTR(ret);
5120
5121         if (location.objectid == 0)
5122                 return ERR_PTR(-ENOENT);
5123
5124         if (location.type == BTRFS_INODE_ITEM_KEY) {
5125                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5126                 return inode;
5127         }
5128
5129         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5130
5131         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5132         ret = fixup_tree_root_location(root, dir, dentry,
5133                                        &location, &sub_root);
5134         if (ret < 0) {
5135                 if (ret != -ENOENT)
5136                         inode = ERR_PTR(ret);
5137                 else
5138                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5139         } else {
5140                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5141         }
5142         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5143
5144         if (!IS_ERR(inode) && root != sub_root) {
5145                 down_read(&root->fs_info->cleanup_work_sem);
5146                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5147                         ret = btrfs_orphan_cleanup(sub_root);
5148                 up_read(&root->fs_info->cleanup_work_sem);
5149                 if (ret) {
5150                         iput(inode);
5151                         inode = ERR_PTR(ret);
5152                 }
5153         }
5154
5155         return inode;
5156 }
5157
5158 static int btrfs_dentry_delete(const struct dentry *dentry)
5159 {
5160         struct btrfs_root *root;
5161         struct inode *inode = dentry->d_inode;
5162
5163         if (!inode && !IS_ROOT(dentry))
5164                 inode = dentry->d_parent->d_inode;
5165
5166         if (inode) {
5167                 root = BTRFS_I(inode)->root;
5168                 if (btrfs_root_refs(&root->root_item) == 0)
5169                         return 1;
5170
5171                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5172                         return 1;
5173         }
5174         return 0;
5175 }
5176
5177 static void btrfs_dentry_release(struct dentry *dentry)
5178 {
5179         kfree(dentry->d_fsdata);
5180 }
5181
5182 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5183                                    unsigned int flags)
5184 {
5185         struct inode *inode;
5186
5187         inode = btrfs_lookup_dentry(dir, dentry);
5188         if (IS_ERR(inode)) {
5189                 if (PTR_ERR(inode) == -ENOENT)
5190                         inode = NULL;
5191                 else
5192                         return ERR_CAST(inode);
5193         }
5194
5195         return d_materialise_unique(dentry, inode);
5196 }
5197
5198 unsigned char btrfs_filetype_table[] = {
5199         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5200 };
5201
5202 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5203 {
5204         struct inode *inode = file_inode(file);
5205         struct btrfs_root *root = BTRFS_I(inode)->root;
5206         struct btrfs_item *item;
5207         struct btrfs_dir_item *di;
5208         struct btrfs_key key;
5209         struct btrfs_key found_key;
5210         struct btrfs_path *path;
5211         struct list_head ins_list;
5212         struct list_head del_list;
5213         int ret;
5214         struct extent_buffer *leaf;
5215         int slot;
5216         unsigned char d_type;
5217         int over = 0;
5218         u32 di_cur;
5219         u32 di_total;
5220         u32 di_len;
5221         int key_type = BTRFS_DIR_INDEX_KEY;
5222         char tmp_name[32];
5223         char *name_ptr;
5224         int name_len;
5225         int is_curr = 0;        /* ctx->pos points to the current index? */
5226
5227         /* FIXME, use a real flag for deciding about the key type */
5228         if (root->fs_info->tree_root == root)
5229                 key_type = BTRFS_DIR_ITEM_KEY;
5230
5231         if (!dir_emit_dots(file, ctx))
5232                 return 0;
5233
5234         path = btrfs_alloc_path();
5235         if (!path)
5236                 return -ENOMEM;
5237
5238         path->reada = 1;
5239
5240         if (key_type == BTRFS_DIR_INDEX_KEY) {
5241                 INIT_LIST_HEAD(&ins_list);
5242                 INIT_LIST_HEAD(&del_list);
5243                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5244         }
5245
5246         btrfs_set_key_type(&key, key_type);
5247         key.offset = ctx->pos;
5248         key.objectid = btrfs_ino(inode);
5249
5250         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5251         if (ret < 0)
5252                 goto err;
5253
5254         while (1) {
5255                 leaf = path->nodes[0];
5256                 slot = path->slots[0];
5257                 if (slot >= btrfs_header_nritems(leaf)) {
5258                         ret = btrfs_next_leaf(root, path);
5259                         if (ret < 0)
5260                                 goto err;
5261                         else if (ret > 0)
5262                                 break;
5263                         continue;
5264                 }
5265
5266                 item = btrfs_item_nr(slot);
5267                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5268
5269                 if (found_key.objectid != key.objectid)
5270                         break;
5271                 if (btrfs_key_type(&found_key) != key_type)
5272                         break;
5273                 if (found_key.offset < ctx->pos)
5274                         goto next;
5275                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5276                     btrfs_should_delete_dir_index(&del_list,
5277                                                   found_key.offset))
5278                         goto next;
5279
5280                 ctx->pos = found_key.offset;
5281                 is_curr = 1;
5282
5283                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5284                 di_cur = 0;
5285                 di_total = btrfs_item_size(leaf, item);
5286
5287                 while (di_cur < di_total) {
5288                         struct btrfs_key location;
5289
5290                         if (verify_dir_item(root, leaf, di))
5291                                 break;
5292
5293                         name_len = btrfs_dir_name_len(leaf, di);
5294                         if (name_len <= sizeof(tmp_name)) {
5295                                 name_ptr = tmp_name;
5296                         } else {
5297                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5298                                 if (!name_ptr) {
5299                                         ret = -ENOMEM;
5300                                         goto err;
5301                                 }
5302                         }
5303                         read_extent_buffer(leaf, name_ptr,
5304                                            (unsigned long)(di + 1), name_len);
5305
5306                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5307                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5308
5309
5310                         /* is this a reference to our own snapshot? If so
5311                          * skip it.
5312                          *
5313                          * In contrast to old kernels, we insert the snapshot's
5314                          * dir item and dir index after it has been created, so
5315                          * we won't find a reference to our own snapshot. We
5316                          * still keep the following code for backward
5317                          * compatibility.
5318                          */
5319                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5320                             location.objectid == root->root_key.objectid) {
5321                                 over = 0;
5322                                 goto skip;
5323                         }
5324                         over = !dir_emit(ctx, name_ptr, name_len,
5325                                        location.objectid, d_type);
5326
5327 skip:
5328                         if (name_ptr != tmp_name)
5329                                 kfree(name_ptr);
5330
5331                         if (over)
5332                                 goto nopos;
5333                         di_len = btrfs_dir_name_len(leaf, di) +
5334                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5335                         di_cur += di_len;
5336                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5337                 }
5338 next:
5339                 path->slots[0]++;
5340         }
5341
5342         if (key_type == BTRFS_DIR_INDEX_KEY) {
5343                 if (is_curr)
5344                         ctx->pos++;
5345                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5346                 if (ret)
5347                         goto nopos;
5348         }
5349
5350         /* Reached end of directory/root. Bump pos past the last item. */
5351         ctx->pos++;
5352
5353         /*
5354          * Stop new entries from being returned after we return the last
5355          * entry.
5356          *
5357          * New directory entries are assigned a strictly increasing
5358          * offset.  This means that new entries created during readdir
5359          * are *guaranteed* to be seen in the future by that readdir.
5360          * This has broken buggy programs which operate on names as
5361          * they're returned by readdir.  Until we re-use freed offsets
5362          * we have this hack to stop new entries from being returned
5363          * under the assumption that they'll never reach this huge
5364          * offset.
5365          *
5366          * This is being careful not to overflow 32bit loff_t unless the
5367          * last entry requires it because doing so has broken 32bit apps
5368          * in the past.
5369          */
5370         if (key_type == BTRFS_DIR_INDEX_KEY) {
5371                 if (ctx->pos >= INT_MAX)
5372                         ctx->pos = LLONG_MAX;
5373                 else
5374                         ctx->pos = INT_MAX;
5375         }
5376 nopos:
5377         ret = 0;
5378 err:
5379         if (key_type == BTRFS_DIR_INDEX_KEY)
5380                 btrfs_put_delayed_items(&ins_list, &del_list);
5381         btrfs_free_path(path);
5382         return ret;
5383 }
5384
5385 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5386 {
5387         struct btrfs_root *root = BTRFS_I(inode)->root;
5388         struct btrfs_trans_handle *trans;
5389         int ret = 0;
5390         bool nolock = false;
5391
5392         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5393                 return 0;
5394
5395         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5396                 nolock = true;
5397
5398         if (wbc->sync_mode == WB_SYNC_ALL) {
5399                 if (nolock)
5400                         trans = btrfs_join_transaction_nolock(root);
5401                 else
5402                         trans = btrfs_join_transaction(root);
5403                 if (IS_ERR(trans))
5404                         return PTR_ERR(trans);
5405                 ret = btrfs_commit_transaction(trans, root);
5406         }
5407         return ret;
5408 }
5409
5410 /*
5411  * This is somewhat expensive, updating the tree every time the
5412  * inode changes.  But, it is most likely to find the inode in cache.
5413  * FIXME, needs more benchmarking...there are no reasons other than performance
5414  * to keep or drop this code.
5415  */
5416 static int btrfs_dirty_inode(struct inode *inode)
5417 {
5418         struct btrfs_root *root = BTRFS_I(inode)->root;
5419         struct btrfs_trans_handle *trans;
5420         int ret;
5421
5422         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5423                 return 0;
5424
5425         trans = btrfs_join_transaction(root);
5426         if (IS_ERR(trans))
5427                 return PTR_ERR(trans);
5428
5429         ret = btrfs_update_inode(trans, root, inode);
5430         if (ret && ret == -ENOSPC) {
5431                 /* whoops, lets try again with the full transaction */
5432                 btrfs_end_transaction(trans, root);
5433                 trans = btrfs_start_transaction(root, 1);
5434                 if (IS_ERR(trans))
5435                         return PTR_ERR(trans);
5436
5437                 ret = btrfs_update_inode(trans, root, inode);
5438         }
5439         btrfs_end_transaction(trans, root);
5440         if (BTRFS_I(inode)->delayed_node)
5441                 btrfs_balance_delayed_items(root);
5442
5443         return ret;
5444 }
5445
5446 /*
5447  * This is a copy of file_update_time.  We need this so we can return error on
5448  * ENOSPC for updating the inode in the case of file write and mmap writes.
5449  */
5450 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5451                              int flags)
5452 {
5453         struct btrfs_root *root = BTRFS_I(inode)->root;
5454
5455         if (btrfs_root_readonly(root))
5456                 return -EROFS;
5457
5458         if (flags & S_VERSION)
5459                 inode_inc_iversion(inode);
5460         if (flags & S_CTIME)
5461                 inode->i_ctime = *now;
5462         if (flags & S_MTIME)
5463                 inode->i_mtime = *now;
5464         if (flags & S_ATIME)
5465                 inode->i_atime = *now;
5466         return btrfs_dirty_inode(inode);
5467 }
5468
5469 /*
5470  * find the highest existing sequence number in a directory
5471  * and then set the in-memory index_cnt variable to reflect
5472  * free sequence numbers
5473  */
5474 static int btrfs_set_inode_index_count(struct inode *inode)
5475 {
5476         struct btrfs_root *root = BTRFS_I(inode)->root;
5477         struct btrfs_key key, found_key;
5478         struct btrfs_path *path;
5479         struct extent_buffer *leaf;
5480         int ret;
5481
5482         key.objectid = btrfs_ino(inode);
5483         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5484         key.offset = (u64)-1;
5485
5486         path = btrfs_alloc_path();
5487         if (!path)
5488                 return -ENOMEM;
5489
5490         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5491         if (ret < 0)
5492                 goto out;
5493         /* FIXME: we should be able to handle this */
5494         if (ret == 0)
5495                 goto out;
5496         ret = 0;
5497
5498         /*
5499          * MAGIC NUMBER EXPLANATION:
5500          * since we search a directory based on f_pos we have to start at 2
5501          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5502          * else has to start at 2
5503          */
5504         if (path->slots[0] == 0) {
5505                 BTRFS_I(inode)->index_cnt = 2;
5506                 goto out;
5507         }
5508
5509         path->slots[0]--;
5510
5511         leaf = path->nodes[0];
5512         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5513
5514         if (found_key.objectid != btrfs_ino(inode) ||
5515             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5516                 BTRFS_I(inode)->index_cnt = 2;
5517                 goto out;
5518         }
5519
5520         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5521 out:
5522         btrfs_free_path(path);
5523         return ret;
5524 }
5525
5526 /*
5527  * helper to find a free sequence number in a given directory.  This current
5528  * code is very simple, later versions will do smarter things in the btree
5529  */
5530 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5531 {
5532         int ret = 0;
5533
5534         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5535                 ret = btrfs_inode_delayed_dir_index_count(dir);
5536                 if (ret) {
5537                         ret = btrfs_set_inode_index_count(dir);
5538                         if (ret)
5539                                 return ret;
5540                 }
5541         }
5542
5543         *index = BTRFS_I(dir)->index_cnt;
5544         BTRFS_I(dir)->index_cnt++;
5545
5546         return ret;
5547 }
5548
5549 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5550                                      struct btrfs_root *root,
5551                                      struct inode *dir,
5552                                      const char *name, int name_len,
5553                                      u64 ref_objectid, u64 objectid,
5554                                      umode_t mode, u64 *index)
5555 {
5556         struct inode *inode;
5557         struct btrfs_inode_item *inode_item;
5558         struct btrfs_key *location;
5559         struct btrfs_path *path;
5560         struct btrfs_inode_ref *ref;
5561         struct btrfs_key key[2];
5562         u32 sizes[2];
5563         int nitems = name ? 2 : 1;
5564         unsigned long ptr;
5565         int ret;
5566
5567         path = btrfs_alloc_path();
5568         if (!path)
5569                 return ERR_PTR(-ENOMEM);
5570
5571         inode = new_inode(root->fs_info->sb);
5572         if (!inode) {
5573                 btrfs_free_path(path);
5574                 return ERR_PTR(-ENOMEM);
5575         }
5576
5577         /*
5578          * we have to initialize this early, so we can reclaim the inode
5579          * number if we fail afterwards in this function.
5580          */
5581         inode->i_ino = objectid;
5582
5583         if (dir && name) {
5584                 trace_btrfs_inode_request(dir);
5585
5586                 ret = btrfs_set_inode_index(dir, index);
5587                 if (ret) {
5588                         btrfs_free_path(path);
5589                         iput(inode);
5590                         return ERR_PTR(ret);
5591                 }
5592         } else if (dir) {
5593                 *index = 0;
5594         }
5595         /*
5596          * index_cnt is ignored for everything but a dir,
5597          * btrfs_get_inode_index_count has an explanation for the magic
5598          * number
5599          */
5600         BTRFS_I(inode)->index_cnt = 2;
5601         BTRFS_I(inode)->dir_index = *index;
5602         BTRFS_I(inode)->root = root;
5603         BTRFS_I(inode)->generation = trans->transid;
5604         inode->i_generation = BTRFS_I(inode)->generation;
5605
5606         /*
5607          * We could have gotten an inode number from somebody who was fsynced
5608          * and then removed in this same transaction, so let's just set full
5609          * sync since it will be a full sync anyway and this will blow away the
5610          * old info in the log.
5611          */
5612         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5613
5614         key[0].objectid = objectid;
5615         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5616         key[0].offset = 0;
5617
5618         sizes[0] = sizeof(struct btrfs_inode_item);
5619
5620         if (name) {
5621                 /*
5622                  * Start new inodes with an inode_ref. This is slightly more
5623                  * efficient for small numbers of hard links since they will
5624                  * be packed into one item. Extended refs will kick in if we
5625                  * add more hard links than can fit in the ref item.
5626                  */
5627                 key[1].objectid = objectid;
5628                 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5629                 key[1].offset = ref_objectid;
5630
5631                 sizes[1] = name_len + sizeof(*ref);
5632         }
5633
5634         path->leave_spinning = 1;
5635         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5636         if (ret != 0)
5637                 goto fail;
5638
5639         inode_init_owner(inode, dir, mode);
5640         inode_set_bytes(inode, 0);
5641         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5642         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5643                                   struct btrfs_inode_item);
5644         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5645                              sizeof(*inode_item));
5646         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5647
5648         if (name) {
5649                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5650                                      struct btrfs_inode_ref);
5651                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5652                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5653                 ptr = (unsigned long)(ref + 1);
5654                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5655         }
5656
5657         btrfs_mark_buffer_dirty(path->nodes[0]);
5658         btrfs_free_path(path);
5659
5660         location = &BTRFS_I(inode)->location;
5661         location->objectid = objectid;
5662         location->offset = 0;
5663         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5664
5665         btrfs_inherit_iflags(inode, dir);
5666
5667         if (S_ISREG(mode)) {
5668                 if (btrfs_test_opt(root, NODATASUM))
5669                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5670                 if (btrfs_test_opt(root, NODATACOW))
5671                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5672                                 BTRFS_INODE_NODATASUM;
5673         }
5674
5675         btrfs_insert_inode_hash(inode);
5676         inode_tree_add(inode);
5677
5678         trace_btrfs_inode_new(inode);
5679         btrfs_set_inode_last_trans(trans, inode);
5680
5681         btrfs_update_root_times(trans, root);
5682
5683         ret = btrfs_inode_inherit_props(trans, inode, dir);
5684         if (ret)
5685                 btrfs_err(root->fs_info,
5686                           "error inheriting props for ino %llu (root %llu): %d",
5687                           btrfs_ino(inode), root->root_key.objectid, ret);
5688
5689         return inode;
5690 fail:
5691         if (dir && name)
5692                 BTRFS_I(dir)->index_cnt--;
5693         btrfs_free_path(path);
5694         iput(inode);
5695         return ERR_PTR(ret);
5696 }
5697
5698 static inline u8 btrfs_inode_type(struct inode *inode)
5699 {
5700         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5701 }
5702
5703 /*
5704  * utility function to add 'inode' into 'parent_inode' with
5705  * a give name and a given sequence number.
5706  * if 'add_backref' is true, also insert a backref from the
5707  * inode to the parent directory.
5708  */
5709 int btrfs_add_link(struct btrfs_trans_handle *trans,
5710                    struct inode *parent_inode, struct inode *inode,
5711                    const char *name, int name_len, int add_backref, u64 index)
5712 {
5713         int ret = 0;
5714         struct btrfs_key key;
5715         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5716         u64 ino = btrfs_ino(inode);
5717         u64 parent_ino = btrfs_ino(parent_inode);
5718
5719         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5720                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5721         } else {
5722                 key.objectid = ino;
5723                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5724                 key.offset = 0;
5725         }
5726
5727         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5728                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5729                                          key.objectid, root->root_key.objectid,
5730                                          parent_ino, index, name, name_len);
5731         } else if (add_backref) {
5732                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5733                                              parent_ino, index);
5734         }
5735
5736         /* Nothing to clean up yet */
5737         if (ret)
5738                 return ret;
5739
5740         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5741                                     parent_inode, &key,
5742                                     btrfs_inode_type(inode), index);
5743         if (ret == -EEXIST || ret == -EOVERFLOW)
5744                 goto fail_dir_item;
5745         else if (ret) {
5746                 btrfs_abort_transaction(trans, root, ret);
5747                 return ret;
5748         }
5749
5750         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5751                            name_len * 2);
5752         inode_inc_iversion(parent_inode);
5753         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5754         ret = btrfs_update_inode(trans, root, parent_inode);
5755         if (ret)
5756                 btrfs_abort_transaction(trans, root, ret);
5757         return ret;
5758
5759 fail_dir_item:
5760         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5761                 u64 local_index;
5762                 int err;
5763                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5764                                  key.objectid, root->root_key.objectid,
5765                                  parent_ino, &local_index, name, name_len);
5766
5767         } else if (add_backref) {
5768                 u64 local_index;
5769                 int err;
5770
5771                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5772                                           ino, parent_ino, &local_index);
5773         }
5774         return ret;
5775 }
5776
5777 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5778                             struct inode *dir, struct dentry *dentry,
5779                             struct inode *inode, int backref, u64 index)
5780 {
5781         int err = btrfs_add_link(trans, dir, inode,
5782                                  dentry->d_name.name, dentry->d_name.len,
5783                                  backref, index);
5784         if (err > 0)
5785                 err = -EEXIST;
5786         return err;
5787 }
5788
5789 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5790                         umode_t mode, dev_t rdev)
5791 {
5792         struct btrfs_trans_handle *trans;
5793         struct btrfs_root *root = BTRFS_I(dir)->root;
5794         struct inode *inode = NULL;
5795         int err;
5796         int drop_inode = 0;
5797         u64 objectid;
5798         u64 index = 0;
5799
5800         if (!new_valid_dev(rdev))
5801                 return -EINVAL;
5802
5803         /*
5804          * 2 for inode item and ref
5805          * 2 for dir items
5806          * 1 for xattr if selinux is on
5807          */
5808         trans = btrfs_start_transaction(root, 5);
5809         if (IS_ERR(trans))
5810                 return PTR_ERR(trans);
5811
5812         err = btrfs_find_free_ino(root, &objectid);
5813         if (err)
5814                 goto out_unlock;
5815
5816         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5817                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5818                                 mode, &index);
5819         if (IS_ERR(inode)) {
5820                 err = PTR_ERR(inode);
5821                 goto out_unlock;
5822         }
5823
5824         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5825         if (err) {
5826                 drop_inode = 1;
5827                 goto out_unlock;
5828         }
5829
5830         /*
5831         * If the active LSM wants to access the inode during
5832         * d_instantiate it needs these. Smack checks to see
5833         * if the filesystem supports xattrs by looking at the
5834         * ops vector.
5835         */
5836
5837         inode->i_op = &btrfs_special_inode_operations;
5838         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5839         if (err)
5840                 drop_inode = 1;
5841         else {
5842                 init_special_inode(inode, inode->i_mode, rdev);
5843                 btrfs_update_inode(trans, root, inode);
5844                 d_instantiate(dentry, inode);
5845         }
5846 out_unlock:
5847         btrfs_end_transaction(trans, root);
5848         btrfs_balance_delayed_items(root);
5849         btrfs_btree_balance_dirty(root);
5850         if (drop_inode) {
5851                 inode_dec_link_count(inode);
5852                 iput(inode);
5853         }
5854         return err;
5855 }
5856
5857 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5858                         umode_t mode, bool excl)
5859 {
5860         struct btrfs_trans_handle *trans;
5861         struct btrfs_root *root = BTRFS_I(dir)->root;
5862         struct inode *inode = NULL;
5863         int drop_inode_on_err = 0;
5864         int err;
5865         u64 objectid;
5866         u64 index = 0;
5867
5868         /*
5869          * 2 for inode item and ref
5870          * 2 for dir items
5871          * 1 for xattr if selinux is on
5872          */
5873         trans = btrfs_start_transaction(root, 5);
5874         if (IS_ERR(trans))
5875                 return PTR_ERR(trans);
5876
5877         err = btrfs_find_free_ino(root, &objectid);
5878         if (err)
5879                 goto out_unlock;
5880
5881         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5882                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5883                                 mode, &index);
5884         if (IS_ERR(inode)) {
5885                 err = PTR_ERR(inode);
5886                 goto out_unlock;
5887         }
5888         drop_inode_on_err = 1;
5889
5890         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5891         if (err)
5892                 goto out_unlock;
5893
5894         err = btrfs_update_inode(trans, root, inode);
5895         if (err)
5896                 goto out_unlock;
5897
5898         /*
5899         * If the active LSM wants to access the inode during
5900         * d_instantiate it needs these. Smack checks to see
5901         * if the filesystem supports xattrs by looking at the
5902         * ops vector.
5903         */
5904         inode->i_fop = &btrfs_file_operations;
5905         inode->i_op = &btrfs_file_inode_operations;
5906
5907         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5908         if (err)
5909                 goto out_unlock;
5910
5911         inode->i_mapping->a_ops = &btrfs_aops;
5912         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5913         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5914         d_instantiate(dentry, inode);
5915
5916 out_unlock:
5917         btrfs_end_transaction(trans, root);
5918         if (err && drop_inode_on_err) {
5919                 inode_dec_link_count(inode);
5920                 iput(inode);
5921         }
5922         btrfs_balance_delayed_items(root);
5923         btrfs_btree_balance_dirty(root);
5924         return err;
5925 }
5926
5927 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5928                       struct dentry *dentry)
5929 {
5930         struct btrfs_trans_handle *trans;
5931         struct btrfs_root *root = BTRFS_I(dir)->root;
5932         struct inode *inode = old_dentry->d_inode;
5933         u64 index;
5934         int err;
5935         int drop_inode = 0;
5936
5937         /* do not allow sys_link's with other subvols of the same device */
5938         if (root->objectid != BTRFS_I(inode)->root->objectid)
5939                 return -EXDEV;
5940
5941         if (inode->i_nlink >= BTRFS_LINK_MAX)
5942                 return -EMLINK;
5943
5944         err = btrfs_set_inode_index(dir, &index);
5945         if (err)
5946                 goto fail;
5947
5948         /*
5949          * 2 items for inode and inode ref
5950          * 2 items for dir items
5951          * 1 item for parent inode
5952          */
5953         trans = btrfs_start_transaction(root, 5);
5954         if (IS_ERR(trans)) {
5955                 err = PTR_ERR(trans);
5956                 goto fail;
5957         }
5958
5959         /* There are several dir indexes for this inode, clear the cache. */
5960         BTRFS_I(inode)->dir_index = 0ULL;
5961         inc_nlink(inode);
5962         inode_inc_iversion(inode);
5963         inode->i_ctime = CURRENT_TIME;
5964         ihold(inode);
5965         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5966
5967         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5968
5969         if (err) {
5970                 drop_inode = 1;
5971         } else {
5972                 struct dentry *parent = dentry->d_parent;
5973                 err = btrfs_update_inode(trans, root, inode);
5974                 if (err)
5975                         goto fail;
5976                 if (inode->i_nlink == 1) {
5977                         /*
5978                          * If new hard link count is 1, it's a file created
5979                          * with open(2) O_TMPFILE flag.
5980                          */
5981                         err = btrfs_orphan_del(trans, inode);
5982                         if (err)
5983                                 goto fail;
5984                 }
5985                 d_instantiate(dentry, inode);
5986                 btrfs_log_new_name(trans, inode, NULL, parent);
5987         }
5988
5989         btrfs_end_transaction(trans, root);
5990         btrfs_balance_delayed_items(root);
5991 fail:
5992         if (drop_inode) {
5993                 inode_dec_link_count(inode);
5994                 iput(inode);
5995         }
5996         btrfs_btree_balance_dirty(root);
5997         return err;
5998 }
5999
6000 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6001 {
6002         struct inode *inode = NULL;
6003         struct btrfs_trans_handle *trans;
6004         struct btrfs_root *root = BTRFS_I(dir)->root;
6005         int err = 0;
6006         int drop_on_err = 0;
6007         u64 objectid = 0;
6008         u64 index = 0;
6009
6010         /*
6011          * 2 items for inode and ref
6012          * 2 items for dir items
6013          * 1 for xattr if selinux is on
6014          */
6015         trans = btrfs_start_transaction(root, 5);
6016         if (IS_ERR(trans))
6017                 return PTR_ERR(trans);
6018
6019         err = btrfs_find_free_ino(root, &objectid);
6020         if (err)
6021                 goto out_fail;
6022
6023         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6024                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6025                                 S_IFDIR | mode, &index);
6026         if (IS_ERR(inode)) {
6027                 err = PTR_ERR(inode);
6028                 goto out_fail;
6029         }
6030
6031         drop_on_err = 1;
6032
6033         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6034         if (err)
6035                 goto out_fail;
6036
6037         inode->i_op = &btrfs_dir_inode_operations;
6038         inode->i_fop = &btrfs_dir_file_operations;
6039
6040         btrfs_i_size_write(inode, 0);
6041         err = btrfs_update_inode(trans, root, inode);
6042         if (err)
6043                 goto out_fail;
6044
6045         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6046                              dentry->d_name.len, 0, index);
6047         if (err)
6048                 goto out_fail;
6049
6050         d_instantiate(dentry, inode);
6051         drop_on_err = 0;
6052
6053 out_fail:
6054         btrfs_end_transaction(trans, root);
6055         if (drop_on_err)
6056                 iput(inode);
6057         btrfs_balance_delayed_items(root);
6058         btrfs_btree_balance_dirty(root);
6059         return err;
6060 }
6061
6062 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6063  * and an extent that you want to insert, deal with overlap and insert
6064  * the new extent into the tree.
6065  */
6066 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6067                                 struct extent_map *existing,
6068                                 struct extent_map *em,
6069                                 u64 map_start, u64 map_len)
6070 {
6071         u64 start_diff;
6072
6073         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6074         start_diff = map_start - em->start;
6075         em->start = map_start;
6076         em->len = map_len;
6077         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6078             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6079                 em->block_start += start_diff;
6080                 em->block_len -= start_diff;
6081         }
6082         return add_extent_mapping(em_tree, em, 0);
6083 }
6084
6085 static noinline int uncompress_inline(struct btrfs_path *path,
6086                                       struct inode *inode, struct page *page,
6087                                       size_t pg_offset, u64 extent_offset,
6088                                       struct btrfs_file_extent_item *item)
6089 {
6090         int ret;
6091         struct extent_buffer *leaf = path->nodes[0];
6092         char *tmp;
6093         size_t max_size;
6094         unsigned long inline_size;
6095         unsigned long ptr;
6096         int compress_type;
6097
6098         WARN_ON(pg_offset != 0);
6099         compress_type = btrfs_file_extent_compression(leaf, item);
6100         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6101         inline_size = btrfs_file_extent_inline_item_len(leaf,
6102                                         btrfs_item_nr(path->slots[0]));
6103         tmp = kmalloc(inline_size, GFP_NOFS);
6104         if (!tmp)
6105                 return -ENOMEM;
6106         ptr = btrfs_file_extent_inline_start(item);
6107
6108         read_extent_buffer(leaf, tmp, ptr, inline_size);
6109
6110         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6111         ret = btrfs_decompress(compress_type, tmp, page,
6112                                extent_offset, inline_size, max_size);
6113         kfree(tmp);
6114         return ret;
6115 }
6116
6117 /*
6118  * a bit scary, this does extent mapping from logical file offset to the disk.
6119  * the ugly parts come from merging extents from the disk with the in-ram
6120  * representation.  This gets more complex because of the data=ordered code,
6121  * where the in-ram extents might be locked pending data=ordered completion.
6122  *
6123  * This also copies inline extents directly into the page.
6124  */
6125
6126 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6127                                     size_t pg_offset, u64 start, u64 len,
6128                                     int create)
6129 {
6130         int ret;
6131         int err = 0;
6132         u64 bytenr;
6133         u64 extent_start = 0;
6134         u64 extent_end = 0;
6135         u64 objectid = btrfs_ino(inode);
6136         u32 found_type;
6137         struct btrfs_path *path = NULL;
6138         struct btrfs_root *root = BTRFS_I(inode)->root;
6139         struct btrfs_file_extent_item *item;
6140         struct extent_buffer *leaf;
6141         struct btrfs_key found_key;
6142         struct extent_map *em = NULL;
6143         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6144         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6145         struct btrfs_trans_handle *trans = NULL;
6146         int compress_type;
6147
6148 again:
6149         read_lock(&em_tree->lock);
6150         em = lookup_extent_mapping(em_tree, start, len);
6151         if (em)
6152                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6153         read_unlock(&em_tree->lock);
6154
6155         if (em) {
6156                 if (em->start > start || em->start + em->len <= start)
6157                         free_extent_map(em);
6158                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6159                         free_extent_map(em);
6160                 else
6161                         goto out;
6162         }
6163         em = alloc_extent_map();
6164         if (!em) {
6165                 err = -ENOMEM;
6166                 goto out;
6167         }
6168         em->bdev = root->fs_info->fs_devices->latest_bdev;
6169         em->start = EXTENT_MAP_HOLE;
6170         em->orig_start = EXTENT_MAP_HOLE;
6171         em->len = (u64)-1;
6172         em->block_len = (u64)-1;
6173
6174         if (!path) {
6175                 path = btrfs_alloc_path();
6176                 if (!path) {
6177                         err = -ENOMEM;
6178                         goto out;
6179                 }
6180                 /*
6181                  * Chances are we'll be called again, so go ahead and do
6182                  * readahead
6183                  */
6184                 path->reada = 1;
6185         }
6186
6187         ret = btrfs_lookup_file_extent(trans, root, path,
6188                                        objectid, start, trans != NULL);
6189         if (ret < 0) {
6190                 err = ret;
6191                 goto out;
6192         }
6193
6194         if (ret != 0) {
6195                 if (path->slots[0] == 0)
6196                         goto not_found;
6197                 path->slots[0]--;
6198         }
6199
6200         leaf = path->nodes[0];
6201         item = btrfs_item_ptr(leaf, path->slots[0],
6202                               struct btrfs_file_extent_item);
6203         /* are we inside the extent that was found? */
6204         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6205         found_type = btrfs_key_type(&found_key);
6206         if (found_key.objectid != objectid ||
6207             found_type != BTRFS_EXTENT_DATA_KEY) {
6208                 /*
6209                  * If we backup past the first extent we want to move forward
6210                  * and see if there is an extent in front of us, otherwise we'll
6211                  * say there is a hole for our whole search range which can
6212                  * cause problems.
6213                  */
6214                 extent_end = start;
6215                 goto next;
6216         }
6217
6218         found_type = btrfs_file_extent_type(leaf, item);
6219         extent_start = found_key.offset;
6220         compress_type = btrfs_file_extent_compression(leaf, item);
6221         if (found_type == BTRFS_FILE_EXTENT_REG ||
6222             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6223                 extent_end = extent_start +
6224                        btrfs_file_extent_num_bytes(leaf, item);
6225         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6226                 size_t size;
6227                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6228                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6229         }
6230 next:
6231         if (start >= extent_end) {
6232                 path->slots[0]++;
6233                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6234                         ret = btrfs_next_leaf(root, path);
6235                         if (ret < 0) {
6236                                 err = ret;
6237                                 goto out;
6238                         }
6239                         if (ret > 0)
6240                                 goto not_found;
6241                         leaf = path->nodes[0];
6242                 }
6243                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6244                 if (found_key.objectid != objectid ||
6245                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6246                         goto not_found;
6247                 if (start + len <= found_key.offset)
6248                         goto not_found;
6249                 em->start = start;
6250                 em->orig_start = start;
6251                 em->len = found_key.offset - start;
6252                 goto not_found_em;
6253         }
6254
6255         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6256         if (found_type == BTRFS_FILE_EXTENT_REG ||
6257             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6258                 em->start = extent_start;
6259                 em->len = extent_end - extent_start;
6260                 em->orig_start = extent_start -
6261                                  btrfs_file_extent_offset(leaf, item);
6262                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6263                                                                       item);
6264                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6265                 if (bytenr == 0) {
6266                         em->block_start = EXTENT_MAP_HOLE;
6267                         goto insert;
6268                 }
6269                 if (compress_type != BTRFS_COMPRESS_NONE) {
6270                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6271                         em->compress_type = compress_type;
6272                         em->block_start = bytenr;
6273                         em->block_len = em->orig_block_len;
6274                 } else {
6275                         bytenr += btrfs_file_extent_offset(leaf, item);
6276                         em->block_start = bytenr;
6277                         em->block_len = em->len;
6278                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6279                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6280                 }
6281                 goto insert;
6282         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6283                 unsigned long ptr;
6284                 char *map;
6285                 size_t size;
6286                 size_t extent_offset;
6287                 size_t copy_size;
6288
6289                 em->block_start = EXTENT_MAP_INLINE;
6290                 if (!page || create) {
6291                         em->start = extent_start;
6292                         em->len = extent_end - extent_start;
6293                         goto out;
6294                 }
6295
6296                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6297                 extent_offset = page_offset(page) + pg_offset - extent_start;
6298                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6299                                 size - extent_offset);
6300                 em->start = extent_start + extent_offset;
6301                 em->len = ALIGN(copy_size, root->sectorsize);
6302                 em->orig_block_len = em->len;
6303                 em->orig_start = em->start;
6304                 if (compress_type) {
6305                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6306                         em->compress_type = compress_type;
6307                 }
6308                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6309                 if (create == 0 && !PageUptodate(page)) {
6310                         if (btrfs_file_extent_compression(leaf, item) !=
6311                             BTRFS_COMPRESS_NONE) {
6312                                 ret = uncompress_inline(path, inode, page,
6313                                                         pg_offset,
6314                                                         extent_offset, item);
6315                                 if (ret) {
6316                                         err = ret;
6317                                         goto out;
6318                                 }
6319                         } else {
6320                                 map = kmap(page);
6321                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6322                                                    copy_size);
6323                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6324                                         memset(map + pg_offset + copy_size, 0,
6325                                                PAGE_CACHE_SIZE - pg_offset -
6326                                                copy_size);
6327                                 }
6328                                 kunmap(page);
6329                         }
6330                         flush_dcache_page(page);
6331                 } else if (create && PageUptodate(page)) {
6332                         BUG();
6333                         if (!trans) {
6334                                 kunmap(page);
6335                                 free_extent_map(em);
6336                                 em = NULL;
6337
6338                                 btrfs_release_path(path);
6339                                 trans = btrfs_join_transaction(root);
6340
6341                                 if (IS_ERR(trans))
6342                                         return ERR_CAST(trans);
6343                                 goto again;
6344                         }
6345                         map = kmap(page);
6346                         write_extent_buffer(leaf, map + pg_offset, ptr,
6347                                             copy_size);
6348                         kunmap(page);
6349                         btrfs_mark_buffer_dirty(leaf);
6350                 }
6351                 set_extent_uptodate(io_tree, em->start,
6352                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6353                 goto insert;
6354         } else {
6355                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6356         }
6357 not_found:
6358         em->start = start;
6359         em->orig_start = start;
6360         em->len = len;
6361 not_found_em:
6362         em->block_start = EXTENT_MAP_HOLE;
6363         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6364 insert:
6365         btrfs_release_path(path);
6366         if (em->start > start || extent_map_end(em) <= start) {
6367                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6368                         em->start, em->len, start, len);
6369                 err = -EIO;
6370                 goto out;
6371         }
6372
6373         err = 0;
6374         write_lock(&em_tree->lock);
6375         ret = add_extent_mapping(em_tree, em, 0);
6376         /* it is possible that someone inserted the extent into the tree
6377          * while we had the lock dropped.  It is also possible that
6378          * an overlapping map exists in the tree
6379          */
6380         if (ret == -EEXIST) {
6381                 struct extent_map *existing;
6382
6383                 ret = 0;
6384
6385                 existing = lookup_extent_mapping(em_tree, start, len);
6386                 if (existing && (existing->start > start ||
6387                     existing->start + existing->len <= start)) {
6388                         free_extent_map(existing);
6389                         existing = NULL;
6390                 }
6391                 if (!existing) {
6392                         existing = lookup_extent_mapping(em_tree, em->start,
6393                                                          em->len);
6394                         if (existing) {
6395                                 err = merge_extent_mapping(em_tree, existing,
6396                                                            em, start,
6397                                                            root->sectorsize);
6398                                 free_extent_map(existing);
6399                                 if (err) {
6400                                         free_extent_map(em);
6401                                         em = NULL;
6402                                 }
6403                         } else {
6404                                 err = -EIO;
6405                                 free_extent_map(em);
6406                                 em = NULL;
6407                         }
6408                 } else {
6409                         free_extent_map(em);
6410                         em = existing;
6411                         err = 0;
6412                 }
6413         }
6414         write_unlock(&em_tree->lock);
6415 out:
6416
6417         trace_btrfs_get_extent(root, em);
6418
6419         if (path)
6420                 btrfs_free_path(path);
6421         if (trans) {
6422                 ret = btrfs_end_transaction(trans, root);
6423                 if (!err)
6424                         err = ret;
6425         }
6426         if (err) {
6427                 free_extent_map(em);
6428                 return ERR_PTR(err);
6429         }
6430         BUG_ON(!em); /* Error is always set */
6431         return em;
6432 }
6433
6434 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6435                                            size_t pg_offset, u64 start, u64 len,
6436                                            int create)
6437 {
6438         struct extent_map *em;
6439         struct extent_map *hole_em = NULL;
6440         u64 range_start = start;
6441         u64 end;
6442         u64 found;
6443         u64 found_end;
6444         int err = 0;
6445
6446         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6447         if (IS_ERR(em))
6448                 return em;
6449         if (em) {
6450                 /*
6451                  * if our em maps to
6452                  * -  a hole or
6453                  * -  a pre-alloc extent,
6454                  * there might actually be delalloc bytes behind it.
6455                  */
6456                 if (em->block_start != EXTENT_MAP_HOLE &&
6457                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6458                         return em;
6459                 else
6460                         hole_em = em;
6461         }
6462
6463         /* check to see if we've wrapped (len == -1 or similar) */
6464         end = start + len;
6465         if (end < start)
6466                 end = (u64)-1;
6467         else
6468                 end -= 1;
6469
6470         em = NULL;
6471
6472         /* ok, we didn't find anything, lets look for delalloc */
6473         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6474                                  end, len, EXTENT_DELALLOC, 1);
6475         found_end = range_start + found;
6476         if (found_end < range_start)
6477                 found_end = (u64)-1;
6478
6479         /*
6480          * we didn't find anything useful, return
6481          * the original results from get_extent()
6482          */
6483         if (range_start > end || found_end <= start) {
6484                 em = hole_em;
6485                 hole_em = NULL;
6486                 goto out;
6487         }
6488
6489         /* adjust the range_start to make sure it doesn't
6490          * go backwards from the start they passed in
6491          */
6492         range_start = max(start, range_start);
6493         found = found_end - range_start;
6494
6495         if (found > 0) {
6496                 u64 hole_start = start;
6497                 u64 hole_len = len;
6498
6499                 em = alloc_extent_map();
6500                 if (!em) {
6501                         err = -ENOMEM;
6502                         goto out;
6503                 }
6504                 /*
6505                  * when btrfs_get_extent can't find anything it
6506                  * returns one huge hole
6507                  *
6508                  * make sure what it found really fits our range, and
6509                  * adjust to make sure it is based on the start from
6510                  * the caller
6511                  */
6512                 if (hole_em) {
6513                         u64 calc_end = extent_map_end(hole_em);
6514
6515                         if (calc_end <= start || (hole_em->start > end)) {
6516                                 free_extent_map(hole_em);
6517                                 hole_em = NULL;
6518                         } else {
6519                                 hole_start = max(hole_em->start, start);
6520                                 hole_len = calc_end - hole_start;
6521                         }
6522                 }
6523                 em->bdev = NULL;
6524                 if (hole_em && range_start > hole_start) {
6525                         /* our hole starts before our delalloc, so we
6526                          * have to return just the parts of the hole
6527                          * that go until  the delalloc starts
6528                          */
6529                         em->len = min(hole_len,
6530                                       range_start - hole_start);
6531                         em->start = hole_start;
6532                         em->orig_start = hole_start;
6533                         /*
6534                          * don't adjust block start at all,
6535                          * it is fixed at EXTENT_MAP_HOLE
6536                          */
6537                         em->block_start = hole_em->block_start;
6538                         em->block_len = hole_len;
6539                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6540                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6541                 } else {
6542                         em->start = range_start;
6543                         em->len = found;
6544                         em->orig_start = range_start;
6545                         em->block_start = EXTENT_MAP_DELALLOC;
6546                         em->block_len = found;
6547                 }
6548         } else if (hole_em) {
6549                 return hole_em;
6550         }
6551 out:
6552
6553         free_extent_map(hole_em);
6554         if (err) {
6555                 free_extent_map(em);
6556                 return ERR_PTR(err);
6557         }
6558         return em;
6559 }
6560
6561 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6562                                                   u64 start, u64 len)
6563 {
6564         struct btrfs_root *root = BTRFS_I(inode)->root;
6565         struct extent_map *em;
6566         struct btrfs_key ins;
6567         u64 alloc_hint;
6568         int ret;
6569
6570         alloc_hint = get_extent_allocation_hint(inode, start, len);
6571         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6572                                    alloc_hint, &ins, 1);
6573         if (ret)
6574                 return ERR_PTR(ret);
6575
6576         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6577                               ins.offset, ins.offset, ins.offset, 0);
6578         if (IS_ERR(em)) {
6579                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6580                 return em;
6581         }
6582
6583         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6584                                            ins.offset, ins.offset, 0);
6585         if (ret) {
6586                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6587                 free_extent_map(em);
6588                 return ERR_PTR(ret);
6589         }
6590
6591         return em;
6592 }
6593
6594 /*
6595  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6596  * block must be cow'd
6597  */
6598 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6599                               u64 *orig_start, u64 *orig_block_len,
6600                               u64 *ram_bytes)
6601 {
6602         struct btrfs_trans_handle *trans;
6603         struct btrfs_path *path;
6604         int ret;
6605         struct extent_buffer *leaf;
6606         struct btrfs_root *root = BTRFS_I(inode)->root;
6607         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6608         struct btrfs_file_extent_item *fi;
6609         struct btrfs_key key;
6610         u64 disk_bytenr;
6611         u64 backref_offset;
6612         u64 extent_end;
6613         u64 num_bytes;
6614         int slot;
6615         int found_type;
6616         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6617
6618         path = btrfs_alloc_path();
6619         if (!path)
6620                 return -ENOMEM;
6621
6622         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6623                                        offset, 0);
6624         if (ret < 0)
6625                 goto out;
6626
6627         slot = path->slots[0];
6628         if (ret == 1) {
6629                 if (slot == 0) {
6630                         /* can't find the item, must cow */
6631                         ret = 0;
6632                         goto out;
6633                 }
6634                 slot--;
6635         }
6636         ret = 0;
6637         leaf = path->nodes[0];
6638         btrfs_item_key_to_cpu(leaf, &key, slot);
6639         if (key.objectid != btrfs_ino(inode) ||
6640             key.type != BTRFS_EXTENT_DATA_KEY) {
6641                 /* not our file or wrong item type, must cow */
6642                 goto out;
6643         }
6644
6645         if (key.offset > offset) {
6646                 /* Wrong offset, must cow */
6647                 goto out;
6648         }
6649
6650         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6651         found_type = btrfs_file_extent_type(leaf, fi);
6652         if (found_type != BTRFS_FILE_EXTENT_REG &&
6653             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6654                 /* not a regular extent, must cow */
6655                 goto out;
6656         }
6657
6658         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6659                 goto out;
6660
6661         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6662         if (extent_end <= offset)
6663                 goto out;
6664
6665         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6666         if (disk_bytenr == 0)
6667                 goto out;
6668
6669         if (btrfs_file_extent_compression(leaf, fi) ||
6670             btrfs_file_extent_encryption(leaf, fi) ||
6671             btrfs_file_extent_other_encoding(leaf, fi))
6672                 goto out;
6673
6674         backref_offset = btrfs_file_extent_offset(leaf, fi);
6675
6676         if (orig_start) {
6677                 *orig_start = key.offset - backref_offset;
6678                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6679                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6680         }
6681
6682         if (btrfs_extent_readonly(root, disk_bytenr))
6683                 goto out;
6684
6685         num_bytes = min(offset + *len, extent_end) - offset;
6686         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6687                 u64 range_end;
6688
6689                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6690                 ret = test_range_bit(io_tree, offset, range_end,
6691                                      EXTENT_DELALLOC, 0, NULL);
6692                 if (ret) {
6693                         ret = -EAGAIN;
6694                         goto out;
6695                 }
6696         }
6697
6698         btrfs_release_path(path);
6699
6700         /*
6701          * look for other files referencing this extent, if we
6702          * find any we must cow
6703          */
6704         trans = btrfs_join_transaction(root);
6705         if (IS_ERR(trans)) {
6706                 ret = 0;
6707                 goto out;
6708         }
6709
6710         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6711                                     key.offset - backref_offset, disk_bytenr);
6712         btrfs_end_transaction(trans, root);
6713         if (ret) {
6714                 ret = 0;
6715                 goto out;
6716         }
6717
6718         /*
6719          * adjust disk_bytenr and num_bytes to cover just the bytes
6720          * in this extent we are about to write.  If there
6721          * are any csums in that range we have to cow in order
6722          * to keep the csums correct
6723          */
6724         disk_bytenr += backref_offset;
6725         disk_bytenr += offset - key.offset;
6726         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6727                                 goto out;
6728         /*
6729          * all of the above have passed, it is safe to overwrite this extent
6730          * without cow
6731          */
6732         *len = num_bytes;
6733         ret = 1;
6734 out:
6735         btrfs_free_path(path);
6736         return ret;
6737 }
6738
6739 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6740 {
6741         struct radix_tree_root *root = &inode->i_mapping->page_tree;
6742         int found = false;
6743         void **pagep = NULL;
6744         struct page *page = NULL;
6745         int start_idx;
6746         int end_idx;
6747
6748         start_idx = start >> PAGE_CACHE_SHIFT;
6749
6750         /*
6751          * end is the last byte in the last page.  end == start is legal
6752          */
6753         end_idx = end >> PAGE_CACHE_SHIFT;
6754
6755         rcu_read_lock();
6756
6757         /* Most of the code in this while loop is lifted from
6758          * find_get_page.  It's been modified to begin searching from a
6759          * page and return just the first page found in that range.  If the
6760          * found idx is less than or equal to the end idx then we know that
6761          * a page exists.  If no pages are found or if those pages are
6762          * outside of the range then we're fine (yay!) */
6763         while (page == NULL &&
6764                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6765                 page = radix_tree_deref_slot(pagep);
6766                 if (unlikely(!page))
6767                         break;
6768
6769                 if (radix_tree_exception(page)) {
6770                         if (radix_tree_deref_retry(page))
6771                                 continue;
6772                         /*
6773                          * Otherwise, shmem/tmpfs must be storing a swap entry
6774                          * here as an exceptional entry: so return it without
6775                          * attempting to raise page count.
6776                          */
6777                         break; /* TODO: Is this relevant for this use case? */
6778                 }
6779
6780                 if (!page_cache_get_speculative(page)) {
6781                         page = NULL;
6782                         continue;
6783                 }
6784
6785                 /*
6786                  * Has the page moved?
6787                  * This is part of the lockless pagecache protocol. See
6788                  * include/linux/pagemap.h for details.
6789                  */
6790                 if (unlikely(page != *pagep)) {
6791                         page_cache_release(page);
6792                         page = NULL;
6793                 }
6794         }
6795
6796         if (page) {
6797                 if (page->index <= end_idx)
6798                         found = true;
6799                 page_cache_release(page);
6800         }
6801
6802         rcu_read_unlock();
6803         return found;
6804 }
6805
6806 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6807                               struct extent_state **cached_state, int writing)
6808 {
6809         struct btrfs_ordered_extent *ordered;
6810         int ret = 0;
6811
6812         while (1) {
6813                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6814                                  0, cached_state);
6815                 /*
6816                  * We're concerned with the entire range that we're going to be
6817                  * doing DIO to, so we need to make sure theres no ordered
6818                  * extents in this range.
6819                  */
6820                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6821                                                      lockend - lockstart + 1);
6822
6823                 /*
6824                  * We need to make sure there are no buffered pages in this
6825                  * range either, we could have raced between the invalidate in
6826                  * generic_file_direct_write and locking the extent.  The
6827                  * invalidate needs to happen so that reads after a write do not
6828                  * get stale data.
6829                  */
6830                 if (!ordered &&
6831                     (!writing ||
6832                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
6833                         break;
6834
6835                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6836                                      cached_state, GFP_NOFS);
6837
6838                 if (ordered) {
6839                         btrfs_start_ordered_extent(inode, ordered, 1);
6840                         btrfs_put_ordered_extent(ordered);
6841                 } else {
6842                         /* Screw you mmap */
6843                         ret = filemap_write_and_wait_range(inode->i_mapping,
6844                                                            lockstart,
6845                                                            lockend);
6846                         if (ret)
6847                                 break;
6848
6849                         /*
6850                          * If we found a page that couldn't be invalidated just
6851                          * fall back to buffered.
6852                          */
6853                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6854                                         lockstart >> PAGE_CACHE_SHIFT,
6855                                         lockend >> PAGE_CACHE_SHIFT);
6856                         if (ret)
6857                                 break;
6858                 }
6859
6860                 cond_resched();
6861         }
6862
6863         return ret;
6864 }
6865
6866 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6867                                            u64 len, u64 orig_start,
6868                                            u64 block_start, u64 block_len,
6869                                            u64 orig_block_len, u64 ram_bytes,
6870                                            int type)
6871 {
6872         struct extent_map_tree *em_tree;
6873         struct extent_map *em;
6874         struct btrfs_root *root = BTRFS_I(inode)->root;
6875         int ret;
6876
6877         em_tree = &BTRFS_I(inode)->extent_tree;
6878         em = alloc_extent_map();
6879         if (!em)
6880                 return ERR_PTR(-ENOMEM);
6881
6882         em->start = start;
6883         em->orig_start = orig_start;
6884         em->mod_start = start;
6885         em->mod_len = len;
6886         em->len = len;
6887         em->block_len = block_len;
6888         em->block_start = block_start;
6889         em->bdev = root->fs_info->fs_devices->latest_bdev;
6890         em->orig_block_len = orig_block_len;
6891         em->ram_bytes = ram_bytes;
6892         em->generation = -1;
6893         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6894         if (type == BTRFS_ORDERED_PREALLOC)
6895                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6896
6897         do {
6898                 btrfs_drop_extent_cache(inode, em->start,
6899                                 em->start + em->len - 1, 0);
6900                 write_lock(&em_tree->lock);
6901                 ret = add_extent_mapping(em_tree, em, 1);
6902                 write_unlock(&em_tree->lock);
6903         } while (ret == -EEXIST);
6904
6905         if (ret) {
6906                 free_extent_map(em);
6907                 return ERR_PTR(ret);
6908         }
6909
6910         return em;
6911 }
6912
6913
6914 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6915                                    struct buffer_head *bh_result, int create)
6916 {
6917         struct extent_map *em;
6918         struct btrfs_root *root = BTRFS_I(inode)->root;
6919         struct extent_state *cached_state = NULL;
6920         u64 start = iblock << inode->i_blkbits;
6921         u64 lockstart, lockend;
6922         u64 len = bh_result->b_size;
6923         int unlock_bits = EXTENT_LOCKED;
6924         int ret = 0;
6925
6926         if (create)
6927                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6928         else
6929                 len = min_t(u64, len, root->sectorsize);
6930
6931         lockstart = start;
6932         lockend = start + len - 1;
6933
6934         /*
6935          * If this errors out it's because we couldn't invalidate pagecache for
6936          * this range and we need to fallback to buffered.
6937          */
6938         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6939                 return -ENOTBLK;
6940
6941         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6942         if (IS_ERR(em)) {
6943                 ret = PTR_ERR(em);
6944                 goto unlock_err;
6945         }
6946
6947         /*
6948          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6949          * io.  INLINE is special, and we could probably kludge it in here, but
6950          * it's still buffered so for safety lets just fall back to the generic
6951          * buffered path.
6952          *
6953          * For COMPRESSED we _have_ to read the entire extent in so we can
6954          * decompress it, so there will be buffering required no matter what we
6955          * do, so go ahead and fallback to buffered.
6956          *
6957          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6958          * to buffered IO.  Don't blame me, this is the price we pay for using
6959          * the generic code.
6960          */
6961         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6962             em->block_start == EXTENT_MAP_INLINE) {
6963                 free_extent_map(em);
6964                 ret = -ENOTBLK;
6965                 goto unlock_err;
6966         }
6967
6968         /* Just a good old fashioned hole, return */
6969         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6970                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6971                 free_extent_map(em);
6972                 goto unlock_err;
6973         }
6974
6975         /*
6976          * We don't allocate a new extent in the following cases
6977          *
6978          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6979          * existing extent.
6980          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6981          * just use the extent.
6982          *
6983          */
6984         if (!create) {
6985                 len = min(len, em->len - (start - em->start));
6986                 lockstart = start + len;
6987                 goto unlock;
6988         }
6989
6990         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6991             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6992              em->block_start != EXTENT_MAP_HOLE)) {
6993                 int type;
6994                 int ret;
6995                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6996
6997                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6998                         type = BTRFS_ORDERED_PREALLOC;
6999                 else
7000                         type = BTRFS_ORDERED_NOCOW;
7001                 len = min(len, em->len - (start - em->start));
7002                 block_start = em->block_start + (start - em->start);
7003
7004                 if (can_nocow_extent(inode, start, &len, &orig_start,
7005                                      &orig_block_len, &ram_bytes) == 1) {
7006                         if (type == BTRFS_ORDERED_PREALLOC) {
7007                                 free_extent_map(em);
7008                                 em = create_pinned_em(inode, start, len,
7009                                                        orig_start,
7010                                                        block_start, len,
7011                                                        orig_block_len,
7012                                                        ram_bytes, type);
7013                                 if (IS_ERR(em))
7014                                         goto unlock_err;
7015                         }
7016
7017                         ret = btrfs_add_ordered_extent_dio(inode, start,
7018                                            block_start, len, len, type);
7019                         if (ret) {
7020                                 free_extent_map(em);
7021                                 goto unlock_err;
7022                         }
7023                         goto unlock;
7024                 }
7025         }
7026
7027         /*
7028          * this will cow the extent, reset the len in case we changed
7029          * it above
7030          */
7031         len = bh_result->b_size;
7032         free_extent_map(em);
7033         em = btrfs_new_extent_direct(inode, start, len);
7034         if (IS_ERR(em)) {
7035                 ret = PTR_ERR(em);
7036                 goto unlock_err;
7037         }
7038         len = min(len, em->len - (start - em->start));
7039 unlock:
7040         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7041                 inode->i_blkbits;
7042         bh_result->b_size = len;
7043         bh_result->b_bdev = em->bdev;
7044         set_buffer_mapped(bh_result);
7045         if (create) {
7046                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7047                         set_buffer_new(bh_result);
7048
7049                 /*
7050                  * Need to update the i_size under the extent lock so buffered
7051                  * readers will get the updated i_size when we unlock.
7052                  */
7053                 if (start + len > i_size_read(inode))
7054                         i_size_write(inode, start + len);
7055
7056                 spin_lock(&BTRFS_I(inode)->lock);
7057                 BTRFS_I(inode)->outstanding_extents++;
7058                 spin_unlock(&BTRFS_I(inode)->lock);
7059
7060                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7061                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
7062                                      &cached_state, GFP_NOFS);
7063                 BUG_ON(ret);
7064         }
7065
7066         /*
7067          * In the case of write we need to clear and unlock the entire range,
7068          * in the case of read we need to unlock only the end area that we
7069          * aren't using if there is any left over space.
7070          */
7071         if (lockstart < lockend) {
7072                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7073                                  lockend, unlock_bits, 1, 0,
7074                                  &cached_state, GFP_NOFS);
7075         } else {
7076                 free_extent_state(cached_state);
7077         }
7078
7079         free_extent_map(em);
7080
7081         return 0;
7082
7083 unlock_err:
7084         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7085                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7086         return ret;
7087 }
7088
7089 static void btrfs_endio_direct_read(struct bio *bio, int err)
7090 {
7091         struct btrfs_dio_private *dip = bio->bi_private;
7092         struct bio_vec *bvec;
7093         struct inode *inode = dip->inode;
7094         struct btrfs_root *root = BTRFS_I(inode)->root;
7095         struct bio *dio_bio;
7096         u32 *csums = (u32 *)dip->csum;
7097         u64 start;
7098         int i;
7099
7100         start = dip->logical_offset;
7101         bio_for_each_segment_all(bvec, bio, i) {
7102                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7103                         struct page *page = bvec->bv_page;
7104                         char *kaddr;
7105                         u32 csum = ~(u32)0;
7106                         unsigned long flags;
7107
7108                         local_irq_save(flags);
7109                         kaddr = kmap_atomic(page);
7110                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7111                                                csum, bvec->bv_len);
7112                         btrfs_csum_final(csum, (char *)&csum);
7113                         kunmap_atomic(kaddr);
7114                         local_irq_restore(flags);
7115
7116                         flush_dcache_page(bvec->bv_page);
7117                         if (csum != csums[i]) {
7118                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7119                                           btrfs_ino(inode), start, csum,
7120                                           csums[i]);
7121                                 err = -EIO;
7122                         }
7123                 }
7124
7125                 start += bvec->bv_len;
7126         }
7127
7128         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7129                       dip->logical_offset + dip->bytes - 1);
7130         dio_bio = dip->dio_bio;
7131
7132         kfree(dip);
7133
7134         /* If we had a csum failure make sure to clear the uptodate flag */
7135         if (err)
7136                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7137         dio_end_io(dio_bio, err);
7138         bio_put(bio);
7139 }
7140
7141 static void btrfs_endio_direct_write(struct bio *bio, int err)
7142 {
7143         struct btrfs_dio_private *dip = bio->bi_private;
7144         struct inode *inode = dip->inode;
7145         struct btrfs_root *root = BTRFS_I(inode)->root;
7146         struct btrfs_ordered_extent *ordered = NULL;
7147         u64 ordered_offset = dip->logical_offset;
7148         u64 ordered_bytes = dip->bytes;
7149         struct bio *dio_bio;
7150         int ret;
7151
7152         if (err)
7153                 goto out_done;
7154 again:
7155         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7156                                                    &ordered_offset,
7157                                                    ordered_bytes, !err);
7158         if (!ret)
7159                 goto out_test;
7160
7161         btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7162         btrfs_queue_work(root->fs_info->endio_write_workers,
7163                          &ordered->work);
7164 out_test:
7165         /*
7166          * our bio might span multiple ordered extents.  If we haven't
7167          * completed the accounting for the whole dio, go back and try again
7168          */
7169         if (ordered_offset < dip->logical_offset + dip->bytes) {
7170                 ordered_bytes = dip->logical_offset + dip->bytes -
7171                         ordered_offset;
7172                 ordered = NULL;
7173                 goto again;
7174         }
7175 out_done:
7176         dio_bio = dip->dio_bio;
7177
7178         kfree(dip);
7179
7180         /* If we had an error make sure to clear the uptodate flag */
7181         if (err)
7182                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7183         dio_end_io(dio_bio, err);
7184         bio_put(bio);
7185 }
7186
7187 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7188                                     struct bio *bio, int mirror_num,
7189                                     unsigned long bio_flags, u64 offset)
7190 {
7191         int ret;
7192         struct btrfs_root *root = BTRFS_I(inode)->root;
7193         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7194         BUG_ON(ret); /* -ENOMEM */
7195         return 0;
7196 }
7197
7198 static void btrfs_end_dio_bio(struct bio *bio, int err)
7199 {
7200         struct btrfs_dio_private *dip = bio->bi_private;
7201
7202         if (err) {
7203                 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7204                           "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7205                       btrfs_ino(dip->inode), bio->bi_rw,
7206                       (unsigned long long)bio->bi_iter.bi_sector,
7207                       bio->bi_iter.bi_size, err);
7208                 dip->errors = 1;
7209
7210                 /*
7211                  * before atomic variable goto zero, we must make sure
7212                  * dip->errors is perceived to be set.
7213                  */
7214                 smp_mb__before_atomic_dec();
7215         }
7216
7217         /* if there are more bios still pending for this dio, just exit */
7218         if (!atomic_dec_and_test(&dip->pending_bios))
7219                 goto out;
7220
7221         if (dip->errors) {
7222                 bio_io_error(dip->orig_bio);
7223         } else {
7224                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7225                 bio_endio(dip->orig_bio, 0);
7226         }
7227 out:
7228         bio_put(bio);
7229 }
7230
7231 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7232                                        u64 first_sector, gfp_t gfp_flags)
7233 {
7234         int nr_vecs = bio_get_nr_vecs(bdev);
7235         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7236 }
7237
7238 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7239                                          int rw, u64 file_offset, int skip_sum,
7240                                          int async_submit)
7241 {
7242         struct btrfs_dio_private *dip = bio->bi_private;
7243         int write = rw & REQ_WRITE;
7244         struct btrfs_root *root = BTRFS_I(inode)->root;
7245         int ret;
7246
7247         if (async_submit)
7248                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7249
7250         bio_get(bio);
7251
7252         if (!write) {
7253                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7254                 if (ret)
7255                         goto err;
7256         }
7257
7258         if (skip_sum)
7259                 goto map;
7260
7261         if (write && async_submit) {
7262                 ret = btrfs_wq_submit_bio(root->fs_info,
7263                                    inode, rw, bio, 0, 0,
7264                                    file_offset,
7265                                    __btrfs_submit_bio_start_direct_io,
7266                                    __btrfs_submit_bio_done);
7267                 goto err;
7268         } else if (write) {
7269                 /*
7270                  * If we aren't doing async submit, calculate the csum of the
7271                  * bio now.
7272                  */
7273                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7274                 if (ret)
7275                         goto err;
7276         } else if (!skip_sum) {
7277                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7278                                                 file_offset);
7279                 if (ret)
7280                         goto err;
7281         }
7282
7283 map:
7284         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7285 err:
7286         bio_put(bio);
7287         return ret;
7288 }
7289
7290 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7291                                     int skip_sum)
7292 {
7293         struct inode *inode = dip->inode;
7294         struct btrfs_root *root = BTRFS_I(inode)->root;
7295         struct bio *bio;
7296         struct bio *orig_bio = dip->orig_bio;
7297         struct bio_vec *bvec = orig_bio->bi_io_vec;
7298         u64 start_sector = orig_bio->bi_iter.bi_sector;
7299         u64 file_offset = dip->logical_offset;
7300         u64 submit_len = 0;
7301         u64 map_length;
7302         int nr_pages = 0;
7303         int ret = 0;
7304         int async_submit = 0;
7305
7306         map_length = orig_bio->bi_iter.bi_size;
7307         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7308                               &map_length, NULL, 0);
7309         if (ret) {
7310                 bio_put(orig_bio);
7311                 return -EIO;
7312         }
7313
7314         if (map_length >= orig_bio->bi_iter.bi_size) {
7315                 bio = orig_bio;
7316                 goto submit;
7317         }
7318
7319         /* async crcs make it difficult to collect full stripe writes. */
7320         if (btrfs_get_alloc_profile(root, 1) &
7321             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7322                 async_submit = 0;
7323         else
7324                 async_submit = 1;
7325
7326         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7327         if (!bio)
7328                 return -ENOMEM;
7329         bio->bi_private = dip;
7330         bio->bi_end_io = btrfs_end_dio_bio;
7331         atomic_inc(&dip->pending_bios);
7332
7333         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7334                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7335                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7336                                  bvec->bv_offset) < bvec->bv_len)) {
7337                         /*
7338                          * inc the count before we submit the bio so
7339                          * we know the end IO handler won't happen before
7340                          * we inc the count. Otherwise, the dip might get freed
7341                          * before we're done setting it up
7342                          */
7343                         atomic_inc(&dip->pending_bios);
7344                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7345                                                      file_offset, skip_sum,
7346                                                      async_submit);
7347                         if (ret) {
7348                                 bio_put(bio);
7349                                 atomic_dec(&dip->pending_bios);
7350                                 goto out_err;
7351                         }
7352
7353                         start_sector += submit_len >> 9;
7354                         file_offset += submit_len;
7355
7356                         submit_len = 0;
7357                         nr_pages = 0;
7358
7359                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7360                                                   start_sector, GFP_NOFS);
7361                         if (!bio)
7362                                 goto out_err;
7363                         bio->bi_private = dip;
7364                         bio->bi_end_io = btrfs_end_dio_bio;
7365
7366                         map_length = orig_bio->bi_iter.bi_size;
7367                         ret = btrfs_map_block(root->fs_info, rw,
7368                                               start_sector << 9,
7369                                               &map_length, NULL, 0);
7370                         if (ret) {
7371                                 bio_put(bio);
7372                                 goto out_err;
7373                         }
7374                 } else {
7375                         submit_len += bvec->bv_len;
7376                         nr_pages++;
7377                         bvec++;
7378                 }
7379         }
7380
7381 submit:
7382         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7383                                      async_submit);
7384         if (!ret)
7385                 return 0;
7386
7387         bio_put(bio);
7388 out_err:
7389         dip->errors = 1;
7390         /*
7391          * before atomic variable goto zero, we must
7392          * make sure dip->errors is perceived to be set.
7393          */
7394         smp_mb__before_atomic_dec();
7395         if (atomic_dec_and_test(&dip->pending_bios))
7396                 bio_io_error(dip->orig_bio);
7397
7398         /* bio_end_io() will handle error, so we needn't return it */
7399         return 0;
7400 }
7401
7402 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7403                                 struct inode *inode, loff_t file_offset)
7404 {
7405         struct btrfs_root *root = BTRFS_I(inode)->root;
7406         struct btrfs_dio_private *dip;
7407         struct bio *io_bio;
7408         int skip_sum;
7409         int sum_len;
7410         int write = rw & REQ_WRITE;
7411         int ret = 0;
7412         u16 csum_size;
7413
7414         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7415
7416         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7417         if (!io_bio) {
7418                 ret = -ENOMEM;
7419                 goto free_ordered;
7420         }
7421
7422         if (!skip_sum && !write) {
7423                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7424                 sum_len = dio_bio->bi_iter.bi_size >>
7425                         inode->i_sb->s_blocksize_bits;
7426                 sum_len *= csum_size;
7427         } else {
7428                 sum_len = 0;
7429         }
7430
7431         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7432         if (!dip) {
7433                 ret = -ENOMEM;
7434                 goto free_io_bio;
7435         }
7436
7437         dip->private = dio_bio->bi_private;
7438         dip->inode = inode;
7439         dip->logical_offset = file_offset;
7440         dip->bytes = dio_bio->bi_iter.bi_size;
7441         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7442         io_bio->bi_private = dip;
7443         dip->errors = 0;
7444         dip->orig_bio = io_bio;
7445         dip->dio_bio = dio_bio;
7446         atomic_set(&dip->pending_bios, 0);
7447
7448         if (write)
7449                 io_bio->bi_end_io = btrfs_endio_direct_write;
7450         else
7451                 io_bio->bi_end_io = btrfs_endio_direct_read;
7452
7453         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7454         if (!ret)
7455                 return;
7456
7457 free_io_bio:
7458         bio_put(io_bio);
7459
7460 free_ordered:
7461         /*
7462          * If this is a write, we need to clean up the reserved space and kill
7463          * the ordered extent.
7464          */
7465         if (write) {
7466                 struct btrfs_ordered_extent *ordered;
7467                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7468                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7469                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7470                         btrfs_free_reserved_extent(root, ordered->start,
7471                                                    ordered->disk_len);
7472                 btrfs_put_ordered_extent(ordered);
7473                 btrfs_put_ordered_extent(ordered);
7474         }
7475         bio_endio(dio_bio, ret);
7476 }
7477
7478 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7479                         const struct iovec *iov, loff_t offset,
7480                         unsigned long nr_segs)
7481 {
7482         int seg;
7483         int i;
7484         size_t size;
7485         unsigned long addr;
7486         unsigned blocksize_mask = root->sectorsize - 1;
7487         ssize_t retval = -EINVAL;
7488         loff_t end = offset;
7489
7490         if (offset & blocksize_mask)
7491                 goto out;
7492
7493         /* Check the memory alignment.  Blocks cannot straddle pages */
7494         for (seg = 0; seg < nr_segs; seg++) {
7495                 addr = (unsigned long)iov[seg].iov_base;
7496                 size = iov[seg].iov_len;
7497                 end += size;
7498                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7499                         goto out;
7500
7501                 /* If this is a write we don't need to check anymore */
7502                 if (rw & WRITE)
7503                         continue;
7504
7505                 /*
7506                  * Check to make sure we don't have duplicate iov_base's in this
7507                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7508                  * when reading back.
7509                  */
7510                 for (i = seg + 1; i < nr_segs; i++) {
7511                         if (iov[seg].iov_base == iov[i].iov_base)
7512                                 goto out;
7513                 }
7514         }
7515         retval = 0;
7516 out:
7517         return retval;
7518 }
7519
7520 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7521                         const struct iovec *iov, loff_t offset,
7522                         unsigned long nr_segs)
7523 {
7524         struct file *file = iocb->ki_filp;
7525         struct inode *inode = file->f_mapping->host;
7526         size_t count = 0;
7527         int flags = 0;
7528         bool wakeup = true;
7529         bool relock = false;
7530         ssize_t ret;
7531
7532         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7533                             offset, nr_segs))
7534                 return 0;
7535
7536         atomic_inc(&inode->i_dio_count);
7537         smp_mb__after_atomic_inc();
7538
7539         /*
7540          * The generic stuff only does filemap_write_and_wait_range, which
7541          * isn't enough if we've written compressed pages to this area, so
7542          * we need to flush the dirty pages again to make absolutely sure
7543          * that any outstanding dirty pages are on disk.
7544          */
7545         count = iov_length(iov, nr_segs);
7546         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7547                      &BTRFS_I(inode)->runtime_flags))
7548                 filemap_fdatawrite_range(inode->i_mapping, offset, count);
7549
7550         if (rw & WRITE) {
7551                 /*
7552                  * If the write DIO is beyond the EOF, we need update
7553                  * the isize, but it is protected by i_mutex. So we can
7554                  * not unlock the i_mutex at this case.
7555                  */
7556                 if (offset + count <= inode->i_size) {
7557                         mutex_unlock(&inode->i_mutex);
7558                         relock = true;
7559                 }
7560                 ret = btrfs_delalloc_reserve_space(inode, count);
7561                 if (ret)
7562                         goto out;
7563         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7564                                      &BTRFS_I(inode)->runtime_flags))) {
7565                 inode_dio_done(inode);
7566                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7567                 wakeup = false;
7568         }
7569
7570         ret = __blockdev_direct_IO(rw, iocb, inode,
7571                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7572                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7573                         btrfs_submit_direct, flags);
7574         if (rw & WRITE) {
7575                 if (ret < 0 && ret != -EIOCBQUEUED)
7576                         btrfs_delalloc_release_space(inode, count);
7577                 else if (ret >= 0 && (size_t)ret < count)
7578                         btrfs_delalloc_release_space(inode,
7579                                                      count - (size_t)ret);
7580                 else
7581                         btrfs_delalloc_release_metadata(inode, 0);
7582         }
7583 out:
7584         if (wakeup)
7585                 inode_dio_done(inode);
7586         if (relock)
7587                 mutex_lock(&inode->i_mutex);
7588
7589         return ret;
7590 }
7591
7592 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7593
7594 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7595                 __u64 start, __u64 len)
7596 {
7597         int     ret;
7598
7599         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7600         if (ret)
7601                 return ret;
7602
7603         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7604 }
7605
7606 int btrfs_readpage(struct file *file, struct page *page)
7607 {
7608         struct extent_io_tree *tree;
7609         tree = &BTRFS_I(page->mapping->host)->io_tree;
7610         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7611 }
7612
7613 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7614 {
7615         struct extent_io_tree *tree;
7616
7617
7618         if (current->flags & PF_MEMALLOC) {
7619                 redirty_page_for_writepage(wbc, page);
7620                 unlock_page(page);
7621                 return 0;
7622         }
7623         tree = &BTRFS_I(page->mapping->host)->io_tree;
7624         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7625 }
7626
7627 static int btrfs_writepages(struct address_space *mapping,
7628                             struct writeback_control *wbc)
7629 {
7630         struct extent_io_tree *tree;
7631
7632         tree = &BTRFS_I(mapping->host)->io_tree;
7633         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7634 }
7635
7636 static int
7637 btrfs_readpages(struct file *file, struct address_space *mapping,
7638                 struct list_head *pages, unsigned nr_pages)
7639 {
7640         struct extent_io_tree *tree;
7641         tree = &BTRFS_I(mapping->host)->io_tree;
7642         return extent_readpages(tree, mapping, pages, nr_pages,
7643                                 btrfs_get_extent);
7644 }
7645 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7646 {
7647         struct extent_io_tree *tree;
7648         struct extent_map_tree *map;
7649         int ret;
7650
7651         tree = &BTRFS_I(page->mapping->host)->io_tree;
7652         map = &BTRFS_I(page->mapping->host)->extent_tree;
7653         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7654         if (ret == 1) {
7655                 ClearPagePrivate(page);
7656                 set_page_private(page, 0);
7657                 page_cache_release(page);
7658         }
7659         return ret;
7660 }
7661
7662 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7663 {
7664         if (PageWriteback(page) || PageDirty(page))
7665                 return 0;
7666         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7667 }
7668
7669 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7670                                  unsigned int length)
7671 {
7672         struct inode *inode = page->mapping->host;
7673         struct extent_io_tree *tree;
7674         struct btrfs_ordered_extent *ordered;
7675         struct extent_state *cached_state = NULL;
7676         u64 page_start = page_offset(page);
7677         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7678         int inode_evicting = inode->i_state & I_FREEING;
7679
7680         /*
7681          * we have the page locked, so new writeback can't start,
7682          * and the dirty bit won't be cleared while we are here.
7683          *
7684          * Wait for IO on this page so that we can safely clear
7685          * the PagePrivate2 bit and do ordered accounting
7686          */
7687         wait_on_page_writeback(page);
7688
7689         tree = &BTRFS_I(inode)->io_tree;
7690         if (offset) {
7691                 btrfs_releasepage(page, GFP_NOFS);
7692                 return;
7693         }
7694
7695         if (!inode_evicting)
7696                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7697         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7698         if (ordered) {
7699                 /*
7700                  * IO on this page will never be started, so we need
7701                  * to account for any ordered extents now
7702                  */
7703                 if (!inode_evicting)
7704                         clear_extent_bit(tree, page_start, page_end,
7705                                          EXTENT_DIRTY | EXTENT_DELALLOC |
7706                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7707                                          EXTENT_DEFRAG, 1, 0, &cached_state,
7708                                          GFP_NOFS);
7709                 /*
7710                  * whoever cleared the private bit is responsible
7711                  * for the finish_ordered_io
7712                  */
7713                 if (TestClearPagePrivate2(page)) {
7714                         struct btrfs_ordered_inode_tree *tree;
7715                         u64 new_len;
7716
7717                         tree = &BTRFS_I(inode)->ordered_tree;
7718
7719                         spin_lock_irq(&tree->lock);
7720                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7721                         new_len = page_start - ordered->file_offset;
7722                         if (new_len < ordered->truncated_len)
7723                                 ordered->truncated_len = new_len;
7724                         spin_unlock_irq(&tree->lock);
7725
7726                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7727                                                            page_start,
7728                                                            PAGE_CACHE_SIZE, 1))
7729                                 btrfs_finish_ordered_io(ordered);
7730                 }
7731                 btrfs_put_ordered_extent(ordered);
7732                 if (!inode_evicting) {
7733                         cached_state = NULL;
7734                         lock_extent_bits(tree, page_start, page_end, 0,
7735                                          &cached_state);
7736                 }
7737         }
7738
7739         if (!inode_evicting) {
7740                 clear_extent_bit(tree, page_start, page_end,
7741                                  EXTENT_LOCKED | EXTENT_DIRTY |
7742                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7743                                  EXTENT_DEFRAG, 1, 1,
7744                                  &cached_state, GFP_NOFS);
7745
7746                 __btrfs_releasepage(page, GFP_NOFS);
7747         }
7748
7749         ClearPageChecked(page);
7750         if (PagePrivate(page)) {
7751                 ClearPagePrivate(page);
7752                 set_page_private(page, 0);
7753                 page_cache_release(page);
7754         }
7755 }
7756
7757 /*
7758  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7759  * called from a page fault handler when a page is first dirtied. Hence we must
7760  * be careful to check for EOF conditions here. We set the page up correctly
7761  * for a written page which means we get ENOSPC checking when writing into
7762  * holes and correct delalloc and unwritten extent mapping on filesystems that
7763  * support these features.
7764  *
7765  * We are not allowed to take the i_mutex here so we have to play games to
7766  * protect against truncate races as the page could now be beyond EOF.  Because
7767  * vmtruncate() writes the inode size before removing pages, once we have the
7768  * page lock we can determine safely if the page is beyond EOF. If it is not
7769  * beyond EOF, then the page is guaranteed safe against truncation until we
7770  * unlock the page.
7771  */
7772 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7773 {
7774         struct page *page = vmf->page;
7775         struct inode *inode = file_inode(vma->vm_file);
7776         struct btrfs_root *root = BTRFS_I(inode)->root;
7777         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7778         struct btrfs_ordered_extent *ordered;
7779         struct extent_state *cached_state = NULL;
7780         char *kaddr;
7781         unsigned long zero_start;
7782         loff_t size;
7783         int ret;
7784         int reserved = 0;
7785         u64 page_start;
7786         u64 page_end;
7787
7788         sb_start_pagefault(inode->i_sb);
7789         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7790         if (!ret) {
7791                 ret = file_update_time(vma->vm_file);
7792                 reserved = 1;
7793         }
7794         if (ret) {
7795                 if (ret == -ENOMEM)
7796                         ret = VM_FAULT_OOM;
7797                 else /* -ENOSPC, -EIO, etc */
7798                         ret = VM_FAULT_SIGBUS;
7799                 if (reserved)
7800                         goto out;
7801                 goto out_noreserve;
7802         }
7803
7804         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7805 again:
7806         lock_page(page);
7807         size = i_size_read(inode);
7808         page_start = page_offset(page);
7809         page_end = page_start + PAGE_CACHE_SIZE - 1;
7810
7811         if ((page->mapping != inode->i_mapping) ||
7812             (page_start >= size)) {
7813                 /* page got truncated out from underneath us */
7814                 goto out_unlock;
7815         }
7816         wait_on_page_writeback(page);
7817
7818         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7819         set_page_extent_mapped(page);
7820
7821         /*
7822          * we can't set the delalloc bits if there are pending ordered
7823          * extents.  Drop our locks and wait for them to finish
7824          */
7825         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7826         if (ordered) {
7827                 unlock_extent_cached(io_tree, page_start, page_end,
7828                                      &cached_state, GFP_NOFS);
7829                 unlock_page(page);
7830                 btrfs_start_ordered_extent(inode, ordered, 1);
7831                 btrfs_put_ordered_extent(ordered);
7832                 goto again;
7833         }
7834
7835         /*
7836          * XXX - page_mkwrite gets called every time the page is dirtied, even
7837          * if it was already dirty, so for space accounting reasons we need to
7838          * clear any delalloc bits for the range we are fixing to save.  There
7839          * is probably a better way to do this, but for now keep consistent with
7840          * prepare_pages in the normal write path.
7841          */
7842         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7843                           EXTENT_DIRTY | EXTENT_DELALLOC |
7844                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7845                           0, 0, &cached_state, GFP_NOFS);
7846
7847         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7848                                         &cached_state);
7849         if (ret) {
7850                 unlock_extent_cached(io_tree, page_start, page_end,
7851                                      &cached_state, GFP_NOFS);
7852                 ret = VM_FAULT_SIGBUS;
7853                 goto out_unlock;
7854         }
7855         ret = 0;
7856
7857         /* page is wholly or partially inside EOF */
7858         if (page_start + PAGE_CACHE_SIZE > size)
7859                 zero_start = size & ~PAGE_CACHE_MASK;
7860         else
7861                 zero_start = PAGE_CACHE_SIZE;
7862
7863         if (zero_start != PAGE_CACHE_SIZE) {
7864                 kaddr = kmap(page);
7865                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7866                 flush_dcache_page(page);
7867                 kunmap(page);
7868         }
7869         ClearPageChecked(page);
7870         set_page_dirty(page);
7871         SetPageUptodate(page);
7872
7873         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7874         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7875         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7876
7877         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7878
7879 out_unlock:
7880         if (!ret) {
7881                 sb_end_pagefault(inode->i_sb);
7882                 return VM_FAULT_LOCKED;
7883         }
7884         unlock_page(page);
7885 out:
7886         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7887 out_noreserve:
7888         sb_end_pagefault(inode->i_sb);
7889         return ret;
7890 }
7891
7892 static int btrfs_truncate(struct inode *inode)
7893 {
7894         struct btrfs_root *root = BTRFS_I(inode)->root;
7895         struct btrfs_block_rsv *rsv;
7896         int ret = 0;
7897         int err = 0;
7898         struct btrfs_trans_handle *trans;
7899         u64 mask = root->sectorsize - 1;
7900         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7901
7902         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7903                                        (u64)-1);
7904         if (ret)
7905                 return ret;
7906
7907         /*
7908          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7909          * 3 things going on here
7910          *
7911          * 1) We need to reserve space for our orphan item and the space to
7912          * delete our orphan item.  Lord knows we don't want to have a dangling
7913          * orphan item because we didn't reserve space to remove it.
7914          *
7915          * 2) We need to reserve space to update our inode.
7916          *
7917          * 3) We need to have something to cache all the space that is going to
7918          * be free'd up by the truncate operation, but also have some slack
7919          * space reserved in case it uses space during the truncate (thank you
7920          * very much snapshotting).
7921          *
7922          * And we need these to all be seperate.  The fact is we can use alot of
7923          * space doing the truncate, and we have no earthly idea how much space
7924          * we will use, so we need the truncate reservation to be seperate so it
7925          * doesn't end up using space reserved for updating the inode or
7926          * removing the orphan item.  We also need to be able to stop the
7927          * transaction and start a new one, which means we need to be able to
7928          * update the inode several times, and we have no idea of knowing how
7929          * many times that will be, so we can't just reserve 1 item for the
7930          * entirety of the opration, so that has to be done seperately as well.
7931          * Then there is the orphan item, which does indeed need to be held on
7932          * to for the whole operation, and we need nobody to touch this reserved
7933          * space except the orphan code.
7934          *
7935          * So that leaves us with
7936          *
7937          * 1) root->orphan_block_rsv - for the orphan deletion.
7938          * 2) rsv - for the truncate reservation, which we will steal from the
7939          * transaction reservation.
7940          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7941          * updating the inode.
7942          */
7943         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7944         if (!rsv)
7945                 return -ENOMEM;
7946         rsv->size = min_size;
7947         rsv->failfast = 1;
7948
7949         /*
7950          * 1 for the truncate slack space
7951          * 1 for updating the inode.
7952          */
7953         trans = btrfs_start_transaction(root, 2);
7954         if (IS_ERR(trans)) {
7955                 err = PTR_ERR(trans);
7956                 goto out;
7957         }
7958
7959         /* Migrate the slack space for the truncate to our reserve */
7960         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7961                                       min_size);
7962         BUG_ON(ret);
7963
7964         /*
7965          * setattr is responsible for setting the ordered_data_close flag,
7966          * but that is only tested during the last file release.  That
7967          * could happen well after the next commit, leaving a great big
7968          * window where new writes may get lost if someone chooses to write
7969          * to this file after truncating to zero
7970          *
7971          * The inode doesn't have any dirty data here, and so if we commit
7972          * this is a noop.  If someone immediately starts writing to the inode
7973          * it is very likely we'll catch some of their writes in this
7974          * transaction, and the commit will find this file on the ordered
7975          * data list with good things to send down.
7976          *
7977          * This is a best effort solution, there is still a window where
7978          * using truncate to replace the contents of the file will
7979          * end up with a zero length file after a crash.
7980          */
7981         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7982                                            &BTRFS_I(inode)->runtime_flags))
7983                 btrfs_add_ordered_operation(trans, root, inode);
7984
7985         /*
7986          * So if we truncate and then write and fsync we normally would just
7987          * write the extents that changed, which is a problem if we need to
7988          * first truncate that entire inode.  So set this flag so we write out
7989          * all of the extents in the inode to the sync log so we're completely
7990          * safe.
7991          */
7992         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7993         trans->block_rsv = rsv;
7994
7995         while (1) {
7996                 ret = btrfs_truncate_inode_items(trans, root, inode,
7997                                                  inode->i_size,
7998                                                  BTRFS_EXTENT_DATA_KEY);
7999                 if (ret != -ENOSPC) {
8000                         err = ret;
8001                         break;
8002                 }
8003
8004                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8005                 ret = btrfs_update_inode(trans, root, inode);
8006                 if (ret) {
8007                         err = ret;
8008                         break;
8009                 }
8010
8011                 btrfs_end_transaction(trans, root);
8012                 btrfs_btree_balance_dirty(root);
8013
8014                 trans = btrfs_start_transaction(root, 2);
8015                 if (IS_ERR(trans)) {
8016                         ret = err = PTR_ERR(trans);
8017                         trans = NULL;
8018                         break;
8019                 }
8020
8021                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8022                                               rsv, min_size);
8023                 BUG_ON(ret);    /* shouldn't happen */
8024                 trans->block_rsv = rsv;
8025         }
8026
8027         if (ret == 0 && inode->i_nlink > 0) {
8028                 trans->block_rsv = root->orphan_block_rsv;
8029                 ret = btrfs_orphan_del(trans, inode);
8030                 if (ret)
8031                         err = ret;
8032         }
8033
8034         if (trans) {
8035                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8036                 ret = btrfs_update_inode(trans, root, inode);
8037                 if (ret && !err)
8038                         err = ret;
8039
8040                 ret = btrfs_end_transaction(trans, root);
8041                 btrfs_btree_balance_dirty(root);
8042         }
8043
8044 out:
8045         btrfs_free_block_rsv(root, rsv);
8046
8047         if (ret && !err)
8048                 err = ret;
8049
8050         return err;
8051 }
8052
8053 /*
8054  * create a new subvolume directory/inode (helper for the ioctl).
8055  */
8056 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8057                              struct btrfs_root *new_root,
8058                              struct btrfs_root *parent_root,
8059                              u64 new_dirid)
8060 {
8061         struct inode *inode;
8062         int err;
8063         u64 index = 0;
8064
8065         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8066                                 new_dirid, new_dirid,
8067                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
8068                                 &index);
8069         if (IS_ERR(inode))
8070                 return PTR_ERR(inode);
8071         inode->i_op = &btrfs_dir_inode_operations;
8072         inode->i_fop = &btrfs_dir_file_operations;
8073
8074         set_nlink(inode, 1);
8075         btrfs_i_size_write(inode, 0);
8076
8077         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8078         if (err)
8079                 btrfs_err(new_root->fs_info,
8080                           "error inheriting subvolume %llu properties: %d",
8081                           new_root->root_key.objectid, err);
8082
8083         err = btrfs_update_inode(trans, new_root, inode);
8084
8085         iput(inode);
8086         return err;
8087 }
8088
8089 struct inode *btrfs_alloc_inode(struct super_block *sb)
8090 {
8091         struct btrfs_inode *ei;
8092         struct inode *inode;
8093
8094         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8095         if (!ei)
8096                 return NULL;
8097
8098         ei->root = NULL;
8099         ei->generation = 0;
8100         ei->last_trans = 0;
8101         ei->last_sub_trans = 0;
8102         ei->logged_trans = 0;
8103         ei->delalloc_bytes = 0;
8104         ei->disk_i_size = 0;
8105         ei->flags = 0;
8106         ei->csum_bytes = 0;
8107         ei->index_cnt = (u64)-1;
8108         ei->dir_index = 0;
8109         ei->last_unlink_trans = 0;
8110         ei->last_log_commit = 0;
8111
8112         spin_lock_init(&ei->lock);
8113         ei->outstanding_extents = 0;
8114         ei->reserved_extents = 0;
8115
8116         ei->runtime_flags = 0;
8117         ei->force_compress = BTRFS_COMPRESS_NONE;
8118
8119         ei->delayed_node = NULL;
8120
8121         inode = &ei->vfs_inode;
8122         extent_map_tree_init(&ei->extent_tree);
8123         extent_io_tree_init(&ei->io_tree, &inode->i_data);
8124         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8125         ei->io_tree.track_uptodate = 1;
8126         ei->io_failure_tree.track_uptodate = 1;
8127         atomic_set(&ei->sync_writers, 0);
8128         mutex_init(&ei->log_mutex);
8129         mutex_init(&ei->delalloc_mutex);
8130         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8131         INIT_LIST_HEAD(&ei->delalloc_inodes);
8132         INIT_LIST_HEAD(&ei->ordered_operations);
8133         RB_CLEAR_NODE(&ei->rb_node);
8134
8135         return inode;
8136 }
8137
8138 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8139 void btrfs_test_destroy_inode(struct inode *inode)
8140 {
8141         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8142         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8143 }
8144 #endif
8145
8146 static void btrfs_i_callback(struct rcu_head *head)
8147 {
8148         struct inode *inode = container_of(head, struct inode, i_rcu);
8149         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8150 }
8151
8152 void btrfs_destroy_inode(struct inode *inode)
8153 {
8154         struct btrfs_ordered_extent *ordered;
8155         struct btrfs_root *root = BTRFS_I(inode)->root;
8156
8157         WARN_ON(!hlist_empty(&inode->i_dentry));
8158         WARN_ON(inode->i_data.nrpages);
8159         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8160         WARN_ON(BTRFS_I(inode)->reserved_extents);
8161         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8162         WARN_ON(BTRFS_I(inode)->csum_bytes);
8163
8164         /*
8165          * This can happen where we create an inode, but somebody else also
8166          * created the same inode and we need to destroy the one we already
8167          * created.
8168          */
8169         if (!root)
8170                 goto free;
8171
8172         /*
8173          * Make sure we're properly removed from the ordered operation
8174          * lists.
8175          */
8176         smp_mb();
8177         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8178                 spin_lock(&root->fs_info->ordered_root_lock);
8179                 list_del_init(&BTRFS_I(inode)->ordered_operations);
8180                 spin_unlock(&root->fs_info->ordered_root_lock);
8181         }
8182
8183         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8184                      &BTRFS_I(inode)->runtime_flags)) {
8185                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8186                         btrfs_ino(inode));
8187                 atomic_dec(&root->orphan_inodes);
8188         }
8189
8190         while (1) {
8191                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8192                 if (!ordered)
8193                         break;
8194                 else {
8195                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8196                                 ordered->file_offset, ordered->len);
8197                         btrfs_remove_ordered_extent(inode, ordered);
8198                         btrfs_put_ordered_extent(ordered);
8199                         btrfs_put_ordered_extent(ordered);
8200                 }
8201         }
8202         inode_tree_del(inode);
8203         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8204 free:
8205         call_rcu(&inode->i_rcu, btrfs_i_callback);
8206 }
8207
8208 int btrfs_drop_inode(struct inode *inode)
8209 {
8210         struct btrfs_root *root = BTRFS_I(inode)->root;
8211
8212         if (root == NULL)
8213                 return 1;
8214
8215         /* the snap/subvol tree is on deleting */
8216         if (btrfs_root_refs(&root->root_item) == 0)
8217                 return 1;
8218         else
8219                 return generic_drop_inode(inode);
8220 }
8221
8222 static void init_once(void *foo)
8223 {
8224         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8225
8226         inode_init_once(&ei->vfs_inode);
8227 }
8228
8229 void btrfs_destroy_cachep(void)
8230 {
8231         /*
8232          * Make sure all delayed rcu free inodes are flushed before we
8233          * destroy cache.
8234          */
8235         rcu_barrier();
8236         if (btrfs_inode_cachep)
8237                 kmem_cache_destroy(btrfs_inode_cachep);
8238         if (btrfs_trans_handle_cachep)
8239                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8240         if (btrfs_transaction_cachep)
8241                 kmem_cache_destroy(btrfs_transaction_cachep);
8242         if (btrfs_path_cachep)
8243                 kmem_cache_destroy(btrfs_path_cachep);
8244         if (btrfs_free_space_cachep)
8245                 kmem_cache_destroy(btrfs_free_space_cachep);
8246         if (btrfs_delalloc_work_cachep)
8247                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8248 }
8249
8250 int btrfs_init_cachep(void)
8251 {
8252         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8253                         sizeof(struct btrfs_inode), 0,
8254                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8255         if (!btrfs_inode_cachep)
8256                 goto fail;
8257
8258         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8259                         sizeof(struct btrfs_trans_handle), 0,
8260                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8261         if (!btrfs_trans_handle_cachep)
8262                 goto fail;
8263
8264         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8265                         sizeof(struct btrfs_transaction), 0,
8266                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8267         if (!btrfs_transaction_cachep)
8268                 goto fail;
8269
8270         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8271                         sizeof(struct btrfs_path), 0,
8272                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8273         if (!btrfs_path_cachep)
8274                 goto fail;
8275
8276         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8277                         sizeof(struct btrfs_free_space), 0,
8278                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8279         if (!btrfs_free_space_cachep)
8280                 goto fail;
8281
8282         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8283                         sizeof(struct btrfs_delalloc_work), 0,
8284                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8285                         NULL);
8286         if (!btrfs_delalloc_work_cachep)
8287                 goto fail;
8288
8289         return 0;
8290 fail:
8291         btrfs_destroy_cachep();
8292         return -ENOMEM;
8293 }
8294
8295 static int btrfs_getattr(struct vfsmount *mnt,
8296                          struct dentry *dentry, struct kstat *stat)
8297 {
8298         u64 delalloc_bytes;
8299         struct inode *inode = dentry->d_inode;
8300         u32 blocksize = inode->i_sb->s_blocksize;
8301
8302         generic_fillattr(inode, stat);
8303         stat->dev = BTRFS_I(inode)->root->anon_dev;
8304         stat->blksize = PAGE_CACHE_SIZE;
8305
8306         spin_lock(&BTRFS_I(inode)->lock);
8307         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8308         spin_unlock(&BTRFS_I(inode)->lock);
8309         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8310                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8311         return 0;
8312 }
8313
8314 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8315                            struct inode *new_dir, struct dentry *new_dentry)
8316 {
8317         struct btrfs_trans_handle *trans;
8318         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8319         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8320         struct inode *new_inode = new_dentry->d_inode;
8321         struct inode *old_inode = old_dentry->d_inode;
8322         struct timespec ctime = CURRENT_TIME;
8323         u64 index = 0;
8324         u64 root_objectid;
8325         int ret;
8326         u64 old_ino = btrfs_ino(old_inode);
8327
8328         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8329                 return -EPERM;
8330
8331         /* we only allow rename subvolume link between subvolumes */
8332         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8333                 return -EXDEV;
8334
8335         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8336             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8337                 return -ENOTEMPTY;
8338
8339         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8340             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8341                 return -ENOTEMPTY;
8342
8343
8344         /* check for collisions, even if the  name isn't there */
8345         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8346                              new_dentry->d_name.name,
8347                              new_dentry->d_name.len);
8348
8349         if (ret) {
8350                 if (ret == -EEXIST) {
8351                         /* we shouldn't get
8352                          * eexist without a new_inode */
8353                         if (WARN_ON(!new_inode)) {
8354                                 return ret;
8355                         }
8356                 } else {
8357                         /* maybe -EOVERFLOW */
8358                         return ret;
8359                 }
8360         }
8361         ret = 0;
8362
8363         /*
8364          * we're using rename to replace one file with another.
8365          * and the replacement file is large.  Start IO on it now so
8366          * we don't add too much work to the end of the transaction
8367          */
8368         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8369             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8370                 filemap_flush(old_inode->i_mapping);
8371
8372         /* close the racy window with snapshot create/destroy ioctl */
8373         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8374                 down_read(&root->fs_info->subvol_sem);
8375         /*
8376          * We want to reserve the absolute worst case amount of items.  So if
8377          * both inodes are subvols and we need to unlink them then that would
8378          * require 4 item modifications, but if they are both normal inodes it
8379          * would require 5 item modifications, so we'll assume their normal
8380          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8381          * should cover the worst case number of items we'll modify.
8382          */
8383         trans = btrfs_start_transaction(root, 11);
8384         if (IS_ERR(trans)) {
8385                 ret = PTR_ERR(trans);
8386                 goto out_notrans;
8387         }
8388
8389         if (dest != root)
8390                 btrfs_record_root_in_trans(trans, dest);
8391
8392         ret = btrfs_set_inode_index(new_dir, &index);
8393         if (ret)
8394                 goto out_fail;
8395
8396         BTRFS_I(old_inode)->dir_index = 0ULL;
8397         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8398                 /* force full log commit if subvolume involved. */
8399                 btrfs_set_log_full_commit(root->fs_info, trans);
8400         } else {
8401                 ret = btrfs_insert_inode_ref(trans, dest,
8402                                              new_dentry->d_name.name,
8403                                              new_dentry->d_name.len,
8404                                              old_ino,
8405                                              btrfs_ino(new_dir), index);
8406                 if (ret)
8407                         goto out_fail;
8408                 /*
8409                  * this is an ugly little race, but the rename is required
8410                  * to make sure that if we crash, the inode is either at the
8411                  * old name or the new one.  pinning the log transaction lets
8412                  * us make sure we don't allow a log commit to come in after
8413                  * we unlink the name but before we add the new name back in.
8414                  */
8415                 btrfs_pin_log_trans(root);
8416         }
8417         /*
8418          * make sure the inode gets flushed if it is replacing
8419          * something.
8420          */
8421         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8422                 btrfs_add_ordered_operation(trans, root, old_inode);
8423
8424         inode_inc_iversion(old_dir);
8425         inode_inc_iversion(new_dir);
8426         inode_inc_iversion(old_inode);
8427         old_dir->i_ctime = old_dir->i_mtime = ctime;
8428         new_dir->i_ctime = new_dir->i_mtime = ctime;
8429         old_inode->i_ctime = ctime;
8430
8431         if (old_dentry->d_parent != new_dentry->d_parent)
8432                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8433
8434         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8435                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8436                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8437                                         old_dentry->d_name.name,
8438                                         old_dentry->d_name.len);
8439         } else {
8440                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8441                                         old_dentry->d_inode,
8442                                         old_dentry->d_name.name,
8443                                         old_dentry->d_name.len);
8444                 if (!ret)
8445                         ret = btrfs_update_inode(trans, root, old_inode);
8446         }
8447         if (ret) {
8448                 btrfs_abort_transaction(trans, root, ret);
8449                 goto out_fail;
8450         }
8451
8452         if (new_inode) {
8453                 inode_inc_iversion(new_inode);
8454                 new_inode->i_ctime = CURRENT_TIME;
8455                 if (unlikely(btrfs_ino(new_inode) ==
8456                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8457                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8458                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8459                                                 root_objectid,
8460                                                 new_dentry->d_name.name,
8461                                                 new_dentry->d_name.len);
8462                         BUG_ON(new_inode->i_nlink == 0);
8463                 } else {
8464                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8465                                                  new_dentry->d_inode,
8466                                                  new_dentry->d_name.name,
8467                                                  new_dentry->d_name.len);
8468                 }
8469                 if (!ret && new_inode->i_nlink == 0)
8470                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8471                 if (ret) {
8472                         btrfs_abort_transaction(trans, root, ret);
8473                         goto out_fail;
8474                 }
8475         }
8476
8477         ret = btrfs_add_link(trans, new_dir, old_inode,
8478                              new_dentry->d_name.name,
8479                              new_dentry->d_name.len, 0, index);
8480         if (ret) {
8481                 btrfs_abort_transaction(trans, root, ret);
8482                 goto out_fail;
8483         }
8484
8485         if (old_inode->i_nlink == 1)
8486                 BTRFS_I(old_inode)->dir_index = index;
8487
8488         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8489                 struct dentry *parent = new_dentry->d_parent;
8490                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8491                 btrfs_end_log_trans(root);
8492         }
8493 out_fail:
8494         btrfs_end_transaction(trans, root);
8495 out_notrans:
8496         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8497                 up_read(&root->fs_info->subvol_sem);
8498
8499         return ret;
8500 }
8501
8502 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8503 {
8504         struct btrfs_delalloc_work *delalloc_work;
8505         struct inode *inode;
8506
8507         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8508                                      work);
8509         inode = delalloc_work->inode;
8510         if (delalloc_work->wait) {
8511                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8512         } else {
8513                 filemap_flush(inode->i_mapping);
8514                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8515                              &BTRFS_I(inode)->runtime_flags))
8516                         filemap_flush(inode->i_mapping);
8517         }
8518
8519         if (delalloc_work->delay_iput)
8520                 btrfs_add_delayed_iput(inode);
8521         else
8522                 iput(inode);
8523         complete(&delalloc_work->completion);
8524 }
8525
8526 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8527                                                     int wait, int delay_iput)
8528 {
8529         struct btrfs_delalloc_work *work;
8530
8531         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8532         if (!work)
8533                 return NULL;
8534
8535         init_completion(&work->completion);
8536         INIT_LIST_HEAD(&work->list);
8537         work->inode = inode;
8538         work->wait = wait;
8539         work->delay_iput = delay_iput;
8540         btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8541
8542         return work;
8543 }
8544
8545 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8546 {
8547         wait_for_completion(&work->completion);
8548         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8549 }
8550
8551 /*
8552  * some fairly slow code that needs optimization. This walks the list
8553  * of all the inodes with pending delalloc and forces them to disk.
8554  */
8555 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8556                                    int nr)
8557 {
8558         struct btrfs_inode *binode;
8559         struct inode *inode;
8560         struct btrfs_delalloc_work *work, *next;
8561         struct list_head works;
8562         struct list_head splice;
8563         int ret = 0;
8564
8565         INIT_LIST_HEAD(&works);
8566         INIT_LIST_HEAD(&splice);
8567
8568         mutex_lock(&root->delalloc_mutex);
8569         spin_lock(&root->delalloc_lock);
8570         list_splice_init(&root->delalloc_inodes, &splice);
8571         while (!list_empty(&splice)) {
8572                 binode = list_entry(splice.next, struct btrfs_inode,
8573                                     delalloc_inodes);
8574
8575                 list_move_tail(&binode->delalloc_inodes,
8576                                &root->delalloc_inodes);
8577                 inode = igrab(&binode->vfs_inode);
8578                 if (!inode) {
8579                         cond_resched_lock(&root->delalloc_lock);
8580                         continue;
8581                 }
8582                 spin_unlock(&root->delalloc_lock);
8583
8584                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8585                 if (unlikely(!work)) {
8586                         if (delay_iput)
8587                                 btrfs_add_delayed_iput(inode);
8588                         else
8589                                 iput(inode);
8590                         ret = -ENOMEM;
8591                         goto out;
8592                 }
8593                 list_add_tail(&work->list, &works);
8594                 btrfs_queue_work(root->fs_info->flush_workers,
8595                                  &work->work);
8596                 ret++;
8597                 if (nr != -1 && ret >= nr)
8598                         goto out;
8599                 cond_resched();
8600                 spin_lock(&root->delalloc_lock);
8601         }
8602         spin_unlock(&root->delalloc_lock);
8603
8604 out:
8605         list_for_each_entry_safe(work, next, &works, list) {
8606                 list_del_init(&work->list);
8607                 btrfs_wait_and_free_delalloc_work(work);
8608         }
8609
8610         if (!list_empty_careful(&splice)) {
8611                 spin_lock(&root->delalloc_lock);
8612                 list_splice_tail(&splice, &root->delalloc_inodes);
8613                 spin_unlock(&root->delalloc_lock);
8614         }
8615         mutex_unlock(&root->delalloc_mutex);
8616         return ret;
8617 }
8618
8619 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8620 {
8621         int ret;
8622
8623         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8624                 return -EROFS;
8625
8626         ret = __start_delalloc_inodes(root, delay_iput, -1);
8627         if (ret > 0)
8628                 ret = 0;
8629         /*
8630          * the filemap_flush will queue IO into the worker threads, but
8631          * we have to make sure the IO is actually started and that
8632          * ordered extents get created before we return
8633          */
8634         atomic_inc(&root->fs_info->async_submit_draining);
8635         while (atomic_read(&root->fs_info->nr_async_submits) ||
8636               atomic_read(&root->fs_info->async_delalloc_pages)) {
8637                 wait_event(root->fs_info->async_submit_wait,
8638                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8639                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8640         }
8641         atomic_dec(&root->fs_info->async_submit_draining);
8642         return ret;
8643 }
8644
8645 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8646                                int nr)
8647 {
8648         struct btrfs_root *root;
8649         struct list_head splice;
8650         int ret;
8651
8652         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8653                 return -EROFS;
8654
8655         INIT_LIST_HEAD(&splice);
8656
8657         mutex_lock(&fs_info->delalloc_root_mutex);
8658         spin_lock(&fs_info->delalloc_root_lock);
8659         list_splice_init(&fs_info->delalloc_roots, &splice);
8660         while (!list_empty(&splice) && nr) {
8661                 root = list_first_entry(&splice, struct btrfs_root,
8662                                         delalloc_root);
8663                 root = btrfs_grab_fs_root(root);
8664                 BUG_ON(!root);
8665                 list_move_tail(&root->delalloc_root,
8666                                &fs_info->delalloc_roots);
8667                 spin_unlock(&fs_info->delalloc_root_lock);
8668
8669                 ret = __start_delalloc_inodes(root, delay_iput, nr);
8670                 btrfs_put_fs_root(root);
8671                 if (ret < 0)
8672                         goto out;
8673
8674                 if (nr != -1) {
8675                         nr -= ret;
8676                         WARN_ON(nr < 0);
8677                 }
8678                 spin_lock(&fs_info->delalloc_root_lock);
8679         }
8680         spin_unlock(&fs_info->delalloc_root_lock);
8681
8682         ret = 0;
8683         atomic_inc(&fs_info->async_submit_draining);
8684         while (atomic_read(&fs_info->nr_async_submits) ||
8685               atomic_read(&fs_info->async_delalloc_pages)) {
8686                 wait_event(fs_info->async_submit_wait,
8687                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8688                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8689         }
8690         atomic_dec(&fs_info->async_submit_draining);
8691 out:
8692         if (!list_empty_careful(&splice)) {
8693                 spin_lock(&fs_info->delalloc_root_lock);
8694                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8695                 spin_unlock(&fs_info->delalloc_root_lock);
8696         }
8697         mutex_unlock(&fs_info->delalloc_root_mutex);
8698         return ret;
8699 }
8700
8701 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8702                          const char *symname)
8703 {
8704         struct btrfs_trans_handle *trans;
8705         struct btrfs_root *root = BTRFS_I(dir)->root;
8706         struct btrfs_path *path;
8707         struct btrfs_key key;
8708         struct inode *inode = NULL;
8709         int err;
8710         int drop_inode = 0;
8711         u64 objectid;
8712         u64 index = 0;
8713         int name_len;
8714         int datasize;
8715         unsigned long ptr;
8716         struct btrfs_file_extent_item *ei;
8717         struct extent_buffer *leaf;
8718
8719         name_len = strlen(symname);
8720         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8721                 return -ENAMETOOLONG;
8722
8723         /*
8724          * 2 items for inode item and ref
8725          * 2 items for dir items
8726          * 1 item for xattr if selinux is on
8727          */
8728         trans = btrfs_start_transaction(root, 5);
8729         if (IS_ERR(trans))
8730                 return PTR_ERR(trans);
8731
8732         err = btrfs_find_free_ino(root, &objectid);
8733         if (err)
8734                 goto out_unlock;
8735
8736         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8737                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8738                                 S_IFLNK|S_IRWXUGO, &index);
8739         if (IS_ERR(inode)) {
8740                 err = PTR_ERR(inode);
8741                 goto out_unlock;
8742         }
8743
8744         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8745         if (err) {
8746                 drop_inode = 1;
8747                 goto out_unlock;
8748         }
8749
8750         /*
8751         * If the active LSM wants to access the inode during
8752         * d_instantiate it needs these. Smack checks to see
8753         * if the filesystem supports xattrs by looking at the
8754         * ops vector.
8755         */
8756         inode->i_fop = &btrfs_file_operations;
8757         inode->i_op = &btrfs_file_inode_operations;
8758
8759         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8760         if (err)
8761                 drop_inode = 1;
8762         else {
8763                 inode->i_mapping->a_ops = &btrfs_aops;
8764                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8765                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8766         }
8767         if (drop_inode)
8768                 goto out_unlock;
8769
8770         path = btrfs_alloc_path();
8771         if (!path) {
8772                 err = -ENOMEM;
8773                 drop_inode = 1;
8774                 goto out_unlock;
8775         }
8776         key.objectid = btrfs_ino(inode);
8777         key.offset = 0;
8778         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8779         datasize = btrfs_file_extent_calc_inline_size(name_len);
8780         err = btrfs_insert_empty_item(trans, root, path, &key,
8781                                       datasize);
8782         if (err) {
8783                 drop_inode = 1;
8784                 btrfs_free_path(path);
8785                 goto out_unlock;
8786         }
8787         leaf = path->nodes[0];
8788         ei = btrfs_item_ptr(leaf, path->slots[0],
8789                             struct btrfs_file_extent_item);
8790         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8791         btrfs_set_file_extent_type(leaf, ei,
8792                                    BTRFS_FILE_EXTENT_INLINE);
8793         btrfs_set_file_extent_encryption(leaf, ei, 0);
8794         btrfs_set_file_extent_compression(leaf, ei, 0);
8795         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8796         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8797
8798         ptr = btrfs_file_extent_inline_start(ei);
8799         write_extent_buffer(leaf, symname, ptr, name_len);
8800         btrfs_mark_buffer_dirty(leaf);
8801         btrfs_free_path(path);
8802
8803         inode->i_op = &btrfs_symlink_inode_operations;
8804         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8805         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8806         inode_set_bytes(inode, name_len);
8807         btrfs_i_size_write(inode, name_len);
8808         err = btrfs_update_inode(trans, root, inode);
8809         if (err)
8810                 drop_inode = 1;
8811
8812 out_unlock:
8813         if (!err)
8814                 d_instantiate(dentry, inode);
8815         btrfs_end_transaction(trans, root);
8816         if (drop_inode) {
8817                 inode_dec_link_count(inode);
8818                 iput(inode);
8819         }
8820         btrfs_btree_balance_dirty(root);
8821         return err;
8822 }
8823
8824 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8825                                        u64 start, u64 num_bytes, u64 min_size,
8826                                        loff_t actual_len, u64 *alloc_hint,
8827                                        struct btrfs_trans_handle *trans)
8828 {
8829         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8830         struct extent_map *em;
8831         struct btrfs_root *root = BTRFS_I(inode)->root;
8832         struct btrfs_key ins;
8833         u64 cur_offset = start;
8834         u64 i_size;
8835         u64 cur_bytes;
8836         int ret = 0;
8837         bool own_trans = true;
8838
8839         if (trans)
8840                 own_trans = false;
8841         while (num_bytes > 0) {
8842                 if (own_trans) {
8843                         trans = btrfs_start_transaction(root, 3);
8844                         if (IS_ERR(trans)) {
8845                                 ret = PTR_ERR(trans);
8846                                 break;
8847                         }
8848                 }
8849
8850                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8851                 cur_bytes = max(cur_bytes, min_size);
8852                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8853                                            *alloc_hint, &ins, 1);
8854                 if (ret) {
8855                         if (own_trans)
8856                                 btrfs_end_transaction(trans, root);
8857                         break;
8858                 }
8859
8860                 ret = insert_reserved_file_extent(trans, inode,
8861                                                   cur_offset, ins.objectid,
8862                                                   ins.offset, ins.offset,
8863                                                   ins.offset, 0, 0, 0,
8864                                                   BTRFS_FILE_EXTENT_PREALLOC);
8865                 if (ret) {
8866                         btrfs_free_reserved_extent(root, ins.objectid,
8867                                                    ins.offset);
8868                         btrfs_abort_transaction(trans, root, ret);
8869                         if (own_trans)
8870                                 btrfs_end_transaction(trans, root);
8871                         break;
8872                 }
8873                 btrfs_drop_extent_cache(inode, cur_offset,
8874                                         cur_offset + ins.offset -1, 0);
8875
8876                 em = alloc_extent_map();
8877                 if (!em) {
8878                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8879                                 &BTRFS_I(inode)->runtime_flags);
8880                         goto next;
8881                 }
8882
8883                 em->start = cur_offset;
8884                 em->orig_start = cur_offset;
8885                 em->len = ins.offset;
8886                 em->block_start = ins.objectid;
8887                 em->block_len = ins.offset;
8888                 em->orig_block_len = ins.offset;
8889                 em->ram_bytes = ins.offset;
8890                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8891                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8892                 em->generation = trans->transid;
8893
8894                 while (1) {
8895                         write_lock(&em_tree->lock);
8896                         ret = add_extent_mapping(em_tree, em, 1);
8897                         write_unlock(&em_tree->lock);
8898                         if (ret != -EEXIST)
8899                                 break;
8900                         btrfs_drop_extent_cache(inode, cur_offset,
8901                                                 cur_offset + ins.offset - 1,
8902                                                 0);
8903                 }
8904                 free_extent_map(em);
8905 next:
8906                 num_bytes -= ins.offset;
8907                 cur_offset += ins.offset;
8908                 *alloc_hint = ins.objectid + ins.offset;
8909
8910                 inode_inc_iversion(inode);
8911                 inode->i_ctime = CURRENT_TIME;
8912                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8913                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8914                     (actual_len > inode->i_size) &&
8915                     (cur_offset > inode->i_size)) {
8916                         if (cur_offset > actual_len)
8917                                 i_size = actual_len;
8918                         else
8919                                 i_size = cur_offset;
8920                         i_size_write(inode, i_size);
8921                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8922                 }
8923
8924                 ret = btrfs_update_inode(trans, root, inode);
8925
8926                 if (ret) {
8927                         btrfs_abort_transaction(trans, root, ret);
8928                         if (own_trans)
8929                                 btrfs_end_transaction(trans, root);
8930                         break;
8931                 }
8932
8933                 if (own_trans)
8934                         btrfs_end_transaction(trans, root);
8935         }
8936         return ret;
8937 }
8938
8939 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8940                               u64 start, u64 num_bytes, u64 min_size,
8941                               loff_t actual_len, u64 *alloc_hint)
8942 {
8943         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8944                                            min_size, actual_len, alloc_hint,
8945                                            NULL);
8946 }
8947
8948 int btrfs_prealloc_file_range_trans(struct inode *inode,
8949                                     struct btrfs_trans_handle *trans, int mode,
8950                                     u64 start, u64 num_bytes, u64 min_size,
8951                                     loff_t actual_len, u64 *alloc_hint)
8952 {
8953         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8954                                            min_size, actual_len, alloc_hint, trans);
8955 }
8956
8957 static int btrfs_set_page_dirty(struct page *page)
8958 {
8959         return __set_page_dirty_nobuffers(page);
8960 }
8961
8962 static int btrfs_permission(struct inode *inode, int mask)
8963 {
8964         struct btrfs_root *root = BTRFS_I(inode)->root;
8965         umode_t mode = inode->i_mode;
8966
8967         if (mask & MAY_WRITE &&
8968             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8969                 if (btrfs_root_readonly(root))
8970                         return -EROFS;
8971                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8972                         return -EACCES;
8973         }
8974         return generic_permission(inode, mask);
8975 }
8976
8977 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
8978 {
8979         struct btrfs_trans_handle *trans;
8980         struct btrfs_root *root = BTRFS_I(dir)->root;
8981         struct inode *inode = NULL;
8982         u64 objectid;
8983         u64 index;
8984         int ret = 0;
8985
8986         /*
8987          * 5 units required for adding orphan entry
8988          */
8989         trans = btrfs_start_transaction(root, 5);
8990         if (IS_ERR(trans))
8991                 return PTR_ERR(trans);
8992
8993         ret = btrfs_find_free_ino(root, &objectid);
8994         if (ret)
8995                 goto out;
8996
8997         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
8998                                 btrfs_ino(dir), objectid, mode, &index);
8999         if (IS_ERR(inode)) {
9000                 ret = PTR_ERR(inode);
9001                 inode = NULL;
9002                 goto out;
9003         }
9004
9005         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9006         if (ret)
9007                 goto out;
9008
9009         ret = btrfs_update_inode(trans, root, inode);
9010         if (ret)
9011                 goto out;
9012
9013         inode->i_fop = &btrfs_file_operations;
9014         inode->i_op = &btrfs_file_inode_operations;
9015
9016         inode->i_mapping->a_ops = &btrfs_aops;
9017         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9018         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9019
9020         ret = btrfs_orphan_add(trans, inode);
9021         if (ret)
9022                 goto out;
9023
9024         d_tmpfile(dentry, inode);
9025         mark_inode_dirty(inode);
9026
9027 out:
9028         btrfs_end_transaction(trans, root);
9029         if (ret)
9030                 iput(inode);
9031         btrfs_balance_delayed_items(root);
9032         btrfs_btree_balance_dirty(root);
9033
9034         return ret;
9035 }
9036
9037 static const struct inode_operations btrfs_dir_inode_operations = {
9038         .getattr        = btrfs_getattr,
9039         .lookup         = btrfs_lookup,
9040         .create         = btrfs_create,
9041         .unlink         = btrfs_unlink,
9042         .link           = btrfs_link,
9043         .mkdir          = btrfs_mkdir,
9044         .rmdir          = btrfs_rmdir,
9045         .rename         = btrfs_rename,
9046         .symlink        = btrfs_symlink,
9047         .setattr        = btrfs_setattr,
9048         .mknod          = btrfs_mknod,
9049         .setxattr       = btrfs_setxattr,
9050         .getxattr       = btrfs_getxattr,
9051         .listxattr      = btrfs_listxattr,
9052         .removexattr    = btrfs_removexattr,
9053         .permission     = btrfs_permission,
9054         .get_acl        = btrfs_get_acl,
9055         .set_acl        = btrfs_set_acl,
9056         .update_time    = btrfs_update_time,
9057         .tmpfile        = btrfs_tmpfile,
9058 };
9059 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9060         .lookup         = btrfs_lookup,
9061         .permission     = btrfs_permission,
9062         .get_acl        = btrfs_get_acl,
9063         .set_acl        = btrfs_set_acl,
9064         .update_time    = btrfs_update_time,
9065 };
9066
9067 static const struct file_operations btrfs_dir_file_operations = {
9068         .llseek         = generic_file_llseek,
9069         .read           = generic_read_dir,
9070         .iterate        = btrfs_real_readdir,
9071         .unlocked_ioctl = btrfs_ioctl,
9072 #ifdef CONFIG_COMPAT
9073         .compat_ioctl   = btrfs_ioctl,
9074 #endif
9075         .release        = btrfs_release_file,
9076         .fsync          = btrfs_sync_file,
9077 };
9078
9079 static struct extent_io_ops btrfs_extent_io_ops = {
9080         .fill_delalloc = run_delalloc_range,
9081         .submit_bio_hook = btrfs_submit_bio_hook,
9082         .merge_bio_hook = btrfs_merge_bio_hook,
9083         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9084         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9085         .writepage_start_hook = btrfs_writepage_start_hook,
9086         .set_bit_hook = btrfs_set_bit_hook,
9087         .clear_bit_hook = btrfs_clear_bit_hook,
9088         .merge_extent_hook = btrfs_merge_extent_hook,
9089         .split_extent_hook = btrfs_split_extent_hook,
9090 };
9091
9092 /*
9093  * btrfs doesn't support the bmap operation because swapfiles
9094  * use bmap to make a mapping of extents in the file.  They assume
9095  * these extents won't change over the life of the file and they
9096  * use the bmap result to do IO directly to the drive.
9097  *
9098  * the btrfs bmap call would return logical addresses that aren't
9099  * suitable for IO and they also will change frequently as COW
9100  * operations happen.  So, swapfile + btrfs == corruption.
9101  *
9102  * For now we're avoiding this by dropping bmap.
9103  */
9104 static const struct address_space_operations btrfs_aops = {
9105         .readpage       = btrfs_readpage,
9106         .writepage      = btrfs_writepage,
9107         .writepages     = btrfs_writepages,
9108         .readpages      = btrfs_readpages,
9109         .direct_IO      = btrfs_direct_IO,
9110         .invalidatepage = btrfs_invalidatepage,
9111         .releasepage    = btrfs_releasepage,
9112         .set_page_dirty = btrfs_set_page_dirty,
9113         .error_remove_page = generic_error_remove_page,
9114 };
9115
9116 static const struct address_space_operations btrfs_symlink_aops = {
9117         .readpage       = btrfs_readpage,
9118         .writepage      = btrfs_writepage,
9119         .invalidatepage = btrfs_invalidatepage,
9120         .releasepage    = btrfs_releasepage,
9121 };
9122
9123 static const struct inode_operations btrfs_file_inode_operations = {
9124         .getattr        = btrfs_getattr,
9125         .setattr        = btrfs_setattr,
9126         .setxattr       = btrfs_setxattr,
9127         .getxattr       = btrfs_getxattr,
9128         .listxattr      = btrfs_listxattr,
9129         .removexattr    = btrfs_removexattr,
9130         .permission     = btrfs_permission,
9131         .fiemap         = btrfs_fiemap,
9132         .get_acl        = btrfs_get_acl,
9133         .set_acl        = btrfs_set_acl,
9134         .update_time    = btrfs_update_time,
9135 };
9136 static const struct inode_operations btrfs_special_inode_operations = {
9137         .getattr        = btrfs_getattr,
9138         .setattr        = btrfs_setattr,
9139         .permission     = btrfs_permission,
9140         .setxattr       = btrfs_setxattr,
9141         .getxattr       = btrfs_getxattr,
9142         .listxattr      = btrfs_listxattr,
9143         .removexattr    = btrfs_removexattr,
9144         .get_acl        = btrfs_get_acl,
9145         .set_acl        = btrfs_set_acl,
9146         .update_time    = btrfs_update_time,
9147 };
9148 static const struct inode_operations btrfs_symlink_inode_operations = {
9149         .readlink       = generic_readlink,
9150         .follow_link    = page_follow_link_light,
9151         .put_link       = page_put_link,
9152         .getattr        = btrfs_getattr,
9153         .setattr        = btrfs_setattr,
9154         .permission     = btrfs_permission,
9155         .setxattr       = btrfs_setxattr,
9156         .getxattr       = btrfs_getxattr,
9157         .listxattr      = btrfs_listxattr,
9158         .removexattr    = btrfs_removexattr,
9159         .update_time    = btrfs_update_time,
9160 };
9161
9162 const struct dentry_operations btrfs_dentry_operations = {
9163         .d_delete       = btrfs_dentry_delete,
9164         .d_release      = btrfs_dentry_release,
9165 };