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 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1532                                       struct inode *inode)
1533 {
1534         spin_lock(&root->delalloc_lock);
1535         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1536                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1537                               &root->delalloc_inodes);
1538                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1539                         &BTRFS_I(inode)->runtime_flags);
1540                 root->nr_delalloc_inodes++;
1541                 if (root->nr_delalloc_inodes == 1) {
1542                         spin_lock(&root->fs_info->delalloc_root_lock);
1543                         BUG_ON(!list_empty(&root->delalloc_root));
1544                         list_add_tail(&root->delalloc_root,
1545                                       &root->fs_info->delalloc_roots);
1546                         spin_unlock(&root->fs_info->delalloc_root_lock);
1547                 }
1548         }
1549         spin_unlock(&root->delalloc_lock);
1550 }
1551
1552 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1553                                      struct inode *inode)
1554 {
1555         spin_lock(&root->delalloc_lock);
1556         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1557                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1558                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1559                           &BTRFS_I(inode)->runtime_flags);
1560                 root->nr_delalloc_inodes--;
1561                 if (!root->nr_delalloc_inodes) {
1562                         spin_lock(&root->fs_info->delalloc_root_lock);
1563                         BUG_ON(list_empty(&root->delalloc_root));
1564                         list_del_init(&root->delalloc_root);
1565                         spin_unlock(&root->fs_info->delalloc_root_lock);
1566                 }
1567         }
1568         spin_unlock(&root->delalloc_lock);
1569 }
1570
1571 /*
1572  * extent_io.c set_bit_hook, used to track delayed allocation
1573  * bytes in this file, and to maintain the list of inodes that
1574  * have pending delalloc work to be done.
1575  */
1576 static void btrfs_set_bit_hook(struct inode *inode,
1577                                struct extent_state *state, unsigned long *bits)
1578 {
1579
1580         /*
1581          * set_bit and clear bit hooks normally require _irqsave/restore
1582          * but in this case, we are only testing for the DELALLOC
1583          * bit, which is only set or cleared with irqs on
1584          */
1585         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1586                 struct btrfs_root *root = BTRFS_I(inode)->root;
1587                 u64 len = state->end + 1 - state->start;
1588                 bool do_list = !btrfs_is_free_space_inode(inode);
1589
1590                 if (*bits & EXTENT_FIRST_DELALLOC) {
1591                         *bits &= ~EXTENT_FIRST_DELALLOC;
1592                 } else {
1593                         spin_lock(&BTRFS_I(inode)->lock);
1594                         BTRFS_I(inode)->outstanding_extents++;
1595                         spin_unlock(&BTRFS_I(inode)->lock);
1596                 }
1597
1598                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1599                                      root->fs_info->delalloc_batch);
1600                 spin_lock(&BTRFS_I(inode)->lock);
1601                 BTRFS_I(inode)->delalloc_bytes += len;
1602                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1603                                          &BTRFS_I(inode)->runtime_flags))
1604                         btrfs_add_delalloc_inodes(root, inode);
1605                 spin_unlock(&BTRFS_I(inode)->lock);
1606         }
1607 }
1608
1609 /*
1610  * extent_io.c clear_bit_hook, see set_bit_hook for why
1611  */
1612 static void btrfs_clear_bit_hook(struct inode *inode,
1613                                  struct extent_state *state,
1614                                  unsigned long *bits)
1615 {
1616         /*
1617          * set_bit and clear bit hooks normally require _irqsave/restore
1618          * but in this case, we are only testing for the DELALLOC
1619          * bit, which is only set or cleared with irqs on
1620          */
1621         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1622                 struct btrfs_root *root = BTRFS_I(inode)->root;
1623                 u64 len = state->end + 1 - state->start;
1624                 bool do_list = !btrfs_is_free_space_inode(inode);
1625
1626                 if (*bits & EXTENT_FIRST_DELALLOC) {
1627                         *bits &= ~EXTENT_FIRST_DELALLOC;
1628                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1629                         spin_lock(&BTRFS_I(inode)->lock);
1630                         BTRFS_I(inode)->outstanding_extents--;
1631                         spin_unlock(&BTRFS_I(inode)->lock);
1632                 }
1633
1634                 if (*bits & EXTENT_DO_ACCOUNTING)
1635                         btrfs_delalloc_release_metadata(inode, len);
1636
1637                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1638                     && do_list)
1639                         btrfs_free_reserved_data_space(inode, len);
1640
1641                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1642                                      root->fs_info->delalloc_batch);
1643                 spin_lock(&BTRFS_I(inode)->lock);
1644                 BTRFS_I(inode)->delalloc_bytes -= len;
1645                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1646                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1647                              &BTRFS_I(inode)->runtime_flags))
1648                         btrfs_del_delalloc_inode(root, inode);
1649                 spin_unlock(&BTRFS_I(inode)->lock);
1650         }
1651 }
1652
1653 /*
1654  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1655  * we don't create bios that span stripes or chunks
1656  */
1657 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1658                          size_t size, struct bio *bio,
1659                          unsigned long bio_flags)
1660 {
1661         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1662         u64 logical = (u64)bio->bi_sector << 9;
1663         u64 length = 0;
1664         u64 map_length;
1665         int ret;
1666
1667         if (bio_flags & EXTENT_BIO_COMPRESSED)
1668                 return 0;
1669
1670         length = bio->bi_size;
1671         map_length = length;
1672         ret = btrfs_map_block(root->fs_info, rw, logical,
1673                               &map_length, NULL, 0);
1674         /* Will always return 0 with map_multi == NULL */
1675         BUG_ON(ret < 0);
1676         if (map_length < length + size)
1677                 return 1;
1678         return 0;
1679 }
1680
1681 /*
1682  * in order to insert checksums into the metadata in large chunks,
1683  * we wait until bio submission time.   All the pages in the bio are
1684  * checksummed and sums are attached onto the ordered extent record.
1685  *
1686  * At IO completion time the cums attached on the ordered extent record
1687  * are inserted into the btree
1688  */
1689 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1690                                     struct bio *bio, int mirror_num,
1691                                     unsigned long bio_flags,
1692                                     u64 bio_offset)
1693 {
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         int ret = 0;
1696
1697         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1698         BUG_ON(ret); /* -ENOMEM */
1699         return 0;
1700 }
1701
1702 /*
1703  * in order to insert checksums into the metadata in large chunks,
1704  * we wait until bio submission time.   All the pages in the bio are
1705  * checksummed and sums are attached onto the ordered extent record.
1706  *
1707  * At IO completion time the cums attached on the ordered extent record
1708  * are inserted into the btree
1709  */
1710 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1711                           int mirror_num, unsigned long bio_flags,
1712                           u64 bio_offset)
1713 {
1714         struct btrfs_root *root = BTRFS_I(inode)->root;
1715         int ret;
1716
1717         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1718         if (ret)
1719                 bio_endio(bio, ret);
1720         return ret;
1721 }
1722
1723 /*
1724  * extent_io.c submission hook. This does the right thing for csum calculation
1725  * on write, or reading the csums from the tree before a read
1726  */
1727 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1728                           int mirror_num, unsigned long bio_flags,
1729                           u64 bio_offset)
1730 {
1731         struct btrfs_root *root = BTRFS_I(inode)->root;
1732         int ret = 0;
1733         int skip_sum;
1734         int metadata = 0;
1735         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1736
1737         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1738
1739         if (btrfs_is_free_space_inode(inode))
1740                 metadata = 2;
1741
1742         if (!(rw & REQ_WRITE)) {
1743                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1744                 if (ret)
1745                         goto out;
1746
1747                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1748                         ret = btrfs_submit_compressed_read(inode, bio,
1749                                                            mirror_num,
1750                                                            bio_flags);
1751                         goto out;
1752                 } else if (!skip_sum) {
1753                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1754                         if (ret)
1755                                 goto out;
1756                 }
1757                 goto mapit;
1758         } else if (async && !skip_sum) {
1759                 /* csum items have already been cloned */
1760                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1761                         goto mapit;
1762                 /* we're doing a write, do the async checksumming */
1763                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1764                                    inode, rw, bio, mirror_num,
1765                                    bio_flags, bio_offset,
1766                                    __btrfs_submit_bio_start,
1767                                    __btrfs_submit_bio_done);
1768                 goto out;
1769         } else if (!skip_sum) {
1770                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1771                 if (ret)
1772                         goto out;
1773         }
1774
1775 mapit:
1776         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1777
1778 out:
1779         if (ret < 0)
1780                 bio_endio(bio, ret);
1781         return ret;
1782 }
1783
1784 /*
1785  * given a list of ordered sums record them in the inode.  This happens
1786  * at IO completion time based on sums calculated at bio submission time.
1787  */
1788 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1789                              struct inode *inode, u64 file_offset,
1790                              struct list_head *list)
1791 {
1792         struct btrfs_ordered_sum *sum;
1793
1794         list_for_each_entry(sum, list, list) {
1795                 trans->adding_csums = 1;
1796                 btrfs_csum_file_blocks(trans,
1797                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1798                 trans->adding_csums = 0;
1799         }
1800         return 0;
1801 }
1802
1803 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1804                               struct extent_state **cached_state)
1805 {
1806         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1807         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1808                                    cached_state, GFP_NOFS);
1809 }
1810
1811 /* see btrfs_writepage_start_hook for details on why this is required */
1812 struct btrfs_writepage_fixup {
1813         struct page *page;
1814         struct btrfs_work work;
1815 };
1816
1817 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1818 {
1819         struct btrfs_writepage_fixup *fixup;
1820         struct btrfs_ordered_extent *ordered;
1821         struct extent_state *cached_state = NULL;
1822         struct page *page;
1823         struct inode *inode;
1824         u64 page_start;
1825         u64 page_end;
1826         int ret;
1827
1828         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1829         page = fixup->page;
1830 again:
1831         lock_page(page);
1832         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1833                 ClearPageChecked(page);
1834                 goto out_page;
1835         }
1836
1837         inode = page->mapping->host;
1838         page_start = page_offset(page);
1839         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1840
1841         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1842                          &cached_state);
1843
1844         /* already ordered? We're done */
1845         if (PagePrivate2(page))
1846                 goto out;
1847
1848         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1849         if (ordered) {
1850                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1851                                      page_end, &cached_state, GFP_NOFS);
1852                 unlock_page(page);
1853                 btrfs_start_ordered_extent(inode, ordered, 1);
1854                 btrfs_put_ordered_extent(ordered);
1855                 goto again;
1856         }
1857
1858         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1859         if (ret) {
1860                 mapping_set_error(page->mapping, ret);
1861                 end_extent_writepage(page, ret, page_start, page_end);
1862                 ClearPageChecked(page);
1863                 goto out;
1864          }
1865
1866         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1867         ClearPageChecked(page);
1868         set_page_dirty(page);
1869 out:
1870         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1871                              &cached_state, GFP_NOFS);
1872 out_page:
1873         unlock_page(page);
1874         page_cache_release(page);
1875         kfree(fixup);
1876 }
1877
1878 /*
1879  * There are a few paths in the higher layers of the kernel that directly
1880  * set the page dirty bit without asking the filesystem if it is a
1881  * good idea.  This causes problems because we want to make sure COW
1882  * properly happens and the data=ordered rules are followed.
1883  *
1884  * In our case any range that doesn't have the ORDERED bit set
1885  * hasn't been properly setup for IO.  We kick off an async process
1886  * to fix it up.  The async helper will wait for ordered extents, set
1887  * the delalloc bit and make it safe to write the page.
1888  */
1889 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1890 {
1891         struct inode *inode = page->mapping->host;
1892         struct btrfs_writepage_fixup *fixup;
1893         struct btrfs_root *root = BTRFS_I(inode)->root;
1894
1895         /* this page is properly in the ordered list */
1896         if (TestClearPagePrivate2(page))
1897                 return 0;
1898
1899         if (PageChecked(page))
1900                 return -EAGAIN;
1901
1902         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1903         if (!fixup)
1904                 return -EAGAIN;
1905
1906         SetPageChecked(page);
1907         page_cache_get(page);
1908         fixup->work.func = btrfs_writepage_fixup_worker;
1909         fixup->page = page;
1910         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1911         return -EBUSY;
1912 }
1913
1914 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1915                                        struct inode *inode, u64 file_pos,
1916                                        u64 disk_bytenr, u64 disk_num_bytes,
1917                                        u64 num_bytes, u64 ram_bytes,
1918                                        u8 compression, u8 encryption,
1919                                        u16 other_encoding, int extent_type)
1920 {
1921         struct btrfs_root *root = BTRFS_I(inode)->root;
1922         struct btrfs_file_extent_item *fi;
1923         struct btrfs_path *path;
1924         struct extent_buffer *leaf;
1925         struct btrfs_key ins;
1926         int ret;
1927
1928         path = btrfs_alloc_path();
1929         if (!path)
1930                 return -ENOMEM;
1931
1932         path->leave_spinning = 1;
1933
1934         /*
1935          * we may be replacing one extent in the tree with another.
1936          * The new extent is pinned in the extent map, and we don't want
1937          * to drop it from the cache until it is completely in the btree.
1938          *
1939          * So, tell btrfs_drop_extents to leave this extent in the cache.
1940          * the caller is expected to unpin it and allow it to be merged
1941          * with the others.
1942          */
1943         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1944                                  file_pos + num_bytes, 0);
1945         if (ret)
1946                 goto out;
1947
1948         ins.objectid = btrfs_ino(inode);
1949         ins.offset = file_pos;
1950         ins.type = BTRFS_EXTENT_DATA_KEY;
1951         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1952         if (ret)
1953                 goto out;
1954         leaf = path->nodes[0];
1955         fi = btrfs_item_ptr(leaf, path->slots[0],
1956                             struct btrfs_file_extent_item);
1957         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1958         btrfs_set_file_extent_type(leaf, fi, extent_type);
1959         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1960         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1961         btrfs_set_file_extent_offset(leaf, fi, 0);
1962         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1963         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1964         btrfs_set_file_extent_compression(leaf, fi, compression);
1965         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1966         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1967
1968         btrfs_mark_buffer_dirty(leaf);
1969         btrfs_release_path(path);
1970
1971         inode_add_bytes(inode, num_bytes);
1972
1973         ins.objectid = disk_bytenr;
1974         ins.offset = disk_num_bytes;
1975         ins.type = BTRFS_EXTENT_ITEM_KEY;
1976         ret = btrfs_alloc_reserved_file_extent(trans, root,
1977                                         root->root_key.objectid,
1978                                         btrfs_ino(inode), file_pos, &ins);
1979 out:
1980         btrfs_free_path(path);
1981
1982         return ret;
1983 }
1984
1985 /* snapshot-aware defrag */
1986 struct sa_defrag_extent_backref {
1987         struct rb_node node;
1988         struct old_sa_defrag_extent *old;
1989         u64 root_id;
1990         u64 inum;
1991         u64 file_pos;
1992         u64 extent_offset;
1993         u64 num_bytes;
1994         u64 generation;
1995 };
1996
1997 struct old_sa_defrag_extent {
1998         struct list_head list;
1999         struct new_sa_defrag_extent *new;
2000
2001         u64 extent_offset;
2002         u64 bytenr;
2003         u64 offset;
2004         u64 len;
2005         int count;
2006 };
2007
2008 struct new_sa_defrag_extent {
2009         struct rb_root root;
2010         struct list_head head;
2011         struct btrfs_path *path;
2012         struct inode *inode;
2013         u64 file_pos;
2014         u64 len;
2015         u64 bytenr;
2016         u64 disk_len;
2017         u8 compress_type;
2018 };
2019
2020 static int backref_comp(struct sa_defrag_extent_backref *b1,
2021                         struct sa_defrag_extent_backref *b2)
2022 {
2023         if (b1->root_id < b2->root_id)
2024                 return -1;
2025         else if (b1->root_id > b2->root_id)
2026                 return 1;
2027
2028         if (b1->inum < b2->inum)
2029                 return -1;
2030         else if (b1->inum > b2->inum)
2031                 return 1;
2032
2033         if (b1->file_pos < b2->file_pos)
2034                 return -1;
2035         else if (b1->file_pos > b2->file_pos)
2036                 return 1;
2037
2038         /*
2039          * [------------------------------] ===> (a range of space)
2040          *     |<--->|   |<---->| =============> (fs/file tree A)
2041          * |<---------------------------->| ===> (fs/file tree B)
2042          *
2043          * A range of space can refer to two file extents in one tree while
2044          * refer to only one file extent in another tree.
2045          *
2046          * So we may process a disk offset more than one time(two extents in A)
2047          * and locate at the same extent(one extent in B), then insert two same
2048          * backrefs(both refer to the extent in B).
2049          */
2050         return 0;
2051 }
2052
2053 static void backref_insert(struct rb_root *root,
2054                            struct sa_defrag_extent_backref *backref)
2055 {
2056         struct rb_node **p = &root->rb_node;
2057         struct rb_node *parent = NULL;
2058         struct sa_defrag_extent_backref *entry;
2059         int ret;
2060
2061         while (*p) {
2062                 parent = *p;
2063                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2064
2065                 ret = backref_comp(backref, entry);
2066                 if (ret < 0)
2067                         p = &(*p)->rb_left;
2068                 else
2069                         p = &(*p)->rb_right;
2070         }
2071
2072         rb_link_node(&backref->node, parent, p);
2073         rb_insert_color(&backref->node, root);
2074 }
2075
2076 /*
2077  * Note the backref might has changed, and in this case we just return 0.
2078  */
2079 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2080                                        void *ctx)
2081 {
2082         struct btrfs_file_extent_item *extent;
2083         struct btrfs_fs_info *fs_info;
2084         struct old_sa_defrag_extent *old = ctx;
2085         struct new_sa_defrag_extent *new = old->new;
2086         struct btrfs_path *path = new->path;
2087         struct btrfs_key key;
2088         struct btrfs_root *root;
2089         struct sa_defrag_extent_backref *backref;
2090         struct extent_buffer *leaf;
2091         struct inode *inode = new->inode;
2092         int slot;
2093         int ret;
2094         u64 extent_offset;
2095         u64 num_bytes;
2096
2097         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2098             inum == btrfs_ino(inode))
2099                 return 0;
2100
2101         key.objectid = root_id;
2102         key.type = BTRFS_ROOT_ITEM_KEY;
2103         key.offset = (u64)-1;
2104
2105         fs_info = BTRFS_I(inode)->root->fs_info;
2106         root = btrfs_read_fs_root_no_name(fs_info, &key);
2107         if (IS_ERR(root)) {
2108                 if (PTR_ERR(root) == -ENOENT)
2109                         return 0;
2110                 WARN_ON(1);
2111                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2112                          inum, offset, root_id);
2113                 return PTR_ERR(root);
2114         }
2115
2116         key.objectid = inum;
2117         key.type = BTRFS_EXTENT_DATA_KEY;
2118         if (offset > (u64)-1 << 32)
2119                 key.offset = 0;
2120         else
2121                 key.offset = offset;
2122
2123         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2124         if (ret < 0) {
2125                 WARN_ON(1);
2126                 return ret;
2127         }
2128
2129         while (1) {
2130                 cond_resched();
2131
2132                 leaf = path->nodes[0];
2133                 slot = path->slots[0];
2134
2135                 if (slot >= btrfs_header_nritems(leaf)) {
2136                         ret = btrfs_next_leaf(root, path);
2137                         if (ret < 0) {
2138                                 goto out;
2139                         } else if (ret > 0) {
2140                                 ret = 0;
2141                                 goto out;
2142                         }
2143                         continue;
2144                 }
2145
2146                 path->slots[0]++;
2147
2148                 btrfs_item_key_to_cpu(leaf, &key, slot);
2149
2150                 if (key.objectid > inum)
2151                         goto out;
2152
2153                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2154                         continue;
2155
2156                 extent = btrfs_item_ptr(leaf, slot,
2157                                         struct btrfs_file_extent_item);
2158
2159                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2160                         continue;
2161
2162                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2163                 if (key.offset - extent_offset != offset)
2164                         continue;
2165
2166                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2167                 if (extent_offset >= old->extent_offset + old->offset +
2168                     old->len || extent_offset + num_bytes <=
2169                     old->extent_offset + old->offset)
2170                         continue;
2171
2172                 break;
2173         }
2174
2175         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2176         if (!backref) {
2177                 ret = -ENOENT;
2178                 goto out;
2179         }
2180
2181         backref->root_id = root_id;
2182         backref->inum = inum;
2183         backref->file_pos = offset + extent_offset;
2184         backref->num_bytes = num_bytes;
2185         backref->extent_offset = extent_offset;
2186         backref->generation = btrfs_file_extent_generation(leaf, extent);
2187         backref->old = old;
2188         backref_insert(&new->root, backref);
2189         old->count++;
2190 out:
2191         btrfs_release_path(path);
2192         WARN_ON(ret);
2193         return ret;
2194 }
2195
2196 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2197                                    struct new_sa_defrag_extent *new)
2198 {
2199         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2200         struct old_sa_defrag_extent *old, *tmp;
2201         int ret;
2202
2203         new->path = path;
2204
2205         list_for_each_entry_safe(old, tmp, &new->head, list) {
2206                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2207                                                   path, record_one_backref,
2208                                                   old);
2209                 BUG_ON(ret < 0 && ret != -ENOENT);
2210
2211                 /* no backref to be processed for this extent */
2212                 if (!old->count) {
2213                         list_del(&old->list);
2214                         kfree(old);
2215                 }
2216         }
2217
2218         if (list_empty(&new->head))
2219                 return false;
2220
2221         return true;
2222 }
2223
2224 static int relink_is_mergable(struct extent_buffer *leaf,
2225                               struct btrfs_file_extent_item *fi,
2226                               u64 disk_bytenr)
2227 {
2228         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2229                 return 0;
2230
2231         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2232                 return 0;
2233
2234         if (btrfs_file_extent_compression(leaf, fi) ||
2235             btrfs_file_extent_encryption(leaf, fi) ||
2236             btrfs_file_extent_other_encoding(leaf, fi))
2237                 return 0;
2238
2239         return 1;
2240 }
2241
2242 /*
2243  * Note the backref might has changed, and in this case we just return 0.
2244  */
2245 static noinline int relink_extent_backref(struct btrfs_path *path,
2246                                  struct sa_defrag_extent_backref *prev,
2247                                  struct sa_defrag_extent_backref *backref)
2248 {
2249         struct btrfs_file_extent_item *extent;
2250         struct btrfs_file_extent_item *item;
2251         struct btrfs_ordered_extent *ordered;
2252         struct btrfs_trans_handle *trans;
2253         struct btrfs_fs_info *fs_info;
2254         struct btrfs_root *root;
2255         struct btrfs_key key;
2256         struct extent_buffer *leaf;
2257         struct old_sa_defrag_extent *old = backref->old;
2258         struct new_sa_defrag_extent *new = old->new;
2259         struct inode *src_inode = new->inode;
2260         struct inode *inode;
2261         struct extent_state *cached = NULL;
2262         int ret = 0;
2263         u64 start;
2264         u64 len;
2265         u64 lock_start;
2266         u64 lock_end;
2267         bool merge = false;
2268         int index;
2269
2270         if (prev && prev->root_id == backref->root_id &&
2271             prev->inum == backref->inum &&
2272             prev->file_pos + prev->num_bytes == backref->file_pos)
2273                 merge = true;
2274
2275         /* step 1: get root */
2276         key.objectid = backref->root_id;
2277         key.type = BTRFS_ROOT_ITEM_KEY;
2278         key.offset = (u64)-1;
2279
2280         fs_info = BTRFS_I(src_inode)->root->fs_info;
2281         index = srcu_read_lock(&fs_info->subvol_srcu);
2282
2283         root = btrfs_read_fs_root_no_name(fs_info, &key);
2284         if (IS_ERR(root)) {
2285                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2286                 if (PTR_ERR(root) == -ENOENT)
2287                         return 0;
2288                 return PTR_ERR(root);
2289         }
2290
2291         /* step 2: get inode */
2292         key.objectid = backref->inum;
2293         key.type = BTRFS_INODE_ITEM_KEY;
2294         key.offset = 0;
2295
2296         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2297         if (IS_ERR(inode)) {
2298                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2299                 return 0;
2300         }
2301
2302         srcu_read_unlock(&fs_info->subvol_srcu, index);
2303
2304         /* step 3: relink backref */
2305         lock_start = backref->file_pos;
2306         lock_end = backref->file_pos + backref->num_bytes - 1;
2307         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2308                          0, &cached);
2309
2310         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2311         if (ordered) {
2312                 btrfs_put_ordered_extent(ordered);
2313                 goto out_unlock;
2314         }
2315
2316         trans = btrfs_join_transaction(root);
2317         if (IS_ERR(trans)) {
2318                 ret = PTR_ERR(trans);
2319                 goto out_unlock;
2320         }
2321
2322         key.objectid = backref->inum;
2323         key.type = BTRFS_EXTENT_DATA_KEY;
2324         key.offset = backref->file_pos;
2325
2326         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2327         if (ret < 0) {
2328                 goto out_free_path;
2329         } else if (ret > 0) {
2330                 ret = 0;
2331                 goto out_free_path;
2332         }
2333
2334         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2335                                 struct btrfs_file_extent_item);
2336
2337         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2338             backref->generation)
2339                 goto out_free_path;
2340
2341         btrfs_release_path(path);
2342
2343         start = backref->file_pos;
2344         if (backref->extent_offset < old->extent_offset + old->offset)
2345                 start += old->extent_offset + old->offset -
2346                          backref->extent_offset;
2347
2348         len = min(backref->extent_offset + backref->num_bytes,
2349                   old->extent_offset + old->offset + old->len);
2350         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2351
2352         ret = btrfs_drop_extents(trans, root, inode, start,
2353                                  start + len, 1);
2354         if (ret)
2355                 goto out_free_path;
2356 again:
2357         key.objectid = btrfs_ino(inode);
2358         key.type = BTRFS_EXTENT_DATA_KEY;
2359         key.offset = start;
2360
2361         path->leave_spinning = 1;
2362         if (merge) {
2363                 struct btrfs_file_extent_item *fi;
2364                 u64 extent_len;
2365                 struct btrfs_key found_key;
2366
2367                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2368                 if (ret < 0)
2369                         goto out_free_path;
2370
2371                 path->slots[0]--;
2372                 leaf = path->nodes[0];
2373                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2374
2375                 fi = btrfs_item_ptr(leaf, path->slots[0],
2376                                     struct btrfs_file_extent_item);
2377                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2378
2379                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2380                     extent_len + found_key.offset == start) {
2381                         btrfs_set_file_extent_num_bytes(leaf, fi,
2382                                                         extent_len + len);
2383                         btrfs_mark_buffer_dirty(leaf);
2384                         inode_add_bytes(inode, len);
2385
2386                         ret = 1;
2387                         goto out_free_path;
2388                 } else {
2389                         merge = false;
2390                         btrfs_release_path(path);
2391                         goto again;
2392                 }
2393         }
2394
2395         ret = btrfs_insert_empty_item(trans, root, path, &key,
2396                                         sizeof(*extent));
2397         if (ret) {
2398                 btrfs_abort_transaction(trans, root, ret);
2399                 goto out_free_path;
2400         }
2401
2402         leaf = path->nodes[0];
2403         item = btrfs_item_ptr(leaf, path->slots[0],
2404                                 struct btrfs_file_extent_item);
2405         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2406         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2407         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2408         btrfs_set_file_extent_num_bytes(leaf, item, len);
2409         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2410         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2411         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2412         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2413         btrfs_set_file_extent_encryption(leaf, item, 0);
2414         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2415
2416         btrfs_mark_buffer_dirty(leaf);
2417         inode_add_bytes(inode, len);
2418         btrfs_release_path(path);
2419
2420         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2421                         new->disk_len, 0,
2422                         backref->root_id, backref->inum,
2423                         new->file_pos, 0);      /* start - extent_offset */
2424         if (ret) {
2425                 btrfs_abort_transaction(trans, root, ret);
2426                 goto out_free_path;
2427         }
2428
2429         ret = 1;
2430 out_free_path:
2431         btrfs_release_path(path);
2432         path->leave_spinning = 0;
2433         btrfs_end_transaction(trans, root);
2434 out_unlock:
2435         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2436                              &cached, GFP_NOFS);
2437         iput(inode);
2438         return ret;
2439 }
2440
2441 static void relink_file_extents(struct new_sa_defrag_extent *new)
2442 {
2443         struct btrfs_path *path;
2444         struct old_sa_defrag_extent *old, *tmp;
2445         struct sa_defrag_extent_backref *backref;
2446         struct sa_defrag_extent_backref *prev = NULL;
2447         struct inode *inode;
2448         struct btrfs_root *root;
2449         struct rb_node *node;
2450         int ret;
2451
2452         inode = new->inode;
2453         root = BTRFS_I(inode)->root;
2454
2455         path = btrfs_alloc_path();
2456         if (!path)
2457                 return;
2458
2459         if (!record_extent_backrefs(path, new)) {
2460                 btrfs_free_path(path);
2461                 goto out;
2462         }
2463         btrfs_release_path(path);
2464
2465         while (1) {
2466                 node = rb_first(&new->root);
2467                 if (!node)
2468                         break;
2469                 rb_erase(node, &new->root);
2470
2471                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2472
2473                 ret = relink_extent_backref(path, prev, backref);
2474                 WARN_ON(ret < 0);
2475
2476                 kfree(prev);
2477
2478                 if (ret == 1)
2479                         prev = backref;
2480                 else
2481                         prev = NULL;
2482                 cond_resched();
2483         }
2484         kfree(prev);
2485
2486         btrfs_free_path(path);
2487
2488         list_for_each_entry_safe(old, tmp, &new->head, list) {
2489                 list_del(&old->list);
2490                 kfree(old);
2491         }
2492 out:
2493         atomic_dec(&root->fs_info->defrag_running);
2494         wake_up(&root->fs_info->transaction_wait);
2495
2496         kfree(new);
2497 }
2498
2499 static struct new_sa_defrag_extent *
2500 record_old_file_extents(struct inode *inode,
2501                         struct btrfs_ordered_extent *ordered)
2502 {
2503         struct btrfs_root *root = BTRFS_I(inode)->root;
2504         struct btrfs_path *path;
2505         struct btrfs_key key;
2506         struct old_sa_defrag_extent *old, *tmp;
2507         struct new_sa_defrag_extent *new;
2508         int ret;
2509
2510         new = kmalloc(sizeof(*new), GFP_NOFS);
2511         if (!new)
2512                 return NULL;
2513
2514         new->inode = inode;
2515         new->file_pos = ordered->file_offset;
2516         new->len = ordered->len;
2517         new->bytenr = ordered->start;
2518         new->disk_len = ordered->disk_len;
2519         new->compress_type = ordered->compress_type;
2520         new->root = RB_ROOT;
2521         INIT_LIST_HEAD(&new->head);
2522
2523         path = btrfs_alloc_path();
2524         if (!path)
2525                 goto out_kfree;
2526
2527         key.objectid = btrfs_ino(inode);
2528         key.type = BTRFS_EXTENT_DATA_KEY;
2529         key.offset = new->file_pos;
2530
2531         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2532         if (ret < 0)
2533                 goto out_free_path;
2534         if (ret > 0 && path->slots[0] > 0)
2535                 path->slots[0]--;
2536
2537         /* find out all the old extents for the file range */
2538         while (1) {
2539                 struct btrfs_file_extent_item *extent;
2540                 struct extent_buffer *l;
2541                 int slot;
2542                 u64 num_bytes;
2543                 u64 offset;
2544                 u64 end;
2545                 u64 disk_bytenr;
2546                 u64 extent_offset;
2547
2548                 l = path->nodes[0];
2549                 slot = path->slots[0];
2550
2551                 if (slot >= btrfs_header_nritems(l)) {
2552                         ret = btrfs_next_leaf(root, path);
2553                         if (ret < 0)
2554                                 goto out_free_list;
2555                         else if (ret > 0)
2556                                 break;
2557                         continue;
2558                 }
2559
2560                 btrfs_item_key_to_cpu(l, &key, slot);
2561
2562                 if (key.objectid != btrfs_ino(inode))
2563                         break;
2564                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2565                         break;
2566                 if (key.offset >= new->file_pos + new->len)
2567                         break;
2568
2569                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2570
2571                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2572                 if (key.offset + num_bytes < new->file_pos)
2573                         goto next;
2574
2575                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2576                 if (!disk_bytenr)
2577                         goto next;
2578
2579                 extent_offset = btrfs_file_extent_offset(l, extent);
2580
2581                 old = kmalloc(sizeof(*old), GFP_NOFS);
2582                 if (!old)
2583                         goto out_free_list;
2584
2585                 offset = max(new->file_pos, key.offset);
2586                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2587
2588                 old->bytenr = disk_bytenr;
2589                 old->extent_offset = extent_offset;
2590                 old->offset = offset - key.offset;
2591                 old->len = end - offset;
2592                 old->new = new;
2593                 old->count = 0;
2594                 list_add_tail(&old->list, &new->head);
2595 next:
2596                 path->slots[0]++;
2597                 cond_resched();
2598         }
2599
2600         btrfs_free_path(path);
2601         atomic_inc(&root->fs_info->defrag_running);
2602
2603         return new;
2604
2605 out_free_list:
2606         list_for_each_entry_safe(old, tmp, &new->head, list) {
2607                 list_del(&old->list);
2608                 kfree(old);
2609         }
2610 out_free_path:
2611         btrfs_free_path(path);
2612 out_kfree:
2613         kfree(new);
2614         return NULL;
2615 }
2616
2617 /*
2618  * helper function for btrfs_finish_ordered_io, this
2619  * just reads in some of the csum leaves to prime them into ram
2620  * before we start the transaction.  It limits the amount of btree
2621  * reads required while inside the transaction.
2622  */
2623 /* as ordered data IO finishes, this gets called so we can finish
2624  * an ordered extent if the range of bytes in the file it covers are
2625  * fully written.
2626  */
2627 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2628 {
2629         struct inode *inode = ordered_extent->inode;
2630         struct btrfs_root *root = BTRFS_I(inode)->root;
2631         struct btrfs_trans_handle *trans = NULL;
2632         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2633         struct extent_state *cached_state = NULL;
2634         struct new_sa_defrag_extent *new = NULL;
2635         int compress_type = 0;
2636         int ret;
2637         bool nolock;
2638
2639         nolock = btrfs_is_free_space_inode(inode);
2640
2641         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2642                 ret = -EIO;
2643                 goto out;
2644         }
2645
2646         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2647                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2648                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2649                 if (nolock)
2650                         trans = btrfs_join_transaction_nolock(root);
2651                 else
2652                         trans = btrfs_join_transaction(root);
2653                 if (IS_ERR(trans)) {
2654                         ret = PTR_ERR(trans);
2655                         trans = NULL;
2656                         goto out;
2657                 }
2658                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2659                 ret = btrfs_update_inode_fallback(trans, root, inode);
2660                 if (ret) /* -ENOMEM or corruption */
2661                         btrfs_abort_transaction(trans, root, ret);
2662                 goto out;
2663         }
2664
2665         lock_extent_bits(io_tree, ordered_extent->file_offset,
2666                          ordered_extent->file_offset + ordered_extent->len - 1,
2667                          0, &cached_state);
2668
2669         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2670                         ordered_extent->file_offset + ordered_extent->len - 1,
2671                         EXTENT_DEFRAG, 1, cached_state);
2672         if (ret) {
2673                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2674                 if (last_snapshot >= BTRFS_I(inode)->generation)
2675                         /* the inode is shared */
2676                         new = record_old_file_extents(inode, ordered_extent);
2677
2678                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2679                         ordered_extent->file_offset + ordered_extent->len - 1,
2680                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2681         }
2682
2683         if (nolock)
2684                 trans = btrfs_join_transaction_nolock(root);
2685         else
2686                 trans = btrfs_join_transaction(root);
2687         if (IS_ERR(trans)) {
2688                 ret = PTR_ERR(trans);
2689                 trans = NULL;
2690                 goto out_unlock;
2691         }
2692         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2693
2694         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2695                 compress_type = ordered_extent->compress_type;
2696         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2697                 BUG_ON(compress_type);
2698                 ret = btrfs_mark_extent_written(trans, inode,
2699                                                 ordered_extent->file_offset,
2700                                                 ordered_extent->file_offset +
2701                                                 ordered_extent->len);
2702         } else {
2703                 BUG_ON(root == root->fs_info->tree_root);
2704                 ret = insert_reserved_file_extent(trans, inode,
2705                                                 ordered_extent->file_offset,
2706                                                 ordered_extent->start,
2707                                                 ordered_extent->disk_len,
2708                                                 ordered_extent->len,
2709                                                 ordered_extent->len,
2710                                                 compress_type, 0, 0,
2711                                                 BTRFS_FILE_EXTENT_REG);
2712         }
2713         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2714                            ordered_extent->file_offset, ordered_extent->len,
2715                            trans->transid);
2716         if (ret < 0) {
2717                 btrfs_abort_transaction(trans, root, ret);
2718                 goto out_unlock;
2719         }
2720
2721         add_pending_csums(trans, inode, ordered_extent->file_offset,
2722                           &ordered_extent->list);
2723
2724         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2725         ret = btrfs_update_inode_fallback(trans, root, inode);
2726         if (ret) { /* -ENOMEM or corruption */
2727                 btrfs_abort_transaction(trans, root, ret);
2728                 goto out_unlock;
2729         }
2730         ret = 0;
2731 out_unlock:
2732         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2733                              ordered_extent->file_offset +
2734                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2735 out:
2736         if (root != root->fs_info->tree_root)
2737                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2738         if (trans)
2739                 btrfs_end_transaction(trans, root);
2740
2741         if (ret) {
2742                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2743                                       ordered_extent->file_offset +
2744                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2745
2746                 /*
2747                  * If the ordered extent had an IOERR or something else went
2748                  * wrong we need to return the space for this ordered extent
2749                  * back to the allocator.
2750                  */
2751                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2752                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2753                         btrfs_free_reserved_extent(root, ordered_extent->start,
2754                                                    ordered_extent->disk_len);
2755         }
2756
2757
2758         /*
2759          * This needs to be done to make sure anybody waiting knows we are done
2760          * updating everything for this ordered extent.
2761          */
2762         btrfs_remove_ordered_extent(inode, ordered_extent);
2763
2764         /* for snapshot-aware defrag */
2765         if (new)
2766                 relink_file_extents(new);
2767
2768         /* once for us */
2769         btrfs_put_ordered_extent(ordered_extent);
2770         /* once for the tree */
2771         btrfs_put_ordered_extent(ordered_extent);
2772
2773         return ret;
2774 }
2775
2776 static void finish_ordered_fn(struct btrfs_work *work)
2777 {
2778         struct btrfs_ordered_extent *ordered_extent;
2779         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2780         btrfs_finish_ordered_io(ordered_extent);
2781 }
2782
2783 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2784                                 struct extent_state *state, int uptodate)
2785 {
2786         struct inode *inode = page->mapping->host;
2787         struct btrfs_root *root = BTRFS_I(inode)->root;
2788         struct btrfs_ordered_extent *ordered_extent = NULL;
2789         struct btrfs_workers *workers;
2790
2791         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2792
2793         ClearPagePrivate2(page);
2794         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2795                                             end - start + 1, uptodate))
2796                 return 0;
2797
2798         ordered_extent->work.func = finish_ordered_fn;
2799         ordered_extent->work.flags = 0;
2800
2801         if (btrfs_is_free_space_inode(inode))
2802                 workers = &root->fs_info->endio_freespace_worker;
2803         else
2804                 workers = &root->fs_info->endio_write_workers;
2805         btrfs_queue_worker(workers, &ordered_extent->work);
2806
2807         return 0;
2808 }
2809
2810 /*
2811  * when reads are done, we need to check csums to verify the data is correct
2812  * if there's a match, we allow the bio to finish.  If not, the code in
2813  * extent_io.c will try to find good copies for us.
2814  */
2815 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2816                                struct extent_state *state, int mirror)
2817 {
2818         size_t offset = start - page_offset(page);
2819         struct inode *inode = page->mapping->host;
2820         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2821         char *kaddr;
2822         u64 private = ~(u32)0;
2823         int ret;
2824         struct btrfs_root *root = BTRFS_I(inode)->root;
2825         u32 csum = ~(u32)0;
2826         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2827                                       DEFAULT_RATELIMIT_BURST);
2828
2829         if (PageChecked(page)) {
2830                 ClearPageChecked(page);
2831                 goto good;
2832         }
2833
2834         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2835                 goto good;
2836
2837         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2838             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2839                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2840                                   GFP_NOFS);
2841                 return 0;
2842         }
2843
2844         if (state && state->start == start) {
2845                 private = state->private;
2846                 ret = 0;
2847         } else {
2848                 ret = get_state_private(io_tree, start, &private);
2849         }
2850         kaddr = kmap_atomic(page);
2851         if (ret)
2852                 goto zeroit;
2853
2854         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2855         btrfs_csum_final(csum, (char *)&csum);
2856         if (csum != private)
2857                 goto zeroit;
2858
2859         kunmap_atomic(kaddr);
2860 good:
2861         return 0;
2862
2863 zeroit:
2864         if (__ratelimit(&_rs))
2865                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2866                         (unsigned long long)btrfs_ino(page->mapping->host),
2867                         (unsigned long long)start, csum,
2868                         (unsigned long long)private);
2869         memset(kaddr + offset, 1, end - start + 1);
2870         flush_dcache_page(page);
2871         kunmap_atomic(kaddr);
2872         if (private == 0)
2873                 return 0;
2874         return -EIO;
2875 }
2876
2877 struct delayed_iput {
2878         struct list_head list;
2879         struct inode *inode;
2880 };
2881
2882 /* JDM: If this is fs-wide, why can't we add a pointer to
2883  * btrfs_inode instead and avoid the allocation? */
2884 void btrfs_add_delayed_iput(struct inode *inode)
2885 {
2886         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2887         struct delayed_iput *delayed;
2888
2889         if (atomic_add_unless(&inode->i_count, -1, 1))
2890                 return;
2891
2892         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2893         delayed->inode = inode;
2894
2895         spin_lock(&fs_info->delayed_iput_lock);
2896         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2897         spin_unlock(&fs_info->delayed_iput_lock);
2898 }
2899
2900 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2901 {
2902         LIST_HEAD(list);
2903         struct btrfs_fs_info *fs_info = root->fs_info;
2904         struct delayed_iput *delayed;
2905         int empty;
2906
2907         spin_lock(&fs_info->delayed_iput_lock);
2908         empty = list_empty(&fs_info->delayed_iputs);
2909         spin_unlock(&fs_info->delayed_iput_lock);
2910         if (empty)
2911                 return;
2912
2913         spin_lock(&fs_info->delayed_iput_lock);
2914         list_splice_init(&fs_info->delayed_iputs, &list);
2915         spin_unlock(&fs_info->delayed_iput_lock);
2916
2917         while (!list_empty(&list)) {
2918                 delayed = list_entry(list.next, struct delayed_iput, list);
2919                 list_del(&delayed->list);
2920                 iput(delayed->inode);
2921                 kfree(delayed);
2922         }
2923 }
2924
2925 /*
2926  * This is called in transaction commit time. If there are no orphan
2927  * files in the subvolume, it removes orphan item and frees block_rsv
2928  * structure.
2929  */
2930 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2931                               struct btrfs_root *root)
2932 {
2933         struct btrfs_block_rsv *block_rsv;
2934         int ret;
2935
2936         if (atomic_read(&root->orphan_inodes) ||
2937             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2938                 return;
2939
2940         spin_lock(&root->orphan_lock);
2941         if (atomic_read(&root->orphan_inodes)) {
2942                 spin_unlock(&root->orphan_lock);
2943                 return;
2944         }
2945
2946         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2947                 spin_unlock(&root->orphan_lock);
2948                 return;
2949         }
2950
2951         block_rsv = root->orphan_block_rsv;
2952         root->orphan_block_rsv = NULL;
2953         spin_unlock(&root->orphan_lock);
2954
2955         if (root->orphan_item_inserted &&
2956             btrfs_root_refs(&root->root_item) > 0) {
2957                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2958                                             root->root_key.objectid);
2959                 BUG_ON(ret);
2960                 root->orphan_item_inserted = 0;
2961         }
2962
2963         if (block_rsv) {
2964                 WARN_ON(block_rsv->size > 0);
2965                 btrfs_free_block_rsv(root, block_rsv);
2966         }
2967 }
2968
2969 /*
2970  * This creates an orphan entry for the given inode in case something goes
2971  * wrong in the middle of an unlink/truncate.
2972  *
2973  * NOTE: caller of this function should reserve 5 units of metadata for
2974  *       this function.
2975  */
2976 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2977 {
2978         struct btrfs_root *root = BTRFS_I(inode)->root;
2979         struct btrfs_block_rsv *block_rsv = NULL;
2980         int reserve = 0;
2981         int insert = 0;
2982         int ret;
2983
2984         if (!root->orphan_block_rsv) {
2985                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2986                 if (!block_rsv)
2987                         return -ENOMEM;
2988         }
2989
2990         spin_lock(&root->orphan_lock);
2991         if (!root->orphan_block_rsv) {
2992                 root->orphan_block_rsv = block_rsv;
2993         } else if (block_rsv) {
2994                 btrfs_free_block_rsv(root, block_rsv);
2995                 block_rsv = NULL;
2996         }
2997
2998         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2999                               &BTRFS_I(inode)->runtime_flags)) {
3000 #if 0
3001                 /*
3002                  * For proper ENOSPC handling, we should do orphan
3003                  * cleanup when mounting. But this introduces backward
3004                  * compatibility issue.
3005                  */
3006                 if (!xchg(&root->orphan_item_inserted, 1))
3007                         insert = 2;
3008                 else
3009                         insert = 1;
3010 #endif
3011                 insert = 1;
3012                 atomic_inc(&root->orphan_inodes);
3013         }
3014
3015         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3016                               &BTRFS_I(inode)->runtime_flags))
3017                 reserve = 1;
3018         spin_unlock(&root->orphan_lock);
3019
3020         /* grab metadata reservation from transaction handle */
3021         if (reserve) {
3022                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3023                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3024         }
3025
3026         /* insert an orphan item to track this unlinked/truncated file */
3027         if (insert >= 1) {
3028                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3029                 if (ret && ret != -EEXIST) {
3030                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3031                                   &BTRFS_I(inode)->runtime_flags);
3032                         btrfs_abort_transaction(trans, root, ret);
3033                         return ret;
3034                 }
3035                 ret = 0;
3036         }
3037
3038         /* insert an orphan item to track subvolume contains orphan files */
3039         if (insert >= 2) {
3040                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3041                                                root->root_key.objectid);
3042                 if (ret && ret != -EEXIST) {
3043                         btrfs_abort_transaction(trans, root, ret);
3044                         return ret;
3045                 }
3046         }
3047         return 0;
3048 }
3049
3050 /*
3051  * We have done the truncate/delete so we can go ahead and remove the orphan
3052  * item for this particular inode.
3053  */
3054 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3055                             struct inode *inode)
3056 {
3057         struct btrfs_root *root = BTRFS_I(inode)->root;
3058         int delete_item = 0;
3059         int release_rsv = 0;
3060         int ret = 0;
3061
3062         spin_lock(&root->orphan_lock);
3063         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3064                                &BTRFS_I(inode)->runtime_flags))
3065                 delete_item = 1;
3066
3067         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3068                                &BTRFS_I(inode)->runtime_flags))
3069                 release_rsv = 1;
3070         spin_unlock(&root->orphan_lock);
3071
3072         if (trans && delete_item) {
3073                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3074                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3075         }
3076
3077         if (release_rsv) {
3078                 btrfs_orphan_release_metadata(inode);
3079                 atomic_dec(&root->orphan_inodes);
3080         }
3081
3082         return 0;
3083 }
3084
3085 /*
3086  * this cleans up any orphans that may be left on the list from the last use
3087  * of this root.
3088  */
3089 int btrfs_orphan_cleanup(struct btrfs_root *root)
3090 {
3091         struct btrfs_path *path;
3092         struct extent_buffer *leaf;
3093         struct btrfs_key key, found_key;
3094         struct btrfs_trans_handle *trans;
3095         struct inode *inode;
3096         u64 last_objectid = 0;
3097         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3098
3099         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3100                 return 0;
3101
3102         path = btrfs_alloc_path();
3103         if (!path) {
3104                 ret = -ENOMEM;
3105                 goto out;
3106         }
3107         path->reada = -1;
3108
3109         key.objectid = BTRFS_ORPHAN_OBJECTID;
3110         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3111         key.offset = (u64)-1;
3112
3113         while (1) {
3114                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3115                 if (ret < 0)
3116                         goto out;
3117
3118                 /*
3119                  * if ret == 0 means we found what we were searching for, which
3120                  * is weird, but possible, so only screw with path if we didn't
3121                  * find the key and see if we have stuff that matches
3122                  */
3123                 if (ret > 0) {
3124                         ret = 0;
3125                         if (path->slots[0] == 0)
3126                                 break;
3127                         path->slots[0]--;
3128                 }
3129
3130                 /* pull out the item */
3131                 leaf = path->nodes[0];
3132                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3133
3134                 /* make sure the item matches what we want */
3135                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3136                         break;
3137                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3138                         break;
3139
3140                 /* release the path since we're done with it */
3141                 btrfs_release_path(path);
3142
3143                 /*
3144                  * this is where we are basically btrfs_lookup, without the
3145                  * crossing root thing.  we store the inode number in the
3146                  * offset of the orphan item.
3147                  */
3148
3149                 if (found_key.offset == last_objectid) {
3150                         btrfs_err(root->fs_info,
3151                                 "Error removing orphan entry, stopping orphan cleanup");
3152                         ret = -EINVAL;
3153                         goto out;
3154                 }
3155
3156                 last_objectid = found_key.offset;
3157
3158                 found_key.objectid = found_key.offset;
3159                 found_key.type = BTRFS_INODE_ITEM_KEY;
3160                 found_key.offset = 0;
3161                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3162                 ret = PTR_RET(inode);
3163                 if (ret && ret != -ESTALE)
3164                         goto out;
3165
3166                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3167                         struct btrfs_root *dead_root;
3168                         struct btrfs_fs_info *fs_info = root->fs_info;
3169                         int is_dead_root = 0;
3170
3171                         /*
3172                          * this is an orphan in the tree root. Currently these
3173                          * could come from 2 sources:
3174                          *  a) a snapshot deletion in progress
3175                          *  b) a free space cache inode
3176                          * We need to distinguish those two, as the snapshot
3177                          * orphan must not get deleted.
3178                          * find_dead_roots already ran before us, so if this
3179                          * is a snapshot deletion, we should find the root
3180                          * in the dead_roots list
3181                          */
3182                         spin_lock(&fs_info->trans_lock);
3183                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3184                                             root_list) {
3185                                 if (dead_root->root_key.objectid ==
3186                                     found_key.objectid) {
3187                                         is_dead_root = 1;
3188                                         break;
3189                                 }
3190                         }
3191                         spin_unlock(&fs_info->trans_lock);
3192                         if (is_dead_root) {
3193                                 /* prevent this orphan from being found again */
3194                                 key.offset = found_key.objectid - 1;
3195                                 continue;
3196                         }
3197                 }
3198                 /*
3199                  * Inode is already gone but the orphan item is still there,
3200                  * kill the orphan item.
3201                  */
3202                 if (ret == -ESTALE) {
3203                         trans = btrfs_start_transaction(root, 1);
3204                         if (IS_ERR(trans)) {
3205                                 ret = PTR_ERR(trans);
3206                                 goto out;
3207                         }
3208                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3209                                 found_key.objectid);
3210                         ret = btrfs_del_orphan_item(trans, root,
3211                                                     found_key.objectid);
3212                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3213                         btrfs_end_transaction(trans, root);
3214                         continue;
3215                 }
3216
3217                 /*
3218                  * add this inode to the orphan list so btrfs_orphan_del does
3219                  * the proper thing when we hit it
3220                  */
3221                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3222                         &BTRFS_I(inode)->runtime_flags);
3223                 atomic_inc(&root->orphan_inodes);
3224
3225                 /* if we have links, this was a truncate, lets do that */
3226                 if (inode->i_nlink) {
3227                         if (!S_ISREG(inode->i_mode)) {
3228                                 WARN_ON(1);
3229                                 iput(inode);
3230                                 continue;
3231                         }
3232                         nr_truncate++;
3233
3234                         /* 1 for the orphan item deletion. */
3235                         trans = btrfs_start_transaction(root, 1);
3236                         if (IS_ERR(trans)) {
3237                                 ret = PTR_ERR(trans);
3238                                 goto out;
3239                         }
3240                         ret = btrfs_orphan_add(trans, inode);
3241                         btrfs_end_transaction(trans, root);
3242                         if (ret)
3243                                 goto out;
3244
3245                         ret = btrfs_truncate(inode);
3246                         if (ret)
3247                                 btrfs_orphan_del(NULL, inode);
3248                 } else {
3249                         nr_unlink++;
3250                 }
3251
3252                 /* this will do delete_inode and everything for us */
3253                 iput(inode);
3254                 if (ret)
3255                         goto out;
3256         }
3257         /* release the path since we're done with it */
3258         btrfs_release_path(path);
3259
3260         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3261
3262         if (root->orphan_block_rsv)
3263                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3264                                         (u64)-1);
3265
3266         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3267                 trans = btrfs_join_transaction(root);
3268                 if (!IS_ERR(trans))
3269                         btrfs_end_transaction(trans, root);
3270         }
3271
3272         if (nr_unlink)
3273                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3274         if (nr_truncate)
3275                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3276
3277 out:
3278         if (ret)
3279                 btrfs_crit(root->fs_info,
3280                         "could not do orphan cleanup %d", ret);
3281         btrfs_free_path(path);
3282         return ret;
3283 }
3284
3285 /*
3286  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3287  * don't find any xattrs, we know there can't be any acls.
3288  *
3289  * slot is the slot the inode is in, objectid is the objectid of the inode
3290  */
3291 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3292                                           int slot, u64 objectid)
3293 {
3294         u32 nritems = btrfs_header_nritems(leaf);
3295         struct btrfs_key found_key;
3296         int scanned = 0;
3297
3298         slot++;
3299         while (slot < nritems) {
3300                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3301
3302                 /* we found a different objectid, there must not be acls */
3303                 if (found_key.objectid != objectid)
3304                         return 0;
3305
3306                 /* we found an xattr, assume we've got an acl */
3307                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3308                         return 1;
3309
3310                 /*
3311                  * we found a key greater than an xattr key, there can't
3312                  * be any acls later on
3313                  */
3314                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3315                         return 0;
3316
3317                 slot++;
3318                 scanned++;
3319
3320                 /*
3321                  * it goes inode, inode backrefs, xattrs, extents,
3322                  * so if there are a ton of hard links to an inode there can
3323                  * be a lot of backrefs.  Don't waste time searching too hard,
3324                  * this is just an optimization
3325                  */
3326                 if (scanned >= 8)
3327                         break;
3328         }
3329         /* we hit the end of the leaf before we found an xattr or
3330          * something larger than an xattr.  We have to assume the inode
3331          * has acls
3332          */
3333         return 1;
3334 }
3335
3336 /*
3337  * read an inode from the btree into the in-memory inode
3338  */
3339 static void btrfs_read_locked_inode(struct inode *inode)
3340 {
3341         struct btrfs_path *path;
3342         struct extent_buffer *leaf;
3343         struct btrfs_inode_item *inode_item;
3344         struct btrfs_timespec *tspec;
3345         struct btrfs_root *root = BTRFS_I(inode)->root;
3346         struct btrfs_key location;
3347         int maybe_acls;
3348         u32 rdev;
3349         int ret;
3350         bool filled = false;
3351
3352         ret = btrfs_fill_inode(inode, &rdev);
3353         if (!ret)
3354                 filled = true;
3355
3356         path = btrfs_alloc_path();
3357         if (!path)
3358                 goto make_bad;
3359
3360         path->leave_spinning = 1;
3361         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3362
3363         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3364         if (ret)
3365                 goto make_bad;
3366
3367         leaf = path->nodes[0];
3368
3369         if (filled)
3370                 goto cache_acl;
3371
3372         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3373                                     struct btrfs_inode_item);
3374         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3375         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3376         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3377         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3378         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3379
3380         tspec = btrfs_inode_atime(inode_item);
3381         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3382         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3383
3384         tspec = btrfs_inode_mtime(inode_item);
3385         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3386         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387
3388         tspec = btrfs_inode_ctime(inode_item);
3389         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3390         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3391
3392         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3393         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3394         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3395
3396         /*
3397          * If we were modified in the current generation and evicted from memory
3398          * and then re-read we need to do a full sync since we don't have any
3399          * idea about which extents were modified before we were evicted from
3400          * cache.
3401          */
3402         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3403                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3404                         &BTRFS_I(inode)->runtime_flags);
3405
3406         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3407         inode->i_generation = BTRFS_I(inode)->generation;
3408         inode->i_rdev = 0;
3409         rdev = btrfs_inode_rdev(leaf, inode_item);
3410
3411         BTRFS_I(inode)->index_cnt = (u64)-1;
3412         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3413 cache_acl:
3414         /*
3415          * try to precache a NULL acl entry for files that don't have
3416          * any xattrs or acls
3417          */
3418         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3419                                            btrfs_ino(inode));
3420         if (!maybe_acls)
3421                 cache_no_acl(inode);
3422
3423         btrfs_free_path(path);
3424
3425         switch (inode->i_mode & S_IFMT) {
3426         case S_IFREG:
3427                 inode->i_mapping->a_ops = &btrfs_aops;
3428                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3429                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3430                 inode->i_fop = &btrfs_file_operations;
3431                 inode->i_op = &btrfs_file_inode_operations;
3432                 break;
3433         case S_IFDIR:
3434                 inode->i_fop = &btrfs_dir_file_operations;
3435                 if (root == root->fs_info->tree_root)
3436                         inode->i_op = &btrfs_dir_ro_inode_operations;
3437                 else
3438                         inode->i_op = &btrfs_dir_inode_operations;
3439                 break;
3440         case S_IFLNK:
3441                 inode->i_op = &btrfs_symlink_inode_operations;
3442                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3443                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3444                 break;
3445         default:
3446                 inode->i_op = &btrfs_special_inode_operations;
3447                 init_special_inode(inode, inode->i_mode, rdev);
3448                 break;
3449         }
3450
3451         btrfs_update_iflags(inode);
3452         return;
3453
3454 make_bad:
3455         btrfs_free_path(path);
3456         make_bad_inode(inode);
3457 }
3458
3459 /*
3460  * given a leaf and an inode, copy the inode fields into the leaf
3461  */
3462 static void fill_inode_item(struct btrfs_trans_handle *trans,
3463                             struct extent_buffer *leaf,
3464                             struct btrfs_inode_item *item,
3465                             struct inode *inode)
3466 {
3467         struct btrfs_map_token token;
3468
3469         btrfs_init_map_token(&token);
3470
3471         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3472         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3473         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3474                                    &token);
3475         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3476         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3477
3478         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3479                                      inode->i_atime.tv_sec, &token);
3480         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3481                                       inode->i_atime.tv_nsec, &token);
3482
3483         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3484                                      inode->i_mtime.tv_sec, &token);
3485         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3486                                       inode->i_mtime.tv_nsec, &token);
3487
3488         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3489                                      inode->i_ctime.tv_sec, &token);
3490         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3491                                       inode->i_ctime.tv_nsec, &token);
3492
3493         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3494                                      &token);
3495         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3496                                          &token);
3497         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3498         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3499         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3500         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3501         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3502 }
3503
3504 /*
3505  * copy everything in the in-memory inode into the btree.
3506  */
3507 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3508                                 struct btrfs_root *root, struct inode *inode)
3509 {
3510         struct btrfs_inode_item *inode_item;
3511         struct btrfs_path *path;
3512         struct extent_buffer *leaf;
3513         int ret;
3514
3515         path = btrfs_alloc_path();
3516         if (!path)
3517                 return -ENOMEM;
3518
3519         path->leave_spinning = 1;
3520         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3521                                  1);
3522         if (ret) {
3523                 if (ret > 0)
3524                         ret = -ENOENT;
3525                 goto failed;
3526         }
3527
3528         btrfs_unlock_up_safe(path, 1);
3529         leaf = path->nodes[0];
3530         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3531                                     struct btrfs_inode_item);
3532
3533         fill_inode_item(trans, leaf, inode_item, inode);
3534         btrfs_mark_buffer_dirty(leaf);
3535         btrfs_set_inode_last_trans(trans, inode);
3536         ret = 0;
3537 failed:
3538         btrfs_free_path(path);
3539         return ret;
3540 }
3541
3542 /*
3543  * copy everything in the in-memory inode into the btree.
3544  */
3545 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3546                                 struct btrfs_root *root, struct inode *inode)
3547 {
3548         int ret;
3549
3550         /*
3551          * If the inode is a free space inode, we can deadlock during commit
3552          * if we put it into the delayed code.
3553          *
3554          * The data relocation inode should also be directly updated
3555          * without delay
3556          */
3557         if (!btrfs_is_free_space_inode(inode)
3558             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3559                 btrfs_update_root_times(trans, root);
3560
3561                 ret = btrfs_delayed_update_inode(trans, root, inode);
3562                 if (!ret)
3563                         btrfs_set_inode_last_trans(trans, inode);
3564                 return ret;
3565         }
3566
3567         return btrfs_update_inode_item(trans, root, inode);
3568 }
3569
3570 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3571                                          struct btrfs_root *root,
3572                                          struct inode *inode)
3573 {
3574         int ret;
3575
3576         ret = btrfs_update_inode(trans, root, inode);
3577         if (ret == -ENOSPC)
3578                 return btrfs_update_inode_item(trans, root, inode);
3579         return ret;
3580 }
3581
3582 /*
3583  * unlink helper that gets used here in inode.c and in the tree logging
3584  * recovery code.  It remove a link in a directory with a given name, and
3585  * also drops the back refs in the inode to the directory
3586  */
3587 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3588                                 struct btrfs_root *root,
3589                                 struct inode *dir, struct inode *inode,
3590                                 const char *name, int name_len)
3591 {
3592         struct btrfs_path *path;
3593         int ret = 0;
3594         struct extent_buffer *leaf;
3595         struct btrfs_dir_item *di;
3596         struct btrfs_key key;
3597         u64 index;
3598         u64 ino = btrfs_ino(inode);
3599         u64 dir_ino = btrfs_ino(dir);
3600
3601         path = btrfs_alloc_path();
3602         if (!path) {
3603                 ret = -ENOMEM;
3604                 goto out;
3605         }
3606
3607         path->leave_spinning = 1;
3608         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3609                                     name, name_len, -1);
3610         if (IS_ERR(di)) {
3611                 ret = PTR_ERR(di);
3612                 goto err;
3613         }
3614         if (!di) {
3615                 ret = -ENOENT;
3616                 goto err;
3617         }
3618         leaf = path->nodes[0];
3619         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3620         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3621         if (ret)
3622                 goto err;
3623         btrfs_release_path(path);
3624
3625         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3626                                   dir_ino, &index);
3627         if (ret) {
3628                 btrfs_info(root->fs_info,
3629                         "failed to delete reference to %.*s, inode %llu parent %llu",
3630                         name_len, name,
3631                         (unsigned long long)ino, (unsigned long long)dir_ino);
3632                 btrfs_abort_transaction(trans, root, ret);
3633                 goto err;
3634         }
3635
3636         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3637         if (ret) {
3638                 btrfs_abort_transaction(trans, root, ret);
3639                 goto err;
3640         }
3641
3642         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3643                                          inode, dir_ino);
3644         if (ret != 0 && ret != -ENOENT) {
3645                 btrfs_abort_transaction(trans, root, ret);
3646                 goto err;
3647         }
3648
3649         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3650                                            dir, index);
3651         if (ret == -ENOENT)
3652                 ret = 0;
3653         else if (ret)
3654                 btrfs_abort_transaction(trans, root, ret);
3655 err:
3656         btrfs_free_path(path);
3657         if (ret)
3658                 goto out;
3659
3660         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3661         inode_inc_iversion(inode);
3662         inode_inc_iversion(dir);
3663         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3664         ret = btrfs_update_inode(trans, root, dir);
3665 out:
3666         return ret;
3667 }
3668
3669 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3670                        struct btrfs_root *root,
3671                        struct inode *dir, struct inode *inode,
3672                        const char *name, int name_len)
3673 {
3674         int ret;
3675         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3676         if (!ret) {
3677                 btrfs_drop_nlink(inode);
3678                 ret = btrfs_update_inode(trans, root, inode);
3679         }
3680         return ret;
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, so
3687  * if we cannot make our reservations the normal way try and see if there is
3688  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3689  * allow the unlink to occur.
3690  */
3691 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3692 {
3693         struct btrfs_trans_handle *trans;
3694         struct btrfs_root *root = BTRFS_I(dir)->root;
3695         int ret;
3696
3697         /*
3698          * 1 for the possible orphan item
3699          * 1 for the dir item
3700          * 1 for the dir index
3701          * 1 for the inode ref
3702          * 1 for the inode
3703          */
3704         trans = btrfs_start_transaction(root, 5);
3705         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3706                 return trans;
3707
3708         if (PTR_ERR(trans) == -ENOSPC) {
3709                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3710
3711                 trans = btrfs_start_transaction(root, 0);
3712                 if (IS_ERR(trans))
3713                         return trans;
3714                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3715                                                &root->fs_info->trans_block_rsv,
3716                                                num_bytes, 5);
3717                 if (ret) {
3718                         btrfs_end_transaction(trans, root);
3719                         return ERR_PTR(ret);
3720                 }
3721                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3722                 trans->bytes_reserved = num_bytes;
3723         }
3724         return trans;
3725 }
3726
3727 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3728 {
3729         struct btrfs_root *root = BTRFS_I(dir)->root;
3730         struct btrfs_trans_handle *trans;
3731         struct inode *inode = dentry->d_inode;
3732         int ret;
3733
3734         trans = __unlink_start_trans(dir);
3735         if (IS_ERR(trans))
3736                 return PTR_ERR(trans);
3737
3738         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3739
3740         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3741                                  dentry->d_name.name, dentry->d_name.len);
3742         if (ret)
3743                 goto out;
3744
3745         if (inode->i_nlink == 0) {
3746                 ret = btrfs_orphan_add(trans, inode);
3747                 if (ret)
3748                         goto out;
3749         }
3750
3751 out:
3752         btrfs_end_transaction(trans, root);
3753         btrfs_btree_balance_dirty(root);
3754         return ret;
3755 }
3756
3757 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3758                         struct btrfs_root *root,
3759                         struct inode *dir, u64 objectid,
3760                         const char *name, int name_len)
3761 {
3762         struct btrfs_path *path;
3763         struct extent_buffer *leaf;
3764         struct btrfs_dir_item *di;
3765         struct btrfs_key key;
3766         u64 index;
3767         int ret;
3768         u64 dir_ino = btrfs_ino(dir);
3769
3770         path = btrfs_alloc_path();
3771         if (!path)
3772                 return -ENOMEM;
3773
3774         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3775                                    name, name_len, -1);
3776         if (IS_ERR_OR_NULL(di)) {
3777                 if (!di)
3778                         ret = -ENOENT;
3779                 else
3780                         ret = PTR_ERR(di);
3781                 goto out;
3782         }
3783
3784         leaf = path->nodes[0];
3785         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3786         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3787         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3788         if (ret) {
3789                 btrfs_abort_transaction(trans, root, ret);
3790                 goto out;
3791         }
3792         btrfs_release_path(path);
3793
3794         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3795                                  objectid, root->root_key.objectid,
3796                                  dir_ino, &index, name, name_len);
3797         if (ret < 0) {
3798                 if (ret != -ENOENT) {
3799                         btrfs_abort_transaction(trans, root, ret);
3800                         goto out;
3801                 }
3802                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3803                                                  name, name_len);
3804                 if (IS_ERR_OR_NULL(di)) {
3805                         if (!di)
3806                                 ret = -ENOENT;
3807                         else
3808                                 ret = PTR_ERR(di);
3809                         btrfs_abort_transaction(trans, root, ret);
3810                         goto out;
3811                 }
3812
3813                 leaf = path->nodes[0];
3814                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3815                 btrfs_release_path(path);
3816                 index = key.offset;
3817         }
3818         btrfs_release_path(path);
3819
3820         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3821         if (ret) {
3822                 btrfs_abort_transaction(trans, root, ret);
3823                 goto out;
3824         }
3825
3826         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3827         inode_inc_iversion(dir);
3828         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3829         ret = btrfs_update_inode_fallback(trans, root, dir);
3830         if (ret)
3831                 btrfs_abort_transaction(trans, root, ret);
3832 out:
3833         btrfs_free_path(path);
3834         return ret;
3835 }
3836
3837 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3838 {
3839         struct inode *inode = dentry->d_inode;
3840         int err = 0;
3841         struct btrfs_root *root = BTRFS_I(dir)->root;
3842         struct btrfs_trans_handle *trans;
3843
3844         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3845                 return -ENOTEMPTY;
3846         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3847                 return -EPERM;
3848
3849         trans = __unlink_start_trans(dir);
3850         if (IS_ERR(trans))
3851                 return PTR_ERR(trans);
3852
3853         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3854                 err = btrfs_unlink_subvol(trans, root, dir,
3855                                           BTRFS_I(inode)->location.objectid,
3856                                           dentry->d_name.name,
3857                                           dentry->d_name.len);
3858                 goto out;
3859         }
3860
3861         err = btrfs_orphan_add(trans, inode);
3862         if (err)
3863                 goto out;
3864
3865         /* now the directory is empty */
3866         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3867                                  dentry->d_name.name, dentry->d_name.len);
3868         if (!err)
3869                 btrfs_i_size_write(inode, 0);
3870 out:
3871         btrfs_end_transaction(trans, root);
3872         btrfs_btree_balance_dirty(root);
3873
3874         return err;
3875 }
3876
3877 /*
3878  * this can truncate away extent items, csum items and directory items.
3879  * It starts at a high offset and removes keys until it can't find
3880  * any higher than new_size
3881  *
3882  * csum items that cross the new i_size are truncated to the new size
3883  * as well.
3884  *
3885  * min_type is the minimum key type to truncate down to.  If set to 0, this
3886  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3887  */
3888 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3889                                struct btrfs_root *root,
3890                                struct inode *inode,
3891                                u64 new_size, u32 min_type)
3892 {
3893         struct btrfs_path *path;
3894         struct extent_buffer *leaf;
3895         struct btrfs_file_extent_item *fi;
3896         struct btrfs_key key;
3897         struct btrfs_key found_key;
3898         u64 extent_start = 0;
3899         u64 extent_num_bytes = 0;
3900         u64 extent_offset = 0;
3901         u64 item_end = 0;
3902         u32 found_type = (u8)-1;
3903         int found_extent;
3904         int del_item;
3905         int pending_del_nr = 0;
3906         int pending_del_slot = 0;
3907         int extent_type = -1;
3908         int ret;
3909         int err = 0;
3910         u64 ino = btrfs_ino(inode);
3911
3912         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3913
3914         path = btrfs_alloc_path();
3915         if (!path)
3916                 return -ENOMEM;
3917         path->reada = -1;
3918
3919         /*
3920          * We want to drop from the next block forward in case this new size is
3921          * not block aligned since we will be keeping the last block of the
3922          * extent just the way it is.
3923          */
3924         if (root->ref_cows || root == root->fs_info->tree_root)
3925                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3926                                         root->sectorsize), (u64)-1, 0);
3927
3928         /*
3929          * This function is also used to drop the items in the log tree before
3930          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3931          * it is used to drop the loged items. So we shouldn't kill the delayed
3932          * items.
3933          */
3934         if (min_type == 0 && root == BTRFS_I(inode)->root)
3935                 btrfs_kill_delayed_inode_items(inode);
3936
3937         key.objectid = ino;
3938         key.offset = (u64)-1;
3939         key.type = (u8)-1;
3940
3941 search_again:
3942         path->leave_spinning = 1;
3943         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3944         if (ret < 0) {
3945                 err = ret;
3946                 goto out;
3947         }
3948
3949         if (ret > 0) {
3950                 /* there are no items in the tree for us to truncate, we're
3951                  * done
3952                  */
3953                 if (path->slots[0] == 0)
3954                         goto out;
3955                 path->slots[0]--;
3956         }
3957
3958         while (1) {
3959                 fi = NULL;
3960                 leaf = path->nodes[0];
3961                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3962                 found_type = btrfs_key_type(&found_key);
3963
3964                 if (found_key.objectid != ino)
3965                         break;
3966
3967                 if (found_type < min_type)
3968                         break;
3969
3970                 item_end = found_key.offset;
3971                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3972                         fi = btrfs_item_ptr(leaf, path->slots[0],
3973                                             struct btrfs_file_extent_item);
3974                         extent_type = btrfs_file_extent_type(leaf, fi);
3975                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3976                                 item_end +=
3977                                     btrfs_file_extent_num_bytes(leaf, fi);
3978                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3979                                 item_end += btrfs_file_extent_inline_len(leaf,
3980                                                                          fi);
3981                         }
3982                         item_end--;
3983                 }
3984                 if (found_type > min_type) {
3985                         del_item = 1;
3986                 } else {
3987                         if (item_end < new_size)
3988                                 break;
3989                         if (found_key.offset >= new_size)
3990                                 del_item = 1;
3991                         else
3992                                 del_item = 0;
3993                 }
3994                 found_extent = 0;
3995                 /* FIXME, shrink the extent if the ref count is only 1 */
3996                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3997                         goto delete;
3998
3999                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4000                         u64 num_dec;
4001                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4002                         if (!del_item) {
4003                                 u64 orig_num_bytes =
4004                                         btrfs_file_extent_num_bytes(leaf, fi);
4005                                 extent_num_bytes = ALIGN(new_size -
4006                                                 found_key.offset,
4007                                                 root->sectorsize);
4008                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4009                                                          extent_num_bytes);
4010                                 num_dec = (orig_num_bytes -
4011                                            extent_num_bytes);
4012                                 if (root->ref_cows && extent_start != 0)
4013                                         inode_sub_bytes(inode, num_dec);
4014                                 btrfs_mark_buffer_dirty(leaf);
4015                         } else {
4016                                 extent_num_bytes =
4017                                         btrfs_file_extent_disk_num_bytes(leaf,
4018                                                                          fi);
4019                                 extent_offset = found_key.offset -
4020                                         btrfs_file_extent_offset(leaf, fi);
4021
4022                                 /* FIXME blocksize != 4096 */
4023                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4024                                 if (extent_start != 0) {
4025                                         found_extent = 1;
4026                                         if (root->ref_cows)
4027                                                 inode_sub_bytes(inode, num_dec);
4028                                 }
4029                         }
4030                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4031                         /*
4032                          * we can't truncate inline items that have had
4033                          * special encodings
4034                          */
4035                         if (!del_item &&
4036                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4037                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4038                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4039                                 u32 size = new_size - found_key.offset;
4040
4041                                 if (root->ref_cows) {
4042                                         inode_sub_bytes(inode, item_end + 1 -
4043                                                         new_size);
4044                                 }
4045                                 size =
4046                                     btrfs_file_extent_calc_inline_size(size);
4047                                 btrfs_truncate_item(root, path, size, 1);
4048                         } else if (root->ref_cows) {
4049                                 inode_sub_bytes(inode, item_end + 1 -
4050                                                 found_key.offset);
4051                         }
4052                 }
4053 delete:
4054                 if (del_item) {
4055                         if (!pending_del_nr) {
4056                                 /* no pending yet, add ourselves */
4057                                 pending_del_slot = path->slots[0];
4058                                 pending_del_nr = 1;
4059                         } else if (pending_del_nr &&
4060                                    path->slots[0] + 1 == pending_del_slot) {
4061                                 /* hop on the pending chunk */
4062                                 pending_del_nr++;
4063                                 pending_del_slot = path->slots[0];
4064                         } else {
4065                                 BUG();
4066                         }
4067                 } else {
4068                         break;
4069                 }
4070                 if (found_extent && (root->ref_cows ||
4071                                      root == root->fs_info->tree_root)) {
4072                         btrfs_set_path_blocking(path);
4073                         ret = btrfs_free_extent(trans, root, extent_start,
4074                                                 extent_num_bytes, 0,
4075                                                 btrfs_header_owner(leaf),
4076                                                 ino, extent_offset, 0);
4077                         BUG_ON(ret);
4078                 }
4079
4080                 if (found_type == BTRFS_INODE_ITEM_KEY)
4081                         break;
4082
4083                 if (path->slots[0] == 0 ||
4084                     path->slots[0] != pending_del_slot) {
4085                         if (pending_del_nr) {
4086                                 ret = btrfs_del_items(trans, root, path,
4087                                                 pending_del_slot,
4088                                                 pending_del_nr);
4089                                 if (ret) {
4090                                         btrfs_abort_transaction(trans,
4091                                                                 root, ret);
4092                                         goto error;
4093                                 }
4094                                 pending_del_nr = 0;
4095                         }
4096                         btrfs_release_path(path);
4097                         goto search_again;
4098                 } else {
4099                         path->slots[0]--;
4100                 }
4101         }
4102 out:
4103         if (pending_del_nr) {
4104                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4105                                       pending_del_nr);
4106                 if (ret)
4107                         btrfs_abort_transaction(trans, root, ret);
4108         }
4109 error:
4110         btrfs_free_path(path);
4111         return err;
4112 }
4113
4114 /*
4115  * btrfs_truncate_page - read, zero a chunk and write a page
4116  * @inode - inode that we're zeroing
4117  * @from - the offset to start zeroing
4118  * @len - the length to zero, 0 to zero the entire range respective to the
4119  *      offset
4120  * @front - zero up to the offset instead of from the offset on
4121  *
4122  * This will find the page for the "from" offset and cow the page and zero the
4123  * part we want to zero.  This is used with truncate and hole punching.
4124  */
4125 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4126                         int front)
4127 {
4128         struct address_space *mapping = inode->i_mapping;
4129         struct btrfs_root *root = BTRFS_I(inode)->root;
4130         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4131         struct btrfs_ordered_extent *ordered;
4132         struct extent_state *cached_state = NULL;
4133         char *kaddr;
4134         u32 blocksize = root->sectorsize;
4135         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4136         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4137         struct page *page;
4138         gfp_t mask = btrfs_alloc_write_mask(mapping);
4139         int ret = 0;
4140         u64 page_start;
4141         u64 page_end;
4142
4143         if ((offset & (blocksize - 1)) == 0 &&
4144             (!len || ((len & (blocksize - 1)) == 0)))
4145                 goto out;
4146         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4147         if (ret)
4148                 goto out;
4149
4150 again:
4151         page = find_or_create_page(mapping, index, mask);
4152         if (!page) {
4153                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4154                 ret = -ENOMEM;
4155                 goto out;
4156         }
4157
4158         page_start = page_offset(page);
4159         page_end = page_start + PAGE_CACHE_SIZE - 1;
4160
4161         if (!PageUptodate(page)) {
4162                 ret = btrfs_readpage(NULL, page);
4163                 lock_page(page);
4164                 if (page->mapping != mapping) {
4165                         unlock_page(page);
4166                         page_cache_release(page);
4167                         goto again;
4168                 }
4169                 if (!PageUptodate(page)) {
4170                         ret = -EIO;
4171                         goto out_unlock;
4172                 }
4173         }
4174         wait_on_page_writeback(page);
4175
4176         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4177         set_page_extent_mapped(page);
4178
4179         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4180         if (ordered) {
4181                 unlock_extent_cached(io_tree, page_start, page_end,
4182                                      &cached_state, GFP_NOFS);
4183                 unlock_page(page);
4184                 page_cache_release(page);
4185                 btrfs_start_ordered_extent(inode, ordered, 1);
4186                 btrfs_put_ordered_extent(ordered);
4187                 goto again;
4188         }
4189
4190         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4191                           EXTENT_DIRTY | EXTENT_DELALLOC |
4192                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4193                           0, 0, &cached_state, GFP_NOFS);
4194
4195         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4196                                         &cached_state);
4197         if (ret) {
4198                 unlock_extent_cached(io_tree, page_start, page_end,
4199                                      &cached_state, GFP_NOFS);
4200                 goto out_unlock;
4201         }
4202
4203         if (offset != PAGE_CACHE_SIZE) {
4204                 if (!len)
4205                         len = PAGE_CACHE_SIZE - offset;
4206                 kaddr = kmap(page);
4207                 if (front)
4208                         memset(kaddr, 0, offset);
4209                 else
4210                         memset(kaddr + offset, 0, len);
4211                 flush_dcache_page(page);
4212                 kunmap(page);
4213         }
4214         ClearPageChecked(page);
4215         set_page_dirty(page);
4216         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4217                              GFP_NOFS);
4218
4219 out_unlock:
4220         if (ret)
4221                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4222         unlock_page(page);
4223         page_cache_release(page);
4224 out:
4225         return ret;
4226 }
4227
4228 /*
4229  * This function puts in dummy file extents for the area we're creating a hole
4230  * for.  So if we are truncating this file to a larger size we need to insert
4231  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4232  * the range between oldsize and size
4233  */
4234 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4235 {
4236         struct btrfs_trans_handle *trans;
4237         struct btrfs_root *root = BTRFS_I(inode)->root;
4238         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4239         struct extent_map *em = NULL;
4240         struct extent_state *cached_state = NULL;
4241         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4242         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4243         u64 block_end = ALIGN(size, root->sectorsize);
4244         u64 last_byte;
4245         u64 cur_offset;
4246         u64 hole_size;
4247         int err = 0;
4248
4249         if (size <= hole_start)
4250                 return 0;
4251
4252         while (1) {
4253                 struct btrfs_ordered_extent *ordered;
4254                 btrfs_wait_ordered_range(inode, hole_start,
4255                                          block_end - hole_start);
4256                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4257                                  &cached_state);
4258                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4259                 if (!ordered)
4260                         break;
4261                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4262                                      &cached_state, GFP_NOFS);
4263                 btrfs_put_ordered_extent(ordered);
4264         }
4265
4266         cur_offset = hole_start;
4267         while (1) {
4268                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4269                                 block_end - cur_offset, 0);
4270                 if (IS_ERR(em)) {
4271                         err = PTR_ERR(em);
4272                         em = NULL;
4273                         break;
4274                 }
4275                 last_byte = min(extent_map_end(em), block_end);
4276                 last_byte = ALIGN(last_byte , root->sectorsize);
4277                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4278                         struct extent_map *hole_em;
4279                         hole_size = last_byte - cur_offset;
4280
4281                         trans = btrfs_start_transaction(root, 3);
4282                         if (IS_ERR(trans)) {
4283                                 err = PTR_ERR(trans);
4284                                 break;
4285                         }
4286
4287                         err = btrfs_drop_extents(trans, root, inode,
4288                                                  cur_offset,
4289                                                  cur_offset + hole_size, 1);
4290                         if (err) {
4291                                 btrfs_abort_transaction(trans, root, err);
4292                                 btrfs_end_transaction(trans, root);
4293                                 break;
4294                         }
4295
4296                         err = btrfs_insert_file_extent(trans, root,
4297                                         btrfs_ino(inode), cur_offset, 0,
4298                                         0, hole_size, 0, hole_size,
4299                                         0, 0, 0);
4300                         if (err) {
4301                                 btrfs_abort_transaction(trans, root, err);
4302                                 btrfs_end_transaction(trans, root);
4303                                 break;
4304                         }
4305
4306                         btrfs_drop_extent_cache(inode, cur_offset,
4307                                                 cur_offset + hole_size - 1, 0);
4308                         hole_em = alloc_extent_map();
4309                         if (!hole_em) {
4310                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4311                                         &BTRFS_I(inode)->runtime_flags);
4312                                 goto next;
4313                         }
4314                         hole_em->start = cur_offset;
4315                         hole_em->len = hole_size;
4316                         hole_em->orig_start = cur_offset;
4317
4318                         hole_em->block_start = EXTENT_MAP_HOLE;
4319                         hole_em->block_len = 0;
4320                         hole_em->orig_block_len = 0;
4321                         hole_em->ram_bytes = hole_size;
4322                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4323                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4324                         hole_em->generation = trans->transid;
4325
4326                         while (1) {
4327                                 write_lock(&em_tree->lock);
4328                                 err = add_extent_mapping(em_tree, hole_em, 1);
4329                                 write_unlock(&em_tree->lock);
4330                                 if (err != -EEXIST)
4331                                         break;
4332                                 btrfs_drop_extent_cache(inode, cur_offset,
4333                                                         cur_offset +
4334                                                         hole_size - 1, 0);
4335                         }
4336                         free_extent_map(hole_em);
4337 next:
4338                         btrfs_update_inode(trans, root, inode);
4339                         btrfs_end_transaction(trans, root);
4340                 }
4341                 free_extent_map(em);
4342                 em = NULL;
4343                 cur_offset = last_byte;
4344                 if (cur_offset >= block_end)
4345                         break;
4346         }
4347
4348         free_extent_map(em);
4349         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4350                              GFP_NOFS);
4351         return err;
4352 }
4353
4354 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4355 {
4356         struct btrfs_root *root = BTRFS_I(inode)->root;
4357         struct btrfs_trans_handle *trans;
4358         loff_t oldsize = i_size_read(inode);
4359         loff_t newsize = attr->ia_size;
4360         int mask = attr->ia_valid;
4361         int ret;
4362
4363         if (newsize == oldsize)
4364                 return 0;
4365
4366         /*
4367          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4368          * special case where we need to update the times despite not having
4369          * these flags set.  For all other operations the VFS set these flags
4370          * explicitly if it wants a timestamp update.
4371          */
4372         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4373                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4374
4375         if (newsize > oldsize) {
4376                 truncate_pagecache(inode, oldsize, newsize);
4377                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4378                 if (ret)
4379                         return ret;
4380
4381                 trans = btrfs_start_transaction(root, 1);
4382                 if (IS_ERR(trans))
4383                         return PTR_ERR(trans);
4384
4385                 i_size_write(inode, newsize);
4386                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4387                 ret = btrfs_update_inode(trans, root, inode);
4388                 btrfs_end_transaction(trans, root);
4389         } else {
4390
4391                 /*
4392                  * We're truncating a file that used to have good data down to
4393                  * zero. Make sure it gets into the ordered flush list so that
4394                  * any new writes get down to disk quickly.
4395                  */
4396                 if (newsize == 0)
4397                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4398                                 &BTRFS_I(inode)->runtime_flags);
4399
4400                 /*
4401                  * 1 for the orphan item we're going to add
4402                  * 1 for the orphan item deletion.
4403                  */
4404                 trans = btrfs_start_transaction(root, 2);
4405                 if (IS_ERR(trans))
4406                         return PTR_ERR(trans);
4407
4408                 /*
4409                  * We need to do this in case we fail at _any_ point during the
4410                  * actual truncate.  Once we do the truncate_setsize we could
4411                  * invalidate pages which forces any outstanding ordered io to
4412                  * be instantly completed which will give us extents that need
4413                  * to be truncated.  If we fail to get an orphan inode down we
4414                  * could have left over extents that were never meant to live,
4415                  * so we need to garuntee from this point on that everything
4416                  * will be consistent.
4417                  */
4418                 ret = btrfs_orphan_add(trans, inode);
4419                 btrfs_end_transaction(trans, root);
4420                 if (ret)
4421                         return ret;
4422
4423                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4424                 truncate_setsize(inode, newsize);
4425
4426                 /* Disable nonlocked read DIO to avoid the end less truncate */
4427                 btrfs_inode_block_unlocked_dio(inode);
4428                 inode_dio_wait(inode);
4429                 btrfs_inode_resume_unlocked_dio(inode);
4430
4431                 ret = btrfs_truncate(inode);
4432                 if (ret && inode->i_nlink)
4433                         btrfs_orphan_del(NULL, inode);
4434         }
4435
4436         return ret;
4437 }
4438
4439 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4440 {
4441         struct inode *inode = dentry->d_inode;
4442         struct btrfs_root *root = BTRFS_I(inode)->root;
4443         int err;
4444
4445         if (btrfs_root_readonly(root))
4446                 return -EROFS;
4447
4448         err = inode_change_ok(inode, attr);
4449         if (err)
4450                 return err;
4451
4452         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4453                 err = btrfs_setsize(inode, attr);
4454                 if (err)
4455                         return err;
4456         }
4457
4458         if (attr->ia_valid) {
4459                 setattr_copy(inode, attr);
4460                 inode_inc_iversion(inode);
4461                 err = btrfs_dirty_inode(inode);
4462
4463                 if (!err && attr->ia_valid & ATTR_MODE)
4464                         err = btrfs_acl_chmod(inode);
4465         }
4466
4467         return err;
4468 }
4469
4470 void btrfs_evict_inode(struct inode *inode)
4471 {
4472         struct btrfs_trans_handle *trans;
4473         struct btrfs_root *root = BTRFS_I(inode)->root;
4474         struct btrfs_block_rsv *rsv, *global_rsv;
4475         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4476         int ret;
4477
4478         trace_btrfs_inode_evict(inode);
4479
4480         truncate_inode_pages(&inode->i_data, 0);
4481         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4482                                btrfs_is_free_space_inode(inode)))
4483                 goto no_delete;
4484
4485         if (is_bad_inode(inode)) {
4486                 btrfs_orphan_del(NULL, inode);
4487                 goto no_delete;
4488         }
4489         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4490         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4491
4492         if (root->fs_info->log_root_recovering) {
4493                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4494                                  &BTRFS_I(inode)->runtime_flags));
4495                 goto no_delete;
4496         }
4497
4498         if (inode->i_nlink > 0) {
4499                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4500                 goto no_delete;
4501         }
4502
4503         ret = btrfs_commit_inode_delayed_inode(inode);
4504         if (ret) {
4505                 btrfs_orphan_del(NULL, inode);
4506                 goto no_delete;
4507         }
4508
4509         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4510         if (!rsv) {
4511                 btrfs_orphan_del(NULL, inode);
4512                 goto no_delete;
4513         }
4514         rsv->size = min_size;
4515         rsv->failfast = 1;
4516         global_rsv = &root->fs_info->global_block_rsv;
4517
4518         btrfs_i_size_write(inode, 0);
4519
4520         /*
4521          * This is a bit simpler than btrfs_truncate since we've already
4522          * reserved our space for our orphan item in the unlink, so we just
4523          * need to reserve some slack space in case we add bytes and update
4524          * inode item when doing the truncate.
4525          */
4526         while (1) {
4527                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4528                                              BTRFS_RESERVE_FLUSH_LIMIT);
4529
4530                 /*
4531                  * Try and steal from the global reserve since we will
4532                  * likely not use this space anyway, we want to try as
4533                  * hard as possible to get this to work.
4534                  */
4535                 if (ret)
4536                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4537
4538                 if (ret) {
4539                         btrfs_warn(root->fs_info,
4540                                 "Could not get space for a delete, will truncate on mount %d",
4541                                 ret);
4542                         btrfs_orphan_del(NULL, inode);
4543                         btrfs_free_block_rsv(root, rsv);
4544                         goto no_delete;
4545                 }
4546
4547                 trans = btrfs_join_transaction(root);
4548                 if (IS_ERR(trans)) {
4549                         btrfs_orphan_del(NULL, inode);
4550                         btrfs_free_block_rsv(root, rsv);
4551                         goto no_delete;
4552                 }
4553
4554                 trans->block_rsv = rsv;
4555
4556                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4557                 if (ret != -ENOSPC)
4558                         break;
4559
4560                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4561                 btrfs_end_transaction(trans, root);
4562                 trans = NULL;
4563                 btrfs_btree_balance_dirty(root);
4564         }
4565
4566         btrfs_free_block_rsv(root, rsv);
4567
4568         if (ret == 0) {
4569                 trans->block_rsv = root->orphan_block_rsv;
4570                 ret = btrfs_orphan_del(trans, inode);
4571                 BUG_ON(ret);
4572         }
4573
4574         trans->block_rsv = &root->fs_info->trans_block_rsv;
4575         if (!(root == root->fs_info->tree_root ||
4576               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4577                 btrfs_return_ino(root, btrfs_ino(inode));
4578
4579         btrfs_end_transaction(trans, root);
4580         btrfs_btree_balance_dirty(root);
4581 no_delete:
4582         btrfs_remove_delayed_node(inode);
4583         clear_inode(inode);
4584         return;
4585 }
4586
4587 /*
4588  * this returns the key found in the dir entry in the location pointer.
4589  * If no dir entries were found, location->objectid is 0.
4590  */
4591 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4592                                struct btrfs_key *location)
4593 {
4594         const char *name = dentry->d_name.name;
4595         int namelen = dentry->d_name.len;
4596         struct btrfs_dir_item *di;
4597         struct btrfs_path *path;
4598         struct btrfs_root *root = BTRFS_I(dir)->root;
4599         int ret = 0;
4600
4601         path = btrfs_alloc_path();
4602         if (!path)
4603                 return -ENOMEM;
4604
4605         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4606                                     namelen, 0);
4607         if (IS_ERR(di))
4608                 ret = PTR_ERR(di);
4609
4610         if (IS_ERR_OR_NULL(di))
4611                 goto out_err;
4612
4613         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4614 out:
4615         btrfs_free_path(path);
4616         return ret;
4617 out_err:
4618         location->objectid = 0;
4619         goto out;
4620 }
4621
4622 /*
4623  * when we hit a tree root in a directory, the btrfs part of the inode
4624  * needs to be changed to reflect the root directory of the tree root.  This
4625  * is kind of like crossing a mount point.
4626  */
4627 static int fixup_tree_root_location(struct btrfs_root *root,
4628                                     struct inode *dir,
4629                                     struct dentry *dentry,
4630                                     struct btrfs_key *location,
4631                                     struct btrfs_root **sub_root)
4632 {
4633         struct btrfs_path *path;
4634         struct btrfs_root *new_root;
4635         struct btrfs_root_ref *ref;
4636         struct extent_buffer *leaf;
4637         int ret;
4638         int err = 0;
4639
4640         path = btrfs_alloc_path();
4641         if (!path) {
4642                 err = -ENOMEM;
4643                 goto out;
4644         }
4645
4646         err = -ENOENT;
4647         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4648                                   BTRFS_I(dir)->root->root_key.objectid,
4649                                   location->objectid);
4650         if (ret) {
4651                 if (ret < 0)
4652                         err = ret;
4653                 goto out;
4654         }
4655
4656         leaf = path->nodes[0];
4657         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4658         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4659             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4660                 goto out;
4661
4662         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4663                                    (unsigned long)(ref + 1),
4664                                    dentry->d_name.len);
4665         if (ret)
4666                 goto out;
4667
4668         btrfs_release_path(path);
4669
4670         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4671         if (IS_ERR(new_root)) {
4672                 err = PTR_ERR(new_root);
4673                 goto out;
4674         }
4675
4676         *sub_root = new_root;
4677         location->objectid = btrfs_root_dirid(&new_root->root_item);
4678         location->type = BTRFS_INODE_ITEM_KEY;
4679         location->offset = 0;
4680         err = 0;
4681 out:
4682         btrfs_free_path(path);
4683         return err;
4684 }
4685
4686 static void inode_tree_add(struct inode *inode)
4687 {
4688         struct btrfs_root *root = BTRFS_I(inode)->root;
4689         struct btrfs_inode *entry;
4690         struct rb_node **p;
4691         struct rb_node *parent;
4692         u64 ino = btrfs_ino(inode);
4693
4694         if (inode_unhashed(inode))
4695                 return;
4696 again:
4697         parent = NULL;
4698         spin_lock(&root->inode_lock);
4699         p = &root->inode_tree.rb_node;
4700         while (*p) {
4701                 parent = *p;
4702                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4703
4704                 if (ino < btrfs_ino(&entry->vfs_inode))
4705                         p = &parent->rb_left;
4706                 else if (ino > btrfs_ino(&entry->vfs_inode))
4707                         p = &parent->rb_right;
4708                 else {
4709                         WARN_ON(!(entry->vfs_inode.i_state &
4710                                   (I_WILL_FREE | I_FREEING)));
4711                         rb_erase(parent, &root->inode_tree);
4712                         RB_CLEAR_NODE(parent);
4713                         spin_unlock(&root->inode_lock);
4714                         goto again;
4715                 }
4716         }
4717         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4718         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4719         spin_unlock(&root->inode_lock);
4720 }
4721
4722 static void inode_tree_del(struct inode *inode)
4723 {
4724         struct btrfs_root *root = BTRFS_I(inode)->root;
4725         int empty = 0;
4726
4727         spin_lock(&root->inode_lock);
4728         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4729                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4730                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4731                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4732         }
4733         spin_unlock(&root->inode_lock);
4734
4735         /*
4736          * Free space cache has inodes in the tree root, but the tree root has a
4737          * root_refs of 0, so this could end up dropping the tree root as a
4738          * snapshot, so we need the extra !root->fs_info->tree_root check to
4739          * make sure we don't drop it.
4740          */
4741         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4742             root != root->fs_info->tree_root) {
4743                 synchronize_srcu(&root->fs_info->subvol_srcu);
4744                 spin_lock(&root->inode_lock);
4745                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4746                 spin_unlock(&root->inode_lock);
4747                 if (empty)
4748                         btrfs_add_dead_root(root);
4749         }
4750 }
4751
4752 void btrfs_invalidate_inodes(struct btrfs_root *root)
4753 {
4754         struct rb_node *node;
4755         struct rb_node *prev;
4756         struct btrfs_inode *entry;
4757         struct inode *inode;
4758         u64 objectid = 0;
4759
4760         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4761
4762         spin_lock(&root->inode_lock);
4763 again:
4764         node = root->inode_tree.rb_node;
4765         prev = NULL;
4766         while (node) {
4767                 prev = node;
4768                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4769
4770                 if (objectid < btrfs_ino(&entry->vfs_inode))
4771                         node = node->rb_left;
4772                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4773                         node = node->rb_right;
4774                 else
4775                         break;
4776         }
4777         if (!node) {
4778                 while (prev) {
4779                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4780                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4781                                 node = prev;
4782                                 break;
4783                         }
4784                         prev = rb_next(prev);
4785                 }
4786         }
4787         while (node) {
4788                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4789                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4790                 inode = igrab(&entry->vfs_inode);
4791                 if (inode) {
4792                         spin_unlock(&root->inode_lock);
4793                         if (atomic_read(&inode->i_count) > 1)
4794                                 d_prune_aliases(inode);
4795                         /*
4796                          * btrfs_drop_inode will have it removed from
4797                          * the inode cache when its usage count
4798                          * hits zero.
4799                          */
4800                         iput(inode);
4801                         cond_resched();
4802                         spin_lock(&root->inode_lock);
4803                         goto again;
4804                 }
4805
4806                 if (cond_resched_lock(&root->inode_lock))
4807                         goto again;
4808
4809                 node = rb_next(node);
4810         }
4811         spin_unlock(&root->inode_lock);
4812 }
4813
4814 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4815 {
4816         struct btrfs_iget_args *args = p;
4817         inode->i_ino = args->ino;
4818         BTRFS_I(inode)->root = args->root;
4819         return 0;
4820 }
4821
4822 static int btrfs_find_actor(struct inode *inode, void *opaque)
4823 {
4824         struct btrfs_iget_args *args = opaque;
4825         return args->ino == btrfs_ino(inode) &&
4826                 args->root == BTRFS_I(inode)->root;
4827 }
4828
4829 static struct inode *btrfs_iget_locked(struct super_block *s,
4830                                        u64 objectid,
4831                                        struct btrfs_root *root)
4832 {
4833         struct inode *inode;
4834         struct btrfs_iget_args args;
4835         args.ino = objectid;
4836         args.root = root;
4837
4838         inode = iget5_locked(s, objectid, btrfs_find_actor,
4839                              btrfs_init_locked_inode,
4840                              (void *)&args);
4841         return inode;
4842 }
4843
4844 /* Get an inode object given its location and corresponding root.
4845  * Returns in *is_new if the inode was read from disk
4846  */
4847 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4848                          struct btrfs_root *root, int *new)
4849 {
4850         struct inode *inode;
4851
4852         inode = btrfs_iget_locked(s, location->objectid, root);
4853         if (!inode)
4854                 return ERR_PTR(-ENOMEM);
4855
4856         if (inode->i_state & I_NEW) {
4857                 BTRFS_I(inode)->root = root;
4858                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4859                 btrfs_read_locked_inode(inode);
4860                 if (!is_bad_inode(inode)) {
4861                         inode_tree_add(inode);
4862                         unlock_new_inode(inode);
4863                         if (new)
4864                                 *new = 1;
4865                 } else {
4866                         unlock_new_inode(inode);
4867                         iput(inode);
4868                         inode = ERR_PTR(-ESTALE);
4869                 }
4870         }
4871
4872         return inode;
4873 }
4874
4875 static struct inode *new_simple_dir(struct super_block *s,
4876                                     struct btrfs_key *key,
4877                                     struct btrfs_root *root)
4878 {
4879         struct inode *inode = new_inode(s);
4880
4881         if (!inode)
4882                 return ERR_PTR(-ENOMEM);
4883
4884         BTRFS_I(inode)->root = root;
4885         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4886         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4887
4888         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4889         inode->i_op = &btrfs_dir_ro_inode_operations;
4890         inode->i_fop = &simple_dir_operations;
4891         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4892         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4893
4894         return inode;
4895 }
4896
4897 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4898 {
4899         struct inode *inode;
4900         struct btrfs_root *root = BTRFS_I(dir)->root;
4901         struct btrfs_root *sub_root = root;
4902         struct btrfs_key location;
4903         int index;
4904         int ret = 0;
4905
4906         if (dentry->d_name.len > BTRFS_NAME_LEN)
4907                 return ERR_PTR(-ENAMETOOLONG);
4908
4909         ret = btrfs_inode_by_name(dir, dentry, &location);
4910         if (ret < 0)
4911                 return ERR_PTR(ret);
4912
4913         if (location.objectid == 0)
4914                 return NULL;
4915
4916         if (location.type == BTRFS_INODE_ITEM_KEY) {
4917                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4918                 return inode;
4919         }
4920
4921         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4922
4923         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4924         ret = fixup_tree_root_location(root, dir, dentry,
4925                                        &location, &sub_root);
4926         if (ret < 0) {
4927                 if (ret != -ENOENT)
4928                         inode = ERR_PTR(ret);
4929                 else
4930                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4931         } else {
4932                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4933         }
4934         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4935
4936         if (!IS_ERR(inode) && root != sub_root) {
4937                 down_read(&root->fs_info->cleanup_work_sem);
4938                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4939                         ret = btrfs_orphan_cleanup(sub_root);
4940                 up_read(&root->fs_info->cleanup_work_sem);
4941                 if (ret)
4942                         inode = ERR_PTR(ret);
4943         }
4944
4945         return inode;
4946 }
4947
4948 static int btrfs_dentry_delete(const struct dentry *dentry)
4949 {
4950         struct btrfs_root *root;
4951         struct inode *inode = dentry->d_inode;
4952
4953         if (!inode && !IS_ROOT(dentry))
4954                 inode = dentry->d_parent->d_inode;
4955
4956         if (inode) {
4957                 root = BTRFS_I(inode)->root;
4958                 if (btrfs_root_refs(&root->root_item) == 0)
4959                         return 1;
4960
4961                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4962                         return 1;
4963         }
4964         return 0;
4965 }
4966
4967 static void btrfs_dentry_release(struct dentry *dentry)
4968 {
4969         if (dentry->d_fsdata)
4970                 kfree(dentry->d_fsdata);
4971 }
4972
4973 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4974                                    unsigned int flags)
4975 {
4976         struct dentry *ret;
4977
4978         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4979         return ret;
4980 }
4981
4982 unsigned char btrfs_filetype_table[] = {
4983         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4984 };
4985
4986 static int btrfs_real_readdir(struct file *filp, void *dirent,
4987                               filldir_t filldir)
4988 {
4989         struct inode *inode = file_inode(filp);
4990         struct btrfs_root *root = BTRFS_I(inode)->root;
4991         struct btrfs_item *item;
4992         struct btrfs_dir_item *di;
4993         struct btrfs_key key;
4994         struct btrfs_key found_key;
4995         struct btrfs_path *path;
4996         struct list_head ins_list;
4997         struct list_head del_list;
4998         int ret;
4999         struct extent_buffer *leaf;
5000         int slot;
5001         unsigned char d_type;
5002         int over = 0;
5003         u32 di_cur;
5004         u32 di_total;
5005         u32 di_len;
5006         int key_type = BTRFS_DIR_INDEX_KEY;
5007         char tmp_name[32];
5008         char *name_ptr;
5009         int name_len;
5010         int is_curr = 0;        /* filp->f_pos points to the current index? */
5011
5012         /* FIXME, use a real flag for deciding about the key type */
5013         if (root->fs_info->tree_root == root)
5014                 key_type = BTRFS_DIR_ITEM_KEY;
5015
5016         /* special case for "." */
5017         if (filp->f_pos == 0) {
5018                 over = filldir(dirent, ".", 1,
5019                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5020                 if (over)
5021                         return 0;
5022                 filp->f_pos = 1;
5023         }
5024         /* special case for .., just use the back ref */
5025         if (filp->f_pos == 1) {
5026                 u64 pino = parent_ino(filp->f_path.dentry);
5027                 over = filldir(dirent, "..", 2,
5028                                filp->f_pos, pino, DT_DIR);
5029                 if (over)
5030                         return 0;
5031                 filp->f_pos = 2;
5032         }
5033         path = btrfs_alloc_path();
5034         if (!path)
5035                 return -ENOMEM;
5036
5037         path->reada = 1;
5038
5039         if (key_type == BTRFS_DIR_INDEX_KEY) {
5040                 INIT_LIST_HEAD(&ins_list);
5041                 INIT_LIST_HEAD(&del_list);
5042                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5043         }
5044
5045         btrfs_set_key_type(&key, key_type);
5046         key.offset = filp->f_pos;
5047         key.objectid = btrfs_ino(inode);
5048
5049         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5050         if (ret < 0)
5051                 goto err;
5052
5053         while (1) {
5054                 leaf = path->nodes[0];
5055                 slot = path->slots[0];
5056                 if (slot >= btrfs_header_nritems(leaf)) {
5057                         ret = btrfs_next_leaf(root, path);
5058                         if (ret < 0)
5059                                 goto err;
5060                         else if (ret > 0)
5061                                 break;
5062                         continue;
5063                 }
5064
5065                 item = btrfs_item_nr(leaf, slot);
5066                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5067
5068                 if (found_key.objectid != key.objectid)
5069                         break;
5070                 if (btrfs_key_type(&found_key) != key_type)
5071                         break;
5072                 if (found_key.offset < filp->f_pos)
5073                         goto next;
5074                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5075                     btrfs_should_delete_dir_index(&del_list,
5076                                                   found_key.offset))
5077                         goto next;
5078
5079                 filp->f_pos = found_key.offset;
5080                 is_curr = 1;
5081
5082                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5083                 di_cur = 0;
5084                 di_total = btrfs_item_size(leaf, item);
5085
5086                 while (di_cur < di_total) {
5087                         struct btrfs_key location;
5088
5089                         if (verify_dir_item(root, leaf, di))
5090                                 break;
5091
5092                         name_len = btrfs_dir_name_len(leaf, di);
5093                         if (name_len <= sizeof(tmp_name)) {
5094                                 name_ptr = tmp_name;
5095                         } else {
5096                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5097                                 if (!name_ptr) {
5098                                         ret = -ENOMEM;
5099                                         goto err;
5100                                 }
5101                         }
5102                         read_extent_buffer(leaf, name_ptr,
5103                                            (unsigned long)(di + 1), name_len);
5104
5105                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5106                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5107
5108
5109                         /* is this a reference to our own snapshot? If so
5110                          * skip it.
5111                          *
5112                          * In contrast to old kernels, we insert the snapshot's
5113                          * dir item and dir index after it has been created, so
5114                          * we won't find a reference to our own snapshot. We
5115                          * still keep the following code for backward
5116                          * compatibility.
5117                          */
5118                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5119                             location.objectid == root->root_key.objectid) {
5120                                 over = 0;
5121                                 goto skip;
5122                         }
5123                         over = filldir(dirent, name_ptr, name_len,
5124                                        found_key.offset, location.objectid,
5125                                        d_type);
5126
5127 skip:
5128                         if (name_ptr != tmp_name)
5129                                 kfree(name_ptr);
5130
5131                         if (over)
5132                                 goto nopos;
5133                         di_len = btrfs_dir_name_len(leaf, di) +
5134                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5135                         di_cur += di_len;
5136                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5137                 }
5138 next:
5139                 path->slots[0]++;
5140         }
5141
5142         if (key_type == BTRFS_DIR_INDEX_KEY) {
5143                 if (is_curr)
5144                         filp->f_pos++;
5145                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5146                                                       &ins_list);
5147                 if (ret)
5148                         goto nopos;
5149         }
5150
5151         /* Reached end of directory/root. Bump pos past the last item. */
5152         if (key_type == BTRFS_DIR_INDEX_KEY)
5153                 /*
5154                  * 32-bit glibc will use getdents64, but then strtol -
5155                  * so the last number we can serve is this.
5156                  */
5157                 filp->f_pos = 0x7fffffff;
5158         else
5159                 filp->f_pos++;
5160 nopos:
5161         ret = 0;
5162 err:
5163         if (key_type == BTRFS_DIR_INDEX_KEY)
5164                 btrfs_put_delayed_items(&ins_list, &del_list);
5165         btrfs_free_path(path);
5166         return ret;
5167 }
5168
5169 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5170 {
5171         struct btrfs_root *root = BTRFS_I(inode)->root;
5172         struct btrfs_trans_handle *trans;
5173         int ret = 0;
5174         bool nolock = false;
5175
5176         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5177                 return 0;
5178
5179         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5180                 nolock = true;
5181
5182         if (wbc->sync_mode == WB_SYNC_ALL) {
5183                 if (nolock)
5184                         trans = btrfs_join_transaction_nolock(root);
5185                 else
5186                         trans = btrfs_join_transaction(root);
5187                 if (IS_ERR(trans))
5188                         return PTR_ERR(trans);
5189                 ret = btrfs_commit_transaction(trans, root);
5190         }
5191         return ret;
5192 }
5193
5194 /*
5195  * This is somewhat expensive, updating the tree every time the
5196  * inode changes.  But, it is most likely to find the inode in cache.
5197  * FIXME, needs more benchmarking...there are no reasons other than performance
5198  * to keep or drop this code.
5199  */
5200 static int btrfs_dirty_inode(struct inode *inode)
5201 {
5202         struct btrfs_root *root = BTRFS_I(inode)->root;
5203         struct btrfs_trans_handle *trans;
5204         int ret;
5205
5206         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5207                 return 0;
5208
5209         trans = btrfs_join_transaction(root);
5210         if (IS_ERR(trans))
5211                 return PTR_ERR(trans);
5212
5213         ret = btrfs_update_inode(trans, root, inode);
5214         if (ret && ret == -ENOSPC) {
5215                 /* whoops, lets try again with the full transaction */
5216                 btrfs_end_transaction(trans, root);
5217                 trans = btrfs_start_transaction(root, 1);
5218                 if (IS_ERR(trans))
5219                         return PTR_ERR(trans);
5220
5221                 ret = btrfs_update_inode(trans, root, inode);
5222         }
5223         btrfs_end_transaction(trans, root);
5224         if (BTRFS_I(inode)->delayed_node)
5225                 btrfs_balance_delayed_items(root);
5226
5227         return ret;
5228 }
5229
5230 /*
5231  * This is a copy of file_update_time.  We need this so we can return error on
5232  * ENOSPC for updating the inode in the case of file write and mmap writes.
5233  */
5234 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5235                              int flags)
5236 {
5237         struct btrfs_root *root = BTRFS_I(inode)->root;
5238
5239         if (btrfs_root_readonly(root))
5240                 return -EROFS;
5241
5242         if (flags & S_VERSION)
5243                 inode_inc_iversion(inode);
5244         if (flags & S_CTIME)
5245                 inode->i_ctime = *now;
5246         if (flags & S_MTIME)
5247                 inode->i_mtime = *now;
5248         if (flags & S_ATIME)
5249                 inode->i_atime = *now;
5250         return btrfs_dirty_inode(inode);
5251 }
5252
5253 /*
5254  * find the highest existing sequence number in a directory
5255  * and then set the in-memory index_cnt variable to reflect
5256  * free sequence numbers
5257  */
5258 static int btrfs_set_inode_index_count(struct inode *inode)
5259 {
5260         struct btrfs_root *root = BTRFS_I(inode)->root;
5261         struct btrfs_key key, found_key;
5262         struct btrfs_path *path;
5263         struct extent_buffer *leaf;
5264         int ret;
5265
5266         key.objectid = btrfs_ino(inode);
5267         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5268         key.offset = (u64)-1;
5269
5270         path = btrfs_alloc_path();
5271         if (!path)
5272                 return -ENOMEM;
5273
5274         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5275         if (ret < 0)
5276                 goto out;
5277         /* FIXME: we should be able to handle this */
5278         if (ret == 0)
5279                 goto out;
5280         ret = 0;
5281
5282         /*
5283          * MAGIC NUMBER EXPLANATION:
5284          * since we search a directory based on f_pos we have to start at 2
5285          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5286          * else has to start at 2
5287          */
5288         if (path->slots[0] == 0) {
5289                 BTRFS_I(inode)->index_cnt = 2;
5290                 goto out;
5291         }
5292
5293         path->slots[0]--;
5294
5295         leaf = path->nodes[0];
5296         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5297
5298         if (found_key.objectid != btrfs_ino(inode) ||
5299             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5300                 BTRFS_I(inode)->index_cnt = 2;
5301                 goto out;
5302         }
5303
5304         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5305 out:
5306         btrfs_free_path(path);
5307         return ret;
5308 }
5309
5310 /*
5311  * helper to find a free sequence number in a given directory.  This current
5312  * code is very simple, later versions will do smarter things in the btree
5313  */
5314 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5315 {
5316         int ret = 0;
5317
5318         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5319                 ret = btrfs_inode_delayed_dir_index_count(dir);
5320                 if (ret) {
5321                         ret = btrfs_set_inode_index_count(dir);
5322                         if (ret)
5323                                 return ret;
5324                 }
5325         }
5326
5327         *index = BTRFS_I(dir)->index_cnt;
5328         BTRFS_I(dir)->index_cnt++;
5329
5330         return ret;
5331 }
5332
5333 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5334                                      struct btrfs_root *root,
5335                                      struct inode *dir,
5336                                      const char *name, int name_len,
5337                                      u64 ref_objectid, u64 objectid,
5338                                      umode_t mode, u64 *index)
5339 {
5340         struct inode *inode;
5341         struct btrfs_inode_item *inode_item;
5342         struct btrfs_key *location;
5343         struct btrfs_path *path;
5344         struct btrfs_inode_ref *ref;
5345         struct btrfs_key key[2];
5346         u32 sizes[2];
5347         unsigned long ptr;
5348         int ret;
5349         int owner;
5350
5351         path = btrfs_alloc_path();
5352         if (!path)
5353                 return ERR_PTR(-ENOMEM);
5354
5355         inode = new_inode(root->fs_info->sb);
5356         if (!inode) {
5357                 btrfs_free_path(path);
5358                 return ERR_PTR(-ENOMEM);
5359         }
5360
5361         /*
5362          * we have to initialize this early, so we can reclaim the inode
5363          * number if we fail afterwards in this function.
5364          */
5365         inode->i_ino = objectid;
5366
5367         if (dir) {
5368                 trace_btrfs_inode_request(dir);
5369
5370                 ret = btrfs_set_inode_index(dir, index);
5371                 if (ret) {
5372                         btrfs_free_path(path);
5373                         iput(inode);
5374                         return ERR_PTR(ret);
5375                 }
5376         }
5377         /*
5378          * index_cnt is ignored for everything but a dir,
5379          * btrfs_get_inode_index_count has an explanation for the magic
5380          * number
5381          */
5382         BTRFS_I(inode)->index_cnt = 2;
5383         BTRFS_I(inode)->root = root;
5384         BTRFS_I(inode)->generation = trans->transid;
5385         inode->i_generation = BTRFS_I(inode)->generation;
5386
5387         /*
5388          * We could have gotten an inode number from somebody who was fsynced
5389          * and then removed in this same transaction, so let's just set full
5390          * sync since it will be a full sync anyway and this will blow away the
5391          * old info in the log.
5392          */
5393         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5394
5395         if (S_ISDIR(mode))
5396                 owner = 0;
5397         else
5398                 owner = 1;
5399
5400         key[0].objectid = objectid;
5401         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5402         key[0].offset = 0;
5403
5404         /*
5405          * Start new inodes with an inode_ref. This is slightly more
5406          * efficient for small numbers of hard links since they will
5407          * be packed into one item. Extended refs will kick in if we
5408          * add more hard links than can fit in the ref item.
5409          */
5410         key[1].objectid = objectid;
5411         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5412         key[1].offset = ref_objectid;
5413
5414         sizes[0] = sizeof(struct btrfs_inode_item);
5415         sizes[1] = name_len + sizeof(*ref);
5416
5417         path->leave_spinning = 1;
5418         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5419         if (ret != 0)
5420                 goto fail;
5421
5422         inode_init_owner(inode, dir, mode);
5423         inode_set_bytes(inode, 0);
5424         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5425         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5426                                   struct btrfs_inode_item);
5427         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5428                              sizeof(*inode_item));
5429         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5430
5431         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5432                              struct btrfs_inode_ref);
5433         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5434         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5435         ptr = (unsigned long)(ref + 1);
5436         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5437
5438         btrfs_mark_buffer_dirty(path->nodes[0]);
5439         btrfs_free_path(path);
5440
5441         location = &BTRFS_I(inode)->location;
5442         location->objectid = objectid;
5443         location->offset = 0;
5444         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5445
5446         btrfs_inherit_iflags(inode, dir);
5447
5448         if (S_ISREG(mode)) {
5449                 if (btrfs_test_opt(root, NODATASUM))
5450                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5451                 if (btrfs_test_opt(root, NODATACOW))
5452                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5453                                 BTRFS_INODE_NODATASUM;
5454         }
5455
5456         insert_inode_hash(inode);
5457         inode_tree_add(inode);
5458
5459         trace_btrfs_inode_new(inode);
5460         btrfs_set_inode_last_trans(trans, inode);
5461
5462         btrfs_update_root_times(trans, root);
5463
5464         return inode;
5465 fail:
5466         if (dir)
5467                 BTRFS_I(dir)->index_cnt--;
5468         btrfs_free_path(path);
5469         iput(inode);
5470         return ERR_PTR(ret);
5471 }
5472
5473 static inline u8 btrfs_inode_type(struct inode *inode)
5474 {
5475         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5476 }
5477
5478 /*
5479  * utility function to add 'inode' into 'parent_inode' with
5480  * a give name and a given sequence number.
5481  * if 'add_backref' is true, also insert a backref from the
5482  * inode to the parent directory.
5483  */
5484 int btrfs_add_link(struct btrfs_trans_handle *trans,
5485                    struct inode *parent_inode, struct inode *inode,
5486                    const char *name, int name_len, int add_backref, u64 index)
5487 {
5488         int ret = 0;
5489         struct btrfs_key key;
5490         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5491         u64 ino = btrfs_ino(inode);
5492         u64 parent_ino = btrfs_ino(parent_inode);
5493
5494         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5495                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5496         } else {
5497                 key.objectid = ino;
5498                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5499                 key.offset = 0;
5500         }
5501
5502         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5503                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5504                                          key.objectid, root->root_key.objectid,
5505                                          parent_ino, index, name, name_len);
5506         } else if (add_backref) {
5507                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5508                                              parent_ino, index);
5509         }
5510
5511         /* Nothing to clean up yet */
5512         if (ret)
5513                 return ret;
5514
5515         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5516                                     parent_inode, &key,
5517                                     btrfs_inode_type(inode), index);
5518         if (ret == -EEXIST || ret == -EOVERFLOW)
5519                 goto fail_dir_item;
5520         else if (ret) {
5521                 btrfs_abort_transaction(trans, root, ret);
5522                 return ret;
5523         }
5524
5525         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5526                            name_len * 2);
5527         inode_inc_iversion(parent_inode);
5528         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5529         ret = btrfs_update_inode(trans, root, parent_inode);
5530         if (ret)
5531                 btrfs_abort_transaction(trans, root, ret);
5532         return ret;
5533
5534 fail_dir_item:
5535         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5536                 u64 local_index;
5537                 int err;
5538                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5539                                  key.objectid, root->root_key.objectid,
5540                                  parent_ino, &local_index, name, name_len);
5541
5542         } else if (add_backref) {
5543                 u64 local_index;
5544                 int err;
5545
5546                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5547                                           ino, parent_ino, &local_index);
5548         }
5549         return ret;
5550 }
5551
5552 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5553                             struct inode *dir, struct dentry *dentry,
5554                             struct inode *inode, int backref, u64 index)
5555 {
5556         int err = btrfs_add_link(trans, dir, inode,
5557                                  dentry->d_name.name, dentry->d_name.len,
5558                                  backref, index);
5559         if (err > 0)
5560                 err = -EEXIST;
5561         return err;
5562 }
5563
5564 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5565                         umode_t mode, dev_t rdev)
5566 {
5567         struct btrfs_trans_handle *trans;
5568         struct btrfs_root *root = BTRFS_I(dir)->root;
5569         struct inode *inode = NULL;
5570         int err;
5571         int drop_inode = 0;
5572         u64 objectid;
5573         u64 index = 0;
5574
5575         if (!new_valid_dev(rdev))
5576                 return -EINVAL;
5577
5578         /*
5579          * 2 for inode item and ref
5580          * 2 for dir items
5581          * 1 for xattr if selinux is on
5582          */
5583         trans = btrfs_start_transaction(root, 5);
5584         if (IS_ERR(trans))
5585                 return PTR_ERR(trans);
5586
5587         err = btrfs_find_free_ino(root, &objectid);
5588         if (err)
5589                 goto out_unlock;
5590
5591         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5592                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5593                                 mode, &index);
5594         if (IS_ERR(inode)) {
5595                 err = PTR_ERR(inode);
5596                 goto out_unlock;
5597         }
5598
5599         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5600         if (err) {
5601                 drop_inode = 1;
5602                 goto out_unlock;
5603         }
5604
5605         /*
5606         * If the active LSM wants to access the inode during
5607         * d_instantiate it needs these. Smack checks to see
5608         * if the filesystem supports xattrs by looking at the
5609         * ops vector.
5610         */
5611
5612         inode->i_op = &btrfs_special_inode_operations;
5613         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5614         if (err)
5615                 drop_inode = 1;
5616         else {
5617                 init_special_inode(inode, inode->i_mode, rdev);
5618                 btrfs_update_inode(trans, root, inode);
5619                 d_instantiate(dentry, inode);
5620         }
5621 out_unlock:
5622         btrfs_end_transaction(trans, root);
5623         btrfs_btree_balance_dirty(root);
5624         if (drop_inode) {
5625                 inode_dec_link_count(inode);
5626                 iput(inode);
5627         }
5628         return err;
5629 }
5630
5631 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5632                         umode_t mode, bool excl)
5633 {
5634         struct btrfs_trans_handle *trans;
5635         struct btrfs_root *root = BTRFS_I(dir)->root;
5636         struct inode *inode = NULL;
5637         int drop_inode_on_err = 0;
5638         int err;
5639         u64 objectid;
5640         u64 index = 0;
5641
5642         /*
5643          * 2 for inode item and ref
5644          * 2 for dir items
5645          * 1 for xattr if selinux is on
5646          */
5647         trans = btrfs_start_transaction(root, 5);
5648         if (IS_ERR(trans))
5649                 return PTR_ERR(trans);
5650
5651         err = btrfs_find_free_ino(root, &objectid);
5652         if (err)
5653                 goto out_unlock;
5654
5655         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5656                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5657                                 mode, &index);
5658         if (IS_ERR(inode)) {
5659                 err = PTR_ERR(inode);
5660                 goto out_unlock;
5661         }
5662         drop_inode_on_err = 1;
5663
5664         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5665         if (err)
5666                 goto out_unlock;
5667
5668         err = btrfs_update_inode(trans, root, inode);
5669         if (err)
5670                 goto out_unlock;
5671
5672         /*
5673         * If the active LSM wants to access the inode during
5674         * d_instantiate it needs these. Smack checks to see
5675         * if the filesystem supports xattrs by looking at the
5676         * ops vector.
5677         */
5678         inode->i_fop = &btrfs_file_operations;
5679         inode->i_op = &btrfs_file_inode_operations;
5680
5681         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5682         if (err)
5683                 goto out_unlock;
5684
5685         inode->i_mapping->a_ops = &btrfs_aops;
5686         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5687         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5688         d_instantiate(dentry, inode);
5689
5690 out_unlock:
5691         btrfs_end_transaction(trans, root);
5692         if (err && drop_inode_on_err) {
5693                 inode_dec_link_count(inode);
5694                 iput(inode);
5695         }
5696         btrfs_btree_balance_dirty(root);
5697         return err;
5698 }
5699
5700 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5701                       struct dentry *dentry)
5702 {
5703         struct btrfs_trans_handle *trans;
5704         struct btrfs_root *root = BTRFS_I(dir)->root;
5705         struct inode *inode = old_dentry->d_inode;
5706         u64 index;
5707         int err;
5708         int drop_inode = 0;
5709
5710         /* do not allow sys_link's with other subvols of the same device */
5711         if (root->objectid != BTRFS_I(inode)->root->objectid)
5712                 return -EXDEV;
5713
5714         if (inode->i_nlink >= BTRFS_LINK_MAX)
5715                 return -EMLINK;
5716
5717         err = btrfs_set_inode_index(dir, &index);
5718         if (err)
5719                 goto fail;
5720
5721         /*
5722          * 2 items for inode and inode ref
5723          * 2 items for dir items
5724          * 1 item for parent inode
5725          */
5726         trans = btrfs_start_transaction(root, 5);
5727         if (IS_ERR(trans)) {
5728                 err = PTR_ERR(trans);
5729                 goto fail;
5730         }
5731
5732         btrfs_inc_nlink(inode);
5733         inode_inc_iversion(inode);
5734         inode->i_ctime = CURRENT_TIME;
5735         ihold(inode);
5736         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5737
5738         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5739
5740         if (err) {
5741                 drop_inode = 1;
5742         } else {
5743                 struct dentry *parent = dentry->d_parent;
5744                 err = btrfs_update_inode(trans, root, inode);
5745                 if (err)
5746                         goto fail;
5747                 d_instantiate(dentry, inode);
5748                 btrfs_log_new_name(trans, inode, NULL, parent);
5749         }
5750
5751         btrfs_end_transaction(trans, root);
5752 fail:
5753         if (drop_inode) {
5754                 inode_dec_link_count(inode);
5755                 iput(inode);
5756         }
5757         btrfs_btree_balance_dirty(root);
5758         return err;
5759 }
5760
5761 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5762 {
5763         struct inode *inode = NULL;
5764         struct btrfs_trans_handle *trans;
5765         struct btrfs_root *root = BTRFS_I(dir)->root;
5766         int err = 0;
5767         int drop_on_err = 0;
5768         u64 objectid = 0;
5769         u64 index = 0;
5770
5771         /*
5772          * 2 items for inode and ref
5773          * 2 items for dir items
5774          * 1 for xattr if selinux is on
5775          */
5776         trans = btrfs_start_transaction(root, 5);
5777         if (IS_ERR(trans))
5778                 return PTR_ERR(trans);
5779
5780         err = btrfs_find_free_ino(root, &objectid);
5781         if (err)
5782                 goto out_fail;
5783
5784         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5785                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5786                                 S_IFDIR | mode, &index);
5787         if (IS_ERR(inode)) {
5788                 err = PTR_ERR(inode);
5789                 goto out_fail;
5790         }
5791
5792         drop_on_err = 1;
5793
5794         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5795         if (err)
5796                 goto out_fail;
5797
5798         inode->i_op = &btrfs_dir_inode_operations;
5799         inode->i_fop = &btrfs_dir_file_operations;
5800
5801         btrfs_i_size_write(inode, 0);
5802         err = btrfs_update_inode(trans, root, inode);
5803         if (err)
5804                 goto out_fail;
5805
5806         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5807                              dentry->d_name.len, 0, index);
5808         if (err)
5809                 goto out_fail;
5810
5811         d_instantiate(dentry, inode);
5812         drop_on_err = 0;
5813
5814 out_fail:
5815         btrfs_end_transaction(trans, root);
5816         if (drop_on_err)
5817                 iput(inode);
5818         btrfs_btree_balance_dirty(root);
5819         return err;
5820 }
5821
5822 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5823  * and an extent that you want to insert, deal with overlap and insert
5824  * the new extent into the tree.
5825  */
5826 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5827                                 struct extent_map *existing,
5828                                 struct extent_map *em,
5829                                 u64 map_start, u64 map_len)
5830 {
5831         u64 start_diff;
5832
5833         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5834         start_diff = map_start - em->start;
5835         em->start = map_start;
5836         em->len = map_len;
5837         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5838             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5839                 em->block_start += start_diff;
5840                 em->block_len -= start_diff;
5841         }
5842         return add_extent_mapping(em_tree, em, 0);
5843 }
5844
5845 static noinline int uncompress_inline(struct btrfs_path *path,
5846                                       struct inode *inode, struct page *page,
5847                                       size_t pg_offset, u64 extent_offset,
5848                                       struct btrfs_file_extent_item *item)
5849 {
5850         int ret;
5851         struct extent_buffer *leaf = path->nodes[0];
5852         char *tmp;
5853         size_t max_size;
5854         unsigned long inline_size;
5855         unsigned long ptr;
5856         int compress_type;
5857
5858         WARN_ON(pg_offset != 0);
5859         compress_type = btrfs_file_extent_compression(leaf, item);
5860         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5861         inline_size = btrfs_file_extent_inline_item_len(leaf,
5862                                         btrfs_item_nr(leaf, path->slots[0]));
5863         tmp = kmalloc(inline_size, GFP_NOFS);
5864         if (!tmp)
5865                 return -ENOMEM;
5866         ptr = btrfs_file_extent_inline_start(item);
5867
5868         read_extent_buffer(leaf, tmp, ptr, inline_size);
5869
5870         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5871         ret = btrfs_decompress(compress_type, tmp, page,
5872                                extent_offset, inline_size, max_size);
5873         if (ret) {
5874                 char *kaddr = kmap_atomic(page);
5875                 unsigned long copy_size = min_t(u64,
5876                                   PAGE_CACHE_SIZE - pg_offset,
5877                                   max_size - extent_offset);
5878                 memset(kaddr + pg_offset, 0, copy_size);
5879                 kunmap_atomic(kaddr);
5880         }
5881         kfree(tmp);
5882         return 0;
5883 }
5884
5885 /*
5886  * a bit scary, this does extent mapping from logical file offset to the disk.
5887  * the ugly parts come from merging extents from the disk with the in-ram
5888  * representation.  This gets more complex because of the data=ordered code,
5889  * where the in-ram extents might be locked pending data=ordered completion.
5890  *
5891  * This also copies inline extents directly into the page.
5892  */
5893
5894 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5895                                     size_t pg_offset, u64 start, u64 len,
5896                                     int create)
5897 {
5898         int ret;
5899         int err = 0;
5900         u64 bytenr;
5901         u64 extent_start = 0;
5902         u64 extent_end = 0;
5903         u64 objectid = btrfs_ino(inode);
5904         u32 found_type;
5905         struct btrfs_path *path = NULL;
5906         struct btrfs_root *root = BTRFS_I(inode)->root;
5907         struct btrfs_file_extent_item *item;
5908         struct extent_buffer *leaf;
5909         struct btrfs_key found_key;
5910         struct extent_map *em = NULL;
5911         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5912         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5913         struct btrfs_trans_handle *trans = NULL;
5914         int compress_type;
5915
5916 again:
5917         read_lock(&em_tree->lock);
5918         em = lookup_extent_mapping(em_tree, start, len);
5919         if (em)
5920                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5921         read_unlock(&em_tree->lock);
5922
5923         if (em) {
5924                 if (em->start > start || em->start + em->len <= start)
5925                         free_extent_map(em);
5926                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5927                         free_extent_map(em);
5928                 else
5929                         goto out;
5930         }
5931         em = alloc_extent_map();
5932         if (!em) {
5933                 err = -ENOMEM;
5934                 goto out;
5935         }
5936         em->bdev = root->fs_info->fs_devices->latest_bdev;
5937         em->start = EXTENT_MAP_HOLE;
5938         em->orig_start = EXTENT_MAP_HOLE;
5939         em->len = (u64)-1;
5940         em->block_len = (u64)-1;
5941
5942         if (!path) {
5943                 path = btrfs_alloc_path();
5944                 if (!path) {
5945                         err = -ENOMEM;
5946                         goto out;
5947                 }
5948                 /*
5949                  * Chances are we'll be called again, so go ahead and do
5950                  * readahead
5951                  */
5952                 path->reada = 1;
5953         }
5954
5955         ret = btrfs_lookup_file_extent(trans, root, path,
5956                                        objectid, start, trans != NULL);
5957         if (ret < 0) {
5958                 err = ret;
5959                 goto out;
5960         }
5961
5962         if (ret != 0) {
5963                 if (path->slots[0] == 0)
5964                         goto not_found;
5965                 path->slots[0]--;
5966         }
5967
5968         leaf = path->nodes[0];
5969         item = btrfs_item_ptr(leaf, path->slots[0],
5970                               struct btrfs_file_extent_item);
5971         /* are we inside the extent that was found? */
5972         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5973         found_type = btrfs_key_type(&found_key);
5974         if (found_key.objectid != objectid ||
5975             found_type != BTRFS_EXTENT_DATA_KEY) {
5976                 goto not_found;
5977         }
5978
5979         found_type = btrfs_file_extent_type(leaf, item);
5980         extent_start = found_key.offset;
5981         compress_type = btrfs_file_extent_compression(leaf, item);
5982         if (found_type == BTRFS_FILE_EXTENT_REG ||
5983             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5984                 extent_end = extent_start +
5985                        btrfs_file_extent_num_bytes(leaf, item);
5986         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5987                 size_t size;
5988                 size = btrfs_file_extent_inline_len(leaf, item);
5989                 extent_end = ALIGN(extent_start + size, root->sectorsize);
5990         }
5991
5992         if (start >= extent_end) {
5993                 path->slots[0]++;
5994                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5995                         ret = btrfs_next_leaf(root, path);
5996                         if (ret < 0) {
5997                                 err = ret;
5998                                 goto out;
5999                         }
6000                         if (ret > 0)
6001                                 goto not_found;
6002                         leaf = path->nodes[0];
6003                 }
6004                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6005                 if (found_key.objectid != objectid ||
6006                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6007                         goto not_found;
6008                 if (start + len <= found_key.offset)
6009                         goto not_found;
6010                 em->start = start;
6011                 em->orig_start = start;
6012                 em->len = found_key.offset - start;
6013                 goto not_found_em;
6014         }
6015
6016         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6017         if (found_type == BTRFS_FILE_EXTENT_REG ||
6018             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6019                 em->start = extent_start;
6020                 em->len = extent_end - extent_start;
6021                 em->orig_start = extent_start -
6022                                  btrfs_file_extent_offset(leaf, item);
6023                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6024                                                                       item);
6025                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6026                 if (bytenr == 0) {
6027                         em->block_start = EXTENT_MAP_HOLE;
6028                         goto insert;
6029                 }
6030                 if (compress_type != BTRFS_COMPRESS_NONE) {
6031                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6032                         em->compress_type = compress_type;
6033                         em->block_start = bytenr;
6034                         em->block_len = em->orig_block_len;
6035                 } else {
6036                         bytenr += btrfs_file_extent_offset(leaf, item);
6037                         em->block_start = bytenr;
6038                         em->block_len = em->len;
6039                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6040                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6041                 }
6042                 goto insert;
6043         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6044                 unsigned long ptr;
6045                 char *map;
6046                 size_t size;
6047                 size_t extent_offset;
6048                 size_t copy_size;
6049
6050                 em->block_start = EXTENT_MAP_INLINE;
6051                 if (!page || create) {
6052                         em->start = extent_start;
6053                         em->len = extent_end - extent_start;
6054                         goto out;
6055                 }
6056
6057                 size = btrfs_file_extent_inline_len(leaf, item);
6058                 extent_offset = page_offset(page) + pg_offset - extent_start;
6059                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6060                                 size - extent_offset);
6061                 em->start = extent_start + extent_offset;
6062                 em->len = ALIGN(copy_size, root->sectorsize);
6063                 em->orig_block_len = em->len;
6064                 em->orig_start = em->start;
6065                 if (compress_type) {
6066                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6067                         em->compress_type = compress_type;
6068                 }
6069                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6070                 if (create == 0 && !PageUptodate(page)) {
6071                         if (btrfs_file_extent_compression(leaf, item) !=
6072                             BTRFS_COMPRESS_NONE) {
6073                                 ret = uncompress_inline(path, inode, page,
6074                                                         pg_offset,
6075                                                         extent_offset, item);
6076                                 BUG_ON(ret); /* -ENOMEM */
6077                         } else {
6078                                 map = kmap(page);
6079                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6080                                                    copy_size);
6081                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6082                                         memset(map + pg_offset + copy_size, 0,
6083                                                PAGE_CACHE_SIZE - pg_offset -
6084                                                copy_size);
6085                                 }
6086                                 kunmap(page);
6087                         }
6088                         flush_dcache_page(page);
6089                 } else if (create && PageUptodate(page)) {
6090                         BUG();
6091                         if (!trans) {
6092                                 kunmap(page);
6093                                 free_extent_map(em);
6094                                 em = NULL;
6095
6096                                 btrfs_release_path(path);
6097                                 trans = btrfs_join_transaction(root);
6098
6099                                 if (IS_ERR(trans))
6100                                         return ERR_CAST(trans);
6101                                 goto again;
6102                         }
6103                         map = kmap(page);
6104                         write_extent_buffer(leaf, map + pg_offset, ptr,
6105                                             copy_size);
6106                         kunmap(page);
6107                         btrfs_mark_buffer_dirty(leaf);
6108                 }
6109                 set_extent_uptodate(io_tree, em->start,
6110                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6111                 goto insert;
6112         } else {
6113                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6114         }
6115 not_found:
6116         em->start = start;
6117         em->orig_start = start;
6118         em->len = len;
6119 not_found_em:
6120         em->block_start = EXTENT_MAP_HOLE;
6121         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6122 insert:
6123         btrfs_release_path(path);
6124         if (em->start > start || extent_map_end(em) <= start) {
6125                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6126                         (unsigned long long)em->start,
6127                         (unsigned long long)em->len,
6128                         (unsigned long long)start,
6129                         (unsigned long long)len);
6130                 err = -EIO;
6131                 goto out;
6132         }
6133
6134         err = 0;
6135         write_lock(&em_tree->lock);
6136         ret = add_extent_mapping(em_tree, em, 0);
6137         /* it is possible that someone inserted the extent into the tree
6138          * while we had the lock dropped.  It is also possible that
6139          * an overlapping map exists in the tree
6140          */
6141         if (ret == -EEXIST) {
6142                 struct extent_map *existing;
6143
6144                 ret = 0;
6145
6146                 existing = lookup_extent_mapping(em_tree, start, len);
6147                 if (existing && (existing->start > start ||
6148                     existing->start + existing->len <= start)) {
6149                         free_extent_map(existing);
6150                         existing = NULL;
6151                 }
6152                 if (!existing) {
6153                         existing = lookup_extent_mapping(em_tree, em->start,
6154                                                          em->len);
6155                         if (existing) {
6156                                 err = merge_extent_mapping(em_tree, existing,
6157                                                            em, start,
6158                                                            root->sectorsize);
6159                                 free_extent_map(existing);
6160                                 if (err) {
6161                                         free_extent_map(em);
6162                                         em = NULL;
6163                                 }
6164                         } else {
6165                                 err = -EIO;
6166                                 free_extent_map(em);
6167                                 em = NULL;
6168                         }
6169                 } else {
6170                         free_extent_map(em);
6171                         em = existing;
6172                         err = 0;
6173                 }
6174         }
6175         write_unlock(&em_tree->lock);
6176 out:
6177
6178         if (em)
6179                 trace_btrfs_get_extent(root, em);
6180
6181         if (path)
6182                 btrfs_free_path(path);
6183         if (trans) {
6184                 ret = btrfs_end_transaction(trans, root);
6185                 if (!err)
6186                         err = ret;
6187         }
6188         if (err) {
6189                 free_extent_map(em);
6190                 return ERR_PTR(err);
6191         }
6192         BUG_ON(!em); /* Error is always set */
6193         return em;
6194 }
6195
6196 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6197                                            size_t pg_offset, u64 start, u64 len,
6198                                            int create)
6199 {
6200         struct extent_map *em;
6201         struct extent_map *hole_em = NULL;
6202         u64 range_start = start;
6203         u64 end;
6204         u64 found;
6205         u64 found_end;
6206         int err = 0;
6207
6208         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6209         if (IS_ERR(em))
6210                 return em;
6211         if (em) {
6212                 /*
6213                  * if our em maps to
6214                  * -  a hole or
6215                  * -  a pre-alloc extent,
6216                  * there might actually be delalloc bytes behind it.
6217                  */
6218                 if (em->block_start != EXTENT_MAP_HOLE &&
6219                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6220                         return em;
6221                 else
6222                         hole_em = em;
6223         }
6224
6225         /* check to see if we've wrapped (len == -1 or similar) */
6226         end = start + len;
6227         if (end < start)
6228                 end = (u64)-1;
6229         else
6230                 end -= 1;
6231
6232         em = NULL;
6233
6234         /* ok, we didn't find anything, lets look for delalloc */
6235         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6236                                  end, len, EXTENT_DELALLOC, 1);
6237         found_end = range_start + found;
6238         if (found_end < range_start)
6239                 found_end = (u64)-1;
6240
6241         /*
6242          * we didn't find anything useful, return
6243          * the original results from get_extent()
6244          */
6245         if (range_start > end || found_end <= start) {
6246                 em = hole_em;
6247                 hole_em = NULL;
6248                 goto out;
6249         }
6250
6251         /* adjust the range_start to make sure it doesn't
6252          * go backwards from the start they passed in
6253          */
6254         range_start = max(start,range_start);
6255         found = found_end - range_start;
6256
6257         if (found > 0) {
6258                 u64 hole_start = start;
6259                 u64 hole_len = len;
6260
6261                 em = alloc_extent_map();
6262                 if (!em) {
6263                         err = -ENOMEM;
6264                         goto out;
6265                 }
6266                 /*
6267                  * when btrfs_get_extent can't find anything it
6268                  * returns one huge hole
6269                  *
6270                  * make sure what it found really fits our range, and
6271                  * adjust to make sure it is based on the start from
6272                  * the caller
6273                  */
6274                 if (hole_em) {
6275                         u64 calc_end = extent_map_end(hole_em);
6276
6277                         if (calc_end <= start || (hole_em->start > end)) {
6278                                 free_extent_map(hole_em);
6279                                 hole_em = NULL;
6280                         } else {
6281                                 hole_start = max(hole_em->start, start);
6282                                 hole_len = calc_end - hole_start;
6283                         }
6284                 }
6285                 em->bdev = NULL;
6286                 if (hole_em && range_start > hole_start) {
6287                         /* our hole starts before our delalloc, so we
6288                          * have to return just the parts of the hole
6289                          * that go until  the delalloc starts
6290                          */
6291                         em->len = min(hole_len,
6292                                       range_start - hole_start);
6293                         em->start = hole_start;
6294                         em->orig_start = hole_start;
6295                         /*
6296                          * don't adjust block start at all,
6297                          * it is fixed at EXTENT_MAP_HOLE
6298                          */
6299                         em->block_start = hole_em->block_start;
6300                         em->block_len = hole_len;
6301                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6302                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6303                 } else {
6304                         em->start = range_start;
6305                         em->len = found;
6306                         em->orig_start = range_start;
6307                         em->block_start = EXTENT_MAP_DELALLOC;
6308                         em->block_len = found;
6309                 }
6310         } else if (hole_em) {
6311                 return hole_em;
6312         }
6313 out:
6314
6315         free_extent_map(hole_em);
6316         if (err) {
6317                 free_extent_map(em);
6318                 return ERR_PTR(err);
6319         }
6320         return em;
6321 }
6322
6323 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6324                                                   u64 start, u64 len)
6325 {
6326         struct btrfs_root *root = BTRFS_I(inode)->root;
6327         struct btrfs_trans_handle *trans;
6328         struct extent_map *em;
6329         struct btrfs_key ins;
6330         u64 alloc_hint;
6331         int ret;
6332
6333         trans = btrfs_join_transaction(root);
6334         if (IS_ERR(trans))
6335                 return ERR_CAST(trans);
6336
6337         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6338
6339         alloc_hint = get_extent_allocation_hint(inode, start, len);
6340         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6341                                    alloc_hint, &ins, 1);
6342         if (ret) {
6343                 em = ERR_PTR(ret);
6344                 goto out;
6345         }
6346
6347         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6348                               ins.offset, ins.offset, ins.offset, 0);
6349         if (IS_ERR(em))
6350                 goto out;
6351
6352         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6353                                            ins.offset, ins.offset, 0);
6354         if (ret) {
6355                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6356                 em = ERR_PTR(ret);
6357         }
6358 out:
6359         btrfs_end_transaction(trans, root);
6360         return em;
6361 }
6362
6363 /*
6364  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6365  * block must be cow'd
6366  */
6367 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6368                                       struct inode *inode, u64 offset, u64 *len,
6369                                       u64 *orig_start, u64 *orig_block_len,
6370                                       u64 *ram_bytes)
6371 {
6372         struct btrfs_path *path;
6373         int ret;
6374         struct extent_buffer *leaf;
6375         struct btrfs_root *root = BTRFS_I(inode)->root;
6376         struct btrfs_file_extent_item *fi;
6377         struct btrfs_key key;
6378         u64 disk_bytenr;
6379         u64 backref_offset;
6380         u64 extent_end;
6381         u64 num_bytes;
6382         int slot;
6383         int found_type;
6384
6385         path = btrfs_alloc_path();
6386         if (!path)
6387                 return -ENOMEM;
6388
6389         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6390                                        offset, 0);
6391         if (ret < 0)
6392                 goto out;
6393
6394         slot = path->slots[0];
6395         if (ret == 1) {
6396                 if (slot == 0) {
6397                         /* can't find the item, must cow */
6398                         ret = 0;
6399                         goto out;
6400                 }
6401                 slot--;
6402         }
6403         ret = 0;
6404         leaf = path->nodes[0];
6405         btrfs_item_key_to_cpu(leaf, &key, slot);
6406         if (key.objectid != btrfs_ino(inode) ||
6407             key.type != BTRFS_EXTENT_DATA_KEY) {
6408                 /* not our file or wrong item type, must cow */
6409                 goto out;
6410         }
6411
6412         if (key.offset > offset) {
6413                 /* Wrong offset, must cow */
6414                 goto out;
6415         }
6416
6417         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6418         found_type = btrfs_file_extent_type(leaf, fi);
6419         if (found_type != BTRFS_FILE_EXTENT_REG &&
6420             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6421                 /* not a regular extent, must cow */
6422                 goto out;
6423         }
6424         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6425         backref_offset = btrfs_file_extent_offset(leaf, fi);
6426
6427         *orig_start = key.offset - backref_offset;
6428         *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6429         *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6430
6431         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6432         if (extent_end < offset + *len) {
6433                 /* extent doesn't include our full range, must cow */
6434                 goto out;
6435         }
6436
6437         if (btrfs_extent_readonly(root, disk_bytenr))
6438                 goto out;
6439
6440         /*
6441          * look for other files referencing this extent, if we
6442          * find any we must cow
6443          */
6444         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6445                                   key.offset - backref_offset, disk_bytenr))
6446                 goto out;
6447
6448         /*
6449          * adjust disk_bytenr and num_bytes to cover just the bytes
6450          * in this extent we are about to write.  If there
6451          * are any csums in that range we have to cow in order
6452          * to keep the csums correct
6453          */
6454         disk_bytenr += backref_offset;
6455         disk_bytenr += offset - key.offset;
6456         num_bytes = min(offset + *len, extent_end) - offset;
6457         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6458                                 goto out;
6459         /*
6460          * all of the above have passed, it is safe to overwrite this extent
6461          * without cow
6462          */
6463         *len = num_bytes;
6464         ret = 1;
6465 out:
6466         btrfs_free_path(path);
6467         return ret;
6468 }
6469
6470 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6471                               struct extent_state **cached_state, int writing)
6472 {
6473         struct btrfs_ordered_extent *ordered;
6474         int ret = 0;
6475
6476         while (1) {
6477                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6478                                  0, cached_state);
6479                 /*
6480                  * We're concerned with the entire range that we're going to be
6481                  * doing DIO to, so we need to make sure theres no ordered
6482                  * extents in this range.
6483                  */
6484                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6485                                                      lockend - lockstart + 1);
6486
6487                 /*
6488                  * We need to make sure there are no buffered pages in this
6489                  * range either, we could have raced between the invalidate in
6490                  * generic_file_direct_write and locking the extent.  The
6491                  * invalidate needs to happen so that reads after a write do not
6492                  * get stale data.
6493                  */
6494                 if (!ordered && (!writing ||
6495                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6496                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6497                                     *cached_state)))
6498                         break;
6499
6500                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6501                                      cached_state, GFP_NOFS);
6502
6503                 if (ordered) {
6504                         btrfs_start_ordered_extent(inode, ordered, 1);
6505                         btrfs_put_ordered_extent(ordered);
6506                 } else {
6507                         /* Screw you mmap */
6508                         ret = filemap_write_and_wait_range(inode->i_mapping,
6509                                                            lockstart,
6510                                                            lockend);
6511                         if (ret)
6512                                 break;
6513
6514                         /*
6515                          * If we found a page that couldn't be invalidated just
6516                          * fall back to buffered.
6517                          */
6518                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6519                                         lockstart >> PAGE_CACHE_SHIFT,
6520                                         lockend >> PAGE_CACHE_SHIFT);
6521                         if (ret)
6522                                 break;
6523                 }
6524
6525                 cond_resched();
6526         }
6527
6528         return ret;
6529 }
6530
6531 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6532                                            u64 len, u64 orig_start,
6533                                            u64 block_start, u64 block_len,
6534                                            u64 orig_block_len, u64 ram_bytes,
6535                                            int type)
6536 {
6537         struct extent_map_tree *em_tree;
6538         struct extent_map *em;
6539         struct btrfs_root *root = BTRFS_I(inode)->root;
6540         int ret;
6541
6542         em_tree = &BTRFS_I(inode)->extent_tree;
6543         em = alloc_extent_map();
6544         if (!em)
6545                 return ERR_PTR(-ENOMEM);
6546
6547         em->start = start;
6548         em->orig_start = orig_start;
6549         em->mod_start = start;
6550         em->mod_len = len;
6551         em->len = len;
6552         em->block_len = block_len;
6553         em->block_start = block_start;
6554         em->bdev = root->fs_info->fs_devices->latest_bdev;
6555         em->orig_block_len = orig_block_len;
6556         em->ram_bytes = ram_bytes;
6557         em->generation = -1;
6558         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6559         if (type == BTRFS_ORDERED_PREALLOC)
6560                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6561
6562         do {
6563                 btrfs_drop_extent_cache(inode, em->start,
6564                                 em->start + em->len - 1, 0);
6565                 write_lock(&em_tree->lock);
6566                 ret = add_extent_mapping(em_tree, em, 1);
6567                 write_unlock(&em_tree->lock);
6568         } while (ret == -EEXIST);
6569
6570         if (ret) {
6571                 free_extent_map(em);
6572                 return ERR_PTR(ret);
6573         }
6574
6575         return em;
6576 }
6577
6578
6579 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6580                                    struct buffer_head *bh_result, int create)
6581 {
6582         struct extent_map *em;
6583         struct btrfs_root *root = BTRFS_I(inode)->root;
6584         struct extent_state *cached_state = NULL;
6585         u64 start = iblock << inode->i_blkbits;
6586         u64 lockstart, lockend;
6587         u64 len = bh_result->b_size;
6588         struct btrfs_trans_handle *trans;
6589         int unlock_bits = EXTENT_LOCKED;
6590         int ret = 0;
6591
6592         if (create)
6593                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6594         else
6595                 len = min_t(u64, len, root->sectorsize);
6596
6597         lockstart = start;
6598         lockend = start + len - 1;
6599
6600         /*
6601          * If this errors out it's because we couldn't invalidate pagecache for
6602          * this range and we need to fallback to buffered.
6603          */
6604         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6605                 return -ENOTBLK;
6606
6607         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6608         if (IS_ERR(em)) {
6609                 ret = PTR_ERR(em);
6610                 goto unlock_err;
6611         }
6612
6613         /*
6614          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6615          * io.  INLINE is special, and we could probably kludge it in here, but
6616          * it's still buffered so for safety lets just fall back to the generic
6617          * buffered path.
6618          *
6619          * For COMPRESSED we _have_ to read the entire extent in so we can
6620          * decompress it, so there will be buffering required no matter what we
6621          * do, so go ahead and fallback to buffered.
6622          *
6623          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6624          * to buffered IO.  Don't blame me, this is the price we pay for using
6625          * the generic code.
6626          */
6627         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6628             em->block_start == EXTENT_MAP_INLINE) {
6629                 free_extent_map(em);
6630                 ret = -ENOTBLK;
6631                 goto unlock_err;
6632         }
6633
6634         /* Just a good old fashioned hole, return */
6635         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6636                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6637                 free_extent_map(em);
6638                 goto unlock_err;
6639         }
6640
6641         /*
6642          * We don't allocate a new extent in the following cases
6643          *
6644          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6645          * existing extent.
6646          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6647          * just use the extent.
6648          *
6649          */
6650         if (!create) {
6651                 len = min(len, em->len - (start - em->start));
6652                 lockstart = start + len;
6653                 goto unlock;
6654         }
6655
6656         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6657             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6658              em->block_start != EXTENT_MAP_HOLE)) {
6659                 int type;
6660                 int ret;
6661                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6662
6663                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6664                         type = BTRFS_ORDERED_PREALLOC;
6665                 else
6666                         type = BTRFS_ORDERED_NOCOW;
6667                 len = min(len, em->len - (start - em->start));
6668                 block_start = em->block_start + (start - em->start);
6669
6670                 /*
6671                  * we're not going to log anything, but we do need
6672                  * to make sure the current transaction stays open
6673                  * while we look for nocow cross refs
6674                  */
6675                 trans = btrfs_join_transaction(root);
6676                 if (IS_ERR(trans))
6677                         goto must_cow;
6678
6679                 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6680                                       &orig_block_len, &ram_bytes) == 1) {
6681                         if (type == BTRFS_ORDERED_PREALLOC) {
6682                                 free_extent_map(em);
6683                                 em = create_pinned_em(inode, start, len,
6684                                                        orig_start,
6685                                                        block_start, len,
6686                                                        orig_block_len,
6687                                                        ram_bytes, type);
6688                                 if (IS_ERR(em)) {
6689                                         btrfs_end_transaction(trans, root);
6690                                         goto unlock_err;
6691                                 }
6692                         }
6693
6694                         ret = btrfs_add_ordered_extent_dio(inode, start,
6695                                            block_start, len, len, type);
6696                         btrfs_end_transaction(trans, root);
6697                         if (ret) {
6698                                 free_extent_map(em);
6699                                 goto unlock_err;
6700                         }
6701                         goto unlock;
6702                 }
6703                 btrfs_end_transaction(trans, root);
6704         }
6705 must_cow:
6706         /*
6707          * this will cow the extent, reset the len in case we changed
6708          * it above
6709          */
6710         len = bh_result->b_size;
6711         free_extent_map(em);
6712         em = btrfs_new_extent_direct(inode, start, len);
6713         if (IS_ERR(em)) {
6714                 ret = PTR_ERR(em);
6715                 goto unlock_err;
6716         }
6717         len = min(len, em->len - (start - em->start));
6718 unlock:
6719         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6720                 inode->i_blkbits;
6721         bh_result->b_size = len;
6722         bh_result->b_bdev = em->bdev;
6723         set_buffer_mapped(bh_result);
6724         if (create) {
6725                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6726                         set_buffer_new(bh_result);
6727
6728                 /*
6729                  * Need to update the i_size under the extent lock so buffered
6730                  * readers will get the updated i_size when we unlock.
6731                  */
6732                 if (start + len > i_size_read(inode))
6733                         i_size_write(inode, start + len);
6734
6735                 spin_lock(&BTRFS_I(inode)->lock);
6736                 BTRFS_I(inode)->outstanding_extents++;
6737                 spin_unlock(&BTRFS_I(inode)->lock);
6738
6739                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6740                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6741                                      &cached_state, GFP_NOFS);
6742                 BUG_ON(ret);
6743         }
6744
6745         /*
6746          * In the case of write we need to clear and unlock the entire range,
6747          * in the case of read we need to unlock only the end area that we
6748          * aren't using if there is any left over space.
6749          */
6750         if (lockstart < lockend) {
6751                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6752                                  lockend, unlock_bits, 1, 0,
6753                                  &cached_state, GFP_NOFS);
6754         } else {
6755                 free_extent_state(cached_state);
6756         }
6757
6758         free_extent_map(em);
6759
6760         return 0;
6761
6762 unlock_err:
6763         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6764                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6765         return ret;
6766 }
6767
6768 struct btrfs_dio_private {
6769         struct inode *inode;
6770         u64 logical_offset;
6771         u64 disk_bytenr;
6772         u64 bytes;
6773         void *private;
6774
6775         /* number of bios pending for this dio */
6776         atomic_t pending_bios;
6777
6778         /* IO errors */
6779         int errors;
6780
6781         /* orig_bio is our btrfs_io_bio */
6782         struct bio *orig_bio;
6783
6784         /* dio_bio came from fs/direct-io.c */
6785         struct bio *dio_bio;
6786 };
6787
6788 static void btrfs_endio_direct_read(struct bio *bio, int err)
6789 {
6790         struct btrfs_dio_private *dip = bio->bi_private;
6791         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6792         struct bio_vec *bvec = bio->bi_io_vec;
6793         struct inode *inode = dip->inode;
6794         struct btrfs_root *root = BTRFS_I(inode)->root;
6795         struct bio *dio_bio;
6796         u64 start;
6797
6798         start = dip->logical_offset;
6799         do {
6800                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6801                         struct page *page = bvec->bv_page;
6802                         char *kaddr;
6803                         u32 csum = ~(u32)0;
6804                         u64 private = ~(u32)0;
6805                         unsigned long flags;
6806
6807                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6808                                               start, &private))
6809                                 goto failed;
6810                         local_irq_save(flags);
6811                         kaddr = kmap_atomic(page);
6812                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6813                                                csum, bvec->bv_len);
6814                         btrfs_csum_final(csum, (char *)&csum);
6815                         kunmap_atomic(kaddr);
6816                         local_irq_restore(flags);
6817
6818                         flush_dcache_page(bvec->bv_page);
6819                         if (csum != private) {
6820 failed:
6821                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6822                                         (unsigned long long)btrfs_ino(inode),
6823                                         (unsigned long long)start,
6824                                         csum, (unsigned)private);
6825                                 err = -EIO;
6826                         }
6827                 }
6828
6829                 start += bvec->bv_len;
6830                 bvec++;
6831         } while (bvec <= bvec_end);
6832
6833         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6834                       dip->logical_offset + dip->bytes - 1);
6835         dio_bio = dip->dio_bio;
6836
6837         kfree(dip);
6838
6839         /* If we had a csum failure make sure to clear the uptodate flag */
6840         if (err)
6841                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6842         dio_end_io(dio_bio, err);
6843         bio_put(bio);
6844 }
6845
6846 static void btrfs_endio_direct_write(struct bio *bio, int err)
6847 {
6848         struct btrfs_dio_private *dip = bio->bi_private;
6849         struct inode *inode = dip->inode;
6850         struct btrfs_root *root = BTRFS_I(inode)->root;
6851         struct btrfs_ordered_extent *ordered = NULL;
6852         u64 ordered_offset = dip->logical_offset;
6853         u64 ordered_bytes = dip->bytes;
6854         struct bio *dio_bio;
6855         int ret;
6856
6857         if (err)
6858                 goto out_done;
6859 again:
6860         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6861                                                    &ordered_offset,
6862                                                    ordered_bytes, !err);
6863         if (!ret)
6864                 goto out_test;
6865
6866         ordered->work.func = finish_ordered_fn;
6867         ordered->work.flags = 0;
6868         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6869                            &ordered->work);
6870 out_test:
6871         /*
6872          * our bio might span multiple ordered extents.  If we haven't
6873          * completed the accounting for the whole dio, go back and try again
6874          */
6875         if (ordered_offset < dip->logical_offset + dip->bytes) {
6876                 ordered_bytes = dip->logical_offset + dip->bytes -
6877                         ordered_offset;
6878                 ordered = NULL;
6879                 goto again;
6880         }
6881 out_done:
6882         dio_bio = dip->dio_bio;
6883
6884         kfree(dip);
6885
6886         /* If we had an error make sure to clear the uptodate flag */
6887         if (err)
6888                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6889         dio_end_io(dio_bio, err);
6890         bio_put(bio);
6891 }
6892
6893 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6894                                     struct bio *bio, int mirror_num,
6895                                     unsigned long bio_flags, u64 offset)
6896 {
6897         int ret;
6898         struct btrfs_root *root = BTRFS_I(inode)->root;
6899         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6900         BUG_ON(ret); /* -ENOMEM */
6901         return 0;
6902 }
6903
6904 static void btrfs_end_dio_bio(struct bio *bio, int err)
6905 {
6906         struct btrfs_dio_private *dip = bio->bi_private;
6907
6908         if (err) {
6909                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6910                       "sector %#Lx len %u err no %d\n",
6911                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6912                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6913                 dip->errors = 1;
6914
6915                 /*
6916                  * before atomic variable goto zero, we must make sure
6917                  * dip->errors is perceived to be set.
6918                  */
6919                 smp_mb__before_atomic_dec();
6920         }
6921
6922         /* if there are more bios still pending for this dio, just exit */
6923         if (!atomic_dec_and_test(&dip->pending_bios))
6924                 goto out;
6925
6926         if (dip->errors) {
6927                 bio_io_error(dip->orig_bio);
6928         } else {
6929                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6930                 bio_endio(dip->orig_bio, 0);
6931         }
6932 out:
6933         bio_put(bio);
6934 }
6935
6936 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6937                                        u64 first_sector, gfp_t gfp_flags)
6938 {
6939         int nr_vecs = bio_get_nr_vecs(bdev);
6940         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6941 }
6942
6943 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6944                                          int rw, u64 file_offset, int skip_sum,
6945                                          int async_submit)
6946 {
6947         int write = rw & REQ_WRITE;
6948         struct btrfs_root *root = BTRFS_I(inode)->root;
6949         int ret;
6950
6951         if (async_submit)
6952                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6953
6954         bio_get(bio);
6955
6956         if (!write) {
6957                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6958                 if (ret)
6959                         goto err;
6960         }
6961
6962         if (skip_sum)
6963                 goto map;
6964
6965         if (write && async_submit) {
6966                 ret = btrfs_wq_submit_bio(root->fs_info,
6967                                    inode, rw, bio, 0, 0,
6968                                    file_offset,
6969                                    __btrfs_submit_bio_start_direct_io,
6970                                    __btrfs_submit_bio_done);
6971                 goto err;
6972         } else if (write) {
6973                 /*
6974                  * If we aren't doing async submit, calculate the csum of the
6975                  * bio now.
6976                  */
6977                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6978                 if (ret)
6979                         goto err;
6980         } else if (!skip_sum) {
6981                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6982                 if (ret)
6983                         goto err;
6984         }
6985
6986 map:
6987         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6988 err:
6989         bio_put(bio);
6990         return ret;
6991 }
6992
6993 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6994                                     int skip_sum)
6995 {
6996         struct inode *inode = dip->inode;
6997         struct btrfs_root *root = BTRFS_I(inode)->root;
6998         struct bio *bio;
6999         struct bio *orig_bio = dip->orig_bio;
7000         struct bio_vec *bvec = orig_bio->bi_io_vec;
7001         u64 start_sector = orig_bio->bi_sector;
7002         u64 file_offset = dip->logical_offset;
7003         u64 submit_len = 0;
7004         u64 map_length;
7005         int nr_pages = 0;
7006         int ret = 0;
7007         int async_submit = 0;
7008
7009         map_length = orig_bio->bi_size;
7010         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7011                               &map_length, NULL, 0);
7012         if (ret) {
7013                 bio_put(orig_bio);
7014                 return -EIO;
7015         }
7016         if (map_length >= orig_bio->bi_size) {
7017                 bio = orig_bio;
7018                 goto submit;
7019         }
7020
7021         /* async crcs make it difficult to collect full stripe writes. */
7022         if (btrfs_get_alloc_profile(root, 1) &
7023             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7024                 async_submit = 0;
7025         else
7026                 async_submit = 1;
7027
7028         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7029         if (!bio)
7030                 return -ENOMEM;
7031         bio->bi_private = dip;
7032         bio->bi_end_io = btrfs_end_dio_bio;
7033         atomic_inc(&dip->pending_bios);
7034
7035         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7036                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7037                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7038                                  bvec->bv_offset) < bvec->bv_len)) {
7039                         /*
7040                          * inc the count before we submit the bio so
7041                          * we know the end IO handler won't happen before
7042                          * we inc the count. Otherwise, the dip might get freed
7043                          * before we're done setting it up
7044                          */
7045                         atomic_inc(&dip->pending_bios);
7046                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7047                                                      file_offset, skip_sum,
7048                                                      async_submit);
7049                         if (ret) {
7050                                 bio_put(bio);
7051                                 atomic_dec(&dip->pending_bios);
7052                                 goto out_err;
7053                         }
7054
7055                         start_sector += submit_len >> 9;
7056                         file_offset += submit_len;
7057
7058                         submit_len = 0;
7059                         nr_pages = 0;
7060
7061                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7062                                                   start_sector, GFP_NOFS);
7063                         if (!bio)
7064                                 goto out_err;
7065                         bio->bi_private = dip;
7066                         bio->bi_end_io = btrfs_end_dio_bio;
7067
7068                         map_length = orig_bio->bi_size;
7069                         ret = btrfs_map_block(root->fs_info, rw,
7070                                               start_sector << 9,
7071                                               &map_length, NULL, 0);
7072                         if (ret) {
7073                                 bio_put(bio);
7074                                 goto out_err;
7075                         }
7076                 } else {
7077                         submit_len += bvec->bv_len;
7078                         nr_pages ++;
7079                         bvec++;
7080                 }
7081         }
7082
7083 submit:
7084         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7085                                      async_submit);
7086         if (!ret)
7087                 return 0;
7088
7089         bio_put(bio);
7090 out_err:
7091         dip->errors = 1;
7092         /*
7093          * before atomic variable goto zero, we must
7094          * make sure dip->errors is perceived to be set.
7095          */
7096         smp_mb__before_atomic_dec();
7097         if (atomic_dec_and_test(&dip->pending_bios))
7098                 bio_io_error(dip->orig_bio);
7099
7100         /* bio_end_io() will handle error, so we needn't return it */
7101         return 0;
7102 }
7103
7104 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7105                                 struct inode *inode, loff_t file_offset)
7106 {
7107         struct btrfs_root *root = BTRFS_I(inode)->root;
7108         struct btrfs_dio_private *dip;
7109         struct bio_vec *bvec = dio_bio->bi_io_vec;
7110         struct bio *io_bio;
7111         int skip_sum;
7112         int write = rw & REQ_WRITE;
7113         int ret = 0;
7114
7115         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7116
7117         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7118
7119         if (!io_bio) {
7120                 ret = -ENOMEM;
7121                 goto free_ordered;
7122         }
7123
7124         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7125         if (!dip) {
7126                 ret = -ENOMEM;
7127                 goto free_io_bio;
7128         }
7129
7130         dip->private = dio_bio->bi_private;
7131         io_bio->bi_private = dio_bio->bi_private;
7132         dip->inode = inode;
7133         dip->logical_offset = file_offset;
7134
7135         dip->bytes = 0;
7136         do {
7137                 dip->bytes += bvec->bv_len;
7138                 bvec++;
7139         } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7140
7141         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7142         io_bio->bi_private = dip;
7143         dip->errors = 0;
7144         dip->orig_bio = io_bio;
7145         dip->dio_bio = dio_bio;
7146         atomic_set(&dip->pending_bios, 0);
7147
7148         if (write)
7149                 io_bio->bi_end_io = btrfs_endio_direct_write;
7150         else
7151                 io_bio->bi_end_io = btrfs_endio_direct_read;
7152
7153         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7154         if (!ret)
7155                 return;
7156
7157 free_io_bio:
7158         bio_put(io_bio);
7159
7160 free_ordered:
7161         /*
7162          * If this is a write, we need to clean up the reserved space and kill
7163          * the ordered extent.
7164          */
7165         if (write) {
7166                 struct btrfs_ordered_extent *ordered;
7167                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7168                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7169                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7170                         btrfs_free_reserved_extent(root, ordered->start,
7171                                                    ordered->disk_len);
7172                 btrfs_put_ordered_extent(ordered);
7173                 btrfs_put_ordered_extent(ordered);
7174         }
7175         bio_endio(dio_bio, ret);
7176 }
7177
7178 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7179                         const struct iovec *iov, loff_t offset,
7180                         unsigned long nr_segs)
7181 {
7182         int seg;
7183         int i;
7184         size_t size;
7185         unsigned long addr;
7186         unsigned blocksize_mask = root->sectorsize - 1;
7187         ssize_t retval = -EINVAL;
7188         loff_t end = offset;
7189
7190         if (offset & blocksize_mask)
7191                 goto out;
7192
7193         /* Check the memory alignment.  Blocks cannot straddle pages */
7194         for (seg = 0; seg < nr_segs; seg++) {
7195                 addr = (unsigned long)iov[seg].iov_base;
7196                 size = iov[seg].iov_len;
7197                 end += size;
7198                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7199                         goto out;
7200
7201                 /* If this is a write we don't need to check anymore */
7202                 if (rw & WRITE)
7203                         continue;
7204
7205                 /*
7206                  * Check to make sure we don't have duplicate iov_base's in this
7207                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7208                  * when reading back.
7209                  */
7210                 for (i = seg + 1; i < nr_segs; i++) {
7211                         if (iov[seg].iov_base == iov[i].iov_base)
7212                                 goto out;
7213                 }
7214         }
7215         retval = 0;
7216 out:
7217         return retval;
7218 }
7219
7220 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7221                         const struct iovec *iov, loff_t offset,
7222                         unsigned long nr_segs)
7223 {
7224         struct file *file = iocb->ki_filp;
7225         struct inode *inode = file->f_mapping->host;
7226         size_t count = 0;
7227         int flags = 0;
7228         bool wakeup = true;
7229         bool relock = false;
7230         ssize_t ret;
7231
7232         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7233                             offset, nr_segs))
7234                 return 0;
7235
7236         atomic_inc(&inode->i_dio_count);
7237         smp_mb__after_atomic_inc();
7238
7239         if (rw & WRITE) {
7240                 count = iov_length(iov, nr_segs);
7241                 /*
7242                  * If the write DIO is beyond the EOF, we need update
7243                  * the isize, but it is protected by i_mutex. So we can
7244                  * not unlock the i_mutex at this case.
7245                  */
7246                 if (offset + count <= inode->i_size) {
7247                         mutex_unlock(&inode->i_mutex);
7248                         relock = true;
7249                 }
7250                 ret = btrfs_delalloc_reserve_space(inode, count);
7251                 if (ret)
7252                         goto out;
7253         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7254                                      &BTRFS_I(inode)->runtime_flags))) {
7255                 inode_dio_done(inode);
7256                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7257                 wakeup = false;
7258         }
7259
7260         ret = __blockdev_direct_IO(rw, iocb, inode,
7261                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7262                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7263                         btrfs_submit_direct, flags);
7264         if (rw & WRITE) {
7265                 if (ret < 0 && ret != -EIOCBQUEUED)
7266                         btrfs_delalloc_release_space(inode, count);
7267                 else if (ret >= 0 && (size_t)ret < count)
7268                         btrfs_delalloc_release_space(inode,
7269                                                      count - (size_t)ret);
7270                 else
7271                         btrfs_delalloc_release_metadata(inode, 0);
7272         }
7273 out:
7274         if (wakeup)
7275                 inode_dio_done(inode);
7276         if (relock)
7277                 mutex_lock(&inode->i_mutex);
7278
7279         return ret;
7280 }
7281
7282 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7283
7284 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7285                 __u64 start, __u64 len)
7286 {
7287         int     ret;
7288
7289         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7290         if (ret)
7291                 return ret;
7292
7293         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7294 }
7295
7296 int btrfs_readpage(struct file *file, struct page *page)
7297 {
7298         struct extent_io_tree *tree;
7299         tree = &BTRFS_I(page->mapping->host)->io_tree;
7300         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7301 }
7302
7303 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7304 {
7305         struct extent_io_tree *tree;
7306
7307
7308         if (current->flags & PF_MEMALLOC) {
7309                 redirty_page_for_writepage(wbc, page);
7310                 unlock_page(page);
7311                 return 0;
7312         }
7313         tree = &BTRFS_I(page->mapping->host)->io_tree;
7314         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7315 }
7316
7317 static int btrfs_writepages(struct address_space *mapping,
7318                             struct writeback_control *wbc)
7319 {
7320         struct extent_io_tree *tree;
7321
7322         tree = &BTRFS_I(mapping->host)->io_tree;
7323         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7324 }
7325
7326 static int
7327 btrfs_readpages(struct file *file, struct address_space *mapping,
7328                 struct list_head *pages, unsigned nr_pages)
7329 {
7330         struct extent_io_tree *tree;
7331         tree = &BTRFS_I(mapping->host)->io_tree;
7332         return extent_readpages(tree, mapping, pages, nr_pages,
7333                                 btrfs_get_extent);
7334 }
7335 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7336 {
7337         struct extent_io_tree *tree;
7338         struct extent_map_tree *map;
7339         int ret;
7340
7341         tree = &BTRFS_I(page->mapping->host)->io_tree;
7342         map = &BTRFS_I(page->mapping->host)->extent_tree;
7343         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7344         if (ret == 1) {
7345                 ClearPagePrivate(page);
7346                 set_page_private(page, 0);
7347                 page_cache_release(page);
7348         }
7349         return ret;
7350 }
7351
7352 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7353 {
7354         if (PageWriteback(page) || PageDirty(page))
7355                 return 0;
7356         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7357 }
7358
7359 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7360 {
7361         struct inode *inode = page->mapping->host;
7362         struct extent_io_tree *tree;
7363         struct btrfs_ordered_extent *ordered;
7364         struct extent_state *cached_state = NULL;
7365         u64 page_start = page_offset(page);
7366         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7367
7368         /*
7369          * we have the page locked, so new writeback can't start,
7370          * and the dirty bit won't be cleared while we are here.
7371          *
7372          * Wait for IO on this page so that we can safely clear
7373          * the PagePrivate2 bit and do ordered accounting
7374          */
7375         wait_on_page_writeback(page);
7376
7377         tree = &BTRFS_I(inode)->io_tree;
7378         if (offset) {
7379                 btrfs_releasepage(page, GFP_NOFS);
7380                 return;
7381         }
7382         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7383         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7384         if (ordered) {
7385                 /*
7386                  * IO on this page will never be started, so we need
7387                  * to account for any ordered extents now
7388                  */
7389                 clear_extent_bit(tree, page_start, page_end,
7390                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7391                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7392                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7393                 /*
7394                  * whoever cleared the private bit is responsible
7395                  * for the finish_ordered_io
7396                  */
7397                 if (TestClearPagePrivate2(page) &&
7398                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7399                                                    PAGE_CACHE_SIZE, 1)) {
7400                         btrfs_finish_ordered_io(ordered);
7401                 }
7402                 btrfs_put_ordered_extent(ordered);
7403                 cached_state = NULL;
7404                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7405         }
7406         clear_extent_bit(tree, page_start, page_end,
7407                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7408                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7409                  &cached_state, GFP_NOFS);
7410         __btrfs_releasepage(page, GFP_NOFS);
7411
7412         ClearPageChecked(page);
7413         if (PagePrivate(page)) {
7414                 ClearPagePrivate(page);
7415                 set_page_private(page, 0);
7416                 page_cache_release(page);
7417         }
7418 }
7419
7420 /*
7421  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7422  * called from a page fault handler when a page is first dirtied. Hence we must
7423  * be careful to check for EOF conditions here. We set the page up correctly
7424  * for a written page which means we get ENOSPC checking when writing into
7425  * holes and correct delalloc and unwritten extent mapping on filesystems that
7426  * support these features.
7427  *
7428  * We are not allowed to take the i_mutex here so we have to play games to
7429  * protect against truncate races as the page could now be beyond EOF.  Because
7430  * vmtruncate() writes the inode size before removing pages, once we have the
7431  * page lock we can determine safely if the page is beyond EOF. If it is not
7432  * beyond EOF, then the page is guaranteed safe against truncation until we
7433  * unlock the page.
7434  */
7435 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7436 {
7437         struct page *page = vmf->page;
7438         struct inode *inode = file_inode(vma->vm_file);
7439         struct btrfs_root *root = BTRFS_I(inode)->root;
7440         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7441         struct btrfs_ordered_extent *ordered;
7442         struct extent_state *cached_state = NULL;
7443         char *kaddr;
7444         unsigned long zero_start;
7445         loff_t size;
7446         int ret;
7447         int reserved = 0;
7448         u64 page_start;
7449         u64 page_end;
7450
7451         sb_start_pagefault(inode->i_sb);
7452         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7453         if (!ret) {
7454                 ret = file_update_time(vma->vm_file);
7455                 reserved = 1;
7456         }
7457         if (ret) {
7458                 if (ret == -ENOMEM)
7459                         ret = VM_FAULT_OOM;
7460                 else /* -ENOSPC, -EIO, etc */
7461                         ret = VM_FAULT_SIGBUS;
7462                 if (reserved)
7463                         goto out;
7464                 goto out_noreserve;
7465         }
7466
7467         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7468 again:
7469         lock_page(page);
7470         size = i_size_read(inode);
7471         page_start = page_offset(page);
7472         page_end = page_start + PAGE_CACHE_SIZE - 1;
7473
7474         if ((page->mapping != inode->i_mapping) ||
7475             (page_start >= size)) {
7476                 /* page got truncated out from underneath us */
7477                 goto out_unlock;
7478         }
7479         wait_on_page_writeback(page);
7480
7481         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7482         set_page_extent_mapped(page);
7483
7484         /*
7485          * we can't set the delalloc bits if there are pending ordered
7486          * extents.  Drop our locks and wait for them to finish
7487          */
7488         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7489         if (ordered) {
7490                 unlock_extent_cached(io_tree, page_start, page_end,
7491                                      &cached_state, GFP_NOFS);
7492                 unlock_page(page);
7493                 btrfs_start_ordered_extent(inode, ordered, 1);
7494                 btrfs_put_ordered_extent(ordered);
7495                 goto again;
7496         }
7497
7498         /*
7499          * XXX - page_mkwrite gets called every time the page is dirtied, even
7500          * if it was already dirty, so for space accounting reasons we need to
7501          * clear any delalloc bits for the range we are fixing to save.  There
7502          * is probably a better way to do this, but for now keep consistent with
7503          * prepare_pages in the normal write path.
7504          */
7505         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7506                           EXTENT_DIRTY | EXTENT_DELALLOC |
7507                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7508                           0, 0, &cached_state, GFP_NOFS);
7509
7510         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7511                                         &cached_state);
7512         if (ret) {
7513                 unlock_extent_cached(io_tree, page_start, page_end,
7514                                      &cached_state, GFP_NOFS);
7515                 ret = VM_FAULT_SIGBUS;
7516                 goto out_unlock;
7517         }
7518         ret = 0;
7519
7520         /* page is wholly or partially inside EOF */
7521         if (page_start + PAGE_CACHE_SIZE > size)
7522                 zero_start = size & ~PAGE_CACHE_MASK;
7523         else
7524                 zero_start = PAGE_CACHE_SIZE;
7525
7526         if (zero_start != PAGE_CACHE_SIZE) {
7527                 kaddr = kmap(page);
7528                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7529                 flush_dcache_page(page);
7530                 kunmap(page);
7531         }
7532         ClearPageChecked(page);
7533         set_page_dirty(page);
7534         SetPageUptodate(page);
7535
7536         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7537         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7538         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7539
7540         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7541
7542 out_unlock:
7543         if (!ret) {
7544                 sb_end_pagefault(inode->i_sb);
7545                 return VM_FAULT_LOCKED;
7546         }
7547         unlock_page(page);
7548 out:
7549         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7550 out_noreserve:
7551         sb_end_pagefault(inode->i_sb);
7552         return ret;
7553 }
7554
7555 static int btrfs_truncate(struct inode *inode)
7556 {
7557         struct btrfs_root *root = BTRFS_I(inode)->root;
7558         struct btrfs_block_rsv *rsv;
7559         int ret;
7560         int err = 0;
7561         struct btrfs_trans_handle *trans;
7562         u64 mask = root->sectorsize - 1;
7563         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7564
7565         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7566         if (ret)
7567                 return ret;
7568
7569         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7570         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7571
7572         /*
7573          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7574          * 3 things going on here
7575          *
7576          * 1) We need to reserve space for our orphan item and the space to
7577          * delete our orphan item.  Lord knows we don't want to have a dangling
7578          * orphan item because we didn't reserve space to remove it.
7579          *
7580          * 2) We need to reserve space to update our inode.
7581          *
7582          * 3) We need to have something to cache all the space that is going to
7583          * be free'd up by the truncate operation, but also have some slack
7584          * space reserved in case it uses space during the truncate (thank you
7585          * very much snapshotting).
7586          *
7587          * And we need these to all be seperate.  The fact is we can use alot of
7588          * space doing the truncate, and we have no earthly idea how much space
7589          * we will use, so we need the truncate reservation to be seperate so it
7590          * doesn't end up using space reserved for updating the inode or
7591          * removing the orphan item.  We also need to be able to stop the
7592          * transaction and start a new one, which means we need to be able to
7593          * update the inode several times, and we have no idea of knowing how
7594          * many times that will be, so we can't just reserve 1 item for the
7595          * entirety of the opration, so that has to be done seperately as well.
7596          * Then there is the orphan item, which does indeed need to be held on
7597          * to for the whole operation, and we need nobody to touch this reserved
7598          * space except the orphan code.
7599          *
7600          * So that leaves us with
7601          *
7602          * 1) root->orphan_block_rsv - for the orphan deletion.
7603          * 2) rsv - for the truncate reservation, which we will steal from the
7604          * transaction reservation.
7605          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7606          * updating the inode.
7607          */
7608         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7609         if (!rsv)
7610                 return -ENOMEM;
7611         rsv->size = min_size;
7612         rsv->failfast = 1;
7613
7614         /*
7615          * 1 for the truncate slack space
7616          * 1 for updating the inode.
7617          */
7618         trans = btrfs_start_transaction(root, 2);
7619         if (IS_ERR(trans)) {
7620                 err = PTR_ERR(trans);
7621                 goto out;
7622         }
7623
7624         /* Migrate the slack space for the truncate to our reserve */
7625         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7626                                       min_size);
7627         BUG_ON(ret);
7628
7629         /*
7630          * setattr is responsible for setting the ordered_data_close flag,
7631          * but that is only tested during the last file release.  That
7632          * could happen well after the next commit, leaving a great big
7633          * window where new writes may get lost if someone chooses to write
7634          * to this file after truncating to zero
7635          *
7636          * The inode doesn't have any dirty data here, and so if we commit
7637          * this is a noop.  If someone immediately starts writing to the inode
7638          * it is very likely we'll catch some of their writes in this
7639          * transaction, and the commit will find this file on the ordered
7640          * data list with good things to send down.
7641          *
7642          * This is a best effort solution, there is still a window where
7643          * using truncate to replace the contents of the file will
7644          * end up with a zero length file after a crash.
7645          */
7646         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7647                                            &BTRFS_I(inode)->runtime_flags))
7648                 btrfs_add_ordered_operation(trans, root, inode);
7649
7650         /*
7651          * So if we truncate and then write and fsync we normally would just
7652          * write the extents that changed, which is a problem if we need to
7653          * first truncate that entire inode.  So set this flag so we write out
7654          * all of the extents in the inode to the sync log so we're completely
7655          * safe.
7656          */
7657         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7658         trans->block_rsv = rsv;
7659
7660         while (1) {
7661                 ret = btrfs_truncate_inode_items(trans, root, inode,
7662                                                  inode->i_size,
7663                                                  BTRFS_EXTENT_DATA_KEY);
7664                 if (ret != -ENOSPC) {
7665                         err = ret;
7666                         break;
7667                 }
7668
7669                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7670                 ret = btrfs_update_inode(trans, root, inode);
7671                 if (ret) {
7672                         err = ret;
7673                         break;
7674                 }
7675
7676                 btrfs_end_transaction(trans, root);
7677                 btrfs_btree_balance_dirty(root);
7678
7679                 trans = btrfs_start_transaction(root, 2);
7680                 if (IS_ERR(trans)) {
7681                         ret = err = PTR_ERR(trans);
7682                         trans = NULL;
7683                         break;
7684                 }
7685
7686                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7687                                               rsv, min_size);
7688                 BUG_ON(ret);    /* shouldn't happen */
7689                 trans->block_rsv = rsv;
7690         }
7691
7692         if (ret == 0 && inode->i_nlink > 0) {
7693                 trans->block_rsv = root->orphan_block_rsv;
7694                 ret = btrfs_orphan_del(trans, inode);
7695                 if (ret)
7696                         err = ret;
7697         }
7698
7699         if (trans) {
7700                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7701                 ret = btrfs_update_inode(trans, root, inode);
7702                 if (ret && !err)
7703                         err = ret;
7704
7705                 ret = btrfs_end_transaction(trans, root);
7706                 btrfs_btree_balance_dirty(root);
7707         }
7708
7709 out:
7710         btrfs_free_block_rsv(root, rsv);
7711
7712         if (ret && !err)
7713                 err = ret;
7714
7715         return err;
7716 }
7717
7718 /*
7719  * create a new subvolume directory/inode (helper for the ioctl).
7720  */
7721 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7722                              struct btrfs_root *new_root, u64 new_dirid)
7723 {
7724         struct inode *inode;
7725         int err;
7726         u64 index = 0;
7727
7728         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7729                                 new_dirid, new_dirid,
7730                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7731                                 &index);
7732         if (IS_ERR(inode))
7733                 return PTR_ERR(inode);
7734         inode->i_op = &btrfs_dir_inode_operations;
7735         inode->i_fop = &btrfs_dir_file_operations;
7736
7737         set_nlink(inode, 1);
7738         btrfs_i_size_write(inode, 0);
7739
7740         err = btrfs_update_inode(trans, new_root, inode);
7741
7742         iput(inode);
7743         return err;
7744 }
7745
7746 struct inode *btrfs_alloc_inode(struct super_block *sb)
7747 {
7748         struct btrfs_inode *ei;
7749         struct inode *inode;
7750
7751         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7752         if (!ei)
7753                 return NULL;
7754
7755         ei->root = NULL;
7756         ei->generation = 0;
7757         ei->last_trans = 0;
7758         ei->last_sub_trans = 0;
7759         ei->logged_trans = 0;
7760         ei->delalloc_bytes = 0;
7761         ei->disk_i_size = 0;
7762         ei->flags = 0;
7763         ei->csum_bytes = 0;
7764         ei->index_cnt = (u64)-1;
7765         ei->last_unlink_trans = 0;
7766         ei->last_log_commit = 0;
7767
7768         spin_lock_init(&ei->lock);
7769         ei->outstanding_extents = 0;
7770         ei->reserved_extents = 0;
7771
7772         ei->runtime_flags = 0;
7773         ei->force_compress = BTRFS_COMPRESS_NONE;
7774
7775         ei->delayed_node = NULL;
7776
7777         inode = &ei->vfs_inode;
7778         extent_map_tree_init(&ei->extent_tree);
7779         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7780         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7781         ei->io_tree.track_uptodate = 1;
7782         ei->io_failure_tree.track_uptodate = 1;
7783         atomic_set(&ei->sync_writers, 0);
7784         mutex_init(&ei->log_mutex);
7785         mutex_init(&ei->delalloc_mutex);
7786         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7787         INIT_LIST_HEAD(&ei->delalloc_inodes);
7788         INIT_LIST_HEAD(&ei->ordered_operations);
7789         RB_CLEAR_NODE(&ei->rb_node);
7790
7791         return inode;
7792 }
7793
7794 static void btrfs_i_callback(struct rcu_head *head)
7795 {
7796         struct inode *inode = container_of(head, struct inode, i_rcu);
7797         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7798 }
7799
7800 void btrfs_destroy_inode(struct inode *inode)
7801 {
7802         struct btrfs_ordered_extent *ordered;
7803         struct btrfs_root *root = BTRFS_I(inode)->root;
7804
7805         WARN_ON(!hlist_empty(&inode->i_dentry));
7806         WARN_ON(inode->i_data.nrpages);
7807         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7808         WARN_ON(BTRFS_I(inode)->reserved_extents);
7809         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7810         WARN_ON(BTRFS_I(inode)->csum_bytes);
7811
7812         /*
7813          * This can happen where we create an inode, but somebody else also
7814          * created the same inode and we need to destroy the one we already
7815          * created.
7816          */
7817         if (!root)
7818                 goto free;
7819
7820         /*
7821          * Make sure we're properly removed from the ordered operation
7822          * lists.
7823          */
7824         smp_mb();
7825         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7826                 spin_lock(&root->fs_info->ordered_root_lock);
7827                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7828                 spin_unlock(&root->fs_info->ordered_root_lock);
7829         }
7830
7831         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7832                      &BTRFS_I(inode)->runtime_flags)) {
7833                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7834                         (unsigned long long)btrfs_ino(inode));
7835                 atomic_dec(&root->orphan_inodes);
7836         }
7837
7838         while (1) {
7839                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7840                 if (!ordered)
7841                         break;
7842                 else {
7843                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7844                                 (unsigned long long)ordered->file_offset,
7845                                 (unsigned long long)ordered->len);
7846                         btrfs_remove_ordered_extent(inode, ordered);
7847                         btrfs_put_ordered_extent(ordered);
7848                         btrfs_put_ordered_extent(ordered);
7849                 }
7850         }
7851         inode_tree_del(inode);
7852         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7853 free:
7854         call_rcu(&inode->i_rcu, btrfs_i_callback);
7855 }
7856
7857 int btrfs_drop_inode(struct inode *inode)
7858 {
7859         struct btrfs_root *root = BTRFS_I(inode)->root;
7860
7861         if (root == NULL)
7862                 return 1;
7863
7864         /* the snap/subvol tree is on deleting */
7865         if (btrfs_root_refs(&root->root_item) == 0 &&
7866             root != root->fs_info->tree_root)
7867                 return 1;
7868         else
7869                 return generic_drop_inode(inode);
7870 }
7871
7872 static void init_once(void *foo)
7873 {
7874         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7875
7876         inode_init_once(&ei->vfs_inode);
7877 }
7878
7879 void btrfs_destroy_cachep(void)
7880 {
7881         /*
7882          * Make sure all delayed rcu free inodes are flushed before we
7883          * destroy cache.
7884          */
7885         rcu_barrier();
7886         if (btrfs_inode_cachep)
7887                 kmem_cache_destroy(btrfs_inode_cachep);
7888         if (btrfs_trans_handle_cachep)
7889                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7890         if (btrfs_transaction_cachep)
7891                 kmem_cache_destroy(btrfs_transaction_cachep);
7892         if (btrfs_path_cachep)
7893                 kmem_cache_destroy(btrfs_path_cachep);
7894         if (btrfs_free_space_cachep)
7895                 kmem_cache_destroy(btrfs_free_space_cachep);
7896         if (btrfs_delalloc_work_cachep)
7897                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7898 }
7899
7900 int btrfs_init_cachep(void)
7901 {
7902         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7903                         sizeof(struct btrfs_inode), 0,
7904                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7905         if (!btrfs_inode_cachep)
7906                 goto fail;
7907
7908         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7909                         sizeof(struct btrfs_trans_handle), 0,
7910                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7911         if (!btrfs_trans_handle_cachep)
7912                 goto fail;
7913
7914         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7915                         sizeof(struct btrfs_transaction), 0,
7916                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7917         if (!btrfs_transaction_cachep)
7918                 goto fail;
7919
7920         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7921                         sizeof(struct btrfs_path), 0,
7922                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7923         if (!btrfs_path_cachep)
7924                 goto fail;
7925
7926         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7927                         sizeof(struct btrfs_free_space), 0,
7928                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7929         if (!btrfs_free_space_cachep)
7930                 goto fail;
7931
7932         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7933                         sizeof(struct btrfs_delalloc_work), 0,
7934                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7935                         NULL);
7936         if (!btrfs_delalloc_work_cachep)
7937                 goto fail;
7938
7939         return 0;
7940 fail:
7941         btrfs_destroy_cachep();
7942         return -ENOMEM;
7943 }
7944
7945 static int btrfs_getattr(struct vfsmount *mnt,
7946                          struct dentry *dentry, struct kstat *stat)
7947 {
7948         u64 delalloc_bytes;
7949         struct inode *inode = dentry->d_inode;
7950         u32 blocksize = inode->i_sb->s_blocksize;
7951
7952         generic_fillattr(inode, stat);
7953         stat->dev = BTRFS_I(inode)->root->anon_dev;
7954         stat->blksize = PAGE_CACHE_SIZE;
7955
7956         spin_lock(&BTRFS_I(inode)->lock);
7957         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7958         spin_unlock(&BTRFS_I(inode)->lock);
7959         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7960                         ALIGN(delalloc_bytes, blocksize)) >> 9;
7961         return 0;
7962 }
7963
7964 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7965                            struct inode *new_dir, struct dentry *new_dentry)
7966 {
7967         struct btrfs_trans_handle *trans;
7968         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7969         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7970         struct inode *new_inode = new_dentry->d_inode;
7971         struct inode *old_inode = old_dentry->d_inode;
7972         struct timespec ctime = CURRENT_TIME;
7973         u64 index = 0;
7974         u64 root_objectid;
7975         int ret;
7976         u64 old_ino = btrfs_ino(old_inode);
7977
7978         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7979                 return -EPERM;
7980
7981         /* we only allow rename subvolume link between subvolumes */
7982         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7983                 return -EXDEV;
7984
7985         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7986             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7987                 return -ENOTEMPTY;
7988
7989         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7990             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7991                 return -ENOTEMPTY;
7992
7993
7994         /* check for collisions, even if the  name isn't there */
7995         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7996                              new_dentry->d_name.name,
7997                              new_dentry->d_name.len);
7998
7999         if (ret) {
8000                 if (ret == -EEXIST) {
8001                         /* we shouldn't get
8002                          * eexist without a new_inode */
8003                         if (!new_inode) {
8004                                 WARN_ON(1);
8005                                 return ret;
8006                         }
8007                 } else {
8008                         /* maybe -EOVERFLOW */
8009                         return ret;
8010                 }
8011         }
8012         ret = 0;
8013
8014         /*
8015          * we're using rename to replace one file with another.
8016          * and the replacement file is large.  Start IO on it now so
8017          * we don't add too much work to the end of the transaction
8018          */
8019         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8020             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8021                 filemap_flush(old_inode->i_mapping);
8022
8023         /* close the racy window with snapshot create/destroy ioctl */
8024         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8025                 down_read(&root->fs_info->subvol_sem);
8026         /*
8027          * We want to reserve the absolute worst case amount of items.  So if
8028          * both inodes are subvols and we need to unlink them then that would
8029          * require 4 item modifications, but if they are both normal inodes it
8030          * would require 5 item modifications, so we'll assume their normal
8031          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8032          * should cover the worst case number of items we'll modify.
8033          */
8034         trans = btrfs_start_transaction(root, 11);
8035         if (IS_ERR(trans)) {
8036                 ret = PTR_ERR(trans);
8037                 goto out_notrans;
8038         }
8039
8040         if (dest != root)
8041                 btrfs_record_root_in_trans(trans, dest);
8042
8043         ret = btrfs_set_inode_index(new_dir, &index);
8044         if (ret)
8045                 goto out_fail;
8046
8047         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8048                 /* force full log commit if subvolume involved. */
8049                 root->fs_info->last_trans_log_full_commit = trans->transid;
8050         } else {
8051                 ret = btrfs_insert_inode_ref(trans, dest,
8052                                              new_dentry->d_name.name,
8053                                              new_dentry->d_name.len,
8054                                              old_ino,
8055                                              btrfs_ino(new_dir), index);
8056                 if (ret)
8057                         goto out_fail;
8058                 /*
8059                  * this is an ugly little race, but the rename is required
8060                  * to make sure that if we crash, the inode is either at the
8061                  * old name or the new one.  pinning the log transaction lets
8062                  * us make sure we don't allow a log commit to come in after
8063                  * we unlink the name but before we add the new name back in.
8064                  */
8065                 btrfs_pin_log_trans(root);
8066         }
8067         /*
8068          * make sure the inode gets flushed if it is replacing
8069          * something.
8070          */
8071         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8072                 btrfs_add_ordered_operation(trans, root, old_inode);
8073
8074         inode_inc_iversion(old_dir);
8075         inode_inc_iversion(new_dir);
8076         inode_inc_iversion(old_inode);
8077         old_dir->i_ctime = old_dir->i_mtime = ctime;
8078         new_dir->i_ctime = new_dir->i_mtime = ctime;
8079         old_inode->i_ctime = ctime;
8080
8081         if (old_dentry->d_parent != new_dentry->d_parent)
8082                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8083
8084         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8085                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8086                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8087                                         old_dentry->d_name.name,
8088                                         old_dentry->d_name.len);
8089         } else {
8090                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8091                                         old_dentry->d_inode,
8092                                         old_dentry->d_name.name,
8093                                         old_dentry->d_name.len);
8094                 if (!ret)
8095                         ret = btrfs_update_inode(trans, root, old_inode);
8096         }
8097         if (ret) {
8098                 btrfs_abort_transaction(trans, root, ret);
8099                 goto out_fail;
8100         }
8101
8102         if (new_inode) {
8103                 inode_inc_iversion(new_inode);
8104                 new_inode->i_ctime = CURRENT_TIME;
8105                 if (unlikely(btrfs_ino(new_inode) ==
8106                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8107                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8108                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8109                                                 root_objectid,
8110                                                 new_dentry->d_name.name,
8111                                                 new_dentry->d_name.len);
8112                         BUG_ON(new_inode->i_nlink == 0);
8113                 } else {
8114                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8115                                                  new_dentry->d_inode,
8116                                                  new_dentry->d_name.name,
8117                                                  new_dentry->d_name.len);
8118                 }
8119                 if (!ret && new_inode->i_nlink == 0) {
8120                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8121                         BUG_ON(ret);
8122                 }
8123                 if (ret) {
8124                         btrfs_abort_transaction(trans, root, ret);
8125                         goto out_fail;
8126                 }
8127         }
8128
8129         ret = btrfs_add_link(trans, new_dir, old_inode,
8130                              new_dentry->d_name.name,
8131                              new_dentry->d_name.len, 0, index);
8132         if (ret) {
8133                 btrfs_abort_transaction(trans, root, ret);
8134                 goto out_fail;
8135         }
8136
8137         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8138                 struct dentry *parent = new_dentry->d_parent;
8139                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8140                 btrfs_end_log_trans(root);
8141         }
8142 out_fail:
8143         btrfs_end_transaction(trans, root);
8144 out_notrans:
8145         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8146                 up_read(&root->fs_info->subvol_sem);
8147
8148         return ret;
8149 }
8150
8151 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8152 {
8153         struct btrfs_delalloc_work *delalloc_work;
8154
8155         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8156                                      work);
8157         if (delalloc_work->wait)
8158                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8159         else
8160                 filemap_flush(delalloc_work->inode->i_mapping);
8161
8162         if (delalloc_work->delay_iput)
8163                 btrfs_add_delayed_iput(delalloc_work->inode);
8164         else
8165                 iput(delalloc_work->inode);
8166         complete(&delalloc_work->completion);
8167 }
8168
8169 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8170                                                     int wait, int delay_iput)
8171 {
8172         struct btrfs_delalloc_work *work;
8173
8174         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8175         if (!work)
8176                 return NULL;
8177
8178         init_completion(&work->completion);
8179         INIT_LIST_HEAD(&work->list);
8180         work->inode = inode;
8181         work->wait = wait;
8182         work->delay_iput = delay_iput;
8183         work->work.func = btrfs_run_delalloc_work;
8184
8185         return work;
8186 }
8187
8188 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8189 {
8190         wait_for_completion(&work->completion);
8191         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8192 }
8193
8194 /*
8195  * some fairly slow code that needs optimization. This walks the list
8196  * of all the inodes with pending delalloc and forces them to disk.
8197  */
8198 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8199 {
8200         struct btrfs_inode *binode;
8201         struct inode *inode;
8202         struct btrfs_delalloc_work *work, *next;
8203         struct list_head works;
8204         struct list_head splice;
8205         int ret = 0;
8206
8207         INIT_LIST_HEAD(&works);
8208         INIT_LIST_HEAD(&splice);
8209
8210         spin_lock(&root->delalloc_lock);
8211         list_splice_init(&root->delalloc_inodes, &splice);
8212         while (!list_empty(&splice)) {
8213                 binode = list_entry(splice.next, struct btrfs_inode,
8214                                     delalloc_inodes);
8215
8216                 list_move_tail(&binode->delalloc_inodes,
8217                                &root->delalloc_inodes);
8218                 inode = igrab(&binode->vfs_inode);
8219                 if (!inode) {
8220                         cond_resched_lock(&root->delalloc_lock);
8221                         continue;
8222                 }
8223                 spin_unlock(&root->delalloc_lock);
8224
8225                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8226                 if (unlikely(!work)) {
8227                         ret = -ENOMEM;
8228                         goto out;
8229                 }
8230                 list_add_tail(&work->list, &works);
8231                 btrfs_queue_worker(&root->fs_info->flush_workers,
8232                                    &work->work);
8233
8234                 cond_resched();
8235                 spin_lock(&root->delalloc_lock);
8236         }
8237         spin_unlock(&root->delalloc_lock);
8238
8239         list_for_each_entry_safe(work, next, &works, list) {
8240                 list_del_init(&work->list);
8241                 btrfs_wait_and_free_delalloc_work(work);
8242         }
8243         return 0;
8244 out:
8245         list_for_each_entry_safe(work, next, &works, list) {
8246                 list_del_init(&work->list);
8247                 btrfs_wait_and_free_delalloc_work(work);
8248         }
8249
8250         if (!list_empty_careful(&splice)) {
8251                 spin_lock(&root->delalloc_lock);
8252                 list_splice_tail(&splice, &root->delalloc_inodes);
8253                 spin_unlock(&root->delalloc_lock);
8254         }
8255         return ret;
8256 }
8257
8258 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8259 {
8260         int ret;
8261
8262         if (root->fs_info->sb->s_flags & MS_RDONLY)
8263                 return -EROFS;
8264
8265         ret = __start_delalloc_inodes(root, delay_iput);
8266         /*
8267          * the filemap_flush will queue IO into the worker threads, but
8268          * we have to make sure the IO is actually started and that
8269          * ordered extents get created before we return
8270          */
8271         atomic_inc(&root->fs_info->async_submit_draining);
8272         while (atomic_read(&root->fs_info->nr_async_submits) ||
8273               atomic_read(&root->fs_info->async_delalloc_pages)) {
8274                 wait_event(root->fs_info->async_submit_wait,
8275                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8276                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8277         }
8278         atomic_dec(&root->fs_info->async_submit_draining);
8279         return ret;
8280 }
8281
8282 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8283                                     int delay_iput)
8284 {
8285         struct btrfs_root *root;
8286         struct list_head splice;
8287         int ret;
8288
8289         if (fs_info->sb->s_flags & MS_RDONLY)
8290                 return -EROFS;
8291
8292         INIT_LIST_HEAD(&splice);
8293
8294         spin_lock(&fs_info->delalloc_root_lock);
8295         list_splice_init(&fs_info->delalloc_roots, &splice);
8296         while (!list_empty(&splice)) {
8297                 root = list_first_entry(&splice, struct btrfs_root,
8298                                         delalloc_root);
8299                 root = btrfs_grab_fs_root(root);
8300                 BUG_ON(!root);
8301                 list_move_tail(&root->delalloc_root,
8302                                &fs_info->delalloc_roots);
8303                 spin_unlock(&fs_info->delalloc_root_lock);
8304
8305                 ret = __start_delalloc_inodes(root, delay_iput);
8306                 btrfs_put_fs_root(root);
8307                 if (ret)
8308                         goto out;
8309
8310                 spin_lock(&fs_info->delalloc_root_lock);
8311         }
8312         spin_unlock(&fs_info->delalloc_root_lock);
8313
8314         atomic_inc(&fs_info->async_submit_draining);
8315         while (atomic_read(&fs_info->nr_async_submits) ||
8316               atomic_read(&fs_info->async_delalloc_pages)) {
8317                 wait_event(fs_info->async_submit_wait,
8318                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8319                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8320         }
8321         atomic_dec(&fs_info->async_submit_draining);
8322         return 0;
8323 out:
8324         if (!list_empty_careful(&splice)) {
8325                 spin_lock(&fs_info->delalloc_root_lock);
8326                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8327                 spin_unlock(&fs_info->delalloc_root_lock);
8328         }
8329         return ret;
8330 }
8331
8332 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8333                          const char *symname)
8334 {
8335         struct btrfs_trans_handle *trans;
8336         struct btrfs_root *root = BTRFS_I(dir)->root;
8337         struct btrfs_path *path;
8338         struct btrfs_key key;
8339         struct inode *inode = NULL;
8340         int err;
8341         int drop_inode = 0;
8342         u64 objectid;
8343         u64 index = 0 ;
8344         int name_len;
8345         int datasize;
8346         unsigned long ptr;
8347         struct btrfs_file_extent_item *ei;
8348         struct extent_buffer *leaf;
8349
8350         name_len = strlen(symname) + 1;
8351         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8352                 return -ENAMETOOLONG;
8353
8354         /*
8355          * 2 items for inode item and ref
8356          * 2 items for dir items
8357          * 1 item for xattr if selinux is on
8358          */
8359         trans = btrfs_start_transaction(root, 5);
8360         if (IS_ERR(trans))
8361                 return PTR_ERR(trans);
8362
8363         err = btrfs_find_free_ino(root, &objectid);
8364         if (err)
8365                 goto out_unlock;
8366
8367         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8368                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8369                                 S_IFLNK|S_IRWXUGO, &index);
8370         if (IS_ERR(inode)) {
8371                 err = PTR_ERR(inode);
8372                 goto out_unlock;
8373         }
8374
8375         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8376         if (err) {
8377                 drop_inode = 1;
8378                 goto out_unlock;
8379         }
8380
8381         /*
8382         * If the active LSM wants to access the inode during
8383         * d_instantiate it needs these. Smack checks to see
8384         * if the filesystem supports xattrs by looking at the
8385         * ops vector.
8386         */
8387         inode->i_fop = &btrfs_file_operations;
8388         inode->i_op = &btrfs_file_inode_operations;
8389
8390         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8391         if (err)
8392                 drop_inode = 1;
8393         else {
8394                 inode->i_mapping->a_ops = &btrfs_aops;
8395                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8396                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8397         }
8398         if (drop_inode)
8399                 goto out_unlock;
8400
8401         path = btrfs_alloc_path();
8402         if (!path) {
8403                 err = -ENOMEM;
8404                 drop_inode = 1;
8405                 goto out_unlock;
8406         }
8407         key.objectid = btrfs_ino(inode);
8408         key.offset = 0;
8409         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8410         datasize = btrfs_file_extent_calc_inline_size(name_len);
8411         err = btrfs_insert_empty_item(trans, root, path, &key,
8412                                       datasize);
8413         if (err) {
8414                 drop_inode = 1;
8415                 btrfs_free_path(path);
8416                 goto out_unlock;
8417         }
8418         leaf = path->nodes[0];
8419         ei = btrfs_item_ptr(leaf, path->slots[0],
8420                             struct btrfs_file_extent_item);
8421         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8422         btrfs_set_file_extent_type(leaf, ei,
8423                                    BTRFS_FILE_EXTENT_INLINE);
8424         btrfs_set_file_extent_encryption(leaf, ei, 0);
8425         btrfs_set_file_extent_compression(leaf, ei, 0);
8426         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8427         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8428
8429         ptr = btrfs_file_extent_inline_start(ei);
8430         write_extent_buffer(leaf, symname, ptr, name_len);
8431         btrfs_mark_buffer_dirty(leaf);
8432         btrfs_free_path(path);
8433
8434         inode->i_op = &btrfs_symlink_inode_operations;
8435         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8436         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8437         inode_set_bytes(inode, name_len);
8438         btrfs_i_size_write(inode, name_len - 1);
8439         err = btrfs_update_inode(trans, root, inode);
8440         if (err)
8441                 drop_inode = 1;
8442
8443 out_unlock:
8444         if (!err)
8445                 d_instantiate(dentry, inode);
8446         btrfs_end_transaction(trans, root);
8447         if (drop_inode) {
8448                 inode_dec_link_count(inode);
8449                 iput(inode);
8450         }
8451         btrfs_btree_balance_dirty(root);
8452         return err;
8453 }
8454
8455 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8456                                        u64 start, u64 num_bytes, u64 min_size,
8457                                        loff_t actual_len, u64 *alloc_hint,
8458                                        struct btrfs_trans_handle *trans)
8459 {
8460         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8461         struct extent_map *em;
8462         struct btrfs_root *root = BTRFS_I(inode)->root;
8463         struct btrfs_key ins;
8464         u64 cur_offset = start;
8465         u64 i_size;
8466         u64 cur_bytes;
8467         int ret = 0;
8468         bool own_trans = true;
8469
8470         if (trans)
8471                 own_trans = false;
8472         while (num_bytes > 0) {
8473                 if (own_trans) {
8474                         trans = btrfs_start_transaction(root, 3);
8475                         if (IS_ERR(trans)) {
8476                                 ret = PTR_ERR(trans);
8477                                 break;
8478                         }
8479                 }
8480
8481                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8482                 cur_bytes = max(cur_bytes, min_size);
8483                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8484                                            min_size, 0, *alloc_hint, &ins, 1);
8485                 if (ret) {
8486                         if (own_trans)
8487                                 btrfs_end_transaction(trans, root);
8488                         break;
8489                 }
8490
8491                 ret = insert_reserved_file_extent(trans, inode,
8492                                                   cur_offset, ins.objectid,
8493                                                   ins.offset, ins.offset,
8494                                                   ins.offset, 0, 0, 0,
8495                                                   BTRFS_FILE_EXTENT_PREALLOC);
8496                 if (ret) {
8497                         btrfs_abort_transaction(trans, root, ret);
8498                         if (own_trans)
8499                                 btrfs_end_transaction(trans, root);
8500                         break;
8501                 }
8502                 btrfs_drop_extent_cache(inode, cur_offset,
8503                                         cur_offset + ins.offset -1, 0);
8504
8505                 em = alloc_extent_map();
8506                 if (!em) {
8507                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8508                                 &BTRFS_I(inode)->runtime_flags);
8509                         goto next;
8510                 }
8511
8512                 em->start = cur_offset;
8513                 em->orig_start = cur_offset;
8514                 em->len = ins.offset;
8515                 em->block_start = ins.objectid;
8516                 em->block_len = ins.offset;
8517                 em->orig_block_len = ins.offset;
8518                 em->ram_bytes = ins.offset;
8519                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8520                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8521                 em->generation = trans->transid;
8522
8523                 while (1) {
8524                         write_lock(&em_tree->lock);
8525                         ret = add_extent_mapping(em_tree, em, 1);
8526                         write_unlock(&em_tree->lock);
8527                         if (ret != -EEXIST)
8528                                 break;
8529                         btrfs_drop_extent_cache(inode, cur_offset,
8530                                                 cur_offset + ins.offset - 1,
8531                                                 0);
8532                 }
8533                 free_extent_map(em);
8534 next:
8535                 num_bytes -= ins.offset;
8536                 cur_offset += ins.offset;
8537                 *alloc_hint = ins.objectid + ins.offset;
8538
8539                 inode_inc_iversion(inode);
8540                 inode->i_ctime = CURRENT_TIME;
8541                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8542                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8543                     (actual_len > inode->i_size) &&
8544                     (cur_offset > inode->i_size)) {
8545                         if (cur_offset > actual_len)
8546                                 i_size = actual_len;
8547                         else
8548                                 i_size = cur_offset;
8549                         i_size_write(inode, i_size);
8550                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8551                 }
8552
8553                 ret = btrfs_update_inode(trans, root, inode);
8554
8555                 if (ret) {
8556                         btrfs_abort_transaction(trans, root, ret);
8557                         if (own_trans)
8558                                 btrfs_end_transaction(trans, root);
8559                         break;
8560                 }
8561
8562                 if (own_trans)
8563                         btrfs_end_transaction(trans, root);
8564         }
8565         return ret;
8566 }
8567
8568 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8569                               u64 start, u64 num_bytes, u64 min_size,
8570                               loff_t actual_len, u64 *alloc_hint)
8571 {
8572         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8573                                            min_size, actual_len, alloc_hint,
8574                                            NULL);
8575 }
8576
8577 int btrfs_prealloc_file_range_trans(struct inode *inode,
8578                                     struct btrfs_trans_handle *trans, int mode,
8579                                     u64 start, u64 num_bytes, u64 min_size,
8580                                     loff_t actual_len, u64 *alloc_hint)
8581 {
8582         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8583                                            min_size, actual_len, alloc_hint, trans);
8584 }
8585
8586 static int btrfs_set_page_dirty(struct page *page)
8587 {
8588         return __set_page_dirty_nobuffers(page);
8589 }
8590
8591 static int btrfs_permission(struct inode *inode, int mask)
8592 {
8593         struct btrfs_root *root = BTRFS_I(inode)->root;
8594         umode_t mode = inode->i_mode;
8595
8596         if (mask & MAY_WRITE &&
8597             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8598                 if (btrfs_root_readonly(root))
8599                         return -EROFS;
8600                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8601                         return -EACCES;
8602         }
8603         return generic_permission(inode, mask);
8604 }
8605
8606 static const struct inode_operations btrfs_dir_inode_operations = {
8607         .getattr        = btrfs_getattr,
8608         .lookup         = btrfs_lookup,
8609         .create         = btrfs_create,
8610         .unlink         = btrfs_unlink,
8611         .link           = btrfs_link,
8612         .mkdir          = btrfs_mkdir,
8613         .rmdir          = btrfs_rmdir,
8614         .rename         = btrfs_rename,
8615         .symlink        = btrfs_symlink,
8616         .setattr        = btrfs_setattr,
8617         .mknod          = btrfs_mknod,
8618         .setxattr       = btrfs_setxattr,
8619         .getxattr       = btrfs_getxattr,
8620         .listxattr      = btrfs_listxattr,
8621         .removexattr    = btrfs_removexattr,
8622         .permission     = btrfs_permission,
8623         .get_acl        = btrfs_get_acl,
8624 };
8625 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8626         .lookup         = btrfs_lookup,
8627         .permission     = btrfs_permission,
8628         .get_acl        = btrfs_get_acl,
8629 };
8630
8631 static const struct file_operations btrfs_dir_file_operations = {
8632         .llseek         = generic_file_llseek,
8633         .read           = generic_read_dir,
8634         .readdir        = btrfs_real_readdir,
8635         .unlocked_ioctl = btrfs_ioctl,
8636 #ifdef CONFIG_COMPAT
8637         .compat_ioctl   = btrfs_ioctl,
8638 #endif
8639         .release        = btrfs_release_file,
8640         .fsync          = btrfs_sync_file,
8641 };
8642
8643 static struct extent_io_ops btrfs_extent_io_ops = {
8644         .fill_delalloc = run_delalloc_range,
8645         .submit_bio_hook = btrfs_submit_bio_hook,
8646         .merge_bio_hook = btrfs_merge_bio_hook,
8647         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8648         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8649         .writepage_start_hook = btrfs_writepage_start_hook,
8650         .set_bit_hook = btrfs_set_bit_hook,
8651         .clear_bit_hook = btrfs_clear_bit_hook,
8652         .merge_extent_hook = btrfs_merge_extent_hook,
8653         .split_extent_hook = btrfs_split_extent_hook,
8654 };
8655
8656 /*
8657  * btrfs doesn't support the bmap operation because swapfiles
8658  * use bmap to make a mapping of extents in the file.  They assume
8659  * these extents won't change over the life of the file and they
8660  * use the bmap result to do IO directly to the drive.
8661  *
8662  * the btrfs bmap call would return logical addresses that aren't
8663  * suitable for IO and they also will change frequently as COW
8664  * operations happen.  So, swapfile + btrfs == corruption.
8665  *
8666  * For now we're avoiding this by dropping bmap.
8667  */
8668 static const struct address_space_operations btrfs_aops = {
8669         .readpage       = btrfs_readpage,
8670         .writepage      = btrfs_writepage,
8671         .writepages     = btrfs_writepages,
8672         .readpages      = btrfs_readpages,
8673         .direct_IO      = btrfs_direct_IO,
8674         .invalidatepage = btrfs_invalidatepage,
8675         .releasepage    = btrfs_releasepage,
8676         .set_page_dirty = btrfs_set_page_dirty,
8677         .error_remove_page = generic_error_remove_page,
8678 };
8679
8680 static const struct address_space_operations btrfs_symlink_aops = {
8681         .readpage       = btrfs_readpage,
8682         .writepage      = btrfs_writepage,
8683         .invalidatepage = btrfs_invalidatepage,
8684         .releasepage    = btrfs_releasepage,
8685 };
8686
8687 static const struct inode_operations btrfs_file_inode_operations = {
8688         .getattr        = btrfs_getattr,
8689         .setattr        = btrfs_setattr,
8690         .setxattr       = btrfs_setxattr,
8691         .getxattr       = btrfs_getxattr,
8692         .listxattr      = btrfs_listxattr,
8693         .removexattr    = btrfs_removexattr,
8694         .permission     = btrfs_permission,
8695         .fiemap         = btrfs_fiemap,
8696         .get_acl        = btrfs_get_acl,
8697         .update_time    = btrfs_update_time,
8698 };
8699 static const struct inode_operations btrfs_special_inode_operations = {
8700         .getattr        = btrfs_getattr,
8701         .setattr        = btrfs_setattr,
8702         .permission     = btrfs_permission,
8703         .setxattr       = btrfs_setxattr,
8704         .getxattr       = btrfs_getxattr,
8705         .listxattr      = btrfs_listxattr,
8706         .removexattr    = btrfs_removexattr,
8707         .get_acl        = btrfs_get_acl,
8708         .update_time    = btrfs_update_time,
8709 };
8710 static const struct inode_operations btrfs_symlink_inode_operations = {
8711         .readlink       = generic_readlink,
8712         .follow_link    = page_follow_link_light,
8713         .put_link       = page_put_link,
8714         .getattr        = btrfs_getattr,
8715         .setattr        = btrfs_setattr,
8716         .permission     = btrfs_permission,
8717         .setxattr       = btrfs_setxattr,
8718         .getxattr       = btrfs_getxattr,
8719         .listxattr      = btrfs_listxattr,
8720         .removexattr    = btrfs_removexattr,
8721         .get_acl        = btrfs_get_acl,
8722         .update_time    = btrfs_update_time,
8723 };
8724
8725 const struct dentry_operations btrfs_dentry_operations = {
8726         .d_delete       = btrfs_dentry_delete,
8727         .d_release      = btrfs_dentry_release,
8728 };