fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include <linux/blkdev.h>
  44 #include "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "print-tree.h"
  50 #include "ordered-data.h"
  51 #include "xattr.h"
  52 #include "tree-log.h"
  53 #include "volumes.h"
  54 #include "compression.h"
  55 #include "locking.h"
  56 #include "free-space-cache.h"
  57 #include "inode-map.h"
  58 #include "backref.h"
  59
  60 struct btrfs_iget_args {
  61         u64 ino;
  62         struct btrfs_root *root;
  63 };
  64
  65 static const struct inode_operations btrfs_dir_inode_operations;
  66 static const struct inode_operations btrfs_symlink_inode_operations;
  67 static const struct inode_operations btrfs_dir_ro_inode_operations;
  68 static const struct inode_operations btrfs_special_inode_operations;
  69 static const struct inode_operations btrfs_file_inode_operations;
  70 static const struct address_space_operations btrfs_aops;
  71 static const struct address_space_operations btrfs_symlink_aops;
  72 static const struct file_operations btrfs_dir_file_operations;
  73 static struct extent_io_ops btrfs_extent_io_ops;
  74
  75 static struct kmem_cache *btrfs_inode_cachep;
  76 static struct kmem_cache *btrfs_delalloc_work_cachep;
  77 struct kmem_cache *btrfs_trans_handle_cachep;
  78 struct kmem_cache *btrfs_transaction_cachep;
  79 struct kmem_cache *btrfs_path_cachep;
  80 struct kmem_cache *btrfs_free_space_cachep;
  81
  82 #define S_SHIFT 12
  83 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  84         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  85         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  86         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  87         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  88         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  89         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  90         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  91 };
  92
  93 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  94 static int btrfs_truncate(struct inode *inode);
  95 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  96 static noinline int cow_file_range(struct inode *inode,
  97                                    struct page *locked_page,
  98                                    u64 start, u64 end, int *page_started,
  99                                    unsigned long *nr_written, int unlock);
 100 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 101                                            u64 len, u64 orig_start,
 102                                            u64 block_start, u64 block_len,
 103                                            u64 orig_block_len, u64 ram_bytes,
 104                                            int type);
 105
 106 static int btrfs_dirty_inode(struct inode *inode);
 107
 108 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 109                                      struct inode *inode,  struct inode *dir,
 110                                      const struct qstr *qstr)
 111 {
 112         int err;
 113
 114         err = btrfs_init_acl(trans, inode, dir);
 115         if (!err)
 116                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 117         return err;
 118 }
 119
 120 /*
 121  * this does all the hard work for inserting an inline extent into
 122  * the btree.  The caller should have done a btrfs_drop_extents so that
 123  * no overlapping inline items exist in the btree
 124  */
 125 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 126                                 struct btrfs_root *root, struct inode *inode,
 127                                 u64 start, size_t size, size_t compressed_size,
 128                                 int compress_type,
 129                                 struct page **compressed_pages)
 130 {
 131         struct btrfs_key key;
 132         struct btrfs_path *path;
 133         struct extent_buffer *leaf;
 134         struct page *page = NULL;
 135         char *kaddr;
 136         unsigned long ptr;
 137         struct btrfs_file_extent_item *ei;
 138         int err = 0;
 139         int ret;
 140         size_t cur_size = size;
 141         size_t datasize;
 142         unsigned long offset;
 143
 144         if (compressed_size && compressed_pages)
 145                 cur_size = compressed_size;
 146
 147         path = btrfs_alloc_path();
 148         if (!path)
 149                 return -ENOMEM;
 150
 151         path->leave_spinning = 1;
 152
 153         key.objectid = btrfs_ino(inode);
 154         key.offset = start;
 155         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 156         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 157
 158         inode_add_bytes(inode, size);
 159         ret = btrfs_insert_empty_item(trans, root, path, &key,
 160                                       datasize);
 161         if (ret) {
 162                 err = ret;
 163                 goto fail;
 164         }
 165         leaf = path->nodes[0];
 166         ei = btrfs_item_ptr(leaf, path->slots[0],
 167                             struct btrfs_file_extent_item);
 168         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 169         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 170         btrfs_set_file_extent_encryption(leaf, ei, 0);
 171         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 172         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 173         ptr = btrfs_file_extent_inline_start(ei);
 174
 175         if (compress_type != BTRFS_COMPRESS_NONE) {
 176                 struct page *cpage;
 177                 int i = 0;
 178                 while (compressed_size > 0) {
 179                         cpage = compressed_pages[i];
 180                         cur_size = min_t(unsigned long, compressed_size,
 181                                        PAGE_CACHE_SIZE);
 182
 183                         kaddr = kmap_atomic(cpage);
 184                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 185                         kunmap_atomic(kaddr);
 186
 187                         i++;
 188                         ptr += cur_size;
 189                         compressed_size -= cur_size;
 190                 }
 191                 btrfs_set_file_extent_compression(leaf, ei,
 192                                                   compress_type);
 193         } else {
 194                 page = find_get_page(inode->i_mapping,
 195                                      start >> PAGE_CACHE_SHIFT);
 196                 btrfs_set_file_extent_compression(leaf, ei, 0);
 197                 kaddr = kmap_atomic(page);
 198                 offset = start & (PAGE_CACHE_SIZE - 1);
 199                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 200                 kunmap_atomic(kaddr);
 201                 page_cache_release(page);
 202         }
 203         btrfs_mark_buffer_dirty(leaf);
 204         btrfs_free_path(path);
 205
 206         /*
 207          * we're an inline extent, so nobody can
 208          * extend the file past i_size without locking
 209          * a page we already have locked.
 210          *
 211          * We must do any isize and inode updates
 212          * before we unlock the pages.  Otherwise we
 213          * could end up racing with unlink.
 214          */
 215         BTRFS_I(inode)->disk_i_size = inode->i_size;
 216         ret = btrfs_update_inode(trans, root, inode);
 217
 218         return ret;
 219 fail:
 220         btrfs_free_path(path);
 221         return err;
 222 }
 223
 224
 225 /*
 226  * conditionally insert an inline extent into the file.  This
 227  * does the checks required to make sure the data is small enough
 228  * to fit as an inline extent.
 229  */
 230 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 231                                  struct btrfs_root *root,
 232                                  struct inode *inode, u64 start, u64 end,
 233                                  size_t compressed_size, int compress_type,
 234                                  struct page **compressed_pages)
 235 {
 236         u64 isize = i_size_read(inode);
 237         u64 actual_end = min(end + 1, isize);
 238         u64 inline_len = actual_end - start;
 239         u64 aligned_end = ALIGN(end, root->sectorsize);
 240         u64 data_len = inline_len;
 241         int ret;
 242
 243         if (compressed_size)
 244                 data_len = compressed_size;
 245
 246         if (start > 0 ||
 247             actual_end >= PAGE_CACHE_SIZE ||
 248             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 249             (!compressed_size &&
 250             (actual_end & (root->sectorsize - 1)) == 0) ||
 251             end + 1 < isize ||
 252             data_len > root->fs_info->max_inline) {
 253                 return 1;
 254         }
 255
 256         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 257         if (ret)
 258                 return ret;
 259
 260         if (isize > actual_end)
 261                 inline_len = min_t(u64, isize, actual_end);
 262         ret = insert_inline_extent(trans, root, inode, start,
 263                                    inline_len, compressed_size,
 264                                    compress_type, compressed_pages);
 265         if (ret && ret != -ENOSPC) {
 266                 btrfs_abort_transaction(trans, root, ret);
 267                 return ret;
 268         } else if (ret == -ENOSPC) {
 269                 return 1;
 270         }
 271
 272         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 273         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 274         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 275         return 0;
 276 }
 277
 278 struct async_extent {
 279         u64 start;
 280         u64 ram_size;
 281         u64 compressed_size;
 282         struct page **pages;
 283         unsigned long nr_pages;
 284         int compress_type;
 285         struct list_head list;
 286 };
 287
 288 struct async_cow {
 289         struct inode *inode;
 290         struct btrfs_root *root;
 291         struct page *locked_page;
 292         u64 start;
 293         u64 end;
 294         struct list_head extents;
 295         struct btrfs_work work;
 296 };
 297
 298 static noinline int add_async_extent(struct async_cow *cow,
 299                                      u64 start, u64 ram_size,
 300                                      u64 compressed_size,
 301                                      struct page **pages,
 302                                      unsigned long nr_pages,
 303                                      int compress_type)
 304 {
 305         struct async_extent *async_extent;
 306
 307         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 308         BUG_ON(!async_extent); /* -ENOMEM */
 309         async_extent->start = start;
 310         async_extent->ram_size = ram_size;
 311         async_extent->compressed_size = compressed_size;
 312         async_extent->pages = pages;
 313         async_extent->nr_pages = nr_pages;
 314         async_extent->compress_type = compress_type;
 315         list_add_tail(&async_extent->list, &cow->extents);
 316         return 0;
 317 }
 318
 319 /*
 320  * we create compressed extents in two phases.  The first
 321  * phase compresses a range of pages that have already been
 322  * locked (both pages and state bits are locked).
 323  *
 324  * This is done inside an ordered work queue, and the compression
 325  * is spread across many cpus.  The actual IO submission is step
 326  * two, and the ordered work queue takes care of making sure that
 327  * happens in the same order things were put onto the queue by
 328  * writepages and friends.
 329  *
 330  * If this code finds it can't get good compression, it puts an
 331  * entry onto the work queue to write the uncompressed bytes.  This
 332  * makes sure that both compressed inodes and uncompressed inodes
 333  * are written in the same order that the flusher thread sent them
 334  * down.
 335  */
 336 static noinline int compress_file_range(struct inode *inode,
 337                                         struct page *locked_page,
 338                                         u64 start, u64 end,
 339                                         struct async_cow *async_cow,
 340                                         int *num_added)
 341 {
 342         struct btrfs_root *root = BTRFS_I(inode)->root;
 343         struct btrfs_trans_handle *trans;
 344         u64 num_bytes;
 345         u64 blocksize = root->sectorsize;
 346         u64 actual_end;
 347         u64 isize = i_size_read(inode);
 348         int ret = 0;
 349         struct page **pages = NULL;
 350         unsigned long nr_pages;
 351         unsigned long nr_pages_ret = 0;
 352         unsigned long total_compressed = 0;
 353         unsigned long total_in = 0;
 354         unsigned long max_compressed = 128 * 1024;
 355         unsigned long max_uncompressed = 128 * 1024;
 356         int i;
 357         int will_compress;
 358         int compress_type = root->fs_info->compress_type;
 359         int redirty = 0;
 360
 361         /* if this is a small write inside eof, kick off a defrag */
 362         if ((end - start + 1) < 16 * 1024 &&
 363             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 364                 btrfs_add_inode_defrag(NULL, inode);
 365
 366         actual_end = min_t(u64, isize, end + 1);
 367 again:
 368         will_compress = 0;
 369         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 370         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 371
 372         /*
 373          * we don't want to send crud past the end of i_size through
 374          * compression, that's just a waste of CPU time.  So, if the
 375          * end of the file is before the start of our current
 376          * requested range of bytes, we bail out to the uncompressed
 377          * cleanup code that can deal with all of this.
 378          *
 379          * It isn't really the fastest way to fix things, but this is a
 380          * very uncommon corner.
 381          */
 382         if (actual_end <= start)
 383                 goto cleanup_and_bail_uncompressed;
 384
 385         total_compressed = actual_end - start;
 386
 387         /* we want to make sure that amount of ram required to uncompress
 388          * an extent is reasonable, so we limit the total size in ram
 389          * of a compressed extent to 128k.  This is a crucial number
 390          * because it also controls how easily we can spread reads across
 391          * cpus for decompression.
 392          *
 393          * We also want to make sure the amount of IO required to do
 394          * a random read is reasonably small, so we limit the size of
 395          * a compressed extent to 128k.
 396          */
 397         total_compressed = min(total_compressed, max_uncompressed);
 398         num_bytes = ALIGN(end - start + 1, blocksize);
 399         num_bytes = max(blocksize,  num_bytes);
 400         total_in = 0;
 401         ret = 0;
 402
 403         /*
 404          * we do compression for mount -o compress and when the
 405          * inode has not been flagged as nocompress.  This flag can
 406          * change at any time if we discover bad compression ratios.
 407          */
 408         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 409             (btrfs_test_opt(root, COMPRESS) ||
 410              (BTRFS_I(inode)->force_compress) ||
 411              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 412                 WARN_ON(pages);
 413                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 414                 if (!pages) {
 415                         /* just bail out to the uncompressed code */
 416                         goto cont;
 417                 }
 418
 419                 if (BTRFS_I(inode)->force_compress)
 420                         compress_type = BTRFS_I(inode)->force_compress;
 421
 422                 /*
 423                  * we need to call clear_page_dirty_for_io on each
 424                  * page in the range.  Otherwise applications with the file
 425                  * mmap'd can wander in and change the page contents while
 426                  * we are compressing them.
 427                  *
 428                  * If the compression fails for any reason, we set the pages
 429                  * dirty again later on.
 430                  */
 431                 extent_range_clear_dirty_for_io(inode, start, end);
 432                 redirty = 1;
 433                 ret = btrfs_compress_pages(compress_type,
 434                                            inode->i_mapping, start,
 435                                            total_compressed, pages,
 436                                            nr_pages, &nr_pages_ret,
 437                                            &total_in,
 438                                            &total_compressed,
 439                                            max_compressed);
 440
 441                 if (!ret) {
 442                         unsigned long offset = total_compressed &
 443                                 (PAGE_CACHE_SIZE - 1);
 444                         struct page *page = pages[nr_pages_ret - 1];
 445                         char *kaddr;
 446
 447                         /* zero the tail end of the last page, we might be
 448                          * sending it down to disk
 449                          */
 450                         if (offset) {
 451                                 kaddr = kmap_atomic(page);
 452                                 memset(kaddr + offset, 0,
 453                                        PAGE_CACHE_SIZE - offset);
 454                                 kunmap_atomic(kaddr);
 455                         }
 456                         will_compress = 1;
 457                 }
 458         }
 459 cont:
 460         if (start == 0) {
 461                 trans = btrfs_join_transaction(root);
 462                 if (IS_ERR(trans)) {
 463                         ret = PTR_ERR(trans);
 464                         trans = NULL;
 465                         goto cleanup_and_out;
 466                 }
 467                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 468
 469                 /* lets try to make an inline extent */
 470                 if (ret || total_in < (actual_end - start)) {
 471                         /* we didn't compress the entire range, try
 472                          * to make an uncompressed inline extent.
 473                          */
 474                         ret = cow_file_range_inline(trans, root, inode,
 475                                                     start, end, 0, 0, NULL);
 476                 } else {
 477                         /* try making a compressed inline extent */
 478                         ret = cow_file_range_inline(trans, root, inode,
 479                                                     start, end,
 480                                                     total_compressed,
 481                                                     compress_type, pages);
 482                 }
 483                 if (ret <= 0) {
 484                         /*
 485                          * inline extent creation worked or returned error,
 486                          * we don't need to create any more async work items.
 487                          * Unlock and free up our temp pages.
 488                          */
 489                         extent_clear_unlock_delalloc(inode,
 490                              &BTRFS_I(inode)->io_tree,
 491                              start, end, NULL,
 492                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 493                              EXTENT_CLEAR_DELALLOC |
 494                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 495
 496                         btrfs_end_transaction(trans, root);
 497                         goto free_pages_out;
 498                 }
 499                 btrfs_end_transaction(trans, root);
 500         }
 501
 502         if (will_compress) {
 503                 /*
 504                  * we aren't doing an inline extent round the compressed size
 505                  * up to a block size boundary so the allocator does sane
 506                  * things
 507                  */
 508                 total_compressed = ALIGN(total_compressed, blocksize);
 509
 510                 /*
 511                  * one last check to make sure the compression is really a
 512                  * win, compare the page count read with the blocks on disk
 513                  */
 514                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 515                 if (total_compressed >= total_in) {
 516                         will_compress = 0;
 517                 } else {
 518                         num_bytes = total_in;
 519                 }
 520         }
 521         if (!will_compress && pages) {
 522                 /*
 523                  * the compression code ran but failed to make things smaller,
 524                  * free any pages it allocated and our page pointer array
 525                  */
 526                 for (i = 0; i < nr_pages_ret; i++) {
 527                         WARN_ON(pages[i]->mapping);
 528                         page_cache_release(pages[i]);
 529                 }
 530                 kfree(pages);
 531                 pages = NULL;
 532                 total_compressed = 0;
 533                 nr_pages_ret = 0;
 534
 535                 /* flag the file so we don't compress in the future */
 536                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 537                     !(BTRFS_I(inode)->force_compress)) {
 538                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 539                 }
 540         }
 541         if (will_compress) {
 542                 *num_added += 1;
 543
 544                 /* the async work queues will take care of doing actual
 545                  * allocation on disk for these compressed pages,
 546                  * and will submit them to the elevator.
 547                  */
 548                 add_async_extent(async_cow, start, num_bytes,
 549                                  total_compressed, pages, nr_pages_ret,
 550                                  compress_type);
 551
 552                 if (start + num_bytes < end) {
 553                         start += num_bytes;
 554                         pages = NULL;
 555                         cond_resched();
 556                         goto again;
 557                 }
 558         } else {
 559 cleanup_and_bail_uncompressed:
 560                 /*
 561                  * No compression, but we still need to write the pages in
 562                  * the file we've been given so far.  redirty the locked
 563                  * page if it corresponds to our extent and set things up
 564                  * for the async work queue to run cow_file_range to do
 565                  * the normal delalloc dance
 566                  */
 567                 if (page_offset(locked_page) >= start &&
 568                     page_offset(locked_page) <= end) {
 569                         __set_page_dirty_nobuffers(locked_page);
 570                         /* unlocked later on in the async handlers */
 571                 }
 572                 if (redirty)
 573                         extent_range_redirty_for_io(inode, start, end);
 574                 add_async_extent(async_cow, start, end - start + 1,
 575                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 576                 *num_added += 1;
 577         }
 578
 579 out:
 580         return ret;
 581
 582 free_pages_out:
 583         for (i = 0; i < nr_pages_ret; i++) {
 584                 WARN_ON(pages[i]->mapping);
 585                 page_cache_release(pages[i]);
 586         }
 587         kfree(pages);
 588
 589         goto out;
 590
 591 cleanup_and_out:
 592         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 593                                      start, end, NULL,
 594                                      EXTENT_CLEAR_UNLOCK_PAGE |
 595                                      EXTENT_CLEAR_DIRTY |
 596                                      EXTENT_CLEAR_DELALLOC |
 597                                      EXTENT_SET_WRITEBACK |
 598                                      EXTENT_END_WRITEBACK);
 599         if (!trans || IS_ERR(trans))
 600                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 601         else
 602                 btrfs_abort_transaction(trans, root, ret);
 603         goto free_pages_out;
 604 }
 605
 606 /*
 607  * phase two of compressed writeback.  This is the ordered portion
 608  * of the code, which only gets called in the order the work was
 609  * queued.  We walk all the async extents created by compress_file_range
 610  * and send them down to the disk.
 611  */
 612 static noinline int submit_compressed_extents(struct inode *inode,
 613                                               struct async_cow *async_cow)
 614 {
 615         struct async_extent *async_extent;
 616         u64 alloc_hint = 0;
 617         struct btrfs_trans_handle *trans;
 618         struct btrfs_key ins;
 619         struct extent_map *em;
 620         struct btrfs_root *root = BTRFS_I(inode)->root;
 621         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 622         struct extent_io_tree *io_tree;
 623         int ret = 0;
 624
 625         if (list_empty(&async_cow->extents))
 626                 return 0;
 627
 628 again:
 629         while (!list_empty(&async_cow->extents)) {
 630                 async_extent = list_entry(async_cow->extents.next,
 631                                           struct async_extent, list);
 632                 list_del(&async_extent->list);
 633
 634                 io_tree = &BTRFS_I(inode)->io_tree;
 635
 636 retry:
 637                 /* did the compression code fall back to uncompressed IO? */
 638                 if (!async_extent->pages) {
 639                         int page_started = 0;
 640                         unsigned long nr_written = 0;
 641
 642                         lock_extent(io_tree, async_extent->start,
 643                                          async_extent->start +
 644                                          async_extent->ram_size - 1);
 645
 646                         /* allocate blocks */
 647                         ret = cow_file_range(inode, async_cow->locked_page,
 648                                              async_extent->start,
 649                                              async_extent->start +
 650                                              async_extent->ram_size - 1,
 651                                              &page_started, &nr_written, 0);
 652
 653                         /* JDM XXX */
 654
 655                         /*
 656                          * if page_started, cow_file_range inserted an
 657                          * inline extent and took care of all the unlocking
 658                          * and IO for us.  Otherwise, we need to submit
 659                          * all those pages down to the drive.
 660                          */
 661                         if (!page_started && !ret)
 662                                 extent_write_locked_range(io_tree,
 663                                                   inode, async_extent->start,
 664                                                   async_extent->start +
 665                                                   async_extent->ram_size - 1,
 666                                                   btrfs_get_extent,
 667                                                   WB_SYNC_ALL);
 668                         else if (ret)
 669                                 unlock_page(async_cow->locked_page);
 670                         kfree(async_extent);
 671                         cond_resched();
 672                         continue;
 673                 }
 674
 675                 lock_extent(io_tree, async_extent->start,
 676                             async_extent->start + async_extent->ram_size - 1);
 677
 678                 trans = btrfs_join_transaction(root);
 679                 if (IS_ERR(trans)) {
 680                         ret = PTR_ERR(trans);
 681                 } else {
 682                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 683                         ret = btrfs_reserve_extent(trans, root,
 684                                            async_extent->compressed_size,
 685                                            async_extent->compressed_size,
 686                                            0, alloc_hint, &ins, 1);
 687                         if (ret && ret != -ENOSPC)
 688                                 btrfs_abort_transaction(trans, root, ret);
 689                         btrfs_end_transaction(trans, root);
 690                 }
 691
 692                 if (ret) {
 693                         int i;
 694
 695                         for (i = 0; i < async_extent->nr_pages; i++) {
 696                                 WARN_ON(async_extent->pages[i]->mapping);
 697                                 page_cache_release(async_extent->pages[i]);
 698                         }
 699                         kfree(async_extent->pages);
 700                         async_extent->nr_pages = 0;
 701                         async_extent->pages = NULL;
 702
 703                         if (ret == -ENOSPC)
 704                                 goto retry;
 705                         goto out_free;
 706                 }
 707
 708                 /*
 709                  * here we're doing allocation and writeback of the
 710                  * compressed pages
 711                  */
 712                 btrfs_drop_extent_cache(inode, async_extent->start,
 713                                         async_extent->start +
 714                                         async_extent->ram_size - 1, 0);
 715
 716                 em = alloc_extent_map();
 717                 if (!em) {
 718                         ret = -ENOMEM;
 719                         goto out_free_reserve;
 720                 }
 721                 em->start = async_extent->start;
 722                 em->len = async_extent->ram_size;
 723                 em->orig_start = em->start;
 724                 em->mod_start = em->start;
 725                 em->mod_len = em->len;
 726
 727                 em->block_start = ins.objectid;
 728                 em->block_len = ins.offset;
 729                 em->orig_block_len = ins.offset;
 730                 em->ram_bytes = async_extent->ram_size;
 731                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 732                 em->compress_type = async_extent->compress_type;
 733                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 734                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 735                 em->generation = -1;
 736
 737                 while (1) {
 738                         write_lock(&em_tree->lock);
 739                         ret = add_extent_mapping(em_tree, em, 1);
 740                         write_unlock(&em_tree->lock);
 741                         if (ret != -EEXIST) {
 742                                 free_extent_map(em);
 743                                 break;
 744                         }
 745                         btrfs_drop_extent_cache(inode, async_extent->start,
 746                                                 async_extent->start +
 747                                                 async_extent->ram_size - 1, 0);
 748                 }
 749
 750                 if (ret)
 751                         goto out_free_reserve;
 752
 753                 ret = btrfs_add_ordered_extent_compress(inode,
 754                                                 async_extent->start,
 755                                                 ins.objectid,
 756                                                 async_extent->ram_size,
 757                                                 ins.offset,
 758                                                 BTRFS_ORDERED_COMPRESSED,
 759                                                 async_extent->compress_type);
 760                 if (ret)
 761                         goto out_free_reserve;
 762
 763                 /*
 764                  * clear dirty, set writeback and unlock the pages.
 765                  */
 766                 extent_clear_unlock_delalloc(inode,
 767                                 &BTRFS_I(inode)->io_tree,
 768                                 async_extent->start,
 769                                 async_extent->start +
 770                                 async_extent->ram_size - 1,
 771                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 772                                 EXTENT_CLEAR_UNLOCK |
 773                                 EXTENT_CLEAR_DELALLOC |
 774                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 775
 776                 ret = btrfs_submit_compressed_write(inode,
 777                                     async_extent->start,
 778                                     async_extent->ram_size,
 779                                     ins.objectid,
 780                                     ins.offset, async_extent->pages,
 781                                     async_extent->nr_pages);
 782                 alloc_hint = ins.objectid + ins.offset;
 783                 kfree(async_extent);
 784                 if (ret)
 785                         goto out;
 786                 cond_resched();
 787         }
 788         ret = 0;
 789 out:
 790         return ret;
 791 out_free_reserve:
 792         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 793 out_free:
 794         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 795                                      async_extent->start,
 796                                      async_extent->start +
 797                                      async_extent->ram_size - 1,
 798                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 799                                      EXTENT_CLEAR_UNLOCK |
 800                                      EXTENT_CLEAR_DELALLOC |
 801                                      EXTENT_CLEAR_DIRTY |
 802                                      EXTENT_SET_WRITEBACK |
 803                                      EXTENT_END_WRITEBACK);
 804         kfree(async_extent);
 805         goto again;
 806 }
 807
 808 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 809                                       u64 num_bytes)
 810 {
 811         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 812         struct extent_map *em;
 813         u64 alloc_hint = 0;
 814
 815         read_lock(&em_tree->lock);
 816         em = search_extent_mapping(em_tree, start, num_bytes);
 817         if (em) {
 818                 /*
 819                  * if block start isn't an actual block number then find the
 820                  * first block in this inode and use that as a hint.  If that
 821                  * block is also bogus then just don't worry about it.
 822                  */
 823                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 824                         free_extent_map(em);
 825                         em = search_extent_mapping(em_tree, 0, 0);
 826                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 827                                 alloc_hint = em->block_start;
 828                         if (em)
 829                                 free_extent_map(em);
 830                 } else {
 831                         alloc_hint = em->block_start;
 832                         free_extent_map(em);
 833                 }
 834         }
 835         read_unlock(&em_tree->lock);
 836
 837         return alloc_hint;
 838 }
 839
 840 /*
 841  * when extent_io.c finds a delayed allocation range in the file,
 842  * the call backs end up in this code.  The basic idea is to
 843  * allocate extents on disk for the range, and create ordered data structs
 844  * in ram to track those extents.
 845  *
 846  * locked_page is the page that writepage had locked already.  We use
 847  * it to make sure we don't do extra locks or unlocks.
 848  *
 849  * *page_started is set to one if we unlock locked_page and do everything
 850  * required to start IO on it.  It may be clean and already done with
 851  * IO when we return.
 852  */
 853 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 854                                      struct inode *inode,
 855                                      struct btrfs_root *root,
 856                                      struct page *locked_page,
 857                                      u64 start, u64 end, int *page_started,
 858                                      unsigned long *nr_written,
 859                                      int unlock)
 860 {
 861         u64 alloc_hint = 0;
 862         u64 num_bytes;
 863         unsigned long ram_size;
 864         u64 disk_num_bytes;
 865         u64 cur_alloc_size;
 866         u64 blocksize = root->sectorsize;
 867         struct btrfs_key ins;
 868         struct extent_map *em;
 869         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870         int ret = 0;
 871
 872         BUG_ON(btrfs_is_free_space_inode(inode));
 873
 874         num_bytes = ALIGN(end - start + 1, blocksize);
 875         num_bytes = max(blocksize,  num_bytes);
 876         disk_num_bytes = num_bytes;
 877
 878         /* if this is a small write inside eof, kick off defrag */
 879         if (num_bytes < 64 * 1024 &&
 880             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 881                 btrfs_add_inode_defrag(trans, inode);
 882
 883         if (start == 0) {
 884                 /* lets try to make an inline extent */
 885                 ret = cow_file_range_inline(trans, root, inode,
 886                                             start, end, 0, 0, NULL);
 887                 if (ret == 0) {
 888                         extent_clear_unlock_delalloc(inode,
 889                                      &BTRFS_I(inode)->io_tree,
 890                                      start, end, NULL,
 891                                      EXTENT_CLEAR_UNLOCK_PAGE |
 892                                      EXTENT_CLEAR_UNLOCK |
 893                                      EXTENT_CLEAR_DELALLOC |
 894                                      EXTENT_CLEAR_DIRTY |
 895                                      EXTENT_SET_WRITEBACK |
 896                                      EXTENT_END_WRITEBACK);
 897
 898                         *nr_written = *nr_written +
 899                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 900                         *page_started = 1;
 901                         goto out;
 902                 } else if (ret < 0) {
 903                         btrfs_abort_transaction(trans, root, ret);
 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(trans, root, cur_alloc_size,
 919                                            root->sectorsize, 0, alloc_hint,
 920                                            &ins, 1);
 921                 if (ret < 0) {
 922                         btrfs_abort_transaction(trans, root, ret);
 923                         goto out_unlock;
 924                 }
 925
 926                 em = alloc_extent_map();
 927                 if (!em) {
 928                         ret = -ENOMEM;
 929                         goto out_reserve;
 930                 }
 931                 em->start = start;
 932                 em->orig_start = em->start;
 933                 ram_size = ins.offset;
 934                 em->len = ins.offset;
 935                 em->mod_start = em->start;
 936                 em->mod_len = em->len;
 937
 938                 em->block_start = ins.objectid;
 939                 em->block_len = ins.offset;
 940                 em->orig_block_len = ins.offset;
 941                 em->ram_bytes = ram_size;
 942                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 943                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 944                 em->generation = -1;
 945
 946                 while (1) {
 947                         write_lock(&em_tree->lock);
 948                         ret = add_extent_mapping(em_tree, em, 1);
 949                         write_unlock(&em_tree->lock);
 950                         if (ret != -EEXIST) {
 951                                 free_extent_map(em);
 952                                 break;
 953                         }
 954                         btrfs_drop_extent_cache(inode, start,
 955                                                 start + ram_size - 1, 0);
 956                 }
 957                 if (ret)
 958                         goto out_reserve;
 959
 960                 cur_alloc_size = ins.offset;
 961                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 962                                                ram_size, cur_alloc_size, 0);
 963                 if (ret)
 964                         goto out_reserve;
 965
 966                 if (root->root_key.objectid ==
 967                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 968                         ret = btrfs_reloc_clone_csums(inode, start,
 969                                                       cur_alloc_size);
 970                         if (ret) {
 971                                 btrfs_abort_transaction(trans, root, ret);
 972                                 goto out_reserve;
 973                         }
 974                 }
 975
 976                 if (disk_num_bytes < cur_alloc_size)
 977                         break;
 978
 979                 /* we're not doing compressed IO, don't unlock the first
 980                  * page (which the caller expects to stay locked), don't
 981                  * clear any dirty bits and don't set any writeback bits
 982                  *
 983                  * Do set the Private2 bit so we know this page was properly
 984                  * setup for writepage
 985                  */
 986                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 987                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 988                         EXTENT_SET_PRIVATE2;
 989
 990                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 991                                              start, start + ram_size - 1,
 992                                              locked_page, op);
 993                 disk_num_bytes -= cur_alloc_size;
 994                 num_bytes -= cur_alloc_size;
 995                 alloc_hint = ins.objectid + ins.offset;
 996                 start += cur_alloc_size;
 997         }
 998 out:
 999         return ret;
1000
1001 out_reserve:
1002         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1003 out_unlock:
1004         extent_clear_unlock_delalloc(inode,
1005                      &BTRFS_I(inode)->io_tree,
1006                      start, end, locked_page,
1007                      EXTENT_CLEAR_UNLOCK_PAGE |
1008                      EXTENT_CLEAR_UNLOCK |
1009                      EXTENT_CLEAR_DELALLOC |
1010                      EXTENT_CLEAR_DIRTY |
1011                      EXTENT_SET_WRITEBACK |
1012                      EXTENT_END_WRITEBACK);
1013
1014         goto out;
1015 }
1016
1017 static noinline int cow_file_range(struct inode *inode,
1018                                    struct page *locked_page,
1019                                    u64 start, u64 end, int *page_started,
1020                                    unsigned long *nr_written,
1021                                    int unlock)
1022 {
1023         struct btrfs_trans_handle *trans;
1024         struct btrfs_root *root = BTRFS_I(inode)->root;
1025         int ret;
1026
1027         trans = btrfs_join_transaction(root);
1028         if (IS_ERR(trans)) {
1029                 extent_clear_unlock_delalloc(inode,
1030                              &BTRFS_I(inode)->io_tree,
1031                              start, end, locked_page,
1032                              EXTENT_CLEAR_UNLOCK_PAGE |
1033                              EXTENT_CLEAR_UNLOCK |
1034                              EXTENT_CLEAR_DELALLOC |
1035                              EXTENT_CLEAR_DIRTY |
1036                              EXTENT_SET_WRITEBACK |
1037                              EXTENT_END_WRITEBACK);
1038                 return PTR_ERR(trans);
1039         }
1040         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1041
1042         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1043                                page_started, nr_written, unlock);
1044
1045         btrfs_end_transaction(trans, root);
1046
1047         return ret;
1048 }
1049
1050 /*
1051  * work queue call back to started compression on a file and pages
1052  */
1053 static noinline void async_cow_start(struct btrfs_work *work)
1054 {
1055         struct async_cow *async_cow;
1056         int num_added = 0;
1057         async_cow = container_of(work, struct async_cow, work);
1058
1059         compress_file_range(async_cow->inode, async_cow->locked_page,
1060                             async_cow->start, async_cow->end, async_cow,
1061                             &num_added);
1062         if (num_added == 0) {
1063                 btrfs_add_delayed_iput(async_cow->inode);
1064                 async_cow->inode = NULL;
1065         }
1066 }
1067
1068 /*
1069  * work queue call back to submit previously compressed pages
1070  */
1071 static noinline void async_cow_submit(struct btrfs_work *work)
1072 {
1073         struct async_cow *async_cow;
1074         struct btrfs_root *root;
1075         unsigned long nr_pages;
1076
1077         async_cow = container_of(work, struct async_cow, work);
1078
1079         root = async_cow->root;
1080         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1081                 PAGE_CACHE_SHIFT;
1082
1083         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1084             5 * 1024 * 1024 &&
1085             waitqueue_active(&root->fs_info->async_submit_wait))
1086                 wake_up(&root->fs_info->async_submit_wait);
1087
1088         if (async_cow->inode)
1089                 submit_compressed_extents(async_cow->inode, async_cow);
1090 }
1091
1092 static noinline void async_cow_free(struct btrfs_work *work)
1093 {
1094         struct async_cow *async_cow;
1095         async_cow = container_of(work, struct async_cow, work);
1096         if (async_cow->inode)
1097                 btrfs_add_delayed_iput(async_cow->inode);
1098         kfree(async_cow);
1099 }
1100
1101 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1102                                 u64 start, u64 end, int *page_started,
1103                                 unsigned long *nr_written)
1104 {
1105         struct async_cow *async_cow;
1106         struct btrfs_root *root = BTRFS_I(inode)->root;
1107         unsigned long nr_pages;
1108         u64 cur_end;
1109         int limit = 10 * 1024 * 1024;
1110
1111         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1112                          1, 0, NULL, GFP_NOFS);
1113         while (start < end) {
1114                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1115                 BUG_ON(!async_cow); /* -ENOMEM */
1116                 async_cow->inode = igrab(inode);
1117                 async_cow->root = root;
1118                 async_cow->locked_page = locked_page;
1119                 async_cow->start = start;
1120
1121                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1122                         cur_end = end;
1123                 else
1124                         cur_end = min(end, start + 512 * 1024 - 1);
1125
1126                 async_cow->end = cur_end;
1127                 INIT_LIST_HEAD(&async_cow->extents);
1128
1129                 async_cow->work.func = async_cow_start;
1130                 async_cow->work.ordered_func = async_cow_submit;
1131                 async_cow->work.ordered_free = async_cow_free;
1132                 async_cow->work.flags = 0;
1133
1134                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1135                         PAGE_CACHE_SHIFT;
1136                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1137
1138                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1139                                    &async_cow->work);
1140
1141                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1142                         wait_event(root->fs_info->async_submit_wait,
1143                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1144                             limit));
1145                 }
1146
1147                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1148                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1149                         wait_event(root->fs_info->async_submit_wait,
1150                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1151                            0));
1152                 }
1153
1154                 *nr_written += nr_pages;
1155                 start = cur_end + 1;
1156         }
1157         *page_started = 1;
1158         return 0;
1159 }
1160
1161 static noinline int csum_exist_in_range(struct btrfs_root *root,
1162                                         u64 bytenr, u64 num_bytes)
1163 {
1164         int ret;
1165         struct btrfs_ordered_sum *sums;
1166         LIST_HEAD(list);
1167
1168         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1169                                        bytenr + num_bytes - 1, &list, 0);
1170         if (ret == 0 && list_empty(&list))
1171                 return 0;
1172
1173         while (!list_empty(&list)) {
1174                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1175                 list_del(&sums->list);
1176                 kfree(sums);
1177         }
1178         return 1;
1179 }
1180
1181 /*
1182  * when nowcow writeback call back.  This checks for snapshots or COW copies
1183  * of the extents that exist in the file, and COWs the file as required.
1184  *
1185  * If no cow copies or snapshots exist, we write directly to the existing
1186  * blocks on disk
1187  */
1188 static noinline int run_delalloc_nocow(struct inode *inode,
1189                                        struct page *locked_page,
1190                               u64 start, u64 end, int *page_started, int force,
1191                               unsigned long *nr_written)
1192 {
1193         struct btrfs_root *root = BTRFS_I(inode)->root;
1194         struct btrfs_trans_handle *trans;
1195         struct extent_buffer *leaf;
1196         struct btrfs_path *path;
1197         struct btrfs_file_extent_item *fi;
1198         struct btrfs_key found_key;
1199         u64 cow_start;
1200         u64 cur_offset;
1201         u64 extent_end;
1202         u64 extent_offset;
1203         u64 disk_bytenr;
1204         u64 num_bytes;
1205         u64 disk_num_bytes;
1206         u64 ram_bytes;
1207         int extent_type;
1208         int ret, err;
1209         int type;
1210         int nocow;
1211         int check_prev = 1;
1212         bool nolock;
1213         u64 ino = btrfs_ino(inode);
1214
1215         path = btrfs_alloc_path();
1216         if (!path) {
1217                 extent_clear_unlock_delalloc(inode,
1218                              &BTRFS_I(inode)->io_tree,
1219                              start, end, locked_page,
1220                              EXTENT_CLEAR_UNLOCK_PAGE |
1221                              EXTENT_CLEAR_UNLOCK |
1222                              EXTENT_CLEAR_DELALLOC |
1223                              EXTENT_CLEAR_DIRTY |
1224                              EXTENT_SET_WRITEBACK |
1225                              EXTENT_END_WRITEBACK);
1226                 return -ENOMEM;
1227         }
1228
1229         nolock = btrfs_is_free_space_inode(inode);
1230
1231         if (nolock)
1232                 trans = btrfs_join_transaction_nolock(root);
1233         else
1234                 trans = btrfs_join_transaction(root);
1235
1236         if (IS_ERR(trans)) {
1237                 extent_clear_unlock_delalloc(inode,
1238                              &BTRFS_I(inode)->io_tree,
1239                              start, end, locked_page,
1240                              EXTENT_CLEAR_UNLOCK_PAGE |
1241                              EXTENT_CLEAR_UNLOCK |
1242                              EXTENT_CLEAR_DELALLOC |
1243                              EXTENT_CLEAR_DIRTY |
1244                              EXTENT_SET_WRITEBACK |
1245                              EXTENT_END_WRITEBACK);
1246                 btrfs_free_path(path);
1247                 return PTR_ERR(trans);
1248         }
1249
1250         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1251
1252         cow_start = (u64)-1;
1253         cur_offset = start;
1254         while (1) {
1255                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1256                                                cur_offset, 0);
1257                 if (ret < 0) {
1258                         btrfs_abort_transaction(trans, root, ret);
1259                         goto error;
1260                 }
1261                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1262                         leaf = path->nodes[0];
1263                         btrfs_item_key_to_cpu(leaf, &found_key,
1264                                               path->slots[0] - 1);
1265                         if (found_key.objectid == ino &&
1266                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1267                                 path->slots[0]--;
1268                 }
1269                 check_prev = 0;
1270 next_slot:
1271                 leaf = path->nodes[0];
1272                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1273                         ret = btrfs_next_leaf(root, path);
1274                         if (ret < 0) {
1275                                 btrfs_abort_transaction(trans, root, ret);
1276                                 goto error;
1277                         }
1278                         if (ret > 0)
1279                                 break;
1280                         leaf = path->nodes[0];
1281                 }
1282
1283                 nocow = 0;
1284                 disk_bytenr = 0;
1285                 num_bytes = 0;
1286                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1287
1288                 if (found_key.objectid > ino ||
1289                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1290                     found_key.offset > end)
1291                         break;
1292
1293                 if (found_key.offset > cur_offset) {
1294                         extent_end = found_key.offset;
1295                         extent_type = 0;
1296                         goto out_check;
1297                 }
1298
1299                 fi = btrfs_item_ptr(leaf, path->slots[0],
1300                                     struct btrfs_file_extent_item);
1301                 extent_type = btrfs_file_extent_type(leaf, fi);
1302
1303                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1304                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1305                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1306                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1307                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1308                         extent_end = found_key.offset +
1309                                 btrfs_file_extent_num_bytes(leaf, fi);
1310                         disk_num_bytes =
1311                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1312                         if (extent_end <= start) {
1313                                 path->slots[0]++;
1314                                 goto next_slot;
1315                         }
1316                         if (disk_bytenr == 0)
1317                                 goto out_check;
1318                         if (btrfs_file_extent_compression(leaf, fi) ||
1319                             btrfs_file_extent_encryption(leaf, fi) ||
1320                             btrfs_file_extent_other_encoding(leaf, fi))
1321                                 goto out_check;
1322                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1323                                 goto out_check;
1324                         if (btrfs_extent_readonly(root, disk_bytenr))
1325                                 goto out_check;
1326                         if (btrfs_cross_ref_exist(trans, root, ino,
1327                                                   found_key.offset -
1328                                                   extent_offset, disk_bytenr))
1329                                 goto out_check;
1330                         disk_bytenr += extent_offset;
1331                         disk_bytenr += cur_offset - found_key.offset;
1332                         num_bytes = min(end + 1, extent_end) - cur_offset;
1333                         /*
1334                          * force cow if csum exists in the range.
1335                          * this ensure that csum for a given extent are
1336                          * either valid or do not exist.
1337                          */
1338                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1339                                 goto out_check;
1340                         nocow = 1;
1341                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1342                         extent_end = found_key.offset +
1343                                 btrfs_file_extent_inline_len(leaf, fi);
1344                         extent_end = ALIGN(extent_end, root->sectorsize);
1345                 } else {
1346                         BUG_ON(1);
1347                 }
1348 out_check:
1349                 if (extent_end <= start) {
1350                         path->slots[0]++;
1351                         goto next_slot;
1352                 }
1353                 if (!nocow) {
1354                         if (cow_start == (u64)-1)
1355                                 cow_start = cur_offset;
1356                         cur_offset = extent_end;
1357                         if (cur_offset > end)
1358                                 break;
1359                         path->slots[0]++;
1360                         goto next_slot;
1361                 }
1362
1363                 btrfs_release_path(path);
1364                 if (cow_start != (u64)-1) {
1365                         ret = __cow_file_range(trans, inode, root, locked_page,
1366                                                cow_start, found_key.offset - 1,
1367                                                page_started, nr_written, 1);
1368                         if (ret) {
1369                                 btrfs_abort_transaction(trans, root, ret);
1370                                 goto error;
1371                         }
1372                         cow_start = (u64)-1;
1373                 }
1374
1375                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376                         struct extent_map *em;
1377                         struct extent_map_tree *em_tree;
1378                         em_tree = &BTRFS_I(inode)->extent_tree;
1379                         em = alloc_extent_map();
1380                         BUG_ON(!em); /* -ENOMEM */
1381                         em->start = cur_offset;
1382                         em->orig_start = found_key.offset - extent_offset;
1383                         em->len = num_bytes;
1384                         em->block_len = num_bytes;
1385                         em->block_start = disk_bytenr;
1386                         em->orig_block_len = disk_num_bytes;
1387                         em->ram_bytes = ram_bytes;
1388                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1389                         em->mod_start = em->start;
1390                         em->mod_len = em->len;
1391                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1392                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1393                         em->generation = -1;
1394                         while (1) {
1395                                 write_lock(&em_tree->lock);
1396                                 ret = add_extent_mapping(em_tree, em, 1);
1397                                 write_unlock(&em_tree->lock);
1398                                 if (ret != -EEXIST) {
1399                                         free_extent_map(em);
1400                                         break;
1401                                 }
1402                                 btrfs_drop_extent_cache(inode, em->start,
1403                                                 em->start + em->len - 1, 0);
1404                         }
1405                         type = BTRFS_ORDERED_PREALLOC;
1406                 } else {
1407                         type = BTRFS_ORDERED_NOCOW;
1408                 }
1409
1410                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1411                                                num_bytes, num_bytes, type);
1412                 BUG_ON(ret); /* -ENOMEM */
1413
1414                 if (root->root_key.objectid ==
1415                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1416                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1417                                                       num_bytes);
1418                         if (ret) {
1419                                 btrfs_abort_transaction(trans, root, ret);
1420                                 goto error;
1421                         }
1422                 }
1423
1424                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1425                                 cur_offset, cur_offset + num_bytes - 1,
1426                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1427                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1428                                 EXTENT_SET_PRIVATE2);
1429                 cur_offset = extent_end;
1430                 if (cur_offset > end)
1431                         break;
1432         }
1433         btrfs_release_path(path);
1434
1435         if (cur_offset <= end && cow_start == (u64)-1) {
1436                 cow_start = cur_offset;
1437                 cur_offset = end;
1438         }
1439
1440         if (cow_start != (u64)-1) {
1441                 ret = __cow_file_range(trans, inode, root, locked_page,
1442                                        cow_start, end,
1443                                        page_started, nr_written, 1);
1444                 if (ret) {
1445                         btrfs_abort_transaction(trans, root, ret);
1446                         goto error;
1447                 }
1448         }
1449
1450 error:
1451         err = btrfs_end_transaction(trans, root);
1452         if (!ret)
1453                 ret = err;
1454
1455         if (ret && cur_offset < end)
1456                 extent_clear_unlock_delalloc(inode,
1457                              &BTRFS_I(inode)->io_tree,
1458                              cur_offset, end, locked_page,
1459                              EXTENT_CLEAR_UNLOCK_PAGE |
1460                              EXTENT_CLEAR_UNLOCK |
1461                              EXTENT_CLEAR_DELALLOC |
1462                              EXTENT_CLEAR_DIRTY |
1463                              EXTENT_SET_WRITEBACK |
1464                              EXTENT_END_WRITEBACK);
1465
1466         btrfs_free_path(path);
1467         return ret;
1468 }
1469
1470 /*
1471  * extent_io.c call back to do delayed allocation processing
1472  */
1473 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1474                               u64 start, u64 end, int *page_started,
1475                               unsigned long *nr_written)
1476 {
1477         int ret;
1478         struct btrfs_root *root = BTRFS_I(inode)->root;
1479
1480         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1481                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1482                                          page_started, 1, nr_written);
1483         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1484                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1485                                          page_started, 0, nr_written);
1486         } else if (!btrfs_test_opt(root, COMPRESS) &&
1487                    !(BTRFS_I(inode)->force_compress) &&
1488                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1489                 ret = cow_file_range(inode, locked_page, start, end,
1490                                       page_started, nr_written, 1);
1491         } else {
1492                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1493                         &BTRFS_I(inode)->runtime_flags);
1494                 ret = cow_file_range_async(inode, locked_page, start, end,
1495                                            page_started, nr_written);
1496         }
1497         return ret;
1498 }
1499
1500 static void btrfs_split_extent_hook(struct inode *inode,
1501                                     struct extent_state *orig, u64 split)
1502 {
1503         /* not delalloc, ignore it */
1504         if (!(orig->state & EXTENT_DELALLOC))
1505                 return;
1506
1507         spin_lock(&BTRFS_I(inode)->lock);
1508         BTRFS_I(inode)->outstanding_extents++;
1509         spin_unlock(&BTRFS_I(inode)->lock);
1510 }
1511
1512 /*
1513  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514  * extents so we can keep track of new extents that are just merged onto old
1515  * extents, such as when we are doing sequential writes, so we can properly
1516  * account for the metadata space we'll need.
1517  */
1518 static void btrfs_merge_extent_hook(struct inode *inode,
1519                                     struct extent_state *new,
1520                                     struct extent_state *other)
1521 {
1522         /* not delalloc, ignore it */
1523         if (!(other->state & EXTENT_DELALLOC))
1524                 return;
1525
1526         spin_lock(&BTRFS_I(inode)->lock);
1527         BTRFS_I(inode)->outstanding_extents--;
1528         spin_unlock(&BTRFS_I(inode)->lock);
1529 }
1530
1531 /*
1532  * extent_io.c set_bit_hook, used to track delayed allocation
1533  * bytes in this file, and to maintain the list of inodes that
1534  * have pending delalloc work to be done.
1535  */
1536 static void btrfs_set_bit_hook(struct inode *inode,
1537                                struct extent_state *state, unsigned long *bits)
1538 {
1539
1540         /*
1541          * set_bit and clear bit hooks normally require _irqsave/restore
1542          * but in this case, we are only testing for the DELALLOC
1543          * bit, which is only set or cleared with irqs on
1544          */
1545         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1546                 struct btrfs_root *root = BTRFS_I(inode)->root;
1547                 u64 len = state->end + 1 - state->start;
1548                 bool do_list = !btrfs_is_free_space_inode(inode);
1549
1550                 if (*bits & EXTENT_FIRST_DELALLOC) {
1551                         *bits &= ~EXTENT_FIRST_DELALLOC;
1552                 } else {
1553                         spin_lock(&BTRFS_I(inode)->lock);
1554                         BTRFS_I(inode)->outstanding_extents++;
1555                         spin_unlock(&BTRFS_I(inode)->lock);
1556                 }
1557
1558                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1559                                      root->fs_info->delalloc_batch);
1560                 spin_lock(&BTRFS_I(inode)->lock);
1561                 BTRFS_I(inode)->delalloc_bytes += len;
1562                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1563                                          &BTRFS_I(inode)->runtime_flags)) {
1564                         spin_lock(&root->fs_info->delalloc_lock);
1565                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567                                               &root->fs_info->delalloc_inodes);
1568                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569                                         &BTRFS_I(inode)->runtime_flags);
1570                         }
1571                         spin_unlock(&root->fs_info->delalloc_lock);
1572                 }
1573                 spin_unlock(&BTRFS_I(inode)->lock);
1574         }
1575 }
1576
1577 /*
1578  * extent_io.c clear_bit_hook, see set_bit_hook for why
1579  */
1580 static void btrfs_clear_bit_hook(struct inode *inode,
1581                                  struct extent_state *state,
1582                                  unsigned long *bits)
1583 {
1584         /*
1585          * set_bit and clear bit hooks normally require _irqsave/restore
1586          * but in this case, we are only testing for the DELALLOC
1587          * bit, which is only set or cleared with irqs on
1588          */
1589         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1590                 struct btrfs_root *root = BTRFS_I(inode)->root;
1591                 u64 len = state->end + 1 - state->start;
1592                 bool do_list = !btrfs_is_free_space_inode(inode);
1593
1594                 if (*bits & EXTENT_FIRST_DELALLOC) {
1595                         *bits &= ~EXTENT_FIRST_DELALLOC;
1596                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1597                         spin_lock(&BTRFS_I(inode)->lock);
1598                         BTRFS_I(inode)->outstanding_extents--;
1599                         spin_unlock(&BTRFS_I(inode)->lock);
1600                 }
1601
1602                 if (*bits & EXTENT_DO_ACCOUNTING)
1603                         btrfs_delalloc_release_metadata(inode, len);
1604
1605                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1606                     && do_list)
1607                         btrfs_free_reserved_data_space(inode, len);
1608
1609                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1610                                      root->fs_info->delalloc_batch);
1611                 spin_lock(&BTRFS_I(inode)->lock);
1612                 BTRFS_I(inode)->delalloc_bytes -= len;
1613                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1614                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1615                              &BTRFS_I(inode)->runtime_flags)) {
1616                         spin_lock(&root->fs_info->delalloc_lock);
1617                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1618                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1619                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1620                                           &BTRFS_I(inode)->runtime_flags);
1621                         }
1622                         spin_unlock(&root->fs_info->delalloc_lock);
1623                 }
1624                 spin_unlock(&BTRFS_I(inode)->lock);
1625         }
1626 }
1627
1628 /*
1629  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1630  * we don't create bios that span stripes or chunks
1631  */
1632 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1633                          size_t size, struct bio *bio,
1634                          unsigned long bio_flags)
1635 {
1636         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1637         u64 logical = (u64)bio->bi_sector << 9;
1638         u64 length = 0;
1639         u64 map_length;
1640         int ret;
1641
1642         if (bio_flags & EXTENT_BIO_COMPRESSED)
1643                 return 0;
1644
1645         length = bio->bi_size;
1646         map_length = length;
1647         ret = btrfs_map_block(root->fs_info, rw, logical,
1648                               &map_length, NULL, 0);
1649         /* Will always return 0 with map_multi == NULL */
1650         BUG_ON(ret < 0);
1651         if (map_length < length + size)
1652                 return 1;
1653         return 0;
1654 }
1655
1656 /*
1657  * in order to insert checksums into the metadata in large chunks,
1658  * we wait until bio submission time.   All the pages in the bio are
1659  * checksummed and sums are attached onto the ordered extent record.
1660  *
1661  * At IO completion time the cums attached on the ordered extent record
1662  * are inserted into the btree
1663  */
1664 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1665                                     struct bio *bio, int mirror_num,
1666                                     unsigned long bio_flags,
1667                                     u64 bio_offset)
1668 {
1669         struct btrfs_root *root = BTRFS_I(inode)->root;
1670         int ret = 0;
1671
1672         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1673         BUG_ON(ret); /* -ENOMEM */
1674         return 0;
1675 }
1676
1677 /*
1678  * in order to insert checksums into the metadata in large chunks,
1679  * we wait until bio submission time.   All the pages in the bio are
1680  * checksummed and sums are attached onto the ordered extent record.
1681  *
1682  * At IO completion time the cums attached on the ordered extent record
1683  * are inserted into the btree
1684  */
1685 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1686                           int mirror_num, unsigned long bio_flags,
1687                           u64 bio_offset)
1688 {
1689         struct btrfs_root *root = BTRFS_I(inode)->root;
1690         int ret;
1691
1692         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1693         if (ret)
1694                 bio_endio(bio, ret);
1695         return ret;
1696 }
1697
1698 /*
1699  * extent_io.c submission hook. This does the right thing for csum calculation
1700  * on write, or reading the csums from the tree before a read
1701  */
1702 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1703                           int mirror_num, unsigned long bio_flags,
1704                           u64 bio_offset)
1705 {
1706         struct btrfs_root *root = BTRFS_I(inode)->root;
1707         int ret = 0;
1708         int skip_sum;
1709         int metadata = 0;
1710         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1711
1712         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1713
1714         if (btrfs_is_free_space_inode(inode))
1715                 metadata = 2;
1716
1717         if (!(rw & REQ_WRITE)) {
1718                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1719                 if (ret)
1720                         goto out;
1721
1722                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1723                         ret = btrfs_submit_compressed_read(inode, bio,
1724                                                            mirror_num,
1725                                                            bio_flags);
1726                         goto out;
1727                 } else if (!skip_sum) {
1728                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1729                         if (ret)
1730                                 goto out;
1731                 }
1732                 goto mapit;
1733         } else if (async && !skip_sum) {
1734                 /* csum items have already been cloned */
1735                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1736                         goto mapit;
1737                 /* we're doing a write, do the async checksumming */
1738                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1739                                    inode, rw, bio, mirror_num,
1740                                    bio_flags, bio_offset,
1741                                    __btrfs_submit_bio_start,
1742                                    __btrfs_submit_bio_done);
1743                 goto out;
1744         } else if (!skip_sum) {
1745                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1746                 if (ret)
1747                         goto out;
1748         }
1749
1750 mapit:
1751         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1752
1753 out:
1754         if (ret < 0)
1755                 bio_endio(bio, ret);
1756         return ret;
1757 }
1758
1759 /*
1760  * given a list of ordered sums record them in the inode.  This happens
1761  * at IO completion time based on sums calculated at bio submission time.
1762  */
1763 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1764                              struct inode *inode, u64 file_offset,
1765                              struct list_head *list)
1766 {
1767         struct btrfs_ordered_sum *sum;
1768
1769         list_for_each_entry(sum, list, list) {
1770                 trans->adding_csums = 1;
1771                 btrfs_csum_file_blocks(trans,
1772                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1773                 trans->adding_csums = 0;
1774         }
1775         return 0;
1776 }
1777
1778 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1779                               struct extent_state **cached_state)
1780 {
1781         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1782         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1783                                    cached_state, GFP_NOFS);
1784 }
1785
1786 /* see btrfs_writepage_start_hook for details on why this is required */
1787 struct btrfs_writepage_fixup {
1788         struct page *page;
1789         struct btrfs_work work;
1790 };
1791
1792 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1793 {
1794         struct btrfs_writepage_fixup *fixup;
1795         struct btrfs_ordered_extent *ordered;
1796         struct extent_state *cached_state = NULL;
1797         struct page *page;
1798         struct inode *inode;
1799         u64 page_start;
1800         u64 page_end;
1801         int ret;
1802
1803         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1804         page = fixup->page;
1805 again:
1806         lock_page(page);
1807         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1808                 ClearPageChecked(page);
1809                 goto out_page;
1810         }
1811
1812         inode = page->mapping->host;
1813         page_start = page_offset(page);
1814         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1815
1816         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1817                          &cached_state);
1818
1819         /* already ordered? We're done */
1820         if (PagePrivate2(page))
1821                 goto out;
1822
1823         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1824         if (ordered) {
1825                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1826                                      page_end, &cached_state, GFP_NOFS);
1827                 unlock_page(page);
1828                 btrfs_start_ordered_extent(inode, ordered, 1);
1829                 btrfs_put_ordered_extent(ordered);
1830                 goto again;
1831         }
1832
1833         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1834         if (ret) {
1835                 mapping_set_error(page->mapping, ret);
1836                 end_extent_writepage(page, ret, page_start, page_end);
1837                 ClearPageChecked(page);
1838                 goto out;
1839          }
1840
1841         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1842         ClearPageChecked(page);
1843         set_page_dirty(page);
1844 out:
1845         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1846                              &cached_state, GFP_NOFS);
1847 out_page:
1848         unlock_page(page);
1849         page_cache_release(page);
1850         kfree(fixup);
1851 }
1852
1853 /*
1854  * There are a few paths in the higher layers of the kernel that directly
1855  * set the page dirty bit without asking the filesystem if it is a
1856  * good idea.  This causes problems because we want to make sure COW
1857  * properly happens and the data=ordered rules are followed.
1858  *
1859  * In our case any range that doesn't have the ORDERED bit set
1860  * hasn't been properly setup for IO.  We kick off an async process
1861  * to fix it up.  The async helper will wait for ordered extents, set
1862  * the delalloc bit and make it safe to write the page.
1863  */
1864 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1865 {
1866         struct inode *inode = page->mapping->host;
1867         struct btrfs_writepage_fixup *fixup;
1868         struct btrfs_root *root = BTRFS_I(inode)->root;
1869
1870         /* this page is properly in the ordered list */
1871         if (TestClearPagePrivate2(page))
1872                 return 0;
1873
1874         if (PageChecked(page))
1875                 return -EAGAIN;
1876
1877         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1878         if (!fixup)
1879                 return -EAGAIN;
1880
1881         SetPageChecked(page);
1882         page_cache_get(page);
1883         fixup->work.func = btrfs_writepage_fixup_worker;
1884         fixup->page = page;
1885         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1886         return -EBUSY;
1887 }
1888
1889 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1890                                        struct inode *inode, u64 file_pos,
1891                                        u64 disk_bytenr, u64 disk_num_bytes,
1892                                        u64 num_bytes, u64 ram_bytes,
1893                                        u8 compression, u8 encryption,
1894                                        u16 other_encoding, int extent_type)
1895 {
1896         struct btrfs_root *root = BTRFS_I(inode)->root;
1897         struct btrfs_file_extent_item *fi;
1898         struct btrfs_path *path;
1899         struct extent_buffer *leaf;
1900         struct btrfs_key ins;
1901         int ret;
1902
1903         path = btrfs_alloc_path();
1904         if (!path)
1905                 return -ENOMEM;
1906
1907         path->leave_spinning = 1;
1908
1909         /*
1910          * we may be replacing one extent in the tree with another.
1911          * The new extent is pinned in the extent map, and we don't want
1912          * to drop it from the cache until it is completely in the btree.
1913          *
1914          * So, tell btrfs_drop_extents to leave this extent in the cache.
1915          * the caller is expected to unpin it and allow it to be merged
1916          * with the others.
1917          */
1918         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1919                                  file_pos + num_bytes, 0);
1920         if (ret)
1921                 goto out;
1922
1923         ins.objectid = btrfs_ino(inode);
1924         ins.offset = file_pos;
1925         ins.type = BTRFS_EXTENT_DATA_KEY;
1926         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1927         if (ret)
1928                 goto out;
1929         leaf = path->nodes[0];
1930         fi = btrfs_item_ptr(leaf, path->slots[0],
1931                             struct btrfs_file_extent_item);
1932         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1933         btrfs_set_file_extent_type(leaf, fi, extent_type);
1934         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1935         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1936         btrfs_set_file_extent_offset(leaf, fi, 0);
1937         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1938         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1939         btrfs_set_file_extent_compression(leaf, fi, compression);
1940         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1941         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1942
1943         btrfs_mark_buffer_dirty(leaf);
1944         btrfs_release_path(path);
1945
1946         inode_add_bytes(inode, num_bytes);
1947
1948         ins.objectid = disk_bytenr;
1949         ins.offset = disk_num_bytes;
1950         ins.type = BTRFS_EXTENT_ITEM_KEY;
1951         ret = btrfs_alloc_reserved_file_extent(trans, root,
1952                                         root->root_key.objectid,
1953                                         btrfs_ino(inode), file_pos, &ins);
1954 out:
1955         btrfs_free_path(path);
1956
1957         return ret;
1958 }
1959
1960 /* snapshot-aware defrag */
1961 struct sa_defrag_extent_backref {
1962         struct rb_node node;
1963         struct old_sa_defrag_extent *old;
1964         u64 root_id;
1965         u64 inum;
1966         u64 file_pos;
1967         u64 extent_offset;
1968         u64 num_bytes;
1969         u64 generation;
1970 };
1971
1972 struct old_sa_defrag_extent {
1973         struct list_head list;
1974         struct new_sa_defrag_extent *new;
1975
1976         u64 extent_offset;
1977         u64 bytenr;
1978         u64 offset;
1979         u64 len;
1980         int count;
1981 };
1982
1983 struct new_sa_defrag_extent {
1984         struct rb_root root;
1985         struct list_head head;
1986         struct btrfs_path *path;
1987         struct inode *inode;
1988         u64 file_pos;
1989         u64 len;
1990         u64 bytenr;
1991         u64 disk_len;
1992         u8 compress_type;
1993 };
1994
1995 static int backref_comp(struct sa_defrag_extent_backref *b1,
1996                         struct sa_defrag_extent_backref *b2)
1997 {
1998         if (b1->root_id < b2->root_id)
1999                 return -1;
2000         else if (b1->root_id > b2->root_id)
2001                 return 1;
2002
2003         if (b1->inum < b2->inum)
2004                 return -1;
2005         else if (b1->inum > b2->inum)
2006                 return 1;
2007
2008         if (b1->file_pos < b2->file_pos)
2009                 return -1;
2010         else if (b1->file_pos > b2->file_pos)
2011                 return 1;
2012
2013         /*
2014          * [------------------------------] ===> (a range of space)
2015          *     |<--->|   |<---->| =============> (fs/file tree A)
2016          * |<---------------------------->| ===> (fs/file tree B)
2017          *
2018          * A range of space can refer to two file extents in one tree while
2019          * refer to only one file extent in another tree.
2020          *
2021          * So we may process a disk offset more than one time(two extents in A)
2022          * and locate at the same extent(one extent in B), then insert two same
2023          * backrefs(both refer to the extent in B).
2024          */
2025         return 0;
2026 }
2027
2028 static void backref_insert(struct rb_root *root,
2029                            struct sa_defrag_extent_backref *backref)
2030 {
2031         struct rb_node **p = &root->rb_node;
2032         struct rb_node *parent = NULL;
2033         struct sa_defrag_extent_backref *entry;
2034         int ret;
2035
2036         while (*p) {
2037                 parent = *p;
2038                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2039
2040                 ret = backref_comp(backref, entry);
2041                 if (ret < 0)
2042                         p = &(*p)->rb_left;
2043                 else
2044                         p = &(*p)->rb_right;
2045         }
2046
2047         rb_link_node(&backref->node, parent, p);
2048         rb_insert_color(&backref->node, root);
2049 }
2050
2051 /*
2052  * Note the backref might has changed, and in this case we just return 0.
2053  */
2054 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2055                                        void *ctx)
2056 {
2057         struct btrfs_file_extent_item *extent;
2058         struct btrfs_fs_info *fs_info;
2059         struct old_sa_defrag_extent *old = ctx;
2060         struct new_sa_defrag_extent *new = old->new;
2061         struct btrfs_path *path = new->path;
2062         struct btrfs_key key;
2063         struct btrfs_root *root;
2064         struct sa_defrag_extent_backref *backref;
2065         struct extent_buffer *leaf;
2066         struct inode *inode = new->inode;
2067         int slot;
2068         int ret;
2069         u64 extent_offset;
2070         u64 num_bytes;
2071
2072         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2073             inum == btrfs_ino(inode))
2074                 return 0;
2075
2076         key.objectid = root_id;
2077         key.type = BTRFS_ROOT_ITEM_KEY;
2078         key.offset = (u64)-1;
2079
2080         fs_info = BTRFS_I(inode)->root->fs_info;
2081         root = btrfs_read_fs_root_no_name(fs_info, &key);
2082         if (IS_ERR(root)) {
2083                 if (PTR_ERR(root) == -ENOENT)
2084                         return 0;
2085                 WARN_ON(1);
2086                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2087                          inum, offset, root_id);
2088                 return PTR_ERR(root);
2089         }
2090
2091         key.objectid = inum;
2092         key.type = BTRFS_EXTENT_DATA_KEY;
2093         if (offset > (u64)-1 << 32)
2094                 key.offset = 0;
2095         else
2096                 key.offset = offset;
2097
2098         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2099         if (ret < 0) {
2100                 WARN_ON(1);
2101                 return ret;
2102         }
2103
2104         while (1) {
2105                 cond_resched();
2106
2107                 leaf = path->nodes[0];
2108                 slot = path->slots[0];
2109
2110                 if (slot >= btrfs_header_nritems(leaf)) {
2111                         ret = btrfs_next_leaf(root, path);
2112                         if (ret < 0) {
2113                                 goto out;
2114                         } else if (ret > 0) {
2115                                 ret = 0;
2116                                 goto out;
2117                         }
2118                         continue;
2119                 }
2120
2121                 path->slots[0]++;
2122
2123                 btrfs_item_key_to_cpu(leaf, &key, slot);
2124
2125                 if (key.objectid > inum)
2126                         goto out;
2127
2128                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2129                         continue;
2130
2131                 extent = btrfs_item_ptr(leaf, slot,
2132                                         struct btrfs_file_extent_item);
2133
2134                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2135                         continue;
2136
2137                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2138                 if (key.offset - extent_offset != offset)
2139                         continue;
2140
2141                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2142                 if (extent_offset >= old->extent_offset + old->offset +
2143                     old->len || extent_offset + num_bytes <=
2144                     old->extent_offset + old->offset)
2145                         continue;
2146
2147                 break;
2148         }
2149
2150         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2151         if (!backref) {
2152                 ret = -ENOENT;
2153                 goto out;
2154         }
2155
2156         backref->root_id = root_id;
2157         backref->inum = inum;
2158         backref->file_pos = offset + extent_offset;
2159         backref->num_bytes = num_bytes;
2160         backref->extent_offset = extent_offset;
2161         backref->generation = btrfs_file_extent_generation(leaf, extent);
2162         backref->old = old;
2163         backref_insert(&new->root, backref);
2164         old->count++;
2165 out:
2166         btrfs_release_path(path);
2167         WARN_ON(ret);
2168         return ret;
2169 }
2170
2171 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2172                                    struct new_sa_defrag_extent *new)
2173 {
2174         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2175         struct old_sa_defrag_extent *old, *tmp;
2176         int ret;
2177
2178         new->path = path;
2179
2180         list_for_each_entry_safe(old, tmp, &new->head, list) {
2181                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2182                                                   path, record_one_backref,
2183                                                   old);
2184                 BUG_ON(ret < 0 && ret != -ENOENT);
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                               u64 disk_bytenr)
2202 {
2203         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_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) ||
2210             btrfs_file_extent_encryption(leaf, fi) ||
2211             btrfs_file_extent_other_encoding(leaf, fi))
2212                 return 0;
2213
2214         return 1;
2215 }
2216
2217 /*
2218  * Note the backref might has changed, and in this case we just return 0.
2219  */
2220 static noinline int relink_extent_backref(struct btrfs_path *path,
2221                                  struct sa_defrag_extent_backref *prev,
2222                                  struct sa_defrag_extent_backref *backref)
2223 {
2224         struct btrfs_file_extent_item *extent;
2225         struct btrfs_file_extent_item *item;
2226         struct btrfs_ordered_extent *ordered;
2227         struct btrfs_trans_handle *trans;
2228         struct btrfs_fs_info *fs_info;
2229         struct btrfs_root *root;
2230         struct btrfs_key key;
2231         struct extent_buffer *leaf;
2232         struct old_sa_defrag_extent *old = backref->old;
2233         struct new_sa_defrag_extent *new = old->new;
2234         struct inode *src_inode = new->inode;
2235         struct inode *inode;
2236         struct extent_state *cached = NULL;
2237         int ret = 0;
2238         u64 start;
2239         u64 len;
2240         u64 lock_start;
2241         u64 lock_end;
2242         bool merge = false;
2243         int index;
2244
2245         if (prev && prev->root_id == backref->root_id &&
2246             prev->inum == backref->inum &&
2247             prev->file_pos + prev->num_bytes == backref->file_pos)
2248                 merge = true;
2249
2250         /* step 1: get root */
2251         key.objectid = backref->root_id;
2252         key.type = BTRFS_ROOT_ITEM_KEY;
2253         key.offset = (u64)-1;
2254
2255         fs_info = BTRFS_I(src_inode)->root->fs_info;
2256         index = srcu_read_lock(&fs_info->subvol_srcu);
2257
2258         root = btrfs_read_fs_root_no_name(fs_info, &key);
2259         if (IS_ERR(root)) {
2260                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2261                 if (PTR_ERR(root) == -ENOENT)
2262                         return 0;
2263                 return PTR_ERR(root);
2264         }
2265
2266         /* step 2: get inode */
2267         key.objectid = backref->inum;
2268         key.type = BTRFS_INODE_ITEM_KEY;
2269         key.offset = 0;
2270
2271         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2272         if (IS_ERR(inode)) {
2273                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2274                 return 0;
2275         }
2276
2277         srcu_read_unlock(&fs_info->subvol_srcu, index);
2278
2279         /* step 3: relink backref */
2280         lock_start = backref->file_pos;
2281         lock_end = backref->file_pos + backref->num_bytes - 1;
2282         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2283                          0, &cached);
2284
2285         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2286         if (ordered) {
2287                 btrfs_put_ordered_extent(ordered);
2288                 goto out_unlock;
2289         }
2290
2291         trans = btrfs_join_transaction(root);
2292         if (IS_ERR(trans)) {
2293                 ret = PTR_ERR(trans);
2294                 goto out_unlock;
2295         }
2296
2297         key.objectid = backref->inum;
2298         key.type = BTRFS_EXTENT_DATA_KEY;
2299         key.offset = backref->file_pos;
2300
2301         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2302         if (ret < 0) {
2303                 goto out_free_path;
2304         } else if (ret > 0) {
2305                 ret = 0;
2306                 goto out_free_path;
2307         }
2308
2309         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2310                                 struct btrfs_file_extent_item);
2311
2312         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2313             backref->generation)
2314                 goto out_free_path;
2315
2316         btrfs_release_path(path);
2317
2318         start = backref->file_pos;
2319         if (backref->extent_offset < old->extent_offset + old->offset)
2320                 start += old->extent_offset + old->offset -
2321                          backref->extent_offset;
2322
2323         len = min(backref->extent_offset + backref->num_bytes,
2324                   old->extent_offset + old->offset + old->len);
2325         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2326
2327         ret = btrfs_drop_extents(trans, root, inode, start,
2328                                  start + len, 1);
2329         if (ret)
2330                 goto out_free_path;
2331 again:
2332         key.objectid = btrfs_ino(inode);
2333         key.type = BTRFS_EXTENT_DATA_KEY;
2334         key.offset = start;
2335
2336         path->leave_spinning = 1;
2337         if (merge) {
2338                 struct btrfs_file_extent_item *fi;
2339                 u64 extent_len;
2340                 struct btrfs_key found_key;
2341
2342                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2343                 if (ret < 0)
2344                         goto out_free_path;
2345
2346                 path->slots[0]--;
2347                 leaf = path->nodes[0];
2348                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2349
2350                 fi = btrfs_item_ptr(leaf, path->slots[0],
2351                                     struct btrfs_file_extent_item);
2352                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2353
2354                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2355                     extent_len + found_key.offset == start) {
2356                         btrfs_set_file_extent_num_bytes(leaf, fi,
2357                                                         extent_len + len);
2358                         btrfs_mark_buffer_dirty(leaf);
2359                         inode_add_bytes(inode, len);
2360
2361                         ret = 1;
2362                         goto out_free_path;
2363                 } else {
2364                         merge = false;
2365                         btrfs_release_path(path);
2366                         goto again;
2367                 }
2368         }
2369
2370         ret = btrfs_insert_empty_item(trans, root, path, &key,
2371                                         sizeof(*extent));
2372         if (ret) {
2373                 btrfs_abort_transaction(trans, root, ret);
2374                 goto out_free_path;
2375         }
2376
2377         leaf = path->nodes[0];
2378         item = btrfs_item_ptr(leaf, path->slots[0],
2379                                 struct btrfs_file_extent_item);
2380         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2381         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2382         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2383         btrfs_set_file_extent_num_bytes(leaf, item, len);
2384         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2385         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2386         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2387         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2388         btrfs_set_file_extent_encryption(leaf, item, 0);
2389         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2390
2391         btrfs_mark_buffer_dirty(leaf);
2392         inode_add_bytes(inode, len);
2393         btrfs_release_path(path);
2394
2395         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2396                         new->disk_len, 0,
2397                         backref->root_id, backref->inum,
2398                         new->file_pos, 0);      /* start - extent_offset */
2399         if (ret) {
2400                 btrfs_abort_transaction(trans, root, ret);
2401                 goto out_free_path;
2402         }
2403
2404         ret = 1;
2405 out_free_path:
2406         btrfs_release_path(path);
2407         path->leave_spinning = 0;
2408         btrfs_end_transaction(trans, root);
2409 out_unlock:
2410         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2411                              &cached, GFP_NOFS);
2412         iput(inode);
2413         return ret;
2414 }
2415
2416 static void relink_file_extents(struct new_sa_defrag_extent *new)
2417 {
2418         struct btrfs_path *path;
2419         struct old_sa_defrag_extent *old, *tmp;
2420         struct sa_defrag_extent_backref *backref;
2421         struct sa_defrag_extent_backref *prev = NULL;
2422         struct inode *inode;
2423         struct btrfs_root *root;
2424         struct rb_node *node;
2425         int ret;
2426
2427         inode = new->inode;
2428         root = BTRFS_I(inode)->root;
2429
2430         path = btrfs_alloc_path();
2431         if (!path)
2432                 return;
2433
2434         if (!record_extent_backrefs(path, new)) {
2435                 btrfs_free_path(path);
2436                 goto out;
2437         }
2438         btrfs_release_path(path);
2439
2440         while (1) {
2441                 node = rb_first(&new->root);
2442                 if (!node)
2443                         break;
2444                 rb_erase(node, &new->root);
2445
2446                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2447
2448                 ret = relink_extent_backref(path, prev, backref);
2449                 WARN_ON(ret < 0);
2450
2451                 kfree(prev);
2452
2453                 if (ret == 1)
2454                         prev = backref;
2455                 else
2456                         prev = NULL;
2457                 cond_resched();
2458         }
2459         kfree(prev);
2460
2461         btrfs_free_path(path);
2462
2463         list_for_each_entry_safe(old, tmp, &new->head, list) {
2464                 list_del(&old->list);
2465                 kfree(old);
2466         }
2467 out:
2468         atomic_dec(&root->fs_info->defrag_running);
2469         wake_up(&root->fs_info->transaction_wait);
2470
2471         kfree(new);
2472 }
2473
2474 static struct new_sa_defrag_extent *
2475 record_old_file_extents(struct inode *inode,
2476                         struct btrfs_ordered_extent *ordered)
2477 {
2478         struct btrfs_root *root = BTRFS_I(inode)->root;
2479         struct btrfs_path *path;
2480         struct btrfs_key key;
2481         struct old_sa_defrag_extent *old, *tmp;
2482         struct new_sa_defrag_extent *new;
2483         int ret;
2484
2485         new = kmalloc(sizeof(*new), GFP_NOFS);
2486         if (!new)
2487                 return NULL;
2488
2489         new->inode = inode;
2490         new->file_pos = ordered->file_offset;
2491         new->len = ordered->len;
2492         new->bytenr = ordered->start;
2493         new->disk_len = ordered->disk_len;
2494         new->compress_type = ordered->compress_type;
2495         new->root = RB_ROOT;
2496         INIT_LIST_HEAD(&new->head);
2497
2498         path = btrfs_alloc_path();
2499         if (!path)
2500                 goto out_kfree;
2501
2502         key.objectid = btrfs_ino(inode);
2503         key.type = BTRFS_EXTENT_DATA_KEY;
2504         key.offset = new->file_pos;
2505
2506         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2507         if (ret < 0)
2508                 goto out_free_path;
2509         if (ret > 0 && path->slots[0] > 0)
2510                 path->slots[0]--;
2511
2512         /* find out all the old extents for the file range */
2513         while (1) {
2514                 struct btrfs_file_extent_item *extent;
2515                 struct extent_buffer *l;
2516                 int slot;
2517                 u64 num_bytes;
2518                 u64 offset;
2519                 u64 end;
2520                 u64 disk_bytenr;
2521                 u64 extent_offset;
2522
2523                 l = path->nodes[0];
2524                 slot = path->slots[0];
2525
2526                 if (slot >= btrfs_header_nritems(l)) {
2527                         ret = btrfs_next_leaf(root, path);
2528                         if (ret < 0)
2529                                 goto out_free_list;
2530                         else if (ret > 0)
2531                                 break;
2532                         continue;
2533                 }
2534
2535                 btrfs_item_key_to_cpu(l, &key, slot);
2536
2537                 if (key.objectid != btrfs_ino(inode))
2538                         break;
2539                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2540                         break;
2541                 if (key.offset >= new->file_pos + new->len)
2542                         break;
2543
2544                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2545
2546                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2547                 if (key.offset + num_bytes < new->file_pos)
2548                         goto next;
2549
2550                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2551                 if (!disk_bytenr)
2552                         goto next;
2553
2554                 extent_offset = btrfs_file_extent_offset(l, extent);
2555
2556                 old = kmalloc(sizeof(*old), GFP_NOFS);
2557                 if (!old)
2558                         goto out_free_list;
2559
2560                 offset = max(new->file_pos, key.offset);
2561                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2562
2563                 old->bytenr = disk_bytenr;
2564                 old->extent_offset = extent_offset;
2565                 old->offset = offset - key.offset;
2566                 old->len = end - offset;
2567                 old->new = new;
2568                 old->count = 0;
2569                 list_add_tail(&old->list, &new->head);
2570 next:
2571                 path->slots[0]++;
2572                 cond_resched();
2573         }
2574
2575         btrfs_free_path(path);
2576         atomic_inc(&root->fs_info->defrag_running);
2577
2578         return new;
2579
2580 out_free_list:
2581         list_for_each_entry_safe(old, tmp, &new->head, list) {
2582                 list_del(&old->list);
2583                 kfree(old);
2584         }
2585 out_free_path:
2586         btrfs_free_path(path);
2587 out_kfree:
2588         kfree(new);
2589         return NULL;
2590 }
2591
2592 /*
2593  * helper function for btrfs_finish_ordered_io, this
2594  * just reads in some of the csum leaves to prime them into ram
2595  * before we start the transaction.  It limits the amount of btree
2596  * reads required while inside the transaction.
2597  */
2598 /* as ordered data IO finishes, this gets called so we can finish
2599  * an ordered extent if the range of bytes in the file it covers are
2600  * fully written.
2601  */
2602 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2603 {
2604         struct inode *inode = ordered_extent->inode;
2605         struct btrfs_root *root = BTRFS_I(inode)->root;
2606         struct btrfs_trans_handle *trans = NULL;
2607         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2608         struct extent_state *cached_state = NULL;
2609         struct new_sa_defrag_extent *new = NULL;
2610         int compress_type = 0;
2611         int ret;
2612         bool nolock;
2613
2614         nolock = btrfs_is_free_space_inode(inode);
2615
2616         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2617                 ret = -EIO;
2618                 goto out;
2619         }
2620
2621         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2622                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2623                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2624                 if (nolock)
2625                         trans = btrfs_join_transaction_nolock(root);
2626                 else
2627                         trans = btrfs_join_transaction(root);
2628                 if (IS_ERR(trans)) {
2629                         ret = PTR_ERR(trans);
2630                         trans = NULL;
2631                         goto out;
2632                 }
2633                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2634                 ret = btrfs_update_inode_fallback(trans, root, inode);
2635                 if (ret) /* -ENOMEM or corruption */
2636                         btrfs_abort_transaction(trans, root, ret);
2637                 goto out;
2638         }
2639
2640         lock_extent_bits(io_tree, ordered_extent->file_offset,
2641                          ordered_extent->file_offset + ordered_extent->len - 1,
2642                          0, &cached_state);
2643
2644         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2645                         ordered_extent->file_offset + ordered_extent->len - 1,
2646                         EXTENT_DEFRAG, 1, cached_state);
2647         if (ret) {
2648                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2649                 if (last_snapshot >= BTRFS_I(inode)->generation)
2650                         /* the inode is shared */
2651                         new = record_old_file_extents(inode, ordered_extent);
2652
2653                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2654                         ordered_extent->file_offset + ordered_extent->len - 1,
2655                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2656         }
2657
2658         if (nolock)
2659                 trans = btrfs_join_transaction_nolock(root);
2660         else
2661                 trans = btrfs_join_transaction(root);
2662         if (IS_ERR(trans)) {
2663                 ret = PTR_ERR(trans);
2664                 trans = NULL;
2665                 goto out_unlock;
2666         }
2667         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2668
2669         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2670                 compress_type = ordered_extent->compress_type;
2671         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2672                 BUG_ON(compress_type);
2673                 ret = btrfs_mark_extent_written(trans, inode,
2674                                                 ordered_extent->file_offset,
2675                                                 ordered_extent->file_offset +
2676                                                 ordered_extent->len);
2677         } else {
2678                 BUG_ON(root == root->fs_info->tree_root);
2679                 ret = insert_reserved_file_extent(trans, inode,
2680                                                 ordered_extent->file_offset,
2681                                                 ordered_extent->start,
2682                                                 ordered_extent->disk_len,
2683                                                 ordered_extent->len,
2684                                                 ordered_extent->len,
2685                                                 compress_type, 0, 0,
2686                                                 BTRFS_FILE_EXTENT_REG);
2687         }
2688         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2689                            ordered_extent->file_offset, ordered_extent->len,
2690                            trans->transid);
2691         if (ret < 0) {
2692                 btrfs_abort_transaction(trans, root, ret);
2693                 goto out_unlock;
2694         }
2695
2696         add_pending_csums(trans, inode, ordered_extent->file_offset,
2697                           &ordered_extent->list);
2698
2699         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2700         ret = btrfs_update_inode_fallback(trans, root, inode);
2701         if (ret) { /* -ENOMEM or corruption */
2702                 btrfs_abort_transaction(trans, root, ret);
2703                 goto out_unlock;
2704         }
2705         ret = 0;
2706 out_unlock:
2707         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2708                              ordered_extent->file_offset +
2709                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2710 out:
2711         if (root != root->fs_info->tree_root)
2712                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2713         if (trans)
2714                 btrfs_end_transaction(trans, root);
2715
2716         if (ret) {
2717                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2718                                       ordered_extent->file_offset +
2719                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2720
2721                 /*
2722                  * If the ordered extent had an IOERR or something else went
2723                  * wrong we need to return the space for this ordered extent
2724                  * back to the allocator.
2725                  */
2726                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2727                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2728                         btrfs_free_reserved_extent(root, ordered_extent->start,
2729                                                    ordered_extent->disk_len);
2730         }
2731
2732
2733         /*
2734          * This needs to be done to make sure anybody waiting knows we are done
2735          * updating everything for this ordered extent.
2736          */
2737         btrfs_remove_ordered_extent(inode, ordered_extent);
2738
2739         /* for snapshot-aware defrag */
2740         if (new)
2741                 relink_file_extents(new);
2742
2743         /* once for us */
2744         btrfs_put_ordered_extent(ordered_extent);
2745         /* once for the tree */
2746         btrfs_put_ordered_extent(ordered_extent);
2747
2748         return ret;
2749 }
2750
2751 static void finish_ordered_fn(struct btrfs_work *work)
2752 {
2753         struct btrfs_ordered_extent *ordered_extent;
2754         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2755         btrfs_finish_ordered_io(ordered_extent);
2756 }
2757
2758 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2759                                 struct extent_state *state, int uptodate)
2760 {
2761         struct inode *inode = page->mapping->host;
2762         struct btrfs_root *root = BTRFS_I(inode)->root;
2763         struct btrfs_ordered_extent *ordered_extent = NULL;
2764         struct btrfs_workers *workers;
2765
2766         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2767
2768         ClearPagePrivate2(page);
2769         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2770                                             end - start + 1, uptodate))
2771                 return 0;
2772
2773         ordered_extent->work.func = finish_ordered_fn;
2774         ordered_extent->work.flags = 0;
2775
2776         if (btrfs_is_free_space_inode(inode))
2777                 workers = &root->fs_info->endio_freespace_worker;
2778         else
2779                 workers = &root->fs_info->endio_write_workers;
2780         btrfs_queue_worker(workers, &ordered_extent->work);
2781
2782         return 0;
2783 }
2784
2785 /*
2786  * when reads are done, we need to check csums to verify the data is correct
2787  * if there's a match, we allow the bio to finish.  If not, the code in
2788  * extent_io.c will try to find good copies for us.
2789  */
2790 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2791                                struct extent_state *state, int mirror)
2792 {
2793         size_t offset = start - page_offset(page);
2794         struct inode *inode = page->mapping->host;
2795         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2796         char *kaddr;
2797         u64 private = ~(u32)0;
2798         int ret;
2799         struct btrfs_root *root = BTRFS_I(inode)->root;
2800         u32 csum = ~(u32)0;
2801         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2802                                       DEFAULT_RATELIMIT_BURST);
2803
2804         if (PageChecked(page)) {
2805                 ClearPageChecked(page);
2806                 goto good;
2807         }
2808
2809         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2810                 goto good;
2811
2812         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2813             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2814                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2815                                   GFP_NOFS);
2816                 return 0;
2817         }
2818
2819         if (state && state->start == start) {
2820                 private = state->private;
2821                 ret = 0;
2822         } else {
2823                 ret = get_state_private(io_tree, start, &private);
2824         }
2825         kaddr = kmap_atomic(page);
2826         if (ret)
2827                 goto zeroit;
2828
2829         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2830         btrfs_csum_final(csum, (char *)&csum);
2831         if (csum != private)
2832                 goto zeroit;
2833
2834         kunmap_atomic(kaddr);
2835 good:
2836         return 0;
2837
2838 zeroit:
2839         if (__ratelimit(&_rs))
2840                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2841                         (unsigned long long)btrfs_ino(page->mapping->host),
2842                         (unsigned long long)start, csum,
2843                         (unsigned long long)private);
2844         memset(kaddr + offset, 1, end - start + 1);
2845         flush_dcache_page(page);
2846         kunmap_atomic(kaddr);
2847         if (private == 0)
2848                 return 0;
2849         return -EIO;
2850 }
2851
2852 struct delayed_iput {
2853         struct list_head list;
2854         struct inode *inode;
2855 };
2856
2857 /* JDM: If this is fs-wide, why can't we add a pointer to
2858  * btrfs_inode instead and avoid the allocation? */
2859 void btrfs_add_delayed_iput(struct inode *inode)
2860 {
2861         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2862         struct delayed_iput *delayed;
2863
2864         if (atomic_add_unless(&inode->i_count, -1, 1))
2865                 return;
2866
2867         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2868         delayed->inode = inode;
2869
2870         spin_lock(&fs_info->delayed_iput_lock);
2871         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2872         spin_unlock(&fs_info->delayed_iput_lock);
2873 }
2874
2875 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2876 {
2877         LIST_HEAD(list);
2878         struct btrfs_fs_info *fs_info = root->fs_info;
2879         struct delayed_iput *delayed;
2880         int empty;
2881
2882         spin_lock(&fs_info->delayed_iput_lock);
2883         empty = list_empty(&fs_info->delayed_iputs);
2884         spin_unlock(&fs_info->delayed_iput_lock);
2885         if (empty)
2886                 return;
2887
2888         spin_lock(&fs_info->delayed_iput_lock);
2889         list_splice_init(&fs_info->delayed_iputs, &list);
2890         spin_unlock(&fs_info->delayed_iput_lock);
2891
2892         while (!list_empty(&list)) {
2893                 delayed = list_entry(list.next, struct delayed_iput, list);
2894                 list_del(&delayed->list);
2895                 iput(delayed->inode);
2896                 kfree(delayed);
2897         }
2898 }
2899
2900 /*
2901  * This is called in transaction commit time. If there are no orphan
2902  * files in the subvolume, it removes orphan item and frees block_rsv
2903  * structure.
2904  */
2905 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2906                               struct btrfs_root *root)
2907 {
2908         struct btrfs_block_rsv *block_rsv;
2909         int ret;
2910
2911         if (atomic_read(&root->orphan_inodes) ||
2912             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2913                 return;
2914
2915         spin_lock(&root->orphan_lock);
2916         if (atomic_read(&root->orphan_inodes)) {
2917                 spin_unlock(&root->orphan_lock);
2918                 return;
2919         }
2920
2921         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2922                 spin_unlock(&root->orphan_lock);
2923                 return;
2924         }
2925
2926         block_rsv = root->orphan_block_rsv;
2927         root->orphan_block_rsv = NULL;
2928         spin_unlock(&root->orphan_lock);
2929
2930         if (root->orphan_item_inserted &&
2931             btrfs_root_refs(&root->root_item) > 0) {
2932                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2933                                             root->root_key.objectid);
2934                 BUG_ON(ret);
2935                 root->orphan_item_inserted = 0;
2936         }
2937
2938         if (block_rsv) {
2939                 WARN_ON(block_rsv->size > 0);
2940                 btrfs_free_block_rsv(root, block_rsv);
2941         }
2942 }
2943
2944 /*
2945  * This creates an orphan entry for the given inode in case something goes
2946  * wrong in the middle of an unlink/truncate.
2947  *
2948  * NOTE: caller of this function should reserve 5 units of metadata for
2949  *       this function.
2950  */
2951 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2952 {
2953         struct btrfs_root *root = BTRFS_I(inode)->root;
2954         struct btrfs_block_rsv *block_rsv = NULL;
2955         int reserve = 0;
2956         int insert = 0;
2957         int ret;
2958
2959         if (!root->orphan_block_rsv) {
2960                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2961                 if (!block_rsv)
2962                         return -ENOMEM;
2963         }
2964
2965         spin_lock(&root->orphan_lock);
2966         if (!root->orphan_block_rsv) {
2967                 root->orphan_block_rsv = block_rsv;
2968         } else if (block_rsv) {
2969                 btrfs_free_block_rsv(root, block_rsv);
2970                 block_rsv = NULL;
2971         }
2972
2973         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2974                               &BTRFS_I(inode)->runtime_flags)) {
2975 #if 0
2976                 /*
2977                  * For proper ENOSPC handling, we should do orphan
2978                  * cleanup when mounting. But this introduces backward
2979                  * compatibility issue.
2980                  */
2981                 if (!xchg(&root->orphan_item_inserted, 1))
2982                         insert = 2;
2983                 else
2984                         insert = 1;
2985 #endif
2986                 insert = 1;
2987                 atomic_inc(&root->orphan_inodes);
2988         }
2989
2990         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2991                               &BTRFS_I(inode)->runtime_flags))
2992                 reserve = 1;
2993         spin_unlock(&root->orphan_lock);
2994
2995         /* grab metadata reservation from transaction handle */
2996         if (reserve) {
2997                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2998                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2999         }
3000
3001         /* insert an orphan item to track this unlinked/truncated file */
3002         if (insert >= 1) {
3003                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3004                 if (ret && ret != -EEXIST) {
3005                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3006                                   &BTRFS_I(inode)->runtime_flags);
3007                         btrfs_abort_transaction(trans, root, ret);
3008                         return ret;
3009                 }
3010                 ret = 0;
3011         }
3012
3013         /* insert an orphan item to track subvolume contains orphan files */
3014         if (insert >= 2) {
3015                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3016                                                root->root_key.objectid);
3017                 if (ret && ret != -EEXIST) {
3018                         btrfs_abort_transaction(trans, root, ret);
3019                         return ret;
3020                 }
3021         }
3022         return 0;
3023 }
3024
3025 /*
3026  * We have done the truncate/delete so we can go ahead and remove the orphan
3027  * item for this particular inode.
3028  */
3029 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3030                             struct inode *inode)
3031 {
3032         struct btrfs_root *root = BTRFS_I(inode)->root;
3033         int delete_item = 0;
3034         int release_rsv = 0;
3035         int ret = 0;
3036
3037         spin_lock(&root->orphan_lock);
3038         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3039                                &BTRFS_I(inode)->runtime_flags))
3040                 delete_item = 1;
3041
3042         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3043                                &BTRFS_I(inode)->runtime_flags))
3044                 release_rsv = 1;
3045         spin_unlock(&root->orphan_lock);
3046
3047         if (trans && delete_item) {
3048                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3049                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3050         }
3051
3052         if (release_rsv) {
3053                 btrfs_orphan_release_metadata(inode);
3054                 atomic_dec(&root->orphan_inodes);
3055         }
3056
3057         return 0;
3058 }
3059
3060 /*
3061  * this cleans up any orphans that may be left on the list from the last use
3062  * of this root.
3063  */
3064 int btrfs_orphan_cleanup(struct btrfs_root *root)
3065 {
3066         struct btrfs_path *path;
3067         struct extent_buffer *leaf;
3068         struct btrfs_key key, found_key;
3069         struct btrfs_trans_handle *trans;
3070         struct inode *inode;
3071         u64 last_objectid = 0;
3072         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3073
3074         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3075                 return 0;
3076
3077         path = btrfs_alloc_path();
3078         if (!path) {
3079                 ret = -ENOMEM;
3080                 goto out;
3081         }
3082         path->reada = -1;
3083
3084         key.objectid = BTRFS_ORPHAN_OBJECTID;
3085         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3086         key.offset = (u64)-1;
3087
3088         while (1) {
3089                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3090                 if (ret < 0)
3091                         goto out;
3092
3093                 /*
3094                  * if ret == 0 means we found what we were searching for, which
3095                  * is weird, but possible, so only screw with path if we didn't
3096                  * find the key and see if we have stuff that matches
3097                  */
3098                 if (ret > 0) {
3099                         ret = 0;
3100                         if (path->slots[0] == 0)
3101                                 break;
3102                         path->slots[0]--;
3103                 }
3104
3105                 /* pull out the item */
3106                 leaf = path->nodes[0];
3107                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3108
3109                 /* make sure the item matches what we want */
3110                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3111                         break;
3112                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3113                         break;
3114
3115                 /* release the path since we're done with it */
3116                 btrfs_release_path(path);
3117
3118                 /*
3119                  * this is where we are basically btrfs_lookup, without the
3120                  * crossing root thing.  we store the inode number in the
3121                  * offset of the orphan item.
3122                  */
3123
3124                 if (found_key.offset == last_objectid) {
3125                         btrfs_err(root->fs_info,
3126                                 "Error removing orphan entry, stopping orphan cleanup");
3127                         ret = -EINVAL;
3128                         goto out;
3129                 }
3130
3131                 last_objectid = found_key.offset;
3132
3133                 found_key.objectid = found_key.offset;
3134                 found_key.type = BTRFS_INODE_ITEM_KEY;
3135                 found_key.offset = 0;
3136                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3137                 ret = PTR_RET(inode);
3138                 if (ret && ret != -ESTALE)
3139                         goto out;
3140
3141                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3142                         struct btrfs_root *dead_root;
3143                         struct btrfs_fs_info *fs_info = root->fs_info;
3144                         int is_dead_root = 0;
3145
3146                         /*
3147                          * this is an orphan in the tree root. Currently these
3148                          * could come from 2 sources:
3149                          *  a) a snapshot deletion in progress
3150                          *  b) a free space cache inode
3151                          * We need to distinguish those two, as the snapshot
3152                          * orphan must not get deleted.
3153                          * find_dead_roots already ran before us, so if this
3154                          * is a snapshot deletion, we should find the root
3155                          * in the dead_roots list
3156                          */
3157                         spin_lock(&fs_info->trans_lock);
3158                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3159                                             root_list) {
3160                                 if (dead_root->root_key.objectid ==
3161                                     found_key.objectid) {
3162                                         is_dead_root = 1;
3163                                         break;
3164                                 }
3165                         }
3166                         spin_unlock(&fs_info->trans_lock);
3167                         if (is_dead_root) {
3168                                 /* prevent this orphan from being found again */
3169                                 key.offset = found_key.objectid - 1;
3170                                 continue;
3171                         }
3172                 }
3173                 /*
3174                  * Inode is already gone but the orphan item is still there,
3175                  * kill the orphan item.
3176                  */
3177                 if (ret == -ESTALE) {
3178                         trans = btrfs_start_transaction(root, 1);
3179                         if (IS_ERR(trans)) {
3180                                 ret = PTR_ERR(trans);
3181                                 goto out;
3182                         }
3183                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3184                                 found_key.objectid);
3185                         ret = btrfs_del_orphan_item(trans, root,
3186                                                     found_key.objectid);
3187                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3188                         btrfs_end_transaction(trans, root);
3189                         continue;
3190                 }
3191
3192                 /*
3193                  * add this inode to the orphan list so btrfs_orphan_del does
3194                  * the proper thing when we hit it
3195                  */
3196                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3197                         &BTRFS_I(inode)->runtime_flags);
3198                 atomic_inc(&root->orphan_inodes);
3199
3200                 /* if we have links, this was a truncate, lets do that */
3201                 if (inode->i_nlink) {
3202                         if (!S_ISREG(inode->i_mode)) {
3203                                 WARN_ON(1);
3204                                 iput(inode);
3205                                 continue;
3206                         }
3207                         nr_truncate++;
3208
3209                         /* 1 for the orphan item deletion. */
3210                         trans = btrfs_start_transaction(root, 1);
3211                         if (IS_ERR(trans)) {
3212                                 ret = PTR_ERR(trans);
3213                                 goto out;
3214                         }
3215                         ret = btrfs_orphan_add(trans, inode);
3216                         btrfs_end_transaction(trans, root);
3217                         if (ret)
3218                                 goto out;
3219
3220                         ret = btrfs_truncate(inode);
3221                         if (ret)
3222                                 btrfs_orphan_del(NULL, inode);
3223                 } else {
3224                         nr_unlink++;
3225                 }
3226
3227                 /* this will do delete_inode and everything for us */
3228                 iput(inode);
3229                 if (ret)
3230                         goto out;
3231         }
3232         /* release the path since we're done with it */
3233         btrfs_release_path(path);
3234
3235         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3236
3237         if (root->orphan_block_rsv)
3238                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3239                                         (u64)-1);
3240
3241         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3242                 trans = btrfs_join_transaction(root);
3243                 if (!IS_ERR(trans))
3244                         btrfs_end_transaction(trans, root);
3245         }
3246
3247         if (nr_unlink)
3248                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3249         if (nr_truncate)
3250                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3251
3252 out:
3253         if (ret)
3254                 btrfs_crit(root->fs_info,
3255                         "could not do orphan cleanup %d", ret);
3256         btrfs_free_path(path);
3257         return ret;
3258 }
3259
3260 /*
3261  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3262  * don't find any xattrs, we know there can't be any acls.
3263  *
3264  * slot is the slot the inode is in, objectid is the objectid of the inode
3265  */
3266 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3267                                           int slot, u64 objectid)
3268 {
3269         u32 nritems = btrfs_header_nritems(leaf);
3270         struct btrfs_key found_key;
3271         int scanned = 0;
3272
3273         slot++;
3274         while (slot < nritems) {
3275                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3276
3277                 /* we found a different objectid, there must not be acls */
3278                 if (found_key.objectid != objectid)
3279                         return 0;
3280
3281                 /* we found an xattr, assume we've got an acl */
3282                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3283                         return 1;
3284
3285                 /*
3286                  * we found a key greater than an xattr key, there can't
3287                  * be any acls later on
3288                  */
3289                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3290                         return 0;
3291
3292                 slot++;
3293                 scanned++;
3294
3295                 /*
3296                  * it goes inode, inode backrefs, xattrs, extents,
3297                  * so if there are a ton of hard links to an inode there can
3298                  * be a lot of backrefs.  Don't waste time searching too hard,
3299                  * this is just an optimization
3300                  */
3301                 if (scanned >= 8)
3302                         break;
3303         }
3304         /* we hit the end of the leaf before we found an xattr or
3305          * something larger than an xattr.  We have to assume the inode
3306          * has acls
3307          */
3308         return 1;
3309 }
3310
3311 /*
3312  * read an inode from the btree into the in-memory inode
3313  */
3314 static void btrfs_read_locked_inode(struct inode *inode)
3315 {
3316         struct btrfs_path *path;
3317         struct extent_buffer *leaf;
3318         struct btrfs_inode_item *inode_item;
3319         struct btrfs_timespec *tspec;
3320         struct btrfs_root *root = BTRFS_I(inode)->root;
3321         struct btrfs_key location;
3322         int maybe_acls;
3323         u32 rdev;
3324         int ret;
3325         bool filled = false;
3326
3327         ret = btrfs_fill_inode(inode, &rdev);
3328         if (!ret)
3329                 filled = true;
3330
3331         path = btrfs_alloc_path();
3332         if (!path)
3333                 goto make_bad;
3334
3335         path->leave_spinning = 1;
3336         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3337
3338         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3339         if (ret)
3340                 goto make_bad;
3341
3342         leaf = path->nodes[0];
3343
3344         if (filled)
3345                 goto cache_acl;
3346
3347         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3348                                     struct btrfs_inode_item);
3349         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3350         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3351         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3352         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3353         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3354
3355         tspec = btrfs_inode_atime(inode_item);
3356         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3357         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3358
3359         tspec = btrfs_inode_mtime(inode_item);
3360         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3361         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3362
3363         tspec = btrfs_inode_ctime(inode_item);
3364         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3365         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3366
3367         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3368         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3369         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3370
3371         /*
3372          * If we were modified in the current generation and evicted from memory
3373          * and then re-read we need to do a full sync since we don't have any
3374          * idea about which extents were modified before we were evicted from
3375          * cache.
3376          */
3377         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3378                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3379                         &BTRFS_I(inode)->runtime_flags);
3380
3381         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3382         inode->i_generation = BTRFS_I(inode)->generation;
3383         inode->i_rdev = 0;
3384         rdev = btrfs_inode_rdev(leaf, inode_item);
3385
3386         BTRFS_I(inode)->index_cnt = (u64)-1;
3387         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3388 cache_acl:
3389         /*
3390          * try to precache a NULL acl entry for files that don't have
3391          * any xattrs or acls
3392          */
3393         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3394                                            btrfs_ino(inode));
3395         if (!maybe_acls)
3396                 cache_no_acl(inode);
3397
3398         btrfs_free_path(path);
3399
3400         switch (inode->i_mode & S_IFMT) {
3401         case S_IFREG:
3402                 inode->i_mapping->a_ops = &btrfs_aops;
3403                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3404                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3405                 inode->i_fop = &btrfs_file_operations;
3406                 inode->i_op = &btrfs_file_inode_operations;
3407                 break;
3408         case S_IFDIR:
3409                 inode->i_fop = &btrfs_dir_file_operations;
3410                 if (root == root->fs_info->tree_root)
3411                         inode->i_op = &btrfs_dir_ro_inode_operations;
3412                 else
3413                         inode->i_op = &btrfs_dir_inode_operations;
3414                 break;
3415         case S_IFLNK:
3416                 inode->i_op = &btrfs_symlink_inode_operations;
3417                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3418                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3419                 break;
3420         default:
3421                 inode->i_op = &btrfs_special_inode_operations;
3422                 init_special_inode(inode, inode->i_mode, rdev);
3423                 break;
3424         }
3425
3426         btrfs_update_iflags(inode);
3427         return;
3428
3429 make_bad:
3430         btrfs_free_path(path);
3431         make_bad_inode(inode);
3432 }
3433
3434 /*
3435  * given a leaf and an inode, copy the inode fields into the leaf
3436  */
3437 static void fill_inode_item(struct btrfs_trans_handle *trans,
3438                             struct extent_buffer *leaf,
3439                             struct btrfs_inode_item *item,
3440                             struct inode *inode)
3441 {
3442         struct btrfs_map_token token;
3443
3444         btrfs_init_map_token(&token);
3445
3446         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3447         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3448         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3449                                    &token);
3450         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3451         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3452
3453         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3454                                      inode->i_atime.tv_sec, &token);
3455         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3456                                       inode->i_atime.tv_nsec, &token);
3457
3458         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3459                                      inode->i_mtime.tv_sec, &token);
3460         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3461                                       inode->i_mtime.tv_nsec, &token);
3462
3463         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3464                                      inode->i_ctime.tv_sec, &token);
3465         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3466                                       inode->i_ctime.tv_nsec, &token);
3467
3468         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3469                                      &token);
3470         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3471                                          &token);
3472         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3473         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3474         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3475         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3476         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3477 }
3478
3479 /*
3480  * copy everything in the in-memory inode into the btree.
3481  */
3482 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3483                                 struct btrfs_root *root, struct inode *inode)
3484 {
3485         struct btrfs_inode_item *inode_item;
3486         struct btrfs_path *path;
3487         struct extent_buffer *leaf;
3488         int ret;
3489
3490         path = btrfs_alloc_path();
3491         if (!path)
3492                 return -ENOMEM;
3493
3494         path->leave_spinning = 1;
3495         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3496                                  1);
3497         if (ret) {
3498                 if (ret > 0)
3499                         ret = -ENOENT;
3500                 goto failed;
3501         }
3502
3503         btrfs_unlock_up_safe(path, 1);
3504         leaf = path->nodes[0];
3505         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3506                                     struct btrfs_inode_item);
3507
3508         fill_inode_item(trans, leaf, inode_item, inode);
3509         btrfs_mark_buffer_dirty(leaf);
3510         btrfs_set_inode_last_trans(trans, inode);
3511         ret = 0;
3512 failed:
3513         btrfs_free_path(path);
3514         return ret;
3515 }
3516
3517 /*
3518  * copy everything in the in-memory inode into the btree.
3519  */
3520 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3521                                 struct btrfs_root *root, struct inode *inode)
3522 {
3523         int ret;
3524
3525         /*
3526          * If the inode is a free space inode, we can deadlock during commit
3527          * if we put it into the delayed code.
3528          *
3529          * The data relocation inode should also be directly updated
3530          * without delay
3531          */
3532         if (!btrfs_is_free_space_inode(inode)
3533             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3534                 btrfs_update_root_times(trans, root);
3535
3536                 ret = btrfs_delayed_update_inode(trans, root, inode);
3537                 if (!ret)
3538                         btrfs_set_inode_last_trans(trans, inode);
3539                 return ret;
3540         }
3541
3542         return btrfs_update_inode_item(trans, root, inode);
3543 }
3544
3545 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3546                                          struct btrfs_root *root,
3547                                          struct inode *inode)
3548 {
3549         int ret;
3550
3551         ret = btrfs_update_inode(trans, root, inode);
3552         if (ret == -ENOSPC)
3553                 return btrfs_update_inode_item(trans, root, inode);
3554         return ret;
3555 }
3556
3557 /*
3558  * unlink helper that gets used here in inode.c and in the tree logging
3559  * recovery code.  It remove a link in a directory with a given name, and
3560  * also drops the back refs in the inode to the directory
3561  */
3562 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3563                                 struct btrfs_root *root,
3564                                 struct inode *dir, struct inode *inode,
3565                                 const char *name, int name_len)
3566 {
3567         struct btrfs_path *path;
3568         int ret = 0;
3569         struct extent_buffer *leaf;
3570         struct btrfs_dir_item *di;
3571         struct btrfs_key key;
3572         u64 index;
3573         u64 ino = btrfs_ino(inode);
3574         u64 dir_ino = btrfs_ino(dir);
3575
3576         path = btrfs_alloc_path();
3577         if (!path) {
3578                 ret = -ENOMEM;
3579                 goto out;
3580         }
3581
3582         path->leave_spinning = 1;
3583         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3584                                     name, name_len, -1);
3585         if (IS_ERR(di)) {
3586                 ret = PTR_ERR(di);
3587                 goto err;
3588         }
3589         if (!di) {
3590                 ret = -ENOENT;
3591                 goto err;
3592         }
3593         leaf = path->nodes[0];
3594         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3595         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3596         if (ret)
3597                 goto err;
3598         btrfs_release_path(path);
3599
3600         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3601                                   dir_ino, &index);
3602         if (ret) {
3603                 btrfs_info(root->fs_info,
3604                         "failed to delete reference to %.*s, inode %llu parent %llu",
3605                         name_len, name,
3606                         (unsigned long long)ino, (unsigned long long)dir_ino);
3607                 btrfs_abort_transaction(trans, root, ret);
3608                 goto err;
3609         }
3610
3611         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3612         if (ret) {
3613                 btrfs_abort_transaction(trans, root, ret);
3614                 goto err;
3615         }
3616
3617         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3618                                          inode, dir_ino);
3619         if (ret != 0 && ret != -ENOENT) {
3620                 btrfs_abort_transaction(trans, root, ret);
3621                 goto err;
3622         }
3623
3624         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3625                                            dir, index);
3626         if (ret == -ENOENT)
3627                 ret = 0;
3628         else if (ret)
3629                 btrfs_abort_transaction(trans, root, ret);
3630 err:
3631         btrfs_free_path(path);
3632         if (ret)
3633                 goto out;
3634
3635         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3636         inode_inc_iversion(inode);
3637         inode_inc_iversion(dir);
3638         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3639         ret = btrfs_update_inode(trans, root, dir);
3640 out:
3641         return ret;
3642 }
3643
3644 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3645                        struct btrfs_root *root,
3646                        struct inode *dir, struct inode *inode,
3647                        const char *name, int name_len)
3648 {
3649         int ret;
3650         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3651         if (!ret) {
3652                 btrfs_drop_nlink(inode);
3653                 ret = btrfs_update_inode(trans, root, inode);
3654         }
3655         return ret;
3656 }
3657
3658
3659 /* helper to check if there is any shared block in the path */
3660 static int check_path_shared(struct btrfs_root *root,
3661                              struct btrfs_path *path)
3662 {
3663         struct extent_buffer *eb;
3664         int level;
3665         u64 refs = 1;
3666
3667         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3668                 int ret;
3669
3670                 if (!path->nodes[level])
3671                         break;
3672                 eb = path->nodes[level];
3673                 if (!btrfs_block_can_be_shared(root, eb))
3674                         continue;
3675                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3676                                                &refs, NULL);
3677                 if (refs > 1)
3678                         return 1;
3679         }
3680         return 0;
3681 }
3682
3683 /*
3684  * helper to start transaction for unlink and rmdir.
3685  *
3686  * unlink and rmdir are special in btrfs, they do not always free space.
3687  * so in enospc case, we should make sure they will free space before
3688  * allowing them to use the global metadata reservation.
3689  */
3690 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3691                                                        struct dentry *dentry)
3692 {
3693         struct btrfs_trans_handle *trans;
3694         struct btrfs_root *root = BTRFS_I(dir)->root;
3695         struct btrfs_path *path;
3696         struct btrfs_dir_item *di;
3697         struct inode *inode = dentry->d_inode;
3698         u64 index;
3699         int check_link = 1;
3700         int err = -ENOSPC;
3701         int ret;
3702         u64 ino = btrfs_ino(inode);
3703         u64 dir_ino = btrfs_ino(dir);
3704
3705         /*
3706          * 1 for the possible orphan item
3707          * 1 for the dir item
3708          * 1 for the dir index
3709          * 1 for the inode ref
3710          * 1 for the inode
3711          */
3712         trans = btrfs_start_transaction(root, 5);
3713         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3714                 return trans;
3715
3716         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3717                 return ERR_PTR(-ENOSPC);
3718
3719         /* check if there is someone else holds reference */
3720         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3721                 return ERR_PTR(-ENOSPC);
3722
3723         if (atomic_read(&inode->i_count) > 2)
3724                 return ERR_PTR(-ENOSPC);
3725
3726         if (xchg(&root->fs_info->enospc_unlink, 1))
3727                 return ERR_PTR(-ENOSPC);
3728
3729         path = btrfs_alloc_path();
3730         if (!path) {
3731                 root->fs_info->enospc_unlink = 0;
3732                 return ERR_PTR(-ENOMEM);
3733         }
3734
3735         /* 1 for the orphan item */
3736         trans = btrfs_start_transaction(root, 1);
3737         if (IS_ERR(trans)) {
3738                 btrfs_free_path(path);
3739                 root->fs_info->enospc_unlink = 0;
3740                 return trans;
3741         }
3742
3743         path->skip_locking = 1;
3744         path->search_commit_root = 1;
3745
3746         ret = btrfs_lookup_inode(trans, root, path,
3747                                 &BTRFS_I(dir)->location, 0);
3748         if (ret < 0) {
3749                 err = ret;
3750                 goto out;
3751         }
3752         if (ret == 0) {
3753                 if (check_path_shared(root, path))
3754                         goto out;
3755         } else {
3756                 check_link = 0;
3757         }
3758         btrfs_release_path(path);
3759
3760         ret = btrfs_lookup_inode(trans, root, path,
3761                                 &BTRFS_I(inode)->location, 0);
3762         if (ret < 0) {
3763                 err = ret;
3764                 goto out;
3765         }
3766         if (ret == 0) {
3767                 if (check_path_shared(root, path))
3768                         goto out;
3769         } else {
3770                 check_link = 0;
3771         }
3772         btrfs_release_path(path);
3773
3774         if (ret == 0 && S_ISREG(inode->i_mode)) {
3775                 ret = btrfs_lookup_file_extent(trans, root, path,
3776                                                ino, (u64)-1, 0);
3777                 if (ret < 0) {
3778                         err = ret;
3779                         goto out;
3780                 }
3781                 BUG_ON(ret == 0); /* Corruption */
3782                 if (check_path_shared(root, path))
3783                         goto out;
3784                 btrfs_release_path(path);
3785         }
3786
3787         if (!check_link) {
3788                 err = 0;
3789                 goto out;
3790         }
3791
3792         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3793                                 dentry->d_name.name, dentry->d_name.len, 0);
3794         if (IS_ERR(di)) {
3795                 err = PTR_ERR(di);
3796                 goto out;
3797         }
3798         if (di) {
3799                 if (check_path_shared(root, path))
3800                         goto out;
3801         } else {
3802                 err = 0;
3803                 goto out;
3804         }
3805         btrfs_release_path(path);
3806
3807         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3808                                         dentry->d_name.len, ino, dir_ino, 0,
3809                                         &index);
3810         if (ret) {
3811                 err = ret;
3812                 goto out;
3813         }
3814
3815         if (check_path_shared(root, path))
3816                 goto out;
3817
3818         btrfs_release_path(path);
3819
3820         /*
3821          * This is a commit root search, if we can lookup inode item and other
3822          * relative items in the commit root, it means the transaction of
3823          * dir/file creation has been committed, and the dir index item that we
3824          * delay to insert has also been inserted into the commit root. So
3825          * we needn't worry about the delayed insertion of the dir index item
3826          * here.
3827          */
3828         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3829                                 dentry->d_name.name, dentry->d_name.len, 0);
3830         if (IS_ERR(di)) {
3831                 err = PTR_ERR(di);
3832                 goto out;
3833         }
3834         BUG_ON(ret == -ENOENT);
3835         if (check_path_shared(root, path))
3836                 goto out;
3837
3838         err = 0;
3839 out:
3840         btrfs_free_path(path);
3841         /* Migrate the orphan reservation over */
3842         if (!err)
3843                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3844                                 &root->fs_info->global_block_rsv,
3845                                 trans->bytes_reserved);
3846
3847         if (err) {
3848                 btrfs_end_transaction(trans, root);
3849                 root->fs_info->enospc_unlink = 0;
3850                 return ERR_PTR(err);
3851         }
3852
3853         trans->block_rsv = &root->fs_info->global_block_rsv;
3854         return trans;
3855 }
3856
3857 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3858                                struct btrfs_root *root)
3859 {
3860         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3861                 btrfs_block_rsv_release(root, trans->block_rsv,
3862                                         trans->bytes_reserved);
3863                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3864                 BUG_ON(!root->fs_info->enospc_unlink);
3865                 root->fs_info->enospc_unlink = 0;
3866         }
3867         btrfs_end_transaction(trans, root);
3868 }
3869
3870 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3871 {
3872         struct btrfs_root *root = BTRFS_I(dir)->root;
3873         struct btrfs_trans_handle *trans;
3874         struct inode *inode = dentry->d_inode;
3875         int ret;
3876
3877         trans = __unlink_start_trans(dir, dentry);
3878         if (IS_ERR(trans))
3879                 return PTR_ERR(trans);
3880
3881         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3882
3883         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3884                                  dentry->d_name.name, dentry->d_name.len);
3885         if (ret)
3886                 goto out;
3887
3888         if (inode->i_nlink == 0) {
3889                 ret = btrfs_orphan_add(trans, inode);
3890                 if (ret)
3891                         goto out;
3892         }
3893
3894 out:
3895         __unlink_end_trans(trans, root);
3896         btrfs_btree_balance_dirty(root);
3897         return ret;
3898 }
3899
3900 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3901                         struct btrfs_root *root,
3902                         struct inode *dir, u64 objectid,
3903                         const char *name, int name_len)
3904 {
3905         struct btrfs_path *path;
3906         struct extent_buffer *leaf;
3907         struct btrfs_dir_item *di;
3908         struct btrfs_key key;
3909         u64 index;
3910         int ret;
3911         u64 dir_ino = btrfs_ino(dir);
3912
3913         path = btrfs_alloc_path();
3914         if (!path)
3915                 return -ENOMEM;
3916
3917         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3918                                    name, name_len, -1);
3919         if (IS_ERR_OR_NULL(di)) {
3920                 if (!di)
3921                         ret = -ENOENT;
3922                 else
3923                         ret = PTR_ERR(di);
3924                 goto out;
3925         }
3926
3927         leaf = path->nodes[0];
3928         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3929         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3930         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3931         if (ret) {
3932                 btrfs_abort_transaction(trans, root, ret);
3933                 goto out;
3934         }
3935         btrfs_release_path(path);
3936
3937         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3938                                  objectid, root->root_key.objectid,
3939                                  dir_ino, &index, name, name_len);
3940         if (ret < 0) {
3941                 if (ret != -ENOENT) {
3942                         btrfs_abort_transaction(trans, root, ret);
3943                         goto out;
3944                 }
3945                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3946                                                  name, name_len);
3947                 if (IS_ERR_OR_NULL(di)) {
3948                         if (!di)
3949                                 ret = -ENOENT;
3950                         else
3951                                 ret = PTR_ERR(di);
3952                         btrfs_abort_transaction(trans, root, ret);
3953                         goto out;
3954                 }
3955
3956                 leaf = path->nodes[0];
3957                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3958                 btrfs_release_path(path);
3959                 index = key.offset;
3960         }
3961         btrfs_release_path(path);
3962
3963         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3964         if (ret) {
3965                 btrfs_abort_transaction(trans, root, ret);
3966                 goto out;
3967         }
3968
3969         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3970         inode_inc_iversion(dir);
3971         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3972         ret = btrfs_update_inode_fallback(trans, root, dir);
3973         if (ret)
3974                 btrfs_abort_transaction(trans, root, ret);
3975 out:
3976         btrfs_free_path(path);
3977         return ret;
3978 }
3979
3980 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3981 {
3982         struct inode *inode = dentry->d_inode;
3983         int err = 0;
3984         struct btrfs_root *root = BTRFS_I(dir)->root;
3985         struct btrfs_trans_handle *trans;
3986
3987         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3988                 return -ENOTEMPTY;
3989         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3990                 return -EPERM;
3991
3992         trans = __unlink_start_trans(dir, dentry);
3993         if (IS_ERR(trans))
3994                 return PTR_ERR(trans);
3995
3996         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3997                 err = btrfs_unlink_subvol(trans, root, dir,
3998                                           BTRFS_I(inode)->location.objectid,
3999                                           dentry->d_name.name,
4000                                           dentry->d_name.len);
4001                 goto out;
4002         }
4003
4004         err = btrfs_orphan_add(trans, inode);
4005         if (err)
4006                 goto out;
4007
4008         /* now the directory is empty */
4009         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4010                                  dentry->d_name.name, dentry->d_name.len);
4011         if (!err)
4012                 btrfs_i_size_write(inode, 0);
4013 out:
4014         __unlink_end_trans(trans, root);
4015         btrfs_btree_balance_dirty(root);
4016
4017         return err;
4018 }
4019
4020 /*
4021  * this can truncate away extent items, csum items and directory items.
4022  * It starts at a high offset and removes keys until it can't find
4023  * any higher than new_size
4024  *
4025  * csum items that cross the new i_size are truncated to the new size
4026  * as well.
4027  *
4028  * min_type is the minimum key type to truncate down to.  If set to 0, this
4029  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4030  */
4031 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4032                                struct btrfs_root *root,
4033                                struct inode *inode,
4034                                u64 new_size, u32 min_type)
4035 {
4036         struct btrfs_path *path;
4037         struct extent_buffer *leaf;
4038         struct btrfs_file_extent_item *fi;
4039         struct btrfs_key key;
4040         struct btrfs_key found_key;
4041         u64 extent_start = 0;
4042         u64 extent_num_bytes = 0;
4043         u64 extent_offset = 0;
4044         u64 item_end = 0;
4045         u32 found_type = (u8)-1;
4046         int found_extent;
4047         int del_item;
4048         int pending_del_nr = 0;
4049         int pending_del_slot = 0;
4050         int extent_type = -1;
4051         int ret;
4052         int err = 0;
4053         u64 ino = btrfs_ino(inode);
4054
4055         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4056
4057         path = btrfs_alloc_path();
4058         if (!path)
4059                 return -ENOMEM;
4060         path->reada = -1;
4061
4062         /*
4063          * We want to drop from the next block forward in case this new size is
4064          * not block aligned since we will be keeping the last block of the
4065          * extent just the way it is.
4066          */
4067         if (root->ref_cows || root == root->fs_info->tree_root)
4068                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4069                                         root->sectorsize), (u64)-1, 0);
4070
4071         /*
4072          * This function is also used to drop the items in the log tree before
4073          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4074          * it is used to drop the loged items. So we shouldn't kill the delayed
4075          * items.
4076          */
4077         if (min_type == 0 && root == BTRFS_I(inode)->root)
4078                 btrfs_kill_delayed_inode_items(inode);
4079
4080         key.objectid = ino;
4081         key.offset = (u64)-1;
4082         key.type = (u8)-1;
4083
4084 search_again:
4085         path->leave_spinning = 1;
4086         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4087         if (ret < 0) {
4088                 err = ret;
4089                 goto out;
4090         }
4091
4092         if (ret > 0) {
4093                 /* there are no items in the tree for us to truncate, we're
4094                  * done
4095                  */
4096                 if (path->slots[0] == 0)
4097                         goto out;
4098                 path->slots[0]--;
4099         }
4100
4101         while (1) {
4102                 fi = NULL;
4103                 leaf = path->nodes[0];
4104                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4105                 found_type = btrfs_key_type(&found_key);
4106
4107                 if (found_key.objectid != ino)
4108                         break;
4109
4110                 if (found_type < min_type)
4111                         break;
4112
4113                 item_end = found_key.offset;
4114                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4115                         fi = btrfs_item_ptr(leaf, path->slots[0],
4116                                             struct btrfs_file_extent_item);
4117                         extent_type = btrfs_file_extent_type(leaf, fi);
4118                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4119                                 item_end +=
4120                                     btrfs_file_extent_num_bytes(leaf, fi);
4121                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4122                                 item_end += btrfs_file_extent_inline_len(leaf,
4123                                                                          fi);
4124                         }
4125                         item_end--;
4126                 }
4127                 if (found_type > min_type) {
4128                         del_item = 1;
4129                 } else {
4130                         if (item_end < new_size)
4131                                 break;
4132                         if (found_key.offset >= new_size)
4133                                 del_item = 1;
4134                         else
4135                                 del_item = 0;
4136                 }
4137                 found_extent = 0;
4138                 /* FIXME, shrink the extent if the ref count is only 1 */
4139                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4140                         goto delete;
4141
4142                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4143                         u64 num_dec;
4144                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4145                         if (!del_item) {
4146                                 u64 orig_num_bytes =
4147                                         btrfs_file_extent_num_bytes(leaf, fi);
4148                                 extent_num_bytes = ALIGN(new_size -
4149                                                 found_key.offset,
4150                                                 root->sectorsize);
4151                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4152                                                          extent_num_bytes);
4153                                 num_dec = (orig_num_bytes -
4154                                            extent_num_bytes);
4155                                 if (root->ref_cows && extent_start != 0)
4156                                         inode_sub_bytes(inode, num_dec);
4157                                 btrfs_mark_buffer_dirty(leaf);
4158                         } else {
4159                                 extent_num_bytes =
4160                                         btrfs_file_extent_disk_num_bytes(leaf,
4161                                                                          fi);
4162                                 extent_offset = found_key.offset -
4163                                         btrfs_file_extent_offset(leaf, fi);
4164
4165                                 /* FIXME blocksize != 4096 */
4166                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4167                                 if (extent_start != 0) {
4168                                         found_extent = 1;
4169                                         if (root->ref_cows)
4170                                                 inode_sub_bytes(inode, num_dec);
4171                                 }
4172                         }
4173                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4174                         /*
4175                          * we can't truncate inline items that have had
4176                          * special encodings
4177                          */
4178                         if (!del_item &&
4179                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4180                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4181                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4182                                 u32 size = new_size - found_key.offset;
4183
4184                                 if (root->ref_cows) {
4185                                         inode_sub_bytes(inode, item_end + 1 -
4186                                                         new_size);
4187                                 }
4188                                 size =
4189                                     btrfs_file_extent_calc_inline_size(size);
4190                                 btrfs_truncate_item(root, path, size, 1);
4191                         } else if (root->ref_cows) {
4192                                 inode_sub_bytes(inode, item_end + 1 -
4193                                                 found_key.offset);
4194                         }
4195                 }
4196 delete:
4197                 if (del_item) {
4198                         if (!pending_del_nr) {
4199                                 /* no pending yet, add ourselves */
4200                                 pending_del_slot = path->slots[0];
4201                                 pending_del_nr = 1;
4202                         } else if (pending_del_nr &&
4203                                    path->slots[0] + 1 == pending_del_slot) {
4204                                 /* hop on the pending chunk */
4205                                 pending_del_nr++;
4206                                 pending_del_slot = path->slots[0];
4207                         } else {
4208                                 BUG();
4209                         }
4210                 } else {
4211                         break;
4212                 }
4213                 if (found_extent && (root->ref_cows ||
4214                                      root == root->fs_info->tree_root)) {
4215                         btrfs_set_path_blocking(path);
4216                         ret = btrfs_free_extent(trans, root, extent_start,
4217                                                 extent_num_bytes, 0,
4218                                                 btrfs_header_owner(leaf),
4219                                                 ino, extent_offset, 0);
4220                         BUG_ON(ret);
4221                 }
4222
4223                 if (found_type == BTRFS_INODE_ITEM_KEY)
4224                         break;
4225
4226                 if (path->slots[0] == 0 ||
4227                     path->slots[0] != pending_del_slot) {
4228                         if (pending_del_nr) {
4229                                 ret = btrfs_del_items(trans, root, path,
4230                                                 pending_del_slot,
4231                                                 pending_del_nr);
4232                                 if (ret) {
4233                                         btrfs_abort_transaction(trans,
4234                                                                 root, ret);
4235                                         goto error;
4236                                 }
4237                                 pending_del_nr = 0;
4238                         }
4239                         btrfs_release_path(path);
4240                         goto search_again;
4241                 } else {
4242                         path->slots[0]--;
4243                 }
4244         }
4245 out:
4246         if (pending_del_nr) {
4247                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4248                                       pending_del_nr);
4249                 if (ret)
4250                         btrfs_abort_transaction(trans, root, ret);
4251         }
4252 error:
4253         btrfs_free_path(path);
4254         return err;
4255 }
4256
4257 /*
4258  * btrfs_truncate_page - read, zero a chunk and write a page
4259  * @inode - inode that we're zeroing
4260  * @from - the offset to start zeroing
4261  * @len - the length to zero, 0 to zero the entire range respective to the
4262  *      offset
4263  * @front - zero up to the offset instead of from the offset on
4264  *
4265  * This will find the page for the "from" offset and cow the page and zero the
4266  * part we want to zero.  This is used with truncate and hole punching.
4267  */
4268 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4269                         int front)
4270 {
4271         struct address_space *mapping = inode->i_mapping;
4272         struct btrfs_root *root = BTRFS_I(inode)->root;
4273         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4274         struct btrfs_ordered_extent *ordered;
4275         struct extent_state *cached_state = NULL;
4276         char *kaddr;
4277         u32 blocksize = root->sectorsize;
4278         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4279         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4280         struct page *page;
4281         gfp_t mask = btrfs_alloc_write_mask(mapping);
4282         int ret = 0;
4283         u64 page_start;
4284         u64 page_end;
4285
4286         if ((offset & (blocksize - 1)) == 0 &&
4287             (!len || ((len & (blocksize - 1)) == 0)))
4288                 goto out;
4289         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4290         if (ret)
4291                 goto out;
4292
4293 again:
4294         page = find_or_create_page(mapping, index, mask);
4295         if (!page) {
4296                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4297                 ret = -ENOMEM;
4298                 goto out;
4299         }
4300
4301         page_start = page_offset(page);
4302         page_end = page_start + PAGE_CACHE_SIZE - 1;
4303
4304         if (!PageUptodate(page)) {
4305                 ret = btrfs_readpage(NULL, page);
4306                 lock_page(page);
4307                 if (page->mapping != mapping) {
4308                         unlock_page(page);
4309                         page_cache_release(page);
4310                         goto again;
4311                 }
4312                 if (!PageUptodate(page)) {
4313                         ret = -EIO;
4314                         goto out_unlock;
4315                 }
4316         }
4317         wait_on_page_writeback(page);
4318
4319         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4320         set_page_extent_mapped(page);
4321
4322         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4323         if (ordered) {
4324                 unlock_extent_cached(io_tree, page_start, page_end,
4325                                      &cached_state, GFP_NOFS);
4326                 unlock_page(page);
4327                 page_cache_release(page);
4328                 btrfs_start_ordered_extent(inode, ordered, 1);
4329                 btrfs_put_ordered_extent(ordered);
4330                 goto again;
4331         }
4332
4333         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4334                           EXTENT_DIRTY | EXTENT_DELALLOC |
4335                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4336                           0, 0, &cached_state, GFP_NOFS);
4337
4338         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4339                                         &cached_state);
4340         if (ret) {
4341                 unlock_extent_cached(io_tree, page_start, page_end,
4342                                      &cached_state, GFP_NOFS);
4343                 goto out_unlock;
4344         }
4345
4346         if (offset != PAGE_CACHE_SIZE) {
4347                 if (!len)
4348                         len = PAGE_CACHE_SIZE - offset;
4349                 kaddr = kmap(page);
4350                 if (front)
4351                         memset(kaddr, 0, offset);
4352                 else
4353                         memset(kaddr + offset, 0, len);
4354                 flush_dcache_page(page);
4355                 kunmap(page);
4356         }
4357         ClearPageChecked(page);
4358         set_page_dirty(page);
4359         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4360                              GFP_NOFS);
4361
4362 out_unlock:
4363         if (ret)
4364                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4365         unlock_page(page);
4366         page_cache_release(page);
4367 out:
4368         return ret;
4369 }
4370
4371 /*
4372  * This function puts in dummy file extents for the area we're creating a hole
4373  * for.  So if we are truncating this file to a larger size we need to insert
4374  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4375  * the range between oldsize and size
4376  */
4377 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4378 {
4379         struct btrfs_trans_handle *trans;
4380         struct btrfs_root *root = BTRFS_I(inode)->root;
4381         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4382         struct extent_map *em = NULL;
4383         struct extent_state *cached_state = NULL;
4384         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4385         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4386         u64 block_end = ALIGN(size, root->sectorsize);
4387         u64 last_byte;
4388         u64 cur_offset;
4389         u64 hole_size;
4390         int err = 0;
4391
4392         if (size <= hole_start)
4393                 return 0;
4394
4395         while (1) {
4396                 struct btrfs_ordered_extent *ordered;
4397                 btrfs_wait_ordered_range(inode, hole_start,
4398                                          block_end - hole_start);
4399                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4400                                  &cached_state);
4401                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4402                 if (!ordered)
4403                         break;
4404                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4405                                      &cached_state, GFP_NOFS);
4406                 btrfs_put_ordered_extent(ordered);
4407         }
4408
4409         cur_offset = hole_start;
4410         while (1) {
4411                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4412                                 block_end - cur_offset, 0);
4413                 if (IS_ERR(em)) {
4414                         err = PTR_ERR(em);
4415                         em = NULL;
4416                         break;
4417                 }
4418                 last_byte = min(extent_map_end(em), block_end);
4419                 last_byte = ALIGN(last_byte , root->sectorsize);
4420                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4421                         struct extent_map *hole_em;
4422                         hole_size = last_byte - cur_offset;
4423
4424                         trans = btrfs_start_transaction(root, 3);
4425                         if (IS_ERR(trans)) {
4426                                 err = PTR_ERR(trans);
4427                                 break;
4428                         }
4429
4430                         err = btrfs_drop_extents(trans, root, inode,
4431                                                  cur_offset,
4432                                                  cur_offset + hole_size, 1);
4433                         if (err) {
4434                                 btrfs_abort_transaction(trans, root, err);
4435                                 btrfs_end_transaction(trans, root);
4436                                 break;
4437                         }
4438
4439                         err = btrfs_insert_file_extent(trans, root,
4440                                         btrfs_ino(inode), cur_offset, 0,
4441                                         0, hole_size, 0, hole_size,
4442                                         0, 0, 0);
4443                         if (err) {
4444                                 btrfs_abort_transaction(trans, root, err);
4445                                 btrfs_end_transaction(trans, root);
4446                                 break;
4447                         }
4448
4449                         btrfs_drop_extent_cache(inode, cur_offset,
4450                                                 cur_offset + hole_size - 1, 0);
4451                         hole_em = alloc_extent_map();
4452                         if (!hole_em) {
4453                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4454                                         &BTRFS_I(inode)->runtime_flags);
4455                                 goto next;
4456                         }
4457                         hole_em->start = cur_offset;
4458                         hole_em->len = hole_size;
4459                         hole_em->orig_start = cur_offset;
4460
4461                         hole_em->block_start = EXTENT_MAP_HOLE;
4462                         hole_em->block_len = 0;
4463                         hole_em->orig_block_len = 0;
4464                         hole_em->ram_bytes = hole_size;
4465                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4466                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4467                         hole_em->generation = trans->transid;
4468
4469                         while (1) {
4470                                 write_lock(&em_tree->lock);
4471                                 err = add_extent_mapping(em_tree, hole_em, 1);
4472                                 write_unlock(&em_tree->lock);
4473                                 if (err != -EEXIST)
4474                                         break;
4475                                 btrfs_drop_extent_cache(inode, cur_offset,
4476                                                         cur_offset +
4477                                                         hole_size - 1, 0);
4478                         }
4479                         free_extent_map(hole_em);
4480 next:
4481                         btrfs_update_inode(trans, root, inode);
4482                         btrfs_end_transaction(trans, root);
4483                 }
4484                 free_extent_map(em);
4485                 em = NULL;
4486                 cur_offset = last_byte;
4487                 if (cur_offset >= block_end)
4488                         break;
4489         }
4490
4491         free_extent_map(em);
4492         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4493                              GFP_NOFS);
4494         return err;
4495 }
4496
4497 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4498 {
4499         struct btrfs_root *root = BTRFS_I(inode)->root;
4500         struct btrfs_trans_handle *trans;
4501         loff_t oldsize = i_size_read(inode);
4502         loff_t newsize = attr->ia_size;
4503         int mask = attr->ia_valid;
4504         int ret;
4505
4506         if (newsize == oldsize)
4507                 return 0;
4508
4509         /*
4510          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4511          * special case where we need to update the times despite not having
4512          * these flags set.  For all other operations the VFS set these flags
4513          * explicitly if it wants a timestamp update.
4514          */
4515         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4516                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4517
4518         if (newsize > oldsize) {
4519                 truncate_pagecache(inode, oldsize, newsize);
4520                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4521                 if (ret)
4522                         return ret;
4523
4524                 trans = btrfs_start_transaction(root, 1);
4525                 if (IS_ERR(trans))
4526                         return PTR_ERR(trans);
4527
4528                 i_size_write(inode, newsize);
4529                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4530                 ret = btrfs_update_inode(trans, root, inode);
4531                 btrfs_end_transaction(trans, root);
4532         } else {
4533
4534                 /*
4535                  * We're truncating a file that used to have good data down to
4536                  * zero. Make sure it gets into the ordered flush list so that
4537                  * any new writes get down to disk quickly.
4538                  */
4539                 if (newsize == 0)
4540                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4541                                 &BTRFS_I(inode)->runtime_flags);
4542
4543                 /*
4544                  * 1 for the orphan item we're going to add
4545                  * 1 for the orphan item deletion.
4546                  */
4547                 trans = btrfs_start_transaction(root, 2);
4548                 if (IS_ERR(trans))
4549                         return PTR_ERR(trans);
4550
4551                 /*
4552                  * We need to do this in case we fail at _any_ point during the
4553                  * actual truncate.  Once we do the truncate_setsize we could
4554                  * invalidate pages which forces any outstanding ordered io to
4555                  * be instantly completed which will give us extents that need
4556                  * to be truncated.  If we fail to get an orphan inode down we
4557                  * could have left over extents that were never meant to live,
4558                  * so we need to garuntee from this point on that everything
4559                  * will be consistent.
4560                  */
4561                 ret = btrfs_orphan_add(trans, inode);
4562                 btrfs_end_transaction(trans, root);
4563                 if (ret)
4564                         return ret;
4565
4566                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4567                 truncate_setsize(inode, newsize);
4568
4569                 /* Disable nonlocked read DIO to avoid the end less truncate */
4570                 btrfs_inode_block_unlocked_dio(inode);
4571                 inode_dio_wait(inode);
4572                 btrfs_inode_resume_unlocked_dio(inode);
4573
4574                 ret = btrfs_truncate(inode);
4575                 if (ret && inode->i_nlink)
4576                         btrfs_orphan_del(NULL, inode);
4577         }
4578
4579         return ret;
4580 }
4581
4582 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4583 {
4584         struct inode *inode = dentry->d_inode;
4585         struct btrfs_root *root = BTRFS_I(inode)->root;
4586         int err;
4587
4588         if (btrfs_root_readonly(root))
4589                 return -EROFS;
4590
4591         err = inode_change_ok(inode, attr);
4592         if (err)
4593                 return err;
4594
4595         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4596                 err = btrfs_setsize(inode, attr);
4597                 if (err)
4598                         return err;
4599         }
4600
4601         if (attr->ia_valid) {
4602                 setattr_copy(inode, attr);
4603                 inode_inc_iversion(inode);
4604                 err = btrfs_dirty_inode(inode);
4605
4606                 if (!err && attr->ia_valid & ATTR_MODE)
4607                         err = btrfs_acl_chmod(inode);
4608         }
4609
4610         return err;
4611 }
4612
4613 void btrfs_evict_inode(struct inode *inode)
4614 {
4615         struct btrfs_trans_handle *trans;
4616         struct btrfs_root *root = BTRFS_I(inode)->root;
4617         struct btrfs_block_rsv *rsv, *global_rsv;
4618         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4619         int ret;
4620
4621         trace_btrfs_inode_evict(inode);
4622
4623         truncate_inode_pages(&inode->i_data, 0);
4624         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4625                                btrfs_is_free_space_inode(inode)))
4626                 goto no_delete;
4627
4628         if (is_bad_inode(inode)) {
4629                 btrfs_orphan_del(NULL, inode);
4630                 goto no_delete;
4631         }
4632         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4633         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4634
4635         if (root->fs_info->log_root_recovering) {
4636                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4637                                  &BTRFS_I(inode)->runtime_flags));
4638                 goto no_delete;
4639         }
4640
4641         if (inode->i_nlink > 0) {
4642                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4643                 goto no_delete;
4644         }
4645
4646         ret = btrfs_commit_inode_delayed_inode(inode);
4647         if (ret) {
4648                 btrfs_orphan_del(NULL, inode);
4649                 goto no_delete;
4650         }
4651
4652         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4653         if (!rsv) {
4654                 btrfs_orphan_del(NULL, inode);
4655                 goto no_delete;
4656         }
4657         rsv->size = min_size;
4658         rsv->failfast = 1;
4659         global_rsv = &root->fs_info->global_block_rsv;
4660
4661         btrfs_i_size_write(inode, 0);
4662
4663         /*
4664          * This is a bit simpler than btrfs_truncate since we've already
4665          * reserved our space for our orphan item in the unlink, so we just
4666          * need to reserve some slack space in case we add bytes and update
4667          * inode item when doing the truncate.
4668          */
4669         while (1) {
4670                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4671                                              BTRFS_RESERVE_FLUSH_LIMIT);
4672
4673                 /*
4674                  * Try and steal from the global reserve since we will
4675                  * likely not use this space anyway, we want to try as
4676                  * hard as possible to get this to work.
4677                  */
4678                 if (ret)
4679                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4680
4681                 if (ret) {
4682                         btrfs_warn(root->fs_info,
4683                                 "Could not get space for a delete, will truncate on mount %d",
4684                                 ret);
4685                         btrfs_orphan_del(NULL, inode);
4686                         btrfs_free_block_rsv(root, rsv);
4687                         goto no_delete;
4688                 }
4689
4690                 trans = btrfs_join_transaction(root);
4691                 if (IS_ERR(trans)) {
4692                         btrfs_orphan_del(NULL, inode);
4693                         btrfs_free_block_rsv(root, rsv);
4694                         goto no_delete;
4695                 }
4696
4697                 trans->block_rsv = rsv;
4698
4699                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4700                 if (ret != -ENOSPC)
4701                         break;
4702
4703                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4704                 btrfs_end_transaction(trans, root);
4705                 trans = NULL;
4706                 btrfs_btree_balance_dirty(root);
4707         }
4708
4709         btrfs_free_block_rsv(root, rsv);
4710
4711         if (ret == 0) {
4712                 trans->block_rsv = root->orphan_block_rsv;
4713                 ret = btrfs_orphan_del(trans, inode);
4714                 BUG_ON(ret);
4715         }
4716
4717         trans->block_rsv = &root->fs_info->trans_block_rsv;
4718         if (!(root == root->fs_info->tree_root ||
4719               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4720                 btrfs_return_ino(root, btrfs_ino(inode));
4721
4722         btrfs_end_transaction(trans, root);
4723         btrfs_btree_balance_dirty(root);
4724 no_delete:
4725         btrfs_remove_delayed_node(inode);
4726         clear_inode(inode);
4727         return;
4728 }
4729
4730 /*
4731  * this returns the key found in the dir entry in the location pointer.
4732  * If no dir entries were found, location->objectid is 0.
4733  */
4734 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4735                                struct btrfs_key *location)
4736 {
4737         const char *name = dentry->d_name.name;
4738         int namelen = dentry->d_name.len;
4739         struct btrfs_dir_item *di;
4740         struct btrfs_path *path;
4741         struct btrfs_root *root = BTRFS_I(dir)->root;
4742         int ret = 0;
4743
4744         path = btrfs_alloc_path();
4745         if (!path)
4746                 return -ENOMEM;
4747
4748         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4749                                     namelen, 0);
4750         if (IS_ERR(di))
4751                 ret = PTR_ERR(di);
4752
4753         if (IS_ERR_OR_NULL(di))
4754                 goto out_err;
4755
4756         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4757 out:
4758         btrfs_free_path(path);
4759         return ret;
4760 out_err:
4761         location->objectid = 0;
4762         goto out;
4763 }
4764
4765 /*
4766  * when we hit a tree root in a directory, the btrfs part of the inode
4767  * needs to be changed to reflect the root directory of the tree root.  This
4768  * is kind of like crossing a mount point.
4769  */
4770 static int fixup_tree_root_location(struct btrfs_root *root,
4771                                     struct inode *dir,
4772                                     struct dentry *dentry,
4773                                     struct btrfs_key *location,
4774                                     struct btrfs_root **sub_root)
4775 {
4776         struct btrfs_path *path;
4777         struct btrfs_root *new_root;
4778         struct btrfs_root_ref *ref;
4779         struct extent_buffer *leaf;
4780         int ret;
4781         int err = 0;
4782
4783         path = btrfs_alloc_path();
4784         if (!path) {
4785                 err = -ENOMEM;
4786                 goto out;
4787         }
4788
4789         err = -ENOENT;
4790         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4791                                   BTRFS_I(dir)->root->root_key.objectid,
4792                                   location->objectid);
4793         if (ret) {
4794                 if (ret < 0)
4795                         err = ret;
4796                 goto out;
4797         }
4798
4799         leaf = path->nodes[0];
4800         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4801         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4802             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4803                 goto out;
4804
4805         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4806                                    (unsigned long)(ref + 1),
4807                                    dentry->d_name.len);
4808         if (ret)
4809                 goto out;
4810
4811         btrfs_release_path(path);
4812
4813         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4814         if (IS_ERR(new_root)) {
4815                 err = PTR_ERR(new_root);
4816                 goto out;
4817         }
4818
4819         *sub_root = new_root;
4820         location->objectid = btrfs_root_dirid(&new_root->root_item);
4821         location->type = BTRFS_INODE_ITEM_KEY;
4822         location->offset = 0;
4823         err = 0;
4824 out:
4825         btrfs_free_path(path);
4826         return err;
4827 }
4828
4829 static void inode_tree_add(struct inode *inode)
4830 {
4831         struct btrfs_root *root = BTRFS_I(inode)->root;
4832         struct btrfs_inode *entry;
4833         struct rb_node **p;
4834         struct rb_node *parent;
4835         u64 ino = btrfs_ino(inode);
4836
4837         if (inode_unhashed(inode))
4838                 return;
4839 again:
4840         parent = NULL;
4841         spin_lock(&root->inode_lock);
4842         p = &root->inode_tree.rb_node;
4843         while (*p) {
4844                 parent = *p;
4845                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4846
4847                 if (ino < btrfs_ino(&entry->vfs_inode))
4848                         p = &parent->rb_left;
4849                 else if (ino > btrfs_ino(&entry->vfs_inode))
4850                         p = &parent->rb_right;
4851                 else {
4852                         WARN_ON(!(entry->vfs_inode.i_state &
4853                                   (I_WILL_FREE | I_FREEING)));
4854                         rb_erase(parent, &root->inode_tree);
4855                         RB_CLEAR_NODE(parent);
4856                         spin_unlock(&root->inode_lock);
4857                         goto again;
4858                 }
4859         }
4860         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4861         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4862         spin_unlock(&root->inode_lock);
4863 }
4864
4865 static void inode_tree_del(struct inode *inode)
4866 {
4867         struct btrfs_root *root = BTRFS_I(inode)->root;
4868         int empty = 0;
4869
4870         spin_lock(&root->inode_lock);
4871         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4872                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4873                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4874                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4875         }
4876         spin_unlock(&root->inode_lock);
4877
4878         /*
4879          * Free space cache has inodes in the tree root, but the tree root has a
4880          * root_refs of 0, so this could end up dropping the tree root as a
4881          * snapshot, so we need the extra !root->fs_info->tree_root check to
4882          * make sure we don't drop it.
4883          */
4884         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4885             root != root->fs_info->tree_root) {
4886                 synchronize_srcu(&root->fs_info->subvol_srcu);
4887                 spin_lock(&root->inode_lock);
4888                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4889                 spin_unlock(&root->inode_lock);
4890                 if (empty)
4891                         btrfs_add_dead_root(root);
4892         }
4893 }
4894
4895 void btrfs_invalidate_inodes(struct btrfs_root *root)
4896 {
4897         struct rb_node *node;
4898         struct rb_node *prev;
4899         struct btrfs_inode *entry;
4900         struct inode *inode;
4901         u64 objectid = 0;
4902
4903         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4904
4905         spin_lock(&root->inode_lock);
4906 again:
4907         node = root->inode_tree.rb_node;
4908         prev = NULL;
4909         while (node) {
4910                 prev = node;
4911                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4912
4913                 if (objectid < btrfs_ino(&entry->vfs_inode))
4914                         node = node->rb_left;
4915                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4916                         node = node->rb_right;
4917                 else
4918                         break;
4919         }
4920         if (!node) {
4921                 while (prev) {
4922                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4923                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4924                                 node = prev;
4925                                 break;
4926                         }
4927                         prev = rb_next(prev);
4928                 }
4929         }
4930         while (node) {
4931                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4932                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4933                 inode = igrab(&entry->vfs_inode);
4934                 if (inode) {
4935                         spin_unlock(&root->inode_lock);
4936                         if (atomic_read(&inode->i_count) > 1)
4937                                 d_prune_aliases(inode);
4938                         /*
4939                          * btrfs_drop_inode will have it removed from
4940                          * the inode cache when its usage count
4941                          * hits zero.
4942                          */
4943                         iput(inode);
4944                         cond_resched();
4945                         spin_lock(&root->inode_lock);
4946                         goto again;
4947                 }
4948
4949                 if (cond_resched_lock(&root->inode_lock))
4950                         goto again;
4951
4952                 node = rb_next(node);
4953         }
4954         spin_unlock(&root->inode_lock);
4955 }
4956
4957 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4958 {
4959         struct btrfs_iget_args *args = p;
4960         inode->i_ino = args->ino;
4961         BTRFS_I(inode)->root = args->root;
4962         return 0;
4963 }
4964
4965 static int btrfs_find_actor(struct inode *inode, void *opaque)
4966 {
4967         struct btrfs_iget_args *args = opaque;
4968         return args->ino == btrfs_ino(inode) &&
4969                 args->root == BTRFS_I(inode)->root;
4970 }
4971
4972 static struct inode *btrfs_iget_locked(struct super_block *s,
4973                                        u64 objectid,
4974                                        struct btrfs_root *root)
4975 {
4976         struct inode *inode;
4977         struct btrfs_iget_args args;
4978         args.ino = objectid;
4979         args.root = root;
4980
4981         inode = iget5_locked(s, objectid, btrfs_find_actor,
4982                              btrfs_init_locked_inode,
4983                              (void *)&args);
4984         return inode;
4985 }
4986
4987 /* Get an inode object given its location and corresponding root.
4988  * Returns in *is_new if the inode was read from disk
4989  */
4990 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4991                          struct btrfs_root *root, int *new)
4992 {
4993         struct inode *inode;
4994
4995         inode = btrfs_iget_locked(s, location->objectid, root);
4996         if (!inode)
4997                 return ERR_PTR(-ENOMEM);
4998
4999         if (inode->i_state & I_NEW) {
5000                 BTRFS_I(inode)->root = root;
5001                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5002                 btrfs_read_locked_inode(inode);
5003                 if (!is_bad_inode(inode)) {
5004                         inode_tree_add(inode);
5005                         unlock_new_inode(inode);
5006                         if (new)
5007                                 *new = 1;
5008                 } else {
5009                         unlock_new_inode(inode);
5010                         iput(inode);
5011                         inode = ERR_PTR(-ESTALE);
5012                 }
5013         }
5014
5015         return inode;
5016 }
5017
5018 static struct inode *new_simple_dir(struct super_block *s,
5019                                     struct btrfs_key *key,
5020                                     struct btrfs_root *root)
5021 {
5022         struct inode *inode = new_inode(s);
5023
5024         if (!inode)
5025                 return ERR_PTR(-ENOMEM);
5026
5027         BTRFS_I(inode)->root = root;
5028         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5029         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5030
5031         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5032         inode->i_op = &btrfs_dir_ro_inode_operations;
5033         inode->i_fop = &simple_dir_operations;
5034         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5035         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5036
5037         return inode;
5038 }
5039
5040 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5041 {
5042         struct inode *inode;
5043         struct btrfs_root *root = BTRFS_I(dir)->root;
5044         struct btrfs_root *sub_root = root;
5045         struct btrfs_key location;
5046         int index;
5047         int ret = 0;
5048
5049         if (dentry->d_name.len > BTRFS_NAME_LEN)
5050                 return ERR_PTR(-ENAMETOOLONG);
5051
5052         ret = btrfs_inode_by_name(dir, dentry, &location);
5053         if (ret < 0)
5054                 return ERR_PTR(ret);
5055
5056         if (location.objectid == 0)
5057                 return NULL;
5058
5059         if (location.type == BTRFS_INODE_ITEM_KEY) {
5060                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5061                 return inode;
5062         }
5063
5064         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5065
5066         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5067         ret = fixup_tree_root_location(root, dir, dentry,
5068                                        &location, &sub_root);
5069         if (ret < 0) {
5070                 if (ret != -ENOENT)
5071                         inode = ERR_PTR(ret);
5072                 else
5073                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5074         } else {
5075                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5076         }
5077         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5078
5079         if (!IS_ERR(inode) && root != sub_root) {
5080                 down_read(&root->fs_info->cleanup_work_sem);
5081                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5082                         ret = btrfs_orphan_cleanup(sub_root);
5083                 up_read(&root->fs_info->cleanup_work_sem);
5084                 if (ret)
5085                         inode = ERR_PTR(ret);
5086         }
5087
5088         return inode;
5089 }
5090
5091 static int btrfs_dentry_delete(const struct dentry *dentry)
5092 {
5093         struct btrfs_root *root;
5094         struct inode *inode = dentry->d_inode;
5095
5096         if (!inode && !IS_ROOT(dentry))
5097                 inode = dentry->d_parent->d_inode;
5098
5099         if (inode) {
5100                 root = BTRFS_I(inode)->root;
5101                 if (btrfs_root_refs(&root->root_item) == 0)
5102                         return 1;
5103
5104                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5105                         return 1;
5106         }
5107         return 0;
5108 }
5109
5110 static void btrfs_dentry_release(struct dentry *dentry)
5111 {
5112         if (dentry->d_fsdata)
5113                 kfree(dentry->d_fsdata);
5114 }
5115
5116 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5117                                    unsigned int flags)
5118 {
5119         struct dentry *ret;
5120
5121         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5122         return ret;
5123 }
5124
5125 unsigned char btrfs_filetype_table[] = {
5126         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5127 };
5128
5129 static int btrfs_real_readdir(struct file *filp, void *dirent,
5130                               filldir_t filldir)
5131 {
5132         struct inode *inode = file_inode(filp);
5133         struct btrfs_root *root = BTRFS_I(inode)->root;
5134         struct btrfs_item *item;
5135         struct btrfs_dir_item *di;
5136         struct btrfs_key key;
5137         struct btrfs_key found_key;
5138         struct btrfs_path *path;
5139         struct list_head ins_list;
5140         struct list_head del_list;
5141         int ret;
5142         struct extent_buffer *leaf;
5143         int slot;
5144         unsigned char d_type;
5145         int over = 0;
5146         u32 di_cur;
5147         u32 di_total;
5148         u32 di_len;
5149         int key_type = BTRFS_DIR_INDEX_KEY;
5150         char tmp_name[32];
5151         char *name_ptr;
5152         int name_len;
5153         int is_curr = 0;        /* filp->f_pos points to the current index? */
5154
5155         /* FIXME, use a real flag for deciding about the key type */
5156         if (root->fs_info->tree_root == root)
5157                 key_type = BTRFS_DIR_ITEM_KEY;
5158
5159         /* special case for "." */
5160         if (filp->f_pos == 0) {
5161                 over = filldir(dirent, ".", 1,
5162                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5163                 if (over)
5164                         return 0;
5165                 filp->f_pos = 1;
5166         }
5167         /* special case for .., just use the back ref */
5168         if (filp->f_pos == 1) {
5169                 u64 pino = parent_ino(filp->f_path.dentry);
5170                 over = filldir(dirent, "..", 2,
5171                                filp->f_pos, pino, DT_DIR);
5172                 if (over)
5173                         return 0;
5174                 filp->f_pos = 2;
5175         }
5176         path = btrfs_alloc_path();
5177         if (!path)
5178                 return -ENOMEM;
5179
5180         path->reada = 1;
5181
5182         if (key_type == BTRFS_DIR_INDEX_KEY) {
5183                 INIT_LIST_HEAD(&ins_list);
5184                 INIT_LIST_HEAD(&del_list);
5185                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5186         }
5187
5188         btrfs_set_key_type(&key, key_type);
5189         key.offset = filp->f_pos;
5190         key.objectid = btrfs_ino(inode);
5191
5192         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5193         if (ret < 0)
5194                 goto err;
5195
5196         while (1) {
5197                 leaf = path->nodes[0];
5198                 slot = path->slots[0];
5199                 if (slot >= btrfs_header_nritems(leaf)) {
5200                         ret = btrfs_next_leaf(root, path);
5201                         if (ret < 0)
5202                                 goto err;
5203                         else if (ret > 0)
5204                                 break;
5205                         continue;
5206                 }
5207
5208                 item = btrfs_item_nr(leaf, slot);
5209                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5210
5211                 if (found_key.objectid != key.objectid)
5212                         break;
5213                 if (btrfs_key_type(&found_key) != key_type)
5214                         break;
5215                 if (found_key.offset < filp->f_pos)
5216                         goto next;
5217                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5218                     btrfs_should_delete_dir_index(&del_list,
5219                                                   found_key.offset))
5220                         goto next;
5221
5222                 filp->f_pos = found_key.offset;
5223                 is_curr = 1;
5224
5225                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5226                 di_cur = 0;
5227                 di_total = btrfs_item_size(leaf, item);
5228
5229                 while (di_cur < di_total) {
5230                         struct btrfs_key location;
5231
5232                         if (verify_dir_item(root, leaf, di))
5233                                 break;
5234
5235                         name_len = btrfs_dir_name_len(leaf, di);
5236                         if (name_len <= sizeof(tmp_name)) {
5237                                 name_ptr = tmp_name;
5238                         } else {
5239                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5240                                 if (!name_ptr) {
5241                                         ret = -ENOMEM;
5242                                         goto err;
5243                                 }
5244                         }
5245                         read_extent_buffer(leaf, name_ptr,
5246                                            (unsigned long)(di + 1), name_len);
5247
5248                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5249                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5250
5251
5252                         /* is this a reference to our own snapshot? If so
5253                          * skip it.
5254                          *
5255                          * In contrast to old kernels, we insert the snapshot's
5256                          * dir item and dir index after it has been created, so
5257                          * we won't find a reference to our own snapshot. We
5258                          * still keep the following code for backward
5259                          * compatibility.
5260                          */
5261                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5262                             location.objectid == root->root_key.objectid) {
5263                                 over = 0;
5264                                 goto skip;
5265                         }
5266                         over = filldir(dirent, name_ptr, name_len,
5267                                        found_key.offset, location.objectid,
5268                                        d_type);
5269
5270 skip:
5271                         if (name_ptr != tmp_name)
5272                                 kfree(name_ptr);
5273
5274                         if (over)
5275                                 goto nopos;
5276                         di_len = btrfs_dir_name_len(leaf, di) +
5277                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5278                         di_cur += di_len;
5279                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5280                 }
5281 next:
5282                 path->slots[0]++;
5283         }
5284
5285         if (key_type == BTRFS_DIR_INDEX_KEY) {
5286                 if (is_curr)
5287                         filp->f_pos++;
5288                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5289                                                       &ins_list);
5290                 if (ret)
5291                         goto nopos;
5292         }
5293
5294         /* Reached end of directory/root. Bump pos past the last item. */
5295         if (key_type == BTRFS_DIR_INDEX_KEY)
5296                 /*
5297                  * 32-bit glibc will use getdents64, but then strtol -
5298                  * so the last number we can serve is this.
5299                  */
5300                 filp->f_pos = 0x7fffffff;
5301         else
5302                 filp->f_pos++;
5303 nopos:
5304         ret = 0;
5305 err:
5306         if (key_type == BTRFS_DIR_INDEX_KEY)
5307                 btrfs_put_delayed_items(&ins_list, &del_list);
5308         btrfs_free_path(path);
5309         return ret;
5310 }
5311
5312 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5313 {
5314         struct btrfs_root *root = BTRFS_I(inode)->root;
5315         struct btrfs_trans_handle *trans;
5316         int ret = 0;
5317         bool nolock = false;
5318
5319         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5320                 return 0;
5321
5322         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5323                 nolock = true;
5324
5325         if (wbc->sync_mode == WB_SYNC_ALL) {
5326                 if (nolock)
5327                         trans = btrfs_join_transaction_nolock(root);
5328                 else
5329                         trans = btrfs_join_transaction(root);
5330                 if (IS_ERR(trans))
5331                         return PTR_ERR(trans);
5332                 ret = btrfs_commit_transaction(trans, root);
5333         }
5334         return ret;
5335 }
5336
5337 /*
5338  * This is somewhat expensive, updating the tree every time the
5339  * inode changes.  But, it is most likely to find the inode in cache.
5340  * FIXME, needs more benchmarking...there are no reasons other than performance
5341  * to keep or drop this code.
5342  */
5343 static int btrfs_dirty_inode(struct inode *inode)
5344 {
5345         struct btrfs_root *root = BTRFS_I(inode)->root;
5346         struct btrfs_trans_handle *trans;
5347         int ret;
5348
5349         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5350                 return 0;
5351
5352         trans = btrfs_join_transaction(root);
5353         if (IS_ERR(trans))
5354                 return PTR_ERR(trans);
5355
5356         ret = btrfs_update_inode(trans, root, inode);
5357         if (ret && ret == -ENOSPC) {
5358                 /* whoops, lets try again with the full transaction */
5359                 btrfs_end_transaction(trans, root);
5360                 trans = btrfs_start_transaction(root, 1);
5361                 if (IS_ERR(trans))
5362                         return PTR_ERR(trans);
5363
5364                 ret = btrfs_update_inode(trans, root, inode);
5365         }
5366         btrfs_end_transaction(trans, root);
5367         if (BTRFS_I(inode)->delayed_node)
5368                 btrfs_balance_delayed_items(root);
5369
5370         return ret;
5371 }
5372
5373 /*
5374  * This is a copy of file_update_time.  We need this so we can return error on
5375  * ENOSPC for updating the inode in the case of file write and mmap writes.
5376  */
5377 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5378                              int flags)
5379 {
5380         struct btrfs_root *root = BTRFS_I(inode)->root;
5381
5382         if (btrfs_root_readonly(root))
5383                 return -EROFS;
5384
5385         if (flags & S_VERSION)
5386                 inode_inc_iversion(inode);
5387         if (flags & S_CTIME)
5388                 inode->i_ctime = *now;
5389         if (flags & S_MTIME)
5390                 inode->i_mtime = *now;
5391         if (flags & S_ATIME)
5392                 inode->i_atime = *now;
5393         return btrfs_dirty_inode(inode);
5394 }
5395
5396 /*
5397  * find the highest existing sequence number in a directory
5398  * and then set the in-memory index_cnt variable to reflect
5399  * free sequence numbers
5400  */
5401 static int btrfs_set_inode_index_count(struct inode *inode)
5402 {
5403         struct btrfs_root *root = BTRFS_I(inode)->root;
5404         struct btrfs_key key, found_key;
5405         struct btrfs_path *path;
5406         struct extent_buffer *leaf;
5407         int ret;
5408
5409         key.objectid = btrfs_ino(inode);
5410         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5411         key.offset = (u64)-1;
5412
5413         path = btrfs_alloc_path();
5414         if (!path)
5415                 return -ENOMEM;
5416
5417         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5418         if (ret < 0)
5419                 goto out;
5420         /* FIXME: we should be able to handle this */
5421         if (ret == 0)
5422                 goto out;
5423         ret = 0;
5424
5425         /*
5426          * MAGIC NUMBER EXPLANATION:
5427          * since we search a directory based on f_pos we have to start at 2
5428          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5429          * else has to start at 2
5430          */
5431         if (path->slots[0] == 0) {
5432                 BTRFS_I(inode)->index_cnt = 2;
5433                 goto out;
5434         }
5435
5436         path->slots[0]--;
5437
5438         leaf = path->nodes[0];
5439         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5440
5441         if (found_key.objectid != btrfs_ino(inode) ||
5442             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5443                 BTRFS_I(inode)->index_cnt = 2;
5444                 goto out;
5445         }
5446
5447         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5448 out:
5449         btrfs_free_path(path);
5450         return ret;
5451 }
5452
5453 /*
5454  * helper to find a free sequence number in a given directory.  This current
5455  * code is very simple, later versions will do smarter things in the btree
5456  */
5457 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5458 {
5459         int ret = 0;
5460
5461         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5462                 ret = btrfs_inode_delayed_dir_index_count(dir);
5463                 if (ret) {
5464                         ret = btrfs_set_inode_index_count(dir);
5465                         if (ret)
5466                                 return ret;
5467                 }
5468         }
5469
5470         *index = BTRFS_I(dir)->index_cnt;
5471         BTRFS_I(dir)->index_cnt++;
5472
5473         return ret;
5474 }
5475
5476 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5477                                      struct btrfs_root *root,
5478                                      struct inode *dir,
5479                                      const char *name, int name_len,
5480                                      u64 ref_objectid, u64 objectid,
5481                                      umode_t mode, u64 *index)
5482 {
5483         struct inode *inode;
5484         struct btrfs_inode_item *inode_item;
5485         struct btrfs_key *location;
5486         struct btrfs_path *path;
5487         struct btrfs_inode_ref *ref;
5488         struct btrfs_key key[2];
5489         u32 sizes[2];
5490         unsigned long ptr;
5491         int ret;
5492         int owner;
5493
5494         path = btrfs_alloc_path();
5495         if (!path)
5496                 return ERR_PTR(-ENOMEM);
5497
5498         inode = new_inode(root->fs_info->sb);
5499         if (!inode) {
5500                 btrfs_free_path(path);
5501                 return ERR_PTR(-ENOMEM);
5502         }
5503
5504         /*
5505          * we have to initialize this early, so we can reclaim the inode
5506          * number if we fail afterwards in this function.
5507          */
5508         inode->i_ino = objectid;
5509
5510         if (dir) {
5511                 trace_btrfs_inode_request(dir);
5512
5513                 ret = btrfs_set_inode_index(dir, index);
5514                 if (ret) {
5515                         btrfs_free_path(path);
5516                         iput(inode);
5517                         return ERR_PTR(ret);
5518                 }
5519         }
5520         /*
5521          * index_cnt is ignored for everything but a dir,
5522          * btrfs_get_inode_index_count has an explanation for the magic
5523          * number
5524          */
5525         BTRFS_I(inode)->index_cnt = 2;
5526         BTRFS_I(inode)->root = root;
5527         BTRFS_I(inode)->generation = trans->transid;
5528         inode->i_generation = BTRFS_I(inode)->generation;
5529
5530         /*
5531          * We could have gotten an inode number from somebody who was fsynced
5532          * and then removed in this same transaction, so let's just set full
5533          * sync since it will be a full sync anyway and this will blow away the
5534          * old info in the log.
5535          */
5536         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5537
5538         if (S_ISDIR(mode))
5539                 owner = 0;
5540         else
5541                 owner = 1;
5542
5543         key[0].objectid = objectid;
5544         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5545         key[0].offset = 0;
5546
5547         /*
5548          * Start new inodes with an inode_ref. This is slightly more
5549          * efficient for small numbers of hard links since they will
5550          * be packed into one item. Extended refs will kick in if we
5551          * add more hard links than can fit in the ref item.
5552          */
5553         key[1].objectid = objectid;
5554         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5555         key[1].offset = ref_objectid;
5556
5557         sizes[0] = sizeof(struct btrfs_inode_item);
5558         sizes[1] = name_len + sizeof(*ref);
5559
5560         path->leave_spinning = 1;
5561         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5562         if (ret != 0)
5563                 goto fail;
5564
5565         inode_init_owner(inode, dir, mode);
5566         inode_set_bytes(inode, 0);
5567         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5568         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5569                                   struct btrfs_inode_item);
5570         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5571                              sizeof(*inode_item));
5572         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5573
5574         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5575                              struct btrfs_inode_ref);
5576         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5577         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5578         ptr = (unsigned long)(ref + 1);
5579         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5580
5581         btrfs_mark_buffer_dirty(path->nodes[0]);
5582         btrfs_free_path(path);
5583
5584         location = &BTRFS_I(inode)->location;
5585         location->objectid = objectid;
5586         location->offset = 0;
5587         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5588
5589         btrfs_inherit_iflags(inode, dir);
5590
5591         if (S_ISREG(mode)) {
5592                 if (btrfs_test_opt(root, NODATASUM))
5593                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5594                 if (btrfs_test_opt(root, NODATACOW))
5595                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5596                                 BTRFS_INODE_NODATASUM;
5597         }
5598
5599         insert_inode_hash(inode);
5600         inode_tree_add(inode);
5601
5602         trace_btrfs_inode_new(inode);
5603         btrfs_set_inode_last_trans(trans, inode);
5604
5605         btrfs_update_root_times(trans, root);
5606
5607         return inode;
5608 fail:
5609         if (dir)
5610                 BTRFS_I(dir)->index_cnt--;
5611         btrfs_free_path(path);
5612         iput(inode);
5613         return ERR_PTR(ret);
5614 }
5615
5616 static inline u8 btrfs_inode_type(struct inode *inode)
5617 {
5618         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5619 }
5620
5621 /*
5622  * utility function to add 'inode' into 'parent_inode' with
5623  * a give name and a given sequence number.
5624  * if 'add_backref' is true, also insert a backref from the
5625  * inode to the parent directory.
5626  */
5627 int btrfs_add_link(struct btrfs_trans_handle *trans,
5628                    struct inode *parent_inode, struct inode *inode,
5629                    const char *name, int name_len, int add_backref, u64 index)
5630 {
5631         int ret = 0;
5632         struct btrfs_key key;
5633         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5634         u64 ino = btrfs_ino(inode);
5635         u64 parent_ino = btrfs_ino(parent_inode);
5636
5637         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5638                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5639         } else {
5640                 key.objectid = ino;
5641                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5642                 key.offset = 0;
5643         }
5644
5645         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5646                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5647                                          key.objectid, root->root_key.objectid,
5648                                          parent_ino, index, name, name_len);
5649         } else if (add_backref) {
5650                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5651                                              parent_ino, index);
5652         }
5653
5654         /* Nothing to clean up yet */
5655         if (ret)
5656                 return ret;
5657
5658         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5659                                     parent_inode, &key,
5660                                     btrfs_inode_type(inode), index);
5661         if (ret == -EEXIST || ret == -EOVERFLOW)
5662                 goto fail_dir_item;
5663         else if (ret) {
5664                 btrfs_abort_transaction(trans, root, ret);
5665                 return ret;
5666         }
5667
5668         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5669                            name_len * 2);
5670         inode_inc_iversion(parent_inode);
5671         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5672         ret = btrfs_update_inode(trans, root, parent_inode);
5673         if (ret)
5674                 btrfs_abort_transaction(trans, root, ret);
5675         return ret;
5676
5677 fail_dir_item:
5678         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5679                 u64 local_index;
5680                 int err;
5681                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5682                                  key.objectid, root->root_key.objectid,
5683                                  parent_ino, &local_index, name, name_len);
5684
5685         } else if (add_backref) {
5686                 u64 local_index;
5687                 int err;
5688
5689                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5690                                           ino, parent_ino, &local_index);
5691         }
5692         return ret;
5693 }
5694
5695 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5696                             struct inode *dir, struct dentry *dentry,
5697                             struct inode *inode, int backref, u64 index)
5698 {
5699         int err = btrfs_add_link(trans, dir, inode,
5700                                  dentry->d_name.name, dentry->d_name.len,
5701                                  backref, index);
5702         if (err > 0)
5703                 err = -EEXIST;
5704         return err;
5705 }
5706
5707 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5708                         umode_t mode, dev_t rdev)
5709 {
5710         struct btrfs_trans_handle *trans;
5711         struct btrfs_root *root = BTRFS_I(dir)->root;
5712         struct inode *inode = NULL;
5713         int err;
5714         int drop_inode = 0;
5715         u64 objectid;
5716         u64 index = 0;
5717
5718         if (!new_valid_dev(rdev))
5719                 return -EINVAL;
5720
5721         /*
5722          * 2 for inode item and ref
5723          * 2 for dir items
5724          * 1 for xattr if selinux is on
5725          */
5726         trans = btrfs_start_transaction(root, 5);
5727         if (IS_ERR(trans))
5728                 return PTR_ERR(trans);
5729
5730         err = btrfs_find_free_ino(root, &objectid);
5731         if (err)
5732                 goto out_unlock;
5733
5734         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5735                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5736                                 mode, &index);
5737         if (IS_ERR(inode)) {
5738                 err = PTR_ERR(inode);
5739                 goto out_unlock;
5740         }
5741
5742         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5743         if (err) {
5744                 drop_inode = 1;
5745                 goto out_unlock;
5746         }
5747
5748         /*
5749         * If the active LSM wants to access the inode during
5750         * d_instantiate it needs these. Smack checks to see
5751         * if the filesystem supports xattrs by looking at the
5752         * ops vector.
5753         */
5754
5755         inode->i_op = &btrfs_special_inode_operations;
5756         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5757         if (err)
5758                 drop_inode = 1;
5759         else {
5760                 init_special_inode(inode, inode->i_mode, rdev);
5761                 btrfs_update_inode(trans, root, inode);
5762                 d_instantiate(dentry, inode);
5763         }
5764 out_unlock:
5765         btrfs_end_transaction(trans, root);
5766         btrfs_btree_balance_dirty(root);
5767         if (drop_inode) {
5768                 inode_dec_link_count(inode);
5769                 iput(inode);
5770         }
5771         return err;
5772 }
5773
5774 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5775                         umode_t mode, bool excl)
5776 {
5777         struct btrfs_trans_handle *trans;
5778         struct btrfs_root *root = BTRFS_I(dir)->root;
5779         struct inode *inode = NULL;
5780         int drop_inode_on_err = 0;
5781         int err;
5782         u64 objectid;
5783         u64 index = 0;
5784
5785         /*
5786          * 2 for inode item and ref
5787          * 2 for dir items
5788          * 1 for xattr if selinux is on
5789          */
5790         trans = btrfs_start_transaction(root, 5);
5791         if (IS_ERR(trans))
5792                 return PTR_ERR(trans);
5793
5794         err = btrfs_find_free_ino(root, &objectid);
5795         if (err)
5796                 goto out_unlock;
5797
5798         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5799                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5800                                 mode, &index);
5801         if (IS_ERR(inode)) {
5802                 err = PTR_ERR(inode);
5803                 goto out_unlock;
5804         }
5805         drop_inode_on_err = 1;
5806
5807         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5808         if (err)
5809                 goto out_unlock;
5810
5811         err = btrfs_update_inode(trans, root, inode);
5812         if (err)
5813                 goto out_unlock;
5814
5815         /*
5816         * If the active LSM wants to access the inode during
5817         * d_instantiate it needs these. Smack checks to see
5818         * if the filesystem supports xattrs by looking at the
5819         * ops vector.
5820         */
5821         inode->i_fop = &btrfs_file_operations;
5822         inode->i_op = &btrfs_file_inode_operations;
5823
5824         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5825         if (err)
5826                 goto out_unlock;
5827
5828         inode->i_mapping->a_ops = &btrfs_aops;
5829         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5830         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5831         d_instantiate(dentry, inode);
5832
5833 out_unlock:
5834         btrfs_end_transaction(trans, root);
5835         if (err && drop_inode_on_err) {
5836                 inode_dec_link_count(inode);
5837                 iput(inode);
5838         }
5839         btrfs_btree_balance_dirty(root);
5840         return err;
5841 }
5842
5843 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5844                       struct dentry *dentry)
5845 {
5846         struct btrfs_trans_handle *trans;
5847         struct btrfs_root *root = BTRFS_I(dir)->root;
5848         struct inode *inode = old_dentry->d_inode;
5849         u64 index;
5850         int err;
5851         int drop_inode = 0;
5852
5853         /* do not allow sys_link's with other subvols of the same device */
5854         if (root->objectid != BTRFS_I(inode)->root->objectid)
5855                 return -EXDEV;
5856
5857         if (inode->i_nlink >= BTRFS_LINK_MAX)
5858                 return -EMLINK;
5859
5860         err = btrfs_set_inode_index(dir, &index);
5861         if (err)
5862                 goto fail;
5863
5864         /*
5865          * 2 items for inode and inode ref
5866          * 2 items for dir items
5867          * 1 item for parent inode
5868          */
5869         trans = btrfs_start_transaction(root, 5);
5870         if (IS_ERR(trans)) {
5871                 err = PTR_ERR(trans);
5872                 goto fail;
5873         }
5874
5875         btrfs_inc_nlink(inode);
5876         inode_inc_iversion(inode);
5877         inode->i_ctime = CURRENT_TIME;
5878         ihold(inode);
5879         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5880
5881         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5882
5883         if (err) {
5884                 drop_inode = 1;
5885         } else {
5886                 struct dentry *parent = dentry->d_parent;
5887                 err = btrfs_update_inode(trans, root, inode);
5888                 if (err)
5889                         goto fail;
5890                 d_instantiate(dentry, inode);
5891                 btrfs_log_new_name(trans, inode, NULL, parent);
5892         }
5893
5894         btrfs_end_transaction(trans, root);
5895 fail:
5896         if (drop_inode) {
5897                 inode_dec_link_count(inode);
5898                 iput(inode);
5899         }
5900         btrfs_btree_balance_dirty(root);
5901         return err;
5902 }
5903
5904 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5905 {
5906         struct inode *inode = NULL;
5907         struct btrfs_trans_handle *trans;
5908         struct btrfs_root *root = BTRFS_I(dir)->root;
5909         int err = 0;
5910         int drop_on_err = 0;
5911         u64 objectid = 0;
5912         u64 index = 0;
5913
5914         /*
5915          * 2 items for inode and ref
5916          * 2 items for dir items
5917          * 1 for xattr if selinux is on
5918          */
5919         trans = btrfs_start_transaction(root, 5);
5920         if (IS_ERR(trans))
5921                 return PTR_ERR(trans);
5922
5923         err = btrfs_find_free_ino(root, &objectid);
5924         if (err)
5925                 goto out_fail;
5926
5927         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5928                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5929                                 S_IFDIR | mode, &index);
5930         if (IS_ERR(inode)) {
5931                 err = PTR_ERR(inode);
5932                 goto out_fail;
5933         }
5934
5935         drop_on_err = 1;
5936
5937         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5938         if (err)
5939                 goto out_fail;
5940
5941         inode->i_op = &btrfs_dir_inode_operations;
5942         inode->i_fop = &btrfs_dir_file_operations;
5943
5944         btrfs_i_size_write(inode, 0);
5945         err = btrfs_update_inode(trans, root, inode);
5946         if (err)
5947                 goto out_fail;
5948
5949         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5950                              dentry->d_name.len, 0, index);
5951         if (err)
5952                 goto out_fail;
5953
5954         d_instantiate(dentry, inode);
5955         drop_on_err = 0;
5956
5957 out_fail:
5958         btrfs_end_transaction(trans, root);
5959         if (drop_on_err)
5960                 iput(inode);
5961         btrfs_btree_balance_dirty(root);
5962         return err;
5963 }
5964
5965 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5966  * and an extent that you want to insert, deal with overlap and insert
5967  * the new extent into the tree.
5968  */
5969 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5970                                 struct extent_map *existing,
5971                                 struct extent_map *em,
5972                                 u64 map_start, u64 map_len)
5973 {
5974         u64 start_diff;
5975
5976         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5977         start_diff = map_start - em->start;
5978         em->start = map_start;
5979         em->len = map_len;
5980         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5981             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5982                 em->block_start += start_diff;
5983                 em->block_len -= start_diff;
5984         }
5985         return add_extent_mapping(em_tree, em, 0);
5986 }
5987
5988 static noinline int uncompress_inline(struct btrfs_path *path,
5989                                       struct inode *inode, struct page *page,
5990                                       size_t pg_offset, u64 extent_offset,
5991                                       struct btrfs_file_extent_item *item)
5992 {
5993         int ret;
5994         struct extent_buffer *leaf = path->nodes[0];
5995         char *tmp;
5996         size_t max_size;
5997         unsigned long inline_size;
5998         unsigned long ptr;
5999         int compress_type;
6000
6001         WARN_ON(pg_offset != 0);
6002         compress_type = btrfs_file_extent_compression(leaf, item);
6003         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6004         inline_size = btrfs_file_extent_inline_item_len(leaf,
6005                                         btrfs_item_nr(leaf, path->slots[0]));
6006         tmp = kmalloc(inline_size, GFP_NOFS);
6007         if (!tmp)
6008                 return -ENOMEM;
6009         ptr = btrfs_file_extent_inline_start(item);
6010
6011         read_extent_buffer(leaf, tmp, ptr, inline_size);
6012
6013         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6014         ret = btrfs_decompress(compress_type, tmp, page,
6015                                extent_offset, inline_size, max_size);
6016         if (ret) {
6017                 char *kaddr = kmap_atomic(page);
6018                 unsigned long copy_size = min_t(u64,
6019                                   PAGE_CACHE_SIZE - pg_offset,
6020                                   max_size - extent_offset);
6021                 memset(kaddr + pg_offset, 0, copy_size);
6022                 kunmap_atomic(kaddr);
6023         }
6024         kfree(tmp);
6025         return 0;
6026 }
6027
6028 /*
6029  * a bit scary, this does extent mapping from logical file offset to the disk.
6030  * the ugly parts come from merging extents from the disk with the in-ram
6031  * representation.  This gets more complex because of the data=ordered code,
6032  * where the in-ram extents might be locked pending data=ordered completion.
6033  *
6034  * This also copies inline extents directly into the page.
6035  */
6036
6037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6038                                     size_t pg_offset, u64 start, u64 len,
6039                                     int create)
6040 {
6041         int ret;
6042         int err = 0;
6043         u64 bytenr;
6044         u64 extent_start = 0;
6045         u64 extent_end = 0;
6046         u64 objectid = btrfs_ino(inode);
6047         u32 found_type;
6048         struct btrfs_path *path = NULL;
6049         struct btrfs_root *root = BTRFS_I(inode)->root;
6050         struct btrfs_file_extent_item *item;
6051         struct extent_buffer *leaf;
6052         struct btrfs_key found_key;
6053         struct extent_map *em = NULL;
6054         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6055         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6056         struct btrfs_trans_handle *trans = NULL;
6057         int compress_type;
6058
6059 again:
6060         read_lock(&em_tree->lock);
6061         em = lookup_extent_mapping(em_tree, start, len);
6062         if (em)
6063                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6064         read_unlock(&em_tree->lock);
6065
6066         if (em) {
6067                 if (em->start > start || em->start + em->len <= start)
6068                         free_extent_map(em);
6069                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6070                         free_extent_map(em);
6071                 else
6072                         goto out;
6073         }
6074         em = alloc_extent_map();
6075         if (!em) {
6076                 err = -ENOMEM;
6077                 goto out;
6078         }
6079         em->bdev = root->fs_info->fs_devices->latest_bdev;
6080         em->start = EXTENT_MAP_HOLE;
6081         em->orig_start = EXTENT_MAP_HOLE;
6082         em->len = (u64)-1;
6083         em->block_len = (u64)-1;
6084
6085         if (!path) {
6086                 path = btrfs_alloc_path();
6087                 if (!path) {
6088                         err = -ENOMEM;
6089                         goto out;
6090                 }
6091                 /*
6092                  * Chances are we'll be called again, so go ahead and do
6093                  * readahead
6094                  */
6095                 path->reada = 1;
6096         }
6097
6098         ret = btrfs_lookup_file_extent(trans, root, path,
6099                                        objectid, start, trans != NULL);
6100         if (ret < 0) {
6101                 err = ret;
6102                 goto out;
6103         }
6104
6105         if (ret != 0) {
6106                 if (path->slots[0] == 0)
6107                         goto not_found;
6108                 path->slots[0]--;
6109         }
6110
6111         leaf = path->nodes[0];
6112         item = btrfs_item_ptr(leaf, path->slots[0],
6113                               struct btrfs_file_extent_item);
6114         /* are we inside the extent that was found? */
6115         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6116         found_type = btrfs_key_type(&found_key);
6117         if (found_key.objectid != objectid ||
6118             found_type != BTRFS_EXTENT_DATA_KEY) {
6119                 goto not_found;
6120         }
6121
6122         found_type = btrfs_file_extent_type(leaf, item);
6123         extent_start = found_key.offset;
6124         compress_type = btrfs_file_extent_compression(leaf, item);
6125         if (found_type == BTRFS_FILE_EXTENT_REG ||
6126             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6127                 extent_end = extent_start +
6128                        btrfs_file_extent_num_bytes(leaf, item);
6129         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6130                 size_t size;
6131                 size = btrfs_file_extent_inline_len(leaf, item);
6132                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6133         }
6134
6135         if (start >= extent_end) {
6136                 path->slots[0]++;
6137                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6138                         ret = btrfs_next_leaf(root, path);
6139                         if (ret < 0) {
6140                                 err = ret;
6141                                 goto out;
6142                         }
6143                         if (ret > 0)
6144                                 goto not_found;
6145                         leaf = path->nodes[0];
6146                 }
6147                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6148                 if (found_key.objectid != objectid ||
6149                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6150                         goto not_found;
6151                 if (start + len <= found_key.offset)
6152                         goto not_found;
6153                 em->start = start;
6154                 em->orig_start = start;
6155                 em->len = found_key.offset - start;
6156                 goto not_found_em;
6157         }
6158
6159         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6160         if (found_type == BTRFS_FILE_EXTENT_REG ||
6161             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6162                 em->start = extent_start;
6163                 em->len = extent_end - extent_start;
6164                 em->orig_start = extent_start -
6165                                  btrfs_file_extent_offset(leaf, item);
6166                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6167                                                                       item);
6168                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6169                 if (bytenr == 0) {
6170                         em->block_start = EXTENT_MAP_HOLE;
6171                         goto insert;
6172                 }
6173                 if (compress_type != BTRFS_COMPRESS_NONE) {
6174                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6175                         em->compress_type = compress_type;
6176                         em->block_start = bytenr;
6177                         em->block_len = em->orig_block_len;
6178                 } else {
6179                         bytenr += btrfs_file_extent_offset(leaf, item);
6180                         em->block_start = bytenr;
6181                         em->block_len = em->len;
6182                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6183                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6184                 }
6185                 goto insert;
6186         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6187                 unsigned long ptr;
6188                 char *map;
6189                 size_t size;
6190                 size_t extent_offset;
6191                 size_t copy_size;
6192
6193                 em->block_start = EXTENT_MAP_INLINE;
6194                 if (!page || create) {
6195                         em->start = extent_start;
6196                         em->len = extent_end - extent_start;
6197                         goto out;
6198                 }
6199
6200                 size = btrfs_file_extent_inline_len(leaf, item);
6201                 extent_offset = page_offset(page) + pg_offset - extent_start;
6202                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6203                                 size - extent_offset);
6204                 em->start = extent_start + extent_offset;
6205                 em->len = ALIGN(copy_size, root->sectorsize);
6206                 em->orig_block_len = em->len;
6207                 em->orig_start = em->start;
6208                 if (compress_type) {
6209                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6210                         em->compress_type = compress_type;
6211                 }
6212                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6213                 if (create == 0 && !PageUptodate(page)) {
6214                         if (btrfs_file_extent_compression(leaf, item) !=
6215                             BTRFS_COMPRESS_NONE) {
6216                                 ret = uncompress_inline(path, inode, page,
6217                                                         pg_offset,
6218                                                         extent_offset, item);
6219                                 BUG_ON(ret); /* -ENOMEM */
6220                         } else {
6221                                 map = kmap(page);
6222                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6223                                                    copy_size);
6224                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6225                                         memset(map + pg_offset + copy_size, 0,
6226                                                PAGE_CACHE_SIZE - pg_offset -
6227                                                copy_size);
6228                                 }
6229                                 kunmap(page);
6230                         }
6231                         flush_dcache_page(page);
6232                 } else if (create && PageUptodate(page)) {
6233                         BUG();
6234                         if (!trans) {
6235                                 kunmap(page);
6236                                 free_extent_map(em);
6237                                 em = NULL;
6238
6239                                 btrfs_release_path(path);
6240                                 trans = btrfs_join_transaction(root);
6241
6242                                 if (IS_ERR(trans))
6243                                         return ERR_CAST(trans);
6244                                 goto again;
6245                         }
6246                         map = kmap(page);
6247                         write_extent_buffer(leaf, map + pg_offset, ptr,
6248                                             copy_size);
6249                         kunmap(page);
6250                         btrfs_mark_buffer_dirty(leaf);
6251                 }
6252                 set_extent_uptodate(io_tree, em->start,
6253                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6254                 goto insert;
6255         } else {
6256                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6257         }
6258 not_found:
6259         em->start = start;
6260         em->orig_start = start;
6261         em->len = len;
6262 not_found_em:
6263         em->block_start = EXTENT_MAP_HOLE;
6264         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6265 insert:
6266         btrfs_release_path(path);
6267         if (em->start > start || extent_map_end(em) <= start) {
6268                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6269                         (unsigned long long)em->start,
6270                         (unsigned long long)em->len,
6271                         (unsigned long long)start,
6272                         (unsigned long long)len);
6273                 err = -EIO;
6274                 goto out;
6275         }
6276
6277         err = 0;
6278         write_lock(&em_tree->lock);
6279         ret = add_extent_mapping(em_tree, em, 0);
6280         /* it is possible that someone inserted the extent into the tree
6281          * while we had the lock dropped.  It is also possible that
6282          * an overlapping map exists in the tree
6283          */
6284         if (ret == -EEXIST) {
6285                 struct extent_map *existing;
6286
6287                 ret = 0;
6288
6289                 existing = lookup_extent_mapping(em_tree, start, len);
6290                 if (existing && (existing->start > start ||
6291                     existing->start + existing->len <= start)) {
6292                         free_extent_map(existing);
6293                         existing = NULL;
6294                 }
6295                 if (!existing) {
6296                         existing = lookup_extent_mapping(em_tree, em->start,
6297                                                          em->len);
6298                         if (existing) {
6299                                 err = merge_extent_mapping(em_tree, existing,
6300                                                            em, start,
6301                                                            root->sectorsize);
6302                                 free_extent_map(existing);
6303                                 if (err) {
6304                                         free_extent_map(em);
6305                                         em = NULL;
6306                                 }
6307                         } else {
6308                                 err = -EIO;
6309                                 free_extent_map(em);
6310                                 em = NULL;
6311                         }
6312                 } else {
6313                         free_extent_map(em);
6314                         em = existing;
6315                         err = 0;
6316                 }
6317         }
6318         write_unlock(&em_tree->lock);
6319 out:
6320
6321         if (em)
6322                 trace_btrfs_get_extent(root, em);
6323
6324         if (path)
6325                 btrfs_free_path(path);
6326         if (trans) {
6327                 ret = btrfs_end_transaction(trans, root);
6328                 if (!err)
6329                         err = ret;
6330         }
6331         if (err) {
6332                 free_extent_map(em);
6333                 return ERR_PTR(err);
6334         }
6335         BUG_ON(!em); /* Error is always set */
6336         return em;
6337 }
6338
6339 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6340                                            size_t pg_offset, u64 start, u64 len,
6341                                            int create)
6342 {
6343         struct extent_map *em;
6344         struct extent_map *hole_em = NULL;
6345         u64 range_start = start;
6346         u64 end;
6347         u64 found;
6348         u64 found_end;
6349         int err = 0;
6350
6351         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6352         if (IS_ERR(em))
6353                 return em;
6354         if (em) {
6355                 /*
6356                  * if our em maps to
6357                  * -  a hole or
6358                  * -  a pre-alloc extent,
6359                  * there might actually be delalloc bytes behind it.
6360                  */
6361                 if (em->block_start != EXTENT_MAP_HOLE &&
6362                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6363                         return em;
6364                 else
6365                         hole_em = em;
6366         }
6367
6368         /* check to see if we've wrapped (len == -1 or similar) */
6369         end = start + len;
6370         if (end < start)
6371                 end = (u64)-1;
6372         else
6373                 end -= 1;
6374
6375         em = NULL;
6376
6377         /* ok, we didn't find anything, lets look for delalloc */
6378         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6379                                  end, len, EXTENT_DELALLOC, 1);
6380         found_end = range_start + found;
6381         if (found_end < range_start)
6382                 found_end = (u64)-1;
6383
6384         /*
6385          * we didn't find anything useful, return
6386          * the original results from get_extent()
6387          */
6388         if (range_start > end || found_end <= start) {
6389                 em = hole_em;
6390                 hole_em = NULL;
6391                 goto out;
6392         }
6393
6394         /* adjust the range_start to make sure it doesn't
6395          * go backwards from the start they passed in
6396          */
6397         range_start = max(start,range_start);
6398         found = found_end - range_start;
6399
6400         if (found > 0) {
6401                 u64 hole_start = start;
6402                 u64 hole_len = len;
6403
6404                 em = alloc_extent_map();
6405                 if (!em) {
6406                         err = -ENOMEM;
6407                         goto out;
6408                 }
6409                 /*
6410                  * when btrfs_get_extent can't find anything it
6411                  * returns one huge hole
6412                  *
6413                  * make sure what it found really fits our range, and
6414                  * adjust to make sure it is based on the start from
6415                  * the caller
6416                  */
6417                 if (hole_em) {
6418                         u64 calc_end = extent_map_end(hole_em);
6419
6420                         if (calc_end <= start || (hole_em->start > end)) {
6421                                 free_extent_map(hole_em);
6422                                 hole_em = NULL;
6423                         } else {
6424                                 hole_start = max(hole_em->start, start);
6425                                 hole_len = calc_end - hole_start;
6426                         }
6427                 }
6428                 em->bdev = NULL;
6429                 if (hole_em && range_start > hole_start) {
6430                         /* our hole starts before our delalloc, so we
6431                          * have to return just the parts of the hole
6432                          * that go until  the delalloc starts
6433                          */
6434                         em->len = min(hole_len,
6435                                       range_start - hole_start);
6436                         em->start = hole_start;
6437                         em->orig_start = hole_start;
6438                         /*
6439                          * don't adjust block start at all,
6440                          * it is fixed at EXTENT_MAP_HOLE
6441                          */
6442                         em->block_start = hole_em->block_start;
6443                         em->block_len = hole_len;
6444                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6445                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6446                 } else {
6447                         em->start = range_start;
6448                         em->len = found;
6449                         em->orig_start = range_start;
6450                         em->block_start = EXTENT_MAP_DELALLOC;
6451                         em->block_len = found;
6452                 }
6453         } else if (hole_em) {
6454                 return hole_em;
6455         }
6456 out:
6457
6458         free_extent_map(hole_em);
6459         if (err) {
6460                 free_extent_map(em);
6461                 return ERR_PTR(err);
6462         }
6463         return em;
6464 }
6465
6466 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6467                                                   u64 start, u64 len)
6468 {
6469         struct btrfs_root *root = BTRFS_I(inode)->root;
6470         struct btrfs_trans_handle *trans;
6471         struct extent_map *em;
6472         struct btrfs_key ins;
6473         u64 alloc_hint;
6474         int ret;
6475
6476         trans = btrfs_join_transaction(root);
6477         if (IS_ERR(trans))
6478                 return ERR_CAST(trans);
6479
6480         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6481
6482         alloc_hint = get_extent_allocation_hint(inode, start, len);
6483         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6484                                    alloc_hint, &ins, 1);
6485         if (ret) {
6486                 em = ERR_PTR(ret);
6487                 goto out;
6488         }
6489
6490         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6491                               ins.offset, ins.offset, ins.offset, 0);
6492         if (IS_ERR(em))
6493                 goto out;
6494
6495         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6496                                            ins.offset, ins.offset, 0);
6497         if (ret) {
6498                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6499                 em = ERR_PTR(ret);
6500         }
6501 out:
6502         btrfs_end_transaction(trans, root);
6503         return em;
6504 }
6505
6506 /*
6507  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6508  * block must be cow'd
6509  */
6510 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6511                                       struct inode *inode, u64 offset, u64 *len,
6512                                       u64 *orig_start, u64 *orig_block_len,
6513                                       u64 *ram_bytes)
6514 {
6515         struct btrfs_path *path;
6516         int ret;
6517         struct extent_buffer *leaf;
6518         struct btrfs_root *root = BTRFS_I(inode)->root;
6519         struct btrfs_file_extent_item *fi;
6520         struct btrfs_key key;
6521         u64 disk_bytenr;
6522         u64 backref_offset;
6523         u64 extent_end;
6524         u64 num_bytes;
6525         int slot;
6526         int found_type;
6527
6528         path = btrfs_alloc_path();
6529         if (!path)
6530                 return -ENOMEM;
6531
6532         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6533                                        offset, 0);
6534         if (ret < 0)
6535                 goto out;
6536
6537         slot = path->slots[0];
6538         if (ret == 1) {
6539                 if (slot == 0) {
6540                         /* can't find the item, must cow */
6541                         ret = 0;
6542                         goto out;
6543                 }
6544                 slot--;
6545         }
6546         ret = 0;
6547         leaf = path->nodes[0];
6548         btrfs_item_key_to_cpu(leaf, &key, slot);
6549         if (key.objectid != btrfs_ino(inode) ||
6550             key.type != BTRFS_EXTENT_DATA_KEY) {
6551                 /* not our file or wrong item type, must cow */
6552                 goto out;
6553         }
6554
6555         if (key.offset > offset) {
6556                 /* Wrong offset, must cow */
6557                 goto out;
6558         }
6559
6560         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6561         found_type = btrfs_file_extent_type(leaf, fi);
6562         if (found_type != BTRFS_FILE_EXTENT_REG &&
6563             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6564                 /* not a regular extent, must cow */
6565                 goto out;
6566         }
6567         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6568         backref_offset = btrfs_file_extent_offset(leaf, fi);
6569
6570         *orig_start = key.offset - backref_offset;
6571         *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6572         *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6573
6574         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6575         if (extent_end < offset + *len) {
6576                 /* extent doesn't include our full range, must cow */
6577                 goto out;
6578         }
6579
6580         if (btrfs_extent_readonly(root, disk_bytenr))
6581                 goto out;
6582
6583         /*
6584          * look for other files referencing this extent, if we
6585          * find any we must cow
6586          */
6587         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6588                                   key.offset - backref_offset, disk_bytenr))
6589                 goto out;
6590
6591         /*
6592          * adjust disk_bytenr and num_bytes to cover just the bytes
6593          * in this extent we are about to write.  If there
6594          * are any csums in that range we have to cow in order
6595          * to keep the csums correct
6596          */
6597         disk_bytenr += backref_offset;
6598         disk_bytenr += offset - key.offset;
6599         num_bytes = min(offset + *len, extent_end) - offset;
6600         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6601                                 goto out;
6602         /*
6603          * all of the above have passed, it is safe to overwrite this extent
6604          * without cow
6605          */
6606         *len = num_bytes;
6607         ret = 1;
6608 out:
6609         btrfs_free_path(path);
6610         return ret;
6611 }
6612
6613 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6614                               struct extent_state **cached_state, int writing)
6615 {
6616         struct btrfs_ordered_extent *ordered;
6617         int ret = 0;
6618
6619         while (1) {
6620                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6621                                  0, cached_state);
6622                 /*
6623                  * We're concerned with the entire range that we're going to be
6624                  * doing DIO to, so we need to make sure theres no ordered
6625                  * extents in this range.
6626                  */
6627                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6628                                                      lockend - lockstart + 1);
6629
6630                 /*
6631                  * We need to make sure there are no buffered pages in this
6632                  * range either, we could have raced between the invalidate in
6633                  * generic_file_direct_write and locking the extent.  The
6634                  * invalidate needs to happen so that reads after a write do not
6635                  * get stale data.
6636                  */
6637                 if (!ordered && (!writing ||
6638                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6639                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6640                                     *cached_state)))
6641                         break;
6642
6643                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6644                                      cached_state, GFP_NOFS);
6645
6646                 if (ordered) {
6647                         btrfs_start_ordered_extent(inode, ordered, 1);
6648                         btrfs_put_ordered_extent(ordered);
6649                 } else {
6650                         /* Screw you mmap */
6651                         ret = filemap_write_and_wait_range(inode->i_mapping,
6652                                                            lockstart,
6653                                                            lockend);
6654                         if (ret)
6655                                 break;
6656
6657                         /*
6658                          * If we found a page that couldn't be invalidated just
6659                          * fall back to buffered.
6660                          */
6661                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6662                                         lockstart >> PAGE_CACHE_SHIFT,
6663                                         lockend >> PAGE_CACHE_SHIFT);
6664                         if (ret)
6665                                 break;
6666                 }
6667
6668                 cond_resched();
6669         }
6670
6671         return ret;
6672 }
6673
6674 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6675                                            u64 len, u64 orig_start,
6676                                            u64 block_start, u64 block_len,
6677                                            u64 orig_block_len, u64 ram_bytes,
6678                                            int type)
6679 {
6680         struct extent_map_tree *em_tree;
6681         struct extent_map *em;
6682         struct btrfs_root *root = BTRFS_I(inode)->root;
6683         int ret;
6684
6685         em_tree = &BTRFS_I(inode)->extent_tree;
6686         em = alloc_extent_map();
6687         if (!em)
6688                 return ERR_PTR(-ENOMEM);
6689
6690         em->start = start;
6691         em->orig_start = orig_start;
6692         em->mod_start = start;
6693         em->mod_len = len;
6694         em->len = len;
6695         em->block_len = block_len;
6696         em->block_start = block_start;
6697         em->bdev = root->fs_info->fs_devices->latest_bdev;
6698         em->orig_block_len = orig_block_len;
6699         em->ram_bytes = ram_bytes;
6700         em->generation = -1;
6701         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6702         if (type == BTRFS_ORDERED_PREALLOC)
6703                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6704
6705         do {
6706                 btrfs_drop_extent_cache(inode, em->start,
6707                                 em->start + em->len - 1, 0);
6708                 write_lock(&em_tree->lock);
6709                 ret = add_extent_mapping(em_tree, em, 1);
6710                 write_unlock(&em_tree->lock);
6711         } while (ret == -EEXIST);
6712
6713         if (ret) {
6714                 free_extent_map(em);
6715                 return ERR_PTR(ret);
6716         }
6717
6718         return em;
6719 }
6720
6721
6722 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6723                                    struct buffer_head *bh_result, int create)
6724 {
6725         struct extent_map *em;
6726         struct btrfs_root *root = BTRFS_I(inode)->root;
6727         struct extent_state *cached_state = NULL;
6728         u64 start = iblock << inode->i_blkbits;
6729         u64 lockstart, lockend;
6730         u64 len = bh_result->b_size;
6731         struct btrfs_trans_handle *trans;
6732         int unlock_bits = EXTENT_LOCKED;
6733         int ret = 0;
6734
6735         if (create)
6736                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6737         else
6738                 len = min_t(u64, len, root->sectorsize);
6739
6740         lockstart = start;
6741         lockend = start + len - 1;
6742
6743         /*
6744          * If this errors out it's because we couldn't invalidate pagecache for
6745          * this range and we need to fallback to buffered.
6746          */
6747         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6748                 return -ENOTBLK;
6749
6750         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6751         if (IS_ERR(em)) {
6752                 ret = PTR_ERR(em);
6753                 goto unlock_err;
6754         }
6755
6756         /*
6757          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6758          * io.  INLINE is special, and we could probably kludge it in here, but
6759          * it's still buffered so for safety lets just fall back to the generic
6760          * buffered path.
6761          *
6762          * For COMPRESSED we _have_ to read the entire extent in so we can
6763          * decompress it, so there will be buffering required no matter what we
6764          * do, so go ahead and fallback to buffered.
6765          *
6766          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6767          * to buffered IO.  Don't blame me, this is the price we pay for using
6768          * the generic code.
6769          */
6770         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6771             em->block_start == EXTENT_MAP_INLINE) {
6772                 free_extent_map(em);
6773                 ret = -ENOTBLK;
6774                 goto unlock_err;
6775         }
6776
6777         /* Just a good old fashioned hole, return */
6778         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6779                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6780                 free_extent_map(em);
6781                 goto unlock_err;
6782         }
6783
6784         /*
6785          * We don't allocate a new extent in the following cases
6786          *
6787          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6788          * existing extent.
6789          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6790          * just use the extent.
6791          *
6792          */
6793         if (!create) {
6794                 len = min(len, em->len - (start - em->start));
6795                 lockstart = start + len;
6796                 goto unlock;
6797         }
6798
6799         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6800             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6801              em->block_start != EXTENT_MAP_HOLE)) {
6802                 int type;
6803                 int ret;
6804                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6805
6806                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6807                         type = BTRFS_ORDERED_PREALLOC;
6808                 else
6809                         type = BTRFS_ORDERED_NOCOW;
6810                 len = min(len, em->len - (start - em->start));
6811                 block_start = em->block_start + (start - em->start);
6812
6813                 /*
6814                  * we're not going to log anything, but we do need
6815                  * to make sure the current transaction stays open
6816                  * while we look for nocow cross refs
6817                  */
6818                 trans = btrfs_join_transaction(root);
6819                 if (IS_ERR(trans))
6820                         goto must_cow;
6821
6822                 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6823                                       &orig_block_len, &ram_bytes) == 1) {
6824                         if (type == BTRFS_ORDERED_PREALLOC) {
6825                                 free_extent_map(em);
6826                                 em = create_pinned_em(inode, start, len,
6827                                                        orig_start,
6828                                                        block_start, len,
6829                                                        orig_block_len,
6830                                                        ram_bytes, type);
6831                                 if (IS_ERR(em)) {
6832                                         btrfs_end_transaction(trans, root);
6833                                         goto unlock_err;
6834                                 }
6835                         }
6836
6837                         ret = btrfs_add_ordered_extent_dio(inode, start,
6838                                            block_start, len, len, type);
6839                         btrfs_end_transaction(trans, root);
6840                         if (ret) {
6841                                 free_extent_map(em);
6842                                 goto unlock_err;
6843                         }
6844                         goto unlock;
6845                 }
6846                 btrfs_end_transaction(trans, root);
6847         }
6848 must_cow:
6849         /*
6850          * this will cow the extent, reset the len in case we changed
6851          * it above
6852          */
6853         len = bh_result->b_size;
6854         free_extent_map(em);
6855         em = btrfs_new_extent_direct(inode, start, len);
6856         if (IS_ERR(em)) {
6857                 ret = PTR_ERR(em);
6858                 goto unlock_err;
6859         }
6860         len = min(len, em->len - (start - em->start));
6861 unlock:
6862         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6863                 inode->i_blkbits;
6864         bh_result->b_size = len;
6865         bh_result->b_bdev = em->bdev;
6866         set_buffer_mapped(bh_result);
6867         if (create) {
6868                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6869                         set_buffer_new(bh_result);
6870
6871                 /*
6872                  * Need to update the i_size under the extent lock so buffered
6873                  * readers will get the updated i_size when we unlock.
6874                  */
6875                 if (start + len > i_size_read(inode))
6876                         i_size_write(inode, start + len);
6877
6878                 spin_lock(&BTRFS_I(inode)->lock);
6879                 BTRFS_I(inode)->outstanding_extents++;
6880                 spin_unlock(&BTRFS_I(inode)->lock);
6881
6882                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6883                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6884                                      &cached_state, GFP_NOFS);
6885                 BUG_ON(ret);
6886         }
6887
6888         /*
6889          * In the case of write we need to clear and unlock the entire range,
6890          * in the case of read we need to unlock only the end area that we
6891          * aren't using if there is any left over space.
6892          */
6893         if (lockstart < lockend) {
6894                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6895                                  lockend, unlock_bits, 1, 0,
6896                                  &cached_state, GFP_NOFS);
6897         } else {
6898                 free_extent_state(cached_state);
6899         }
6900
6901         free_extent_map(em);
6902
6903         return 0;
6904
6905 unlock_err:
6906         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6907                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6908         return ret;
6909 }
6910
6911 struct btrfs_dio_private {
6912         struct inode *inode;
6913         u64 logical_offset;
6914         u64 disk_bytenr;
6915         u64 bytes;
6916         void *private;
6917
6918         /* number of bios pending for this dio */
6919         atomic_t pending_bios;
6920
6921         /* IO errors */
6922         int errors;
6923
6924         /* orig_bio is our btrfs_io_bio */
6925         struct bio *orig_bio;
6926
6927         /* dio_bio came from fs/direct-io.c */
6928         struct bio *dio_bio;
6929 };
6930
6931 static void btrfs_endio_direct_read(struct bio *bio, int err)
6932 {
6933         struct btrfs_dio_private *dip = bio->bi_private;
6934         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6935         struct bio_vec *bvec = bio->bi_io_vec;
6936         struct inode *inode = dip->inode;
6937         struct btrfs_root *root = BTRFS_I(inode)->root;
6938         struct bio *dio_bio;
6939         u64 start;
6940
6941         start = dip->logical_offset;
6942         do {
6943                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6944                         struct page *page = bvec->bv_page;
6945                         char *kaddr;
6946                         u32 csum = ~(u32)0;
6947                         u64 private = ~(u32)0;
6948                         unsigned long flags;
6949
6950                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6951                                               start, &private))
6952                                 goto failed;
6953                         local_irq_save(flags);
6954                         kaddr = kmap_atomic(page);
6955                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6956                                                csum, bvec->bv_len);
6957                         btrfs_csum_final(csum, (char *)&csum);
6958                         kunmap_atomic(kaddr);
6959                         local_irq_restore(flags);
6960
6961                         flush_dcache_page(bvec->bv_page);
6962                         if (csum != private) {
6963 failed:
6964                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6965                                         (unsigned long long)btrfs_ino(inode),
6966                                         (unsigned long long)start,
6967                                         csum, (unsigned)private);
6968                                 err = -EIO;
6969                         }
6970                 }
6971
6972                 start += bvec->bv_len;
6973                 bvec++;
6974         } while (bvec <= bvec_end);
6975
6976         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6977                       dip->logical_offset + dip->bytes - 1);
6978         dio_bio = dip->dio_bio;
6979
6980         kfree(dip);
6981
6982         /* If we had a csum failure make sure to clear the uptodate flag */
6983         if (err)
6984                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6985         dio_end_io(dio_bio, err);
6986         bio_put(bio);
6987 }
6988
6989 static void btrfs_endio_direct_write(struct bio *bio, int err)
6990 {
6991         struct btrfs_dio_private *dip = bio->bi_private;
6992         struct inode *inode = dip->inode;
6993         struct btrfs_root *root = BTRFS_I(inode)->root;
6994         struct btrfs_ordered_extent *ordered = NULL;
6995         u64 ordered_offset = dip->logical_offset;
6996         u64 ordered_bytes = dip->bytes;
6997         struct bio *dio_bio;
6998         int ret;
6999
7000         if (err)
7001                 goto out_done;
7002 again:
7003         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7004                                                    &ordered_offset,
7005                                                    ordered_bytes, !err);
7006         if (!ret)
7007                 goto out_test;
7008
7009         ordered->work.func = finish_ordered_fn;
7010         ordered->work.flags = 0;
7011         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7012                            &ordered->work);
7013 out_test:
7014         /*
7015          * our bio might span multiple ordered extents.  If we haven't
7016          * completed the accounting for the whole dio, go back and try again
7017          */
7018         if (ordered_offset < dip->logical_offset + dip->bytes) {
7019                 ordered_bytes = dip->logical_offset + dip->bytes -
7020                         ordered_offset;
7021                 ordered = NULL;
7022                 goto again;
7023         }
7024 out_done:
7025         dio_bio = dip->dio_bio;
7026
7027         kfree(dip);
7028
7029         /* If we had an error make sure to clear the uptodate flag */
7030         if (err)
7031                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7032         dio_end_io(dio_bio, err);
7033         bio_put(bio);
7034 }
7035
7036 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7037                                     struct bio *bio, int mirror_num,
7038                                     unsigned long bio_flags, u64 offset)
7039 {
7040         int ret;
7041         struct btrfs_root *root = BTRFS_I(inode)->root;
7042         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7043         BUG_ON(ret); /* -ENOMEM */
7044         return 0;
7045 }
7046
7047 static void btrfs_end_dio_bio(struct bio *bio, int err)
7048 {
7049         struct btrfs_dio_private *dip = bio->bi_private;
7050
7051         if (err) {
7052                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7053                       "sector %#Lx len %u err no %d\n",
7054                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7055                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7056                 dip->errors = 1;
7057
7058                 /*
7059                  * before atomic variable goto zero, we must make sure
7060                  * dip->errors is perceived to be set.
7061                  */
7062                 smp_mb__before_atomic_dec();
7063         }
7064
7065         /* if there are more bios still pending for this dio, just exit */
7066         if (!atomic_dec_and_test(&dip->pending_bios))
7067                 goto out;
7068
7069         if (dip->errors) {
7070                 bio_io_error(dip->orig_bio);
7071         } else {
7072                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7073                 bio_endio(dip->orig_bio, 0);
7074         }
7075 out:
7076         bio_put(bio);
7077 }
7078
7079 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7080                                        u64 first_sector, gfp_t gfp_flags)
7081 {
7082         int nr_vecs = bio_get_nr_vecs(bdev);
7083         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7084 }
7085
7086 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7087                                          int rw, u64 file_offset, int skip_sum,
7088                                          int async_submit)
7089 {
7090         int write = rw & REQ_WRITE;
7091         struct btrfs_root *root = BTRFS_I(inode)->root;
7092         int ret;
7093
7094         if (async_submit)
7095                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7096
7097         bio_get(bio);
7098
7099         if (!write) {
7100                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7101                 if (ret)
7102                         goto err;
7103         }
7104
7105         if (skip_sum)
7106                 goto map;
7107
7108         if (write && async_submit) {
7109                 ret = btrfs_wq_submit_bio(root->fs_info,
7110                                    inode, rw, bio, 0, 0,
7111                                    file_offset,
7112                                    __btrfs_submit_bio_start_direct_io,
7113                                    __btrfs_submit_bio_done);
7114                 goto err;
7115         } else if (write) {
7116                 /*
7117                  * If we aren't doing async submit, calculate the csum of the
7118                  * bio now.
7119                  */
7120                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7121                 if (ret)
7122                         goto err;
7123         } else if (!skip_sum) {
7124                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7125                 if (ret)
7126                         goto err;
7127         }
7128
7129 map:
7130         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7131 err:
7132         bio_put(bio);
7133         return ret;
7134 }
7135
7136 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7137                                     int skip_sum)
7138 {
7139         struct inode *inode = dip->inode;
7140         struct btrfs_root *root = BTRFS_I(inode)->root;
7141         struct bio *bio;
7142         struct bio *orig_bio = dip->orig_bio;
7143         struct bio_vec *bvec = orig_bio->bi_io_vec;
7144         u64 start_sector = orig_bio->bi_sector;
7145         u64 file_offset = dip->logical_offset;
7146         u64 submit_len = 0;
7147         u64 map_length;
7148         int nr_pages = 0;
7149         int ret = 0;
7150         int async_submit = 0;
7151
7152         map_length = orig_bio->bi_size;
7153         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7154                               &map_length, NULL, 0);
7155         if (ret) {
7156                 bio_put(orig_bio);
7157                 return -EIO;
7158         }
7159         if (map_length >= orig_bio->bi_size) {
7160                 bio = orig_bio;
7161                 goto submit;
7162         }
7163
7164         /* async crcs make it difficult to collect full stripe writes. */
7165         if (btrfs_get_alloc_profile(root, 1) &
7166             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7167                 async_submit = 0;
7168         else
7169                 async_submit = 1;
7170
7171         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7172         if (!bio)
7173                 return -ENOMEM;
7174         bio->bi_private = dip;
7175         bio->bi_end_io = btrfs_end_dio_bio;
7176         atomic_inc(&dip->pending_bios);
7177
7178         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7179                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7180                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7181                                  bvec->bv_offset) < bvec->bv_len)) {
7182                         /*
7183                          * inc the count before we submit the bio so
7184                          * we know the end IO handler won't happen before
7185                          * we inc the count. Otherwise, the dip might get freed
7186                          * before we're done setting it up
7187                          */
7188                         atomic_inc(&dip->pending_bios);
7189                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7190                                                      file_offset, skip_sum,
7191                                                      async_submit);
7192                         if (ret) {
7193                                 bio_put(bio);
7194                                 atomic_dec(&dip->pending_bios);
7195                                 goto out_err;
7196                         }
7197
7198                         start_sector += submit_len >> 9;
7199                         file_offset += submit_len;
7200
7201                         submit_len = 0;
7202                         nr_pages = 0;
7203
7204                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7205                                                   start_sector, GFP_NOFS);
7206                         if (!bio)
7207                                 goto out_err;
7208                         bio->bi_private = dip;
7209                         bio->bi_end_io = btrfs_end_dio_bio;
7210
7211                         map_length = orig_bio->bi_size;
7212                         ret = btrfs_map_block(root->fs_info, rw,
7213                                               start_sector << 9,
7214                                               &map_length, NULL, 0);
7215                         if (ret) {
7216                                 bio_put(bio);
7217                                 goto out_err;
7218                         }
7219                 } else {
7220                         submit_len += bvec->bv_len;
7221                         nr_pages ++;
7222                         bvec++;
7223                 }
7224         }
7225
7226 submit:
7227         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7228                                      async_submit);
7229         if (!ret)
7230                 return 0;
7231
7232         bio_put(bio);
7233 out_err:
7234         dip->errors = 1;
7235         /*
7236          * before atomic variable goto zero, we must
7237          * make sure dip->errors is perceived to be set.
7238          */
7239         smp_mb__before_atomic_dec();
7240         if (atomic_dec_and_test(&dip->pending_bios))
7241                 bio_io_error(dip->orig_bio);
7242
7243         /* bio_end_io() will handle error, so we needn't return it */
7244         return 0;
7245 }
7246
7247 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7248                                 struct inode *inode, loff_t file_offset)
7249 {
7250         struct btrfs_root *root = BTRFS_I(inode)->root;
7251         struct btrfs_dio_private *dip;
7252         struct bio_vec *bvec = dio_bio->bi_io_vec;
7253         struct bio *io_bio;
7254         int skip_sum;
7255         int write = rw & REQ_WRITE;
7256         int ret = 0;
7257
7258         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7259
7260         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7261
7262         if (!io_bio) {
7263                 ret = -ENOMEM;
7264                 goto free_ordered;
7265         }
7266
7267         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7268         if (!dip) {
7269                 ret = -ENOMEM;
7270                 goto free_io_bio;
7271         }
7272
7273         dip->private = dio_bio->bi_private;
7274         io_bio->bi_private = dio_bio->bi_private;
7275         dip->inode = inode;
7276         dip->logical_offset = file_offset;
7277
7278         dip->bytes = 0;
7279         do {
7280                 dip->bytes += bvec->bv_len;
7281                 bvec++;
7282         } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7283
7284         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7285         io_bio->bi_private = dip;
7286         dip->errors = 0;
7287         dip->orig_bio = io_bio;
7288         dip->dio_bio = dio_bio;
7289         atomic_set(&dip->pending_bios, 0);
7290
7291         if (write)
7292                 io_bio->bi_end_io = btrfs_endio_direct_write;
7293         else
7294                 io_bio->bi_end_io = btrfs_endio_direct_read;
7295
7296         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7297         if (!ret)
7298                 return;
7299
7300 free_io_bio:
7301         bio_put(io_bio);
7302
7303 free_ordered:
7304         /*
7305          * If this is a write, we need to clean up the reserved space and kill
7306          * the ordered extent.
7307          */
7308         if (write) {
7309                 struct btrfs_ordered_extent *ordered;
7310                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7311                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7312                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7313                         btrfs_free_reserved_extent(root, ordered->start,
7314                                                    ordered->disk_len);
7315                 btrfs_put_ordered_extent(ordered);
7316                 btrfs_put_ordered_extent(ordered);
7317         }
7318         bio_endio(dio_bio, ret);
7319 }
7320
7321 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7322                         const struct iovec *iov, loff_t offset,
7323                         unsigned long nr_segs)
7324 {
7325         int seg;
7326         int i;
7327         size_t size;
7328         unsigned long addr;
7329         unsigned blocksize_mask = root->sectorsize - 1;
7330         ssize_t retval = -EINVAL;
7331         loff_t end = offset;
7332
7333         if (offset & blocksize_mask)
7334                 goto out;
7335
7336         /* Check the memory alignment.  Blocks cannot straddle pages */
7337         for (seg = 0; seg < nr_segs; seg++) {
7338                 addr = (unsigned long)iov[seg].iov_base;
7339                 size = iov[seg].iov_len;
7340                 end += size;
7341                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7342                         goto out;
7343
7344                 /* If this is a write we don't need to check anymore */
7345                 if (rw & WRITE)
7346                         continue;
7347
7348                 /*
7349                  * Check to make sure we don't have duplicate iov_base's in this
7350                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7351                  * when reading back.
7352                  */
7353                 for (i = seg + 1; i < nr_segs; i++) {
7354                         if (iov[seg].iov_base == iov[i].iov_base)
7355                                 goto out;
7356                 }
7357         }
7358         retval = 0;
7359 out:
7360         return retval;
7361 }
7362
7363 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7364                         const struct iovec *iov, loff_t offset,
7365                         unsigned long nr_segs)
7366 {
7367         struct file *file = iocb->ki_filp;
7368         struct inode *inode = file->f_mapping->host;
7369         size_t count = 0;
7370         int flags = 0;
7371         bool wakeup = true;
7372         bool relock = false;
7373         ssize_t ret;
7374
7375         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7376                             offset, nr_segs))
7377                 return 0;
7378
7379         atomic_inc(&inode->i_dio_count);
7380         smp_mb__after_atomic_inc();
7381
7382         if (rw & WRITE) {
7383                 count = iov_length(iov, nr_segs);
7384                 /*
7385                  * If the write DIO is beyond the EOF, we need update
7386                  * the isize, but it is protected by i_mutex. So we can
7387                  * not unlock the i_mutex at this case.
7388                  */
7389                 if (offset + count <= inode->i_size) {
7390                         mutex_unlock(&inode->i_mutex);
7391                         relock = true;
7392                 }
7393                 ret = btrfs_delalloc_reserve_space(inode, count);
7394                 if (ret)
7395                         goto out;
7396         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7397                                      &BTRFS_I(inode)->runtime_flags))) {
7398                 inode_dio_done(inode);
7399                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7400                 wakeup = false;
7401         }
7402
7403         ret = __blockdev_direct_IO(rw, iocb, inode,
7404                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7405                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7406                         btrfs_submit_direct, flags);
7407         if (rw & WRITE) {
7408                 if (ret < 0 && ret != -EIOCBQUEUED)
7409                         btrfs_delalloc_release_space(inode, count);
7410                 else if (ret >= 0 && (size_t)ret < count)
7411                         btrfs_delalloc_release_space(inode,
7412                                                      count - (size_t)ret);
7413                 else
7414                         btrfs_delalloc_release_metadata(inode, 0);
7415         }
7416 out:
7417         if (wakeup)
7418                 inode_dio_done(inode);
7419         if (relock)
7420                 mutex_lock(&inode->i_mutex);
7421
7422         return ret;
7423 }
7424
7425 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7426
7427 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7428                 __u64 start, __u64 len)
7429 {
7430         int     ret;
7431
7432         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7433         if (ret)
7434                 return ret;
7435
7436         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7437 }
7438
7439 int btrfs_readpage(struct file *file, struct page *page)
7440 {
7441         struct extent_io_tree *tree;
7442         tree = &BTRFS_I(page->mapping->host)->io_tree;
7443         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7444 }
7445
7446 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7447 {
7448         struct extent_io_tree *tree;
7449
7450
7451         if (current->flags & PF_MEMALLOC) {
7452                 redirty_page_for_writepage(wbc, page);
7453                 unlock_page(page);
7454                 return 0;
7455         }
7456         tree = &BTRFS_I(page->mapping->host)->io_tree;
7457         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7458 }
7459
7460 static int btrfs_writepages(struct address_space *mapping,
7461                             struct writeback_control *wbc)
7462 {
7463         struct extent_io_tree *tree;
7464
7465         tree = &BTRFS_I(mapping->host)->io_tree;
7466         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7467 }
7468
7469 static int
7470 btrfs_readpages(struct file *file, struct address_space *mapping,
7471                 struct list_head *pages, unsigned nr_pages)
7472 {
7473         struct extent_io_tree *tree;
7474         tree = &BTRFS_I(mapping->host)->io_tree;
7475         return extent_readpages(tree, mapping, pages, nr_pages,
7476                                 btrfs_get_extent);
7477 }
7478 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7479 {
7480         struct extent_io_tree *tree;
7481         struct extent_map_tree *map;
7482         int ret;
7483
7484         tree = &BTRFS_I(page->mapping->host)->io_tree;
7485         map = &BTRFS_I(page->mapping->host)->extent_tree;
7486         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7487         if (ret == 1) {
7488                 ClearPagePrivate(page);
7489                 set_page_private(page, 0);
7490                 page_cache_release(page);
7491         }
7492         return ret;
7493 }
7494
7495 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7496 {
7497         if (PageWriteback(page) || PageDirty(page))
7498                 return 0;
7499         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7500 }
7501
7502 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7503 {
7504         struct inode *inode = page->mapping->host;
7505         struct extent_io_tree *tree;
7506         struct btrfs_ordered_extent *ordered;
7507         struct extent_state *cached_state = NULL;
7508         u64 page_start = page_offset(page);
7509         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7510
7511         /*
7512          * we have the page locked, so new writeback can't start,
7513          * and the dirty bit won't be cleared while we are here.
7514          *
7515          * Wait for IO on this page so that we can safely clear
7516          * the PagePrivate2 bit and do ordered accounting
7517          */
7518         wait_on_page_writeback(page);
7519
7520         tree = &BTRFS_I(inode)->io_tree;
7521         if (offset) {
7522                 btrfs_releasepage(page, GFP_NOFS);
7523                 return;
7524         }
7525         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7526         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7527         if (ordered) {
7528                 /*
7529                  * IO on this page will never be started, so we need
7530                  * to account for any ordered extents now
7531                  */
7532                 clear_extent_bit(tree, page_start, page_end,
7533                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7534                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7535                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7536                 /*
7537                  * whoever cleared the private bit is responsible
7538                  * for the finish_ordered_io
7539                  */
7540                 if (TestClearPagePrivate2(page) &&
7541                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7542                                                    PAGE_CACHE_SIZE, 1)) {
7543                         btrfs_finish_ordered_io(ordered);
7544                 }
7545                 btrfs_put_ordered_extent(ordered);
7546                 cached_state = NULL;
7547                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7548         }
7549         clear_extent_bit(tree, page_start, page_end,
7550                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7551                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7552                  &cached_state, GFP_NOFS);
7553         __btrfs_releasepage(page, GFP_NOFS);
7554
7555         ClearPageChecked(page);
7556         if (PagePrivate(page)) {
7557                 ClearPagePrivate(page);
7558                 set_page_private(page, 0);
7559                 page_cache_release(page);
7560         }
7561 }
7562
7563 /*
7564  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7565  * called from a page fault handler when a page is first dirtied. Hence we must
7566  * be careful to check for EOF conditions here. We set the page up correctly
7567  * for a written page which means we get ENOSPC checking when writing into
7568  * holes and correct delalloc and unwritten extent mapping on filesystems that
7569  * support these features.
7570  *
7571  * We are not allowed to take the i_mutex here so we have to play games to
7572  * protect against truncate races as the page could now be beyond EOF.  Because
7573  * vmtruncate() writes the inode size before removing pages, once we have the
7574  * page lock we can determine safely if the page is beyond EOF. If it is not
7575  * beyond EOF, then the page is guaranteed safe against truncation until we
7576  * unlock the page.
7577  */
7578 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7579 {
7580         struct page *page = vmf->page;
7581         struct inode *inode = file_inode(vma->vm_file);
7582         struct btrfs_root *root = BTRFS_I(inode)->root;
7583         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7584         struct btrfs_ordered_extent *ordered;
7585         struct extent_state *cached_state = NULL;
7586         char *kaddr;
7587         unsigned long zero_start;
7588         loff_t size;
7589         int ret;
7590         int reserved = 0;
7591         u64 page_start;
7592         u64 page_end;
7593
7594         sb_start_pagefault(inode->i_sb);
7595         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7596         if (!ret) {
7597                 ret = file_update_time(vma->vm_file);
7598                 reserved = 1;
7599         }
7600         if (ret) {
7601                 if (ret == -ENOMEM)
7602                         ret = VM_FAULT_OOM;
7603                 else /* -ENOSPC, -EIO, etc */
7604                         ret = VM_FAULT_SIGBUS;
7605                 if (reserved)
7606                         goto out;
7607                 goto out_noreserve;
7608         }
7609
7610         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7611 again:
7612         lock_page(page);
7613         size = i_size_read(inode);
7614         page_start = page_offset(page);
7615         page_end = page_start + PAGE_CACHE_SIZE - 1;
7616
7617         if ((page->mapping != inode->i_mapping) ||
7618             (page_start >= size)) {
7619                 /* page got truncated out from underneath us */
7620                 goto out_unlock;
7621         }
7622         wait_on_page_writeback(page);
7623
7624         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7625         set_page_extent_mapped(page);
7626
7627         /*
7628          * we can't set the delalloc bits if there are pending ordered
7629          * extents.  Drop our locks and wait for them to finish
7630          */
7631         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7632         if (ordered) {
7633                 unlock_extent_cached(io_tree, page_start, page_end,
7634                                      &cached_state, GFP_NOFS);
7635                 unlock_page(page);
7636                 btrfs_start_ordered_extent(inode, ordered, 1);
7637                 btrfs_put_ordered_extent(ordered);
7638                 goto again;
7639         }
7640
7641         /*
7642          * XXX - page_mkwrite gets called every time the page is dirtied, even
7643          * if it was already dirty, so for space accounting reasons we need to
7644          * clear any delalloc bits for the range we are fixing to save.  There
7645          * is probably a better way to do this, but for now keep consistent with
7646          * prepare_pages in the normal write path.
7647          */
7648         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7649                           EXTENT_DIRTY | EXTENT_DELALLOC |
7650                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7651                           0, 0, &cached_state, GFP_NOFS);
7652
7653         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7654                                         &cached_state);
7655         if (ret) {
7656                 unlock_extent_cached(io_tree, page_start, page_end,
7657                                      &cached_state, GFP_NOFS);
7658                 ret = VM_FAULT_SIGBUS;
7659                 goto out_unlock;
7660         }
7661         ret = 0;
7662
7663         /* page is wholly or partially inside EOF */
7664         if (page_start + PAGE_CACHE_SIZE > size)
7665                 zero_start = size & ~PAGE_CACHE_MASK;
7666         else
7667                 zero_start = PAGE_CACHE_SIZE;
7668
7669         if (zero_start != PAGE_CACHE_SIZE) {
7670                 kaddr = kmap(page);
7671                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7672                 flush_dcache_page(page);
7673                 kunmap(page);
7674         }
7675         ClearPageChecked(page);
7676         set_page_dirty(page);
7677         SetPageUptodate(page);
7678
7679         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7680         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7681         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7682
7683         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7684
7685 out_unlock:
7686         if (!ret) {
7687                 sb_end_pagefault(inode->i_sb);
7688                 return VM_FAULT_LOCKED;
7689         }
7690         unlock_page(page);
7691 out:
7692         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7693 out_noreserve:
7694         sb_end_pagefault(inode->i_sb);
7695         return ret;
7696 }
7697
7698 static int btrfs_truncate(struct inode *inode)
7699 {
7700         struct btrfs_root *root = BTRFS_I(inode)->root;
7701         struct btrfs_block_rsv *rsv;
7702         int ret;
7703         int err = 0;
7704         struct btrfs_trans_handle *trans;
7705         u64 mask = root->sectorsize - 1;
7706         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7707
7708         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7709         if (ret)
7710                 return ret;
7711
7712         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7713         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7714
7715         /*
7716          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7717          * 3 things going on here
7718          *
7719          * 1) We need to reserve space for our orphan item and the space to
7720          * delete our orphan item.  Lord knows we don't want to have a dangling
7721          * orphan item because we didn't reserve space to remove it.
7722          *
7723          * 2) We need to reserve space to update our inode.
7724          *
7725          * 3) We need to have something to cache all the space that is going to
7726          * be free'd up by the truncate operation, but also have some slack
7727          * space reserved in case it uses space during the truncate (thank you
7728          * very much snapshotting).
7729          *
7730          * And we need these to all be seperate.  The fact is we can use alot of
7731          * space doing the truncate, and we have no earthly idea how much space
7732          * we will use, so we need the truncate reservation to be seperate so it
7733          * doesn't end up using space reserved for updating the inode or
7734          * removing the orphan item.  We also need to be able to stop the
7735          * transaction and start a new one, which means we need to be able to
7736          * update the inode several times, and we have no idea of knowing how
7737          * many times that will be, so we can't just reserve 1 item for the
7738          * entirety of the opration, so that has to be done seperately as well.
7739          * Then there is the orphan item, which does indeed need to be held on
7740          * to for the whole operation, and we need nobody to touch this reserved
7741          * space except the orphan code.
7742          *
7743          * So that leaves us with
7744          *
7745          * 1) root->orphan_block_rsv - for the orphan deletion.
7746          * 2) rsv - for the truncate reservation, which we will steal from the
7747          * transaction reservation.
7748          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7749          * updating the inode.
7750          */
7751         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7752         if (!rsv)
7753                 return -ENOMEM;
7754         rsv->size = min_size;
7755         rsv->failfast = 1;
7756
7757         /*
7758          * 1 for the truncate slack space
7759          * 1 for updating the inode.
7760          */
7761         trans = btrfs_start_transaction(root, 2);
7762         if (IS_ERR(trans)) {
7763                 err = PTR_ERR(trans);
7764                 goto out;
7765         }
7766
7767         /* Migrate the slack space for the truncate to our reserve */
7768         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7769                                       min_size);
7770         BUG_ON(ret);
7771
7772         /*
7773          * setattr is responsible for setting the ordered_data_close flag,
7774          * but that is only tested during the last file release.  That
7775          * could happen well after the next commit, leaving a great big
7776          * window where new writes may get lost if someone chooses to write
7777          * to this file after truncating to zero
7778          *
7779          * The inode doesn't have any dirty data here, and so if we commit
7780          * this is a noop.  If someone immediately starts writing to the inode
7781          * it is very likely we'll catch some of their writes in this
7782          * transaction, and the commit will find this file on the ordered
7783          * data list with good things to send down.
7784          *
7785          * This is a best effort solution, there is still a window where
7786          * using truncate to replace the contents of the file will
7787          * end up with a zero length file after a crash.
7788          */
7789         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7790                                            &BTRFS_I(inode)->runtime_flags))
7791                 btrfs_add_ordered_operation(trans, root, inode);
7792
7793         /*
7794          * So if we truncate and then write and fsync we normally would just
7795          * write the extents that changed, which is a problem if we need to
7796          * first truncate that entire inode.  So set this flag so we write out
7797          * all of the extents in the inode to the sync log so we're completely
7798          * safe.
7799          */
7800         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7801         trans->block_rsv = rsv;
7802
7803         while (1) {
7804                 ret = btrfs_truncate_inode_items(trans, root, inode,
7805                                                  inode->i_size,
7806                                                  BTRFS_EXTENT_DATA_KEY);
7807                 if (ret != -ENOSPC) {
7808                         err = ret;
7809                         break;
7810                 }
7811
7812                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7813                 ret = btrfs_update_inode(trans, root, inode);
7814                 if (ret) {
7815                         err = ret;
7816                         break;
7817                 }
7818
7819                 btrfs_end_transaction(trans, root);
7820                 btrfs_btree_balance_dirty(root);
7821
7822                 trans = btrfs_start_transaction(root, 2);
7823                 if (IS_ERR(trans)) {
7824                         ret = err = PTR_ERR(trans);
7825                         trans = NULL;
7826                         break;
7827                 }
7828
7829                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7830                                               rsv, min_size);
7831                 BUG_ON(ret);    /* shouldn't happen */
7832                 trans->block_rsv = rsv;
7833         }
7834
7835         if (ret == 0 && inode->i_nlink > 0) {
7836                 trans->block_rsv = root->orphan_block_rsv;
7837                 ret = btrfs_orphan_del(trans, inode);
7838                 if (ret)
7839                         err = ret;
7840         }
7841
7842         if (trans) {
7843                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7844                 ret = btrfs_update_inode(trans, root, inode);
7845                 if (ret && !err)
7846                         err = ret;
7847
7848                 ret = btrfs_end_transaction(trans, root);
7849                 btrfs_btree_balance_dirty(root);
7850         }
7851
7852 out:
7853         btrfs_free_block_rsv(root, rsv);
7854
7855         if (ret && !err)
7856                 err = ret;
7857
7858         return err;
7859 }
7860
7861 /*
7862  * create a new subvolume directory/inode (helper for the ioctl).
7863  */
7864 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7865                              struct btrfs_root *new_root, u64 new_dirid)
7866 {
7867         struct inode *inode;
7868         int err;
7869         u64 index = 0;
7870
7871         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7872                                 new_dirid, new_dirid,
7873                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7874                                 &index);
7875         if (IS_ERR(inode))
7876                 return PTR_ERR(inode);
7877         inode->i_op = &btrfs_dir_inode_operations;
7878         inode->i_fop = &btrfs_dir_file_operations;
7879
7880         set_nlink(inode, 1);
7881         btrfs_i_size_write(inode, 0);
7882
7883         err = btrfs_update_inode(trans, new_root, inode);
7884
7885         iput(inode);
7886         return err;
7887 }
7888
7889 struct inode *btrfs_alloc_inode(struct super_block *sb)
7890 {
7891         struct btrfs_inode *ei;
7892         struct inode *inode;
7893
7894         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7895         if (!ei)
7896                 return NULL;
7897
7898         ei->root = NULL;
7899         ei->generation = 0;
7900         ei->last_trans = 0;
7901         ei->last_sub_trans = 0;
7902         ei->logged_trans = 0;
7903         ei->delalloc_bytes = 0;
7904         ei->disk_i_size = 0;
7905         ei->flags = 0;
7906         ei->csum_bytes = 0;
7907         ei->index_cnt = (u64)-1;
7908         ei->last_unlink_trans = 0;
7909         ei->last_log_commit = 0;
7910
7911         spin_lock_init(&ei->lock);
7912         ei->outstanding_extents = 0;
7913         ei->reserved_extents = 0;
7914
7915         ei->runtime_flags = 0;
7916         ei->force_compress = BTRFS_COMPRESS_NONE;
7917
7918         ei->delayed_node = NULL;
7919
7920         inode = &ei->vfs_inode;
7921         extent_map_tree_init(&ei->extent_tree);
7922         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7923         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7924         ei->io_tree.track_uptodate = 1;
7925         ei->io_failure_tree.track_uptodate = 1;
7926         atomic_set(&ei->sync_writers, 0);
7927         mutex_init(&ei->log_mutex);
7928         mutex_init(&ei->delalloc_mutex);
7929         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7930         INIT_LIST_HEAD(&ei->delalloc_inodes);
7931         INIT_LIST_HEAD(&ei->ordered_operations);
7932         RB_CLEAR_NODE(&ei->rb_node);
7933
7934         return inode;
7935 }
7936
7937 static void btrfs_i_callback(struct rcu_head *head)
7938 {
7939         struct inode *inode = container_of(head, struct inode, i_rcu);
7940         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7941 }
7942
7943 void btrfs_destroy_inode(struct inode *inode)
7944 {
7945         struct btrfs_ordered_extent *ordered;
7946         struct btrfs_root *root = BTRFS_I(inode)->root;
7947
7948         WARN_ON(!hlist_empty(&inode->i_dentry));
7949         WARN_ON(inode->i_data.nrpages);
7950         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7951         WARN_ON(BTRFS_I(inode)->reserved_extents);
7952         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7953         WARN_ON(BTRFS_I(inode)->csum_bytes);
7954
7955         /*
7956          * This can happen where we create an inode, but somebody else also
7957          * created the same inode and we need to destroy the one we already
7958          * created.
7959          */
7960         if (!root)
7961                 goto free;
7962
7963         /*
7964          * Make sure we're properly removed from the ordered operation
7965          * lists.
7966          */
7967         smp_mb();
7968         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7969                 spin_lock(&root->fs_info->ordered_extent_lock);
7970                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7971                 spin_unlock(&root->fs_info->ordered_extent_lock);
7972         }
7973
7974         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7975                      &BTRFS_I(inode)->runtime_flags)) {
7976                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7977                         (unsigned long long)btrfs_ino(inode));
7978                 atomic_dec(&root->orphan_inodes);
7979         }
7980
7981         while (1) {
7982                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7983                 if (!ordered)
7984                         break;
7985                 else {
7986                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7987                                 (unsigned long long)ordered->file_offset,
7988                                 (unsigned long long)ordered->len);
7989                         btrfs_remove_ordered_extent(inode, ordered);
7990                         btrfs_put_ordered_extent(ordered);
7991                         btrfs_put_ordered_extent(ordered);
7992                 }
7993         }
7994         inode_tree_del(inode);
7995         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7996 free:
7997         call_rcu(&inode->i_rcu, btrfs_i_callback);
7998 }
7999
8000 int btrfs_drop_inode(struct inode *inode)
8001 {
8002         struct btrfs_root *root = BTRFS_I(inode)->root;
8003
8004         if (root == NULL)
8005                 return 1;
8006
8007         /* the snap/subvol tree is on deleting */
8008         if (btrfs_root_refs(&root->root_item) == 0 &&
8009             root != root->fs_info->tree_root)
8010                 return 1;
8011         else
8012                 return generic_drop_inode(inode);
8013 }
8014
8015 static void init_once(void *foo)
8016 {
8017         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8018
8019         inode_init_once(&ei->vfs_inode);
8020 }
8021
8022 void btrfs_destroy_cachep(void)
8023 {
8024         /*
8025          * Make sure all delayed rcu free inodes are flushed before we
8026          * destroy cache.
8027          */
8028         rcu_barrier();
8029         if (btrfs_inode_cachep)
8030                 kmem_cache_destroy(btrfs_inode_cachep);
8031         if (btrfs_trans_handle_cachep)
8032                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8033         if (btrfs_transaction_cachep)
8034                 kmem_cache_destroy(btrfs_transaction_cachep);
8035         if (btrfs_path_cachep)
8036                 kmem_cache_destroy(btrfs_path_cachep);
8037         if (btrfs_free_space_cachep)
8038                 kmem_cache_destroy(btrfs_free_space_cachep);
8039         if (btrfs_delalloc_work_cachep)
8040                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8041 }
8042
8043 int btrfs_init_cachep(void)
8044 {
8045         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8046                         sizeof(struct btrfs_inode), 0,
8047                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8048         if (!btrfs_inode_cachep)
8049                 goto fail;
8050
8051         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8052                         sizeof(struct btrfs_trans_handle), 0,
8053                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8054         if (!btrfs_trans_handle_cachep)
8055                 goto fail;
8056
8057         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8058                         sizeof(struct btrfs_transaction), 0,
8059                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8060         if (!btrfs_transaction_cachep)
8061                 goto fail;
8062
8063         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8064                         sizeof(struct btrfs_path), 0,
8065                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8066         if (!btrfs_path_cachep)
8067                 goto fail;
8068
8069         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8070                         sizeof(struct btrfs_free_space), 0,
8071                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8072         if (!btrfs_free_space_cachep)
8073                 goto fail;
8074
8075         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8076                         sizeof(struct btrfs_delalloc_work), 0,
8077                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8078                         NULL);
8079         if (!btrfs_delalloc_work_cachep)
8080                 goto fail;
8081
8082         return 0;
8083 fail:
8084         btrfs_destroy_cachep();
8085         return -ENOMEM;
8086 }
8087
8088 static int btrfs_getattr(struct vfsmount *mnt,
8089                          struct dentry *dentry, struct kstat *stat)
8090 {
8091         u64 delalloc_bytes;
8092         struct inode *inode = dentry->d_inode;
8093         u32 blocksize = inode->i_sb->s_blocksize;
8094
8095         generic_fillattr(inode, stat);
8096         stat->dev = BTRFS_I(inode)->root->anon_dev;
8097         stat->blksize = PAGE_CACHE_SIZE;
8098
8099         spin_lock(&BTRFS_I(inode)->lock);
8100         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8101         spin_unlock(&BTRFS_I(inode)->lock);
8102         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8103                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8104         return 0;
8105 }
8106
8107 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8108                            struct inode *new_dir, struct dentry *new_dentry)
8109 {
8110         struct btrfs_trans_handle *trans;
8111         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8112         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8113         struct inode *new_inode = new_dentry->d_inode;
8114         struct inode *old_inode = old_dentry->d_inode;
8115         struct timespec ctime = CURRENT_TIME;
8116         u64 index = 0;
8117         u64 root_objectid;
8118         int ret;
8119         u64 old_ino = btrfs_ino(old_inode);
8120
8121         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8122                 return -EPERM;
8123
8124         /* we only allow rename subvolume link between subvolumes */
8125         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8126                 return -EXDEV;
8127
8128         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8129             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8130                 return -ENOTEMPTY;
8131
8132         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8133             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8134                 return -ENOTEMPTY;
8135
8136
8137         /* check for collisions, even if the  name isn't there */
8138         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8139                              new_dentry->d_name.name,
8140                              new_dentry->d_name.len);
8141
8142         if (ret) {
8143                 if (ret == -EEXIST) {
8144                         /* we shouldn't get
8145                          * eexist without a new_inode */
8146                         if (!new_inode) {
8147                                 WARN_ON(1);
8148                                 return ret;
8149                         }
8150                 } else {
8151                         /* maybe -EOVERFLOW */
8152                         return ret;
8153                 }
8154         }
8155         ret = 0;
8156
8157         /*
8158          * we're using rename to replace one file with another.
8159          * and the replacement file is large.  Start IO on it now so
8160          * we don't add too much work to the end of the transaction
8161          */
8162         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8163             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8164                 filemap_flush(old_inode->i_mapping);
8165
8166         /* close the racy window with snapshot create/destroy ioctl */
8167         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8168                 down_read(&root->fs_info->subvol_sem);
8169         /*
8170          * We want to reserve the absolute worst case amount of items.  So if
8171          * both inodes are subvols and we need to unlink them then that would
8172          * require 4 item modifications, but if they are both normal inodes it
8173          * would require 5 item modifications, so we'll assume their normal
8174          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8175          * should cover the worst case number of items we'll modify.
8176          */
8177         trans = btrfs_start_transaction(root, 11);
8178         if (IS_ERR(trans)) {
8179                 ret = PTR_ERR(trans);
8180                 goto out_notrans;
8181         }
8182
8183         if (dest != root)
8184                 btrfs_record_root_in_trans(trans, dest);
8185
8186         ret = btrfs_set_inode_index(new_dir, &index);
8187         if (ret)
8188                 goto out_fail;
8189
8190         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8191                 /* force full log commit if subvolume involved. */
8192                 root->fs_info->last_trans_log_full_commit = trans->transid;
8193         } else {
8194                 ret = btrfs_insert_inode_ref(trans, dest,
8195                                              new_dentry->d_name.name,
8196                                              new_dentry->d_name.len,
8197                                              old_ino,
8198                                              btrfs_ino(new_dir), index);
8199                 if (ret)
8200                         goto out_fail;
8201                 /*
8202                  * this is an ugly little race, but the rename is required
8203                  * to make sure that if we crash, the inode is either at the
8204                  * old name or the new one.  pinning the log transaction lets
8205                  * us make sure we don't allow a log commit to come in after
8206                  * we unlink the name but before we add the new name back in.
8207                  */
8208                 btrfs_pin_log_trans(root);
8209         }
8210         /*
8211          * make sure the inode gets flushed if it is replacing
8212          * something.
8213          */
8214         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8215                 btrfs_add_ordered_operation(trans, root, old_inode);
8216
8217         inode_inc_iversion(old_dir);
8218         inode_inc_iversion(new_dir);
8219         inode_inc_iversion(old_inode);
8220         old_dir->i_ctime = old_dir->i_mtime = ctime;
8221         new_dir->i_ctime = new_dir->i_mtime = ctime;
8222         old_inode->i_ctime = ctime;
8223
8224         if (old_dentry->d_parent != new_dentry->d_parent)
8225                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8226
8227         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8228                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8229                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8230                                         old_dentry->d_name.name,
8231                                         old_dentry->d_name.len);
8232         } else {
8233                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8234                                         old_dentry->d_inode,
8235                                         old_dentry->d_name.name,
8236                                         old_dentry->d_name.len);
8237                 if (!ret)
8238                         ret = btrfs_update_inode(trans, root, old_inode);
8239         }
8240         if (ret) {
8241                 btrfs_abort_transaction(trans, root, ret);
8242                 goto out_fail;
8243         }
8244
8245         if (new_inode) {
8246                 inode_inc_iversion(new_inode);
8247                 new_inode->i_ctime = CURRENT_TIME;
8248                 if (unlikely(btrfs_ino(new_inode) ==
8249                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8250                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8251                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8252                                                 root_objectid,
8253                                                 new_dentry->d_name.name,
8254                                                 new_dentry->d_name.len);
8255                         BUG_ON(new_inode->i_nlink == 0);
8256                 } else {
8257                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8258                                                  new_dentry->d_inode,
8259                                                  new_dentry->d_name.name,
8260                                                  new_dentry->d_name.len);
8261                 }
8262                 if (!ret && new_inode->i_nlink == 0) {
8263                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8264                         BUG_ON(ret);
8265                 }
8266                 if (ret) {
8267                         btrfs_abort_transaction(trans, root, ret);
8268                         goto out_fail;
8269                 }
8270         }
8271
8272         ret = btrfs_add_link(trans, new_dir, old_inode,
8273                              new_dentry->d_name.name,
8274                              new_dentry->d_name.len, 0, index);
8275         if (ret) {
8276                 btrfs_abort_transaction(trans, root, ret);
8277                 goto out_fail;
8278         }
8279
8280         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8281                 struct dentry *parent = new_dentry->d_parent;
8282                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8283                 btrfs_end_log_trans(root);
8284         }
8285 out_fail:
8286         btrfs_end_transaction(trans, root);
8287 out_notrans:
8288         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8289                 up_read(&root->fs_info->subvol_sem);
8290
8291         return ret;
8292 }
8293
8294 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8295 {
8296         struct btrfs_delalloc_work *delalloc_work;
8297
8298         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8299                                      work);
8300         if (delalloc_work->wait)
8301                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8302         else
8303                 filemap_flush(delalloc_work->inode->i_mapping);
8304
8305         if (delalloc_work->delay_iput)
8306                 btrfs_add_delayed_iput(delalloc_work->inode);
8307         else
8308                 iput(delalloc_work->inode);
8309         complete(&delalloc_work->completion);
8310 }
8311
8312 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8313                                                     int wait, int delay_iput)
8314 {
8315         struct btrfs_delalloc_work *work;
8316
8317         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8318         if (!work)
8319                 return NULL;
8320
8321         init_completion(&work->completion);
8322         INIT_LIST_HEAD(&work->list);
8323         work->inode = inode;
8324         work->wait = wait;
8325         work->delay_iput = delay_iput;
8326         work->work.func = btrfs_run_delalloc_work;
8327
8328         return work;
8329 }
8330
8331 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8332 {
8333         wait_for_completion(&work->completion);
8334         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8335 }
8336
8337 /*
8338  * some fairly slow code that needs optimization. This walks the list
8339  * of all the inodes with pending delalloc and forces them to disk.
8340  */
8341 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8342 {
8343         struct btrfs_inode *binode;
8344         struct inode *inode;
8345         struct btrfs_delalloc_work *work, *next;
8346         struct list_head works;
8347         struct list_head splice;
8348         int ret = 0;
8349
8350         if (root->fs_info->sb->s_flags & MS_RDONLY)
8351                 return -EROFS;
8352
8353         INIT_LIST_HEAD(&works);
8354         INIT_LIST_HEAD(&splice);
8355
8356         spin_lock(&root->fs_info->delalloc_lock);
8357         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8358         while (!list_empty(&splice)) {
8359                 binode = list_entry(splice.next, struct btrfs_inode,
8360                                     delalloc_inodes);
8361
8362                 list_del_init(&binode->delalloc_inodes);
8363
8364                 inode = igrab(&binode->vfs_inode);
8365                 if (!inode) {
8366                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8367                                   &binode->runtime_flags);
8368                         continue;
8369                 }
8370
8371                 list_add_tail(&binode->delalloc_inodes,
8372                               &root->fs_info->delalloc_inodes);
8373                 spin_unlock(&root->fs_info->delalloc_lock);
8374
8375                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8376                 if (unlikely(!work)) {
8377                         ret = -ENOMEM;
8378                         goto out;
8379                 }
8380                 list_add_tail(&work->list, &works);
8381                 btrfs_queue_worker(&root->fs_info->flush_workers,
8382                                    &work->work);
8383
8384                 cond_resched();
8385                 spin_lock(&root->fs_info->delalloc_lock);
8386         }
8387         spin_unlock(&root->fs_info->delalloc_lock);
8388
8389         list_for_each_entry_safe(work, next, &works, list) {
8390                 list_del_init(&work->list);
8391                 btrfs_wait_and_free_delalloc_work(work);
8392         }
8393
8394         /* the filemap_flush will queue IO into the worker threads, but
8395          * we have to make sure the IO is actually started and that
8396          * ordered extents get created before we return
8397          */
8398         atomic_inc(&root->fs_info->async_submit_draining);
8399         while (atomic_read(&root->fs_info->nr_async_submits) ||
8400               atomic_read(&root->fs_info->async_delalloc_pages)) {
8401                 wait_event(root->fs_info->async_submit_wait,
8402                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8403                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8404         }
8405         atomic_dec(&root->fs_info->async_submit_draining);
8406         return 0;
8407 out:
8408         list_for_each_entry_safe(work, next, &works, list) {
8409                 list_del_init(&work->list);
8410                 btrfs_wait_and_free_delalloc_work(work);
8411         }
8412
8413         if (!list_empty_careful(&splice)) {
8414                 spin_lock(&root->fs_info->delalloc_lock);
8415                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8416                 spin_unlock(&root->fs_info->delalloc_lock);
8417         }
8418         return ret;
8419 }
8420
8421 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8422                          const char *symname)
8423 {
8424         struct btrfs_trans_handle *trans;
8425         struct btrfs_root *root = BTRFS_I(dir)->root;
8426         struct btrfs_path *path;
8427         struct btrfs_key key;
8428         struct inode *inode = NULL;
8429         int err;
8430         int drop_inode = 0;
8431         u64 objectid;
8432         u64 index = 0 ;
8433         int name_len;
8434         int datasize;
8435         unsigned long ptr;
8436         struct btrfs_file_extent_item *ei;
8437         struct extent_buffer *leaf;
8438
8439         name_len = strlen(symname) + 1;
8440         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8441                 return -ENAMETOOLONG;
8442
8443         /*
8444          * 2 items for inode item and ref
8445          * 2 items for dir items
8446          * 1 item for xattr if selinux is on
8447          */
8448         trans = btrfs_start_transaction(root, 5);
8449         if (IS_ERR(trans))
8450                 return PTR_ERR(trans);
8451
8452         err = btrfs_find_free_ino(root, &objectid);
8453         if (err)
8454                 goto out_unlock;
8455
8456         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8457                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8458                                 S_IFLNK|S_IRWXUGO, &index);
8459         if (IS_ERR(inode)) {
8460                 err = PTR_ERR(inode);
8461                 goto out_unlock;
8462         }
8463
8464         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8465         if (err) {
8466                 drop_inode = 1;
8467                 goto out_unlock;
8468         }
8469
8470         /*
8471         * If the active LSM wants to access the inode during
8472         * d_instantiate it needs these. Smack checks to see
8473         * if the filesystem supports xattrs by looking at the
8474         * ops vector.
8475         */
8476         inode->i_fop = &btrfs_file_operations;
8477         inode->i_op = &btrfs_file_inode_operations;
8478
8479         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8480         if (err)
8481                 drop_inode = 1;
8482         else {
8483                 inode->i_mapping->a_ops = &btrfs_aops;
8484                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8485                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8486         }
8487         if (drop_inode)
8488                 goto out_unlock;
8489
8490         path = btrfs_alloc_path();
8491         if (!path) {
8492                 err = -ENOMEM;
8493                 drop_inode = 1;
8494                 goto out_unlock;
8495         }
8496         key.objectid = btrfs_ino(inode);
8497         key.offset = 0;
8498         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8499         datasize = btrfs_file_extent_calc_inline_size(name_len);
8500         err = btrfs_insert_empty_item(trans, root, path, &key,
8501                                       datasize);
8502         if (err) {
8503                 drop_inode = 1;
8504                 btrfs_free_path(path);
8505                 goto out_unlock;
8506         }
8507         leaf = path->nodes[0];
8508         ei = btrfs_item_ptr(leaf, path->slots[0],
8509                             struct btrfs_file_extent_item);
8510         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8511         btrfs_set_file_extent_type(leaf, ei,
8512                                    BTRFS_FILE_EXTENT_INLINE);
8513         btrfs_set_file_extent_encryption(leaf, ei, 0);
8514         btrfs_set_file_extent_compression(leaf, ei, 0);
8515         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8516         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8517
8518         ptr = btrfs_file_extent_inline_start(ei);
8519         write_extent_buffer(leaf, symname, ptr, name_len);
8520         btrfs_mark_buffer_dirty(leaf);
8521         btrfs_free_path(path);
8522
8523         inode->i_op = &btrfs_symlink_inode_operations;
8524         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8525         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8526         inode_set_bytes(inode, name_len);
8527         btrfs_i_size_write(inode, name_len - 1);
8528         err = btrfs_update_inode(trans, root, inode);
8529         if (err)
8530                 drop_inode = 1;
8531
8532 out_unlock:
8533         if (!err)
8534                 d_instantiate(dentry, inode);
8535         btrfs_end_transaction(trans, root);
8536         if (drop_inode) {
8537                 inode_dec_link_count(inode);
8538                 iput(inode);
8539         }
8540         btrfs_btree_balance_dirty(root);
8541         return err;
8542 }
8543
8544 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8545                                        u64 start, u64 num_bytes, u64 min_size,
8546                                        loff_t actual_len, u64 *alloc_hint,
8547                                        struct btrfs_trans_handle *trans)
8548 {
8549         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8550         struct extent_map *em;
8551         struct btrfs_root *root = BTRFS_I(inode)->root;
8552         struct btrfs_key ins;
8553         u64 cur_offset = start;
8554         u64 i_size;
8555         u64 cur_bytes;
8556         int ret = 0;
8557         bool own_trans = true;
8558
8559         if (trans)
8560                 own_trans = false;
8561         while (num_bytes > 0) {
8562                 if (own_trans) {
8563                         trans = btrfs_start_transaction(root, 3);
8564                         if (IS_ERR(trans)) {
8565                                 ret = PTR_ERR(trans);
8566                                 break;
8567                         }
8568                 }
8569
8570                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8571                 cur_bytes = max(cur_bytes, min_size);
8572                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8573                                            min_size, 0, *alloc_hint, &ins, 1);
8574                 if (ret) {
8575                         if (own_trans)
8576                                 btrfs_end_transaction(trans, root);
8577                         break;
8578                 }
8579
8580                 ret = insert_reserved_file_extent(trans, inode,
8581                                                   cur_offset, ins.objectid,
8582                                                   ins.offset, ins.offset,
8583                                                   ins.offset, 0, 0, 0,
8584                                                   BTRFS_FILE_EXTENT_PREALLOC);
8585                 if (ret) {
8586                         btrfs_abort_transaction(trans, root, ret);
8587                         if (own_trans)
8588                                 btrfs_end_transaction(trans, root);
8589                         break;
8590                 }
8591                 btrfs_drop_extent_cache(inode, cur_offset,
8592                                         cur_offset + ins.offset -1, 0);
8593
8594                 em = alloc_extent_map();
8595                 if (!em) {
8596                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8597                                 &BTRFS_I(inode)->runtime_flags);
8598                         goto next;
8599                 }
8600
8601                 em->start = cur_offset;
8602                 em->orig_start = cur_offset;
8603                 em->len = ins.offset;
8604                 em->block_start = ins.objectid;
8605                 em->block_len = ins.offset;
8606                 em->orig_block_len = ins.offset;
8607                 em->ram_bytes = ins.offset;
8608                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8609                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8610                 em->generation = trans->transid;
8611
8612                 while (1) {
8613                         write_lock(&em_tree->lock);
8614                         ret = add_extent_mapping(em_tree, em, 1);
8615                         write_unlock(&em_tree->lock);
8616                         if (ret != -EEXIST)
8617                                 break;
8618                         btrfs_drop_extent_cache(inode, cur_offset,
8619                                                 cur_offset + ins.offset - 1,
8620                                                 0);
8621                 }
8622                 free_extent_map(em);
8623 next:
8624                 num_bytes -= ins.offset;
8625                 cur_offset += ins.offset;
8626                 *alloc_hint = ins.objectid + ins.offset;
8627
8628                 inode_inc_iversion(inode);
8629                 inode->i_ctime = CURRENT_TIME;
8630                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8631                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8632                     (actual_len > inode->i_size) &&
8633                     (cur_offset > inode->i_size)) {
8634                         if (cur_offset > actual_len)
8635                                 i_size = actual_len;
8636                         else
8637                                 i_size = cur_offset;
8638                         i_size_write(inode, i_size);
8639                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8640                 }
8641
8642                 ret = btrfs_update_inode(trans, root, inode);
8643
8644                 if (ret) {
8645                         btrfs_abort_transaction(trans, root, ret);
8646                         if (own_trans)
8647                                 btrfs_end_transaction(trans, root);
8648                         break;
8649                 }
8650
8651                 if (own_trans)
8652                         btrfs_end_transaction(trans, root);
8653         }
8654         return ret;
8655 }
8656
8657 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8658                               u64 start, u64 num_bytes, u64 min_size,
8659                               loff_t actual_len, u64 *alloc_hint)
8660 {
8661         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8662                                            min_size, actual_len, alloc_hint,
8663                                            NULL);
8664 }
8665
8666 int btrfs_prealloc_file_range_trans(struct inode *inode,
8667                                     struct btrfs_trans_handle *trans, int mode,
8668                                     u64 start, u64 num_bytes, u64 min_size,
8669                                     loff_t actual_len, u64 *alloc_hint)
8670 {
8671         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8672                                            min_size, actual_len, alloc_hint, trans);
8673 }
8674
8675 static int btrfs_set_page_dirty(struct page *page)
8676 {
8677         return __set_page_dirty_nobuffers(page);
8678 }
8679
8680 static int btrfs_permission(struct inode *inode, int mask)
8681 {
8682         struct btrfs_root *root = BTRFS_I(inode)->root;
8683         umode_t mode = inode->i_mode;
8684
8685         if (mask & MAY_WRITE &&
8686             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8687                 if (btrfs_root_readonly(root))
8688                         return -EROFS;
8689                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8690                         return -EACCES;
8691         }
8692         return generic_permission(inode, mask);
8693 }
8694
8695 static const struct inode_operations btrfs_dir_inode_operations = {
8696         .getattr        = btrfs_getattr,
8697         .lookup         = btrfs_lookup,
8698         .create         = btrfs_create,
8699         .unlink         = btrfs_unlink,
8700         .link           = btrfs_link,
8701         .mkdir          = btrfs_mkdir,
8702         .rmdir          = btrfs_rmdir,
8703         .rename         = btrfs_rename,
8704         .symlink        = btrfs_symlink,
8705         .setattr        = btrfs_setattr,
8706         .mknod          = btrfs_mknod,
8707         .setxattr       = btrfs_setxattr,
8708         .getxattr       = btrfs_getxattr,
8709         .listxattr      = btrfs_listxattr,
8710         .removexattr    = btrfs_removexattr,
8711         .permission     = btrfs_permission,
8712         .get_acl        = btrfs_get_acl,
8713 };
8714 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8715         .lookup         = btrfs_lookup,
8716         .permission     = btrfs_permission,
8717         .get_acl        = btrfs_get_acl,
8718 };
8719
8720 static const struct file_operations btrfs_dir_file_operations = {
8721         .llseek         = generic_file_llseek,
8722         .read           = generic_read_dir,
8723         .readdir        = btrfs_real_readdir,
8724         .unlocked_ioctl = btrfs_ioctl,
8725 #ifdef CONFIG_COMPAT
8726         .compat_ioctl   = btrfs_ioctl,
8727 #endif
8728         .release        = btrfs_release_file,
8729         .fsync          = btrfs_sync_file,
8730 };
8731
8732 static struct extent_io_ops btrfs_extent_io_ops = {
8733         .fill_delalloc = run_delalloc_range,
8734         .submit_bio_hook = btrfs_submit_bio_hook,
8735         .merge_bio_hook = btrfs_merge_bio_hook,
8736         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8737         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8738         .writepage_start_hook = btrfs_writepage_start_hook,
8739         .set_bit_hook = btrfs_set_bit_hook,
8740         .clear_bit_hook = btrfs_clear_bit_hook,
8741         .merge_extent_hook = btrfs_merge_extent_hook,
8742         .split_extent_hook = btrfs_split_extent_hook,
8743 };
8744
8745 /*
8746  * btrfs doesn't support the bmap operation because swapfiles
8747  * use bmap to make a mapping of extents in the file.  They assume
8748  * these extents won't change over the life of the file and they
8749  * use the bmap result to do IO directly to the drive.
8750  *
8751  * the btrfs bmap call would return logical addresses that aren't
8752  * suitable for IO and they also will change frequently as COW
8753  * operations happen.  So, swapfile + btrfs == corruption.
8754  *
8755  * For now we're avoiding this by dropping bmap.
8756  */
8757 static const struct address_space_operations btrfs_aops = {
8758         .readpage       = btrfs_readpage,
8759         .writepage      = btrfs_writepage,
8760         .writepages     = btrfs_writepages,
8761         .readpages      = btrfs_readpages,
8762         .direct_IO      = btrfs_direct_IO,
8763         .invalidatepage = btrfs_invalidatepage,
8764         .releasepage    = btrfs_releasepage,
8765         .set_page_dirty = btrfs_set_page_dirty,
8766         .error_remove_page = generic_error_remove_page,
8767 };
8768
8769 static const struct address_space_operations btrfs_symlink_aops = {
8770         .readpage       = btrfs_readpage,
8771         .writepage      = btrfs_writepage,
8772         .invalidatepage = btrfs_invalidatepage,
8773         .releasepage    = btrfs_releasepage,
8774 };
8775
8776 static const struct inode_operations btrfs_file_inode_operations = {
8777         .getattr        = btrfs_getattr,
8778         .setattr        = btrfs_setattr,
8779         .setxattr       = btrfs_setxattr,
8780         .getxattr       = btrfs_getxattr,
8781         .listxattr      = btrfs_listxattr,
8782         .removexattr    = btrfs_removexattr,
8783         .permission     = btrfs_permission,
8784         .fiemap         = btrfs_fiemap,
8785         .get_acl        = btrfs_get_acl,
8786         .update_time    = btrfs_update_time,
8787 };
8788 static const struct inode_operations btrfs_special_inode_operations = {
8789         .getattr        = btrfs_getattr,
8790         .setattr        = btrfs_setattr,
8791         .permission     = btrfs_permission,
8792         .setxattr       = btrfs_setxattr,
8793         .getxattr       = btrfs_getxattr,
8794         .listxattr      = btrfs_listxattr,
8795         .removexattr    = btrfs_removexattr,
8796         .get_acl        = btrfs_get_acl,
8797         .update_time    = btrfs_update_time,
8798 };
8799 static const struct inode_operations btrfs_symlink_inode_operations = {
8800         .readlink       = generic_readlink,
8801         .follow_link    = page_follow_link_light,
8802         .put_link       = page_put_link,
8803         .getattr        = btrfs_getattr,
8804         .setattr        = btrfs_setattr,
8805         .permission     = btrfs_permission,
8806         .setxattr       = btrfs_setxattr,
8807         .getxattr       = btrfs_getxattr,
8808         .listxattr      = btrfs_listxattr,
8809         .removexattr    = btrfs_removexattr,
8810         .get_acl        = btrfs_get_acl,
8811         .update_time    = btrfs_update_time,
8812 };
8813
8814 const struct dentry_operations btrfs_dentry_operations = {
8815         .d_delete       = btrfs_dentry_delete,
8816         .d_release      = btrfs_dentry_release,
8817 };