fs/btrfs/file.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/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/mpage.h>
  27 #include <linux/aio.h>
  28 #include <linux/falloc.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/slab.h>
  34 #include <linux/btrfs.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "print-tree.h"
  40 #include "tree-log.h"
  41 #include "locking.h"
  42 #include "volumes.h"
  43
  44 static struct kmem_cache *btrfs_inode_defrag_cachep;
  45 /*
  46  * when auto defrag is enabled we
  47  * queue up these defrag structs to remember which
  48  * inodes need defragging passes
  49  */
  50 struct inode_defrag {
  51         struct rb_node rb_node;
  52         /* objectid */
  53         u64 ino;
  54         /*
  55          * transid where the defrag was added, we search for
  56          * extents newer than this
  57          */
  58         u64 transid;
  59
  60         /* root objectid */
  61         u64 root;
  62
  63         /* last offset we were able to defrag */
  64         u64 last_offset;
  65
  66         /* if we've wrapped around back to zero once already */
  67         int cycled;
  68 };
  69
  70 static int __compare_inode_defrag(struct inode_defrag *defrag1,
  71                                   struct inode_defrag *defrag2)
  72 {
  73         if (defrag1->root > defrag2->root)
  74                 return 1;
  75         else if (defrag1->root < defrag2->root)
  76                 return -1;
  77         else if (defrag1->ino > defrag2->ino)
  78                 return 1;
  79         else if (defrag1->ino < defrag2->ino)
  80                 return -1;
  81         else
  82                 return 0;
  83 }
  84
  85 /* pop a record for an inode into the defrag tree.  The lock
  86  * must be held already
  87  *
  88  * If you're inserting a record for an older transid than an
  89  * existing record, the transid already in the tree is lowered
  90  *
  91  * If an existing record is found the defrag item you
  92  * pass in is freed
  93  */
  94 static int __btrfs_add_inode_defrag(struct inode *inode,
  95                                     struct inode_defrag *defrag)
  96 {
  97         struct btrfs_root *root = BTRFS_I(inode)->root;
  98         struct inode_defrag *entry;
  99         struct rb_node **p;
 100         struct rb_node *parent = NULL;
 101         int ret;
 102
 103         p = &root->fs_info->defrag_inodes.rb_node;
 104         while (*p) {
 105                 parent = *p;
 106                 entry = rb_entry(parent, struct inode_defrag, rb_node);
 107
 108                 ret = __compare_inode_defrag(defrag, entry);
 109                 if (ret < 0)
 110                         p = &parent->rb_left;
 111                 else if (ret > 0)
 112                         p = &parent->rb_right;
 113                 else {
 114                         /* if we're reinserting an entry for
 115                          * an old defrag run, make sure to
 116                          * lower the transid of our existing record
 117                          */
 118                         if (defrag->transid < entry->transid)
 119                                 entry->transid = defrag->transid;
 120                         if (defrag->last_offset > entry->last_offset)
 121                                 entry->last_offset = defrag->last_offset;
 122                         return -EEXIST;
 123                 }
 124         }
 125         set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
 126         rb_link_node(&defrag->rb_node, parent, p);
 127         rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
 128         return 0;
 129 }
 130
 131 static inline int __need_auto_defrag(struct btrfs_root *root)
 132 {
 133         if (!btrfs_test_opt(root, AUTO_DEFRAG))
 134                 return 0;
 135
 136         if (btrfs_fs_closing(root->fs_info))
 137                 return 0;
 138
 139         return 1;
 140 }
 141
 142 /*
 143  * insert a defrag record for this inode if auto defrag is
 144  * enabled
 145  */
 146 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
 147                            struct inode *inode)
 148 {
 149         struct btrfs_root *root = BTRFS_I(inode)->root;
 150         struct inode_defrag *defrag;
 151         u64 transid;
 152         int ret;
 153
 154         if (!__need_auto_defrag(root))
 155                 return 0;
 156
 157         if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
 158                 return 0;
 159
 160         if (trans)
 161                 transid = trans->transid;
 162         else
 163                 transid = BTRFS_I(inode)->root->last_trans;
 164
 165         defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
 166         if (!defrag)
 167                 return -ENOMEM;
 168
 169         defrag->ino = btrfs_ino(inode);
 170         defrag->transid = transid;
 171         defrag->root = root->root_key.objectid;
 172
 173         spin_lock(&root->fs_info->defrag_inodes_lock);
 174         if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
 175                 /*
 176                  * If we set IN_DEFRAG flag and evict the inode from memory,
 177                  * and then re-read this inode, this new inode doesn't have
 178                  * IN_DEFRAG flag. At the case, we may find the existed defrag.
 179                  */
 180                 ret = __btrfs_add_inode_defrag(inode, defrag);
 181                 if (ret)
 182                         kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 183         } else {
 184                 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 185         }
 186         spin_unlock(&root->fs_info->defrag_inodes_lock);
 187         return 0;
 188 }
 189
 190 /*
 191  * Requeue the defrag object. If there is a defrag object that points to
 192  * the same inode in the tree, we will merge them together (by
 193  * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
 194  */
 195 static void btrfs_requeue_inode_defrag(struct inode *inode,
 196                                        struct inode_defrag *defrag)
 197 {
 198         struct btrfs_root *root = BTRFS_I(inode)->root;
 199         int ret;
 200
 201         if (!__need_auto_defrag(root))
 202                 goto out;
 203
 204         /*
 205          * Here we don't check the IN_DEFRAG flag, because we need merge
 206          * them together.
 207          */
 208         spin_lock(&root->fs_info->defrag_inodes_lock);
 209         ret = __btrfs_add_inode_defrag(inode, defrag);
 210         spin_unlock(&root->fs_info->defrag_inodes_lock);
 211         if (ret)
 212                 goto out;
 213         return;
 214 out:
 215         kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 216 }
 217
 218 /*
 219  * pick the defragable inode that we want, if it doesn't exist, we will get
 220  * the next one.
 221  */
 222 static struct inode_defrag *
 223 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
 224 {
 225         struct inode_defrag *entry = NULL;
 226         struct inode_defrag tmp;
 227         struct rb_node *p;
 228         struct rb_node *parent = NULL;
 229         int ret;
 230
 231         tmp.ino = ino;
 232         tmp.root = root;
 233
 234         spin_lock(&fs_info->defrag_inodes_lock);
 235         p = fs_info->defrag_inodes.rb_node;
 236         while (p) {
 237                 parent = p;
 238                 entry = rb_entry(parent, struct inode_defrag, rb_node);
 239
 240                 ret = __compare_inode_defrag(&tmp, entry);
 241                 if (ret < 0)
 242                         p = parent->rb_left;
 243                 else if (ret > 0)
 244                         p = parent->rb_right;
 245                 else
 246                         goto out;
 247         }
 248
 249         if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
 250                 parent = rb_next(parent);
 251                 if (parent)
 252                         entry = rb_entry(parent, struct inode_defrag, rb_node);
 253                 else
 254                         entry = NULL;
 255         }
 256 out:
 257         if (entry)
 258                 rb_erase(parent, &fs_info->defrag_inodes);
 259         spin_unlock(&fs_info->defrag_inodes_lock);
 260         return entry;
 261 }
 262
 263 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
 264 {
 265         struct inode_defrag *defrag;
 266         struct rb_node *node;
 267
 268         spin_lock(&fs_info->defrag_inodes_lock);
 269         node = rb_first(&fs_info->defrag_inodes);
 270         while (node) {
 271                 rb_erase(node, &fs_info->defrag_inodes);
 272                 defrag = rb_entry(node, struct inode_defrag, rb_node);
 273                 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 274
 275                 if (need_resched()) {
 276                         spin_unlock(&fs_info->defrag_inodes_lock);
 277                         cond_resched();
 278                         spin_lock(&fs_info->defrag_inodes_lock);
 279                 }
 280
 281                 node = rb_first(&fs_info->defrag_inodes);
 282         }
 283         spin_unlock(&fs_info->defrag_inodes_lock);
 284 }
 285
 286 #define BTRFS_DEFRAG_BATCH      1024
 287
 288 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
 289                                     struct inode_defrag *defrag)
 290 {
 291         struct btrfs_root *inode_root;
 292         struct inode *inode;
 293         struct btrfs_key key;
 294         struct btrfs_ioctl_defrag_range_args range;
 295         int num_defrag;
 296         int index;
 297         int ret;
 298
 299         /* get the inode */
 300         key.objectid = defrag->root;
 301         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 302         key.offset = (u64)-1;
 303
 304         index = srcu_read_lock(&fs_info->subvol_srcu);
 305
 306         inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
 307         if (IS_ERR(inode_root)) {
 308                 ret = PTR_ERR(inode_root);
 309                 goto cleanup;
 310         }
 311
 312         key.objectid = defrag->ino;
 313         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 314         key.offset = 0;
 315         inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
 316         if (IS_ERR(inode)) {
 317                 ret = PTR_ERR(inode);
 318                 goto cleanup;
 319         }
 320         srcu_read_unlock(&fs_info->subvol_srcu, index);
 321
 322         /* do a chunk of defrag */
 323         clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
 324         memset(&range, 0, sizeof(range));
 325         range.len = (u64)-1;
 326         range.start = defrag->last_offset;
 327
 328         sb_start_write(fs_info->sb);
 329         num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
 330                                        BTRFS_DEFRAG_BATCH);
 331         sb_end_write(fs_info->sb);
 332         /*
 333          * if we filled the whole defrag batch, there
 334          * must be more work to do.  Queue this defrag
 335          * again
 336          */
 337         if (num_defrag == BTRFS_DEFRAG_BATCH) {
 338                 defrag->last_offset = range.start;
 339                 btrfs_requeue_inode_defrag(inode, defrag);
 340         } else if (defrag->last_offset && !defrag->cycled) {
 341                 /*
 342                  * we didn't fill our defrag batch, but
 343                  * we didn't start at zero.  Make sure we loop
 344                  * around to the start of the file.
 345                  */
 346                 defrag->last_offset = 0;
 347                 defrag->cycled = 1;
 348                 btrfs_requeue_inode_defrag(inode, defrag);
 349         } else {
 350                 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 351         }
 352
 353         iput(inode);
 354         return 0;
 355 cleanup:
 356         srcu_read_unlock(&fs_info->subvol_srcu, index);
 357         kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 358         return ret;
 359 }
 360
 361 /*
 362  * run through the list of inodes in the FS that need
 363  * defragging
 364  */
 365 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
 366 {
 367         struct inode_defrag *defrag;
 368         u64 first_ino = 0;
 369         u64 root_objectid = 0;
 370
 371         atomic_inc(&fs_info->defrag_running);
 372         while (1) {
 373                 /* Pause the auto defragger. */
 374                 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
 375                              &fs_info->fs_state))
 376                         break;
 377
 378                 if (!__need_auto_defrag(fs_info->tree_root))
 379                         break;
 380
 381                 /* find an inode to defrag */
 382                 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
 383                                                  first_ino);
 384                 if (!defrag) {
 385                         if (root_objectid || first_ino) {
 386                                 root_objectid = 0;
 387                                 first_ino = 0;
 388                                 continue;
 389                         } else {
 390                                 break;
 391                         }
 392                 }
 393
 394                 first_ino = defrag->ino + 1;
 395                 root_objectid = defrag->root;
 396
 397                 __btrfs_run_defrag_inode(fs_info, defrag);
 398         }
 399         atomic_dec(&fs_info->defrag_running);
 400
 401         /*
 402          * during unmount, we use the transaction_wait queue to
 403          * wait for the defragger to stop
 404          */
 405         wake_up(&fs_info->transaction_wait);
 406         return 0;
 407 }
 408
 409 /* simple helper to fault in pages and copy.  This should go away
 410  * and be replaced with calls into generic code.
 411  */
 412 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
 413                                          size_t write_bytes,
 414                                          struct page **prepared_pages,
 415                                          struct iov_iter *i)
 416 {
 417         size_t copied = 0;
 418         size_t total_copied = 0;
 419         int pg = 0;
 420         int offset = pos & (PAGE_CACHE_SIZE - 1);
 421
 422         while (write_bytes > 0) {
 423                 size_t count = min_t(size_t,
 424                                      PAGE_CACHE_SIZE - offset, write_bytes);
 425                 struct page *page = prepared_pages[pg];
 426                 /*
 427                  * Copy data from userspace to the current page
 428                  *
 429                  * Disable pagefault to avoid recursive lock since
 430                  * the pages are already locked
 431                  */
 432                 pagefault_disable();
 433                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
 434                 pagefault_enable();
 435
 436                 /* Flush processor's dcache for this page */
 437                 flush_dcache_page(page);
 438
 439                 /*
 440                  * if we get a partial write, we can end up with
 441                  * partially up to date pages.  These add
 442                  * a lot of complexity, so make sure they don't
 443                  * happen by forcing this copy to be retried.
 444                  *
 445                  * The rest of the btrfs_file_write code will fall
 446                  * back to page at a time copies after we return 0.
 447                  */
 448                 if (!PageUptodate(page) && copied < count)
 449                         copied = 0;
 450
 451                 iov_iter_advance(i, copied);
 452                 write_bytes -= copied;
 453                 total_copied += copied;
 454
 455                 /* Return to btrfs_file_aio_write to fault page */
 456                 if (unlikely(copied == 0))
 457                         break;
 458
 459                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
 460                         offset += copied;
 461                 } else {
 462                         pg++;
 463                         offset = 0;
 464                 }
 465         }
 466         return total_copied;
 467 }
 468
 469 /*
 470  * unlocks pages after btrfs_file_write is done with them
 471  */
 472 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
 473 {
 474         size_t i;
 475         for (i = 0; i < num_pages; i++) {
 476                 /* page checked is some magic around finding pages that
 477                  * have been modified without going through btrfs_set_page_dirty
 478                  * clear it here
 479                  */
 480                 ClearPageChecked(pages[i]);
 481                 unlock_page(pages[i]);
 482                 mark_page_accessed(pages[i]);
 483                 page_cache_release(pages[i]);
 484         }
 485 }
 486
 487 /*
 488  * after copy_from_user, pages need to be dirtied and we need to make
 489  * sure holes are created between the current EOF and the start of
 490  * any next extents (if required).
 491  *
 492  * this also makes the decision about creating an inline extent vs
 493  * doing real data extents, marking pages dirty and delalloc as required.
 494  */
 495 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
 496                              struct page **pages, size_t num_pages,
 497                              loff_t pos, size_t write_bytes,
 498                              struct extent_state **cached)
 499 {
 500         int err = 0;
 501         int i;
 502         u64 num_bytes;
 503         u64 start_pos;
 504         u64 end_of_last_block;
 505         u64 end_pos = pos + write_bytes;
 506         loff_t isize = i_size_read(inode);
 507
 508         start_pos = pos & ~((u64)root->sectorsize - 1);
 509         num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
 510
 511         end_of_last_block = start_pos + num_bytes - 1;
 512         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 513                                         cached);
 514         if (err)
 515                 return err;
 516
 517         for (i = 0; i < num_pages; i++) {
 518                 struct page *p = pages[i];
 519                 SetPageUptodate(p);
 520                 ClearPageChecked(p);
 521                 set_page_dirty(p);
 522         }
 523
 524         /*
 525          * we've only changed i_size in ram, and we haven't updated
 526          * the disk i_size.  There is no need to log the inode
 527          * at this time.
 528          */
 529         if (end_pos > isize)
 530                 i_size_write(inode, end_pos);
 531         return 0;
 532 }
 533
 534 /*
 535  * this drops all the extents in the cache that intersect the range
 536  * [start, end].  Existing extents are split as required.
 537  */
 538 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 539                              int skip_pinned)
 540 {
 541         struct extent_map *em;
 542         struct extent_map *split = NULL;
 543         struct extent_map *split2 = NULL;
 544         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 545         u64 len = end - start + 1;
 546         u64 gen;
 547         int ret;
 548         int testend = 1;
 549         unsigned long flags;
 550         int compressed = 0;
 551         bool modified;
 552
 553         WARN_ON(end < start);
 554         if (end == (u64)-1) {
 555                 len = (u64)-1;
 556                 testend = 0;
 557         }
 558         while (1) {
 559                 int no_splits = 0;
 560
 561                 modified = false;
 562                 if (!split)
 563                         split = alloc_extent_map();
 564                 if (!split2)
 565                         split2 = alloc_extent_map();
 566                 if (!split || !split2)
 567                         no_splits = 1;
 568
 569                 write_lock(&em_tree->lock);
 570                 em = lookup_extent_mapping(em_tree, start, len);
 571                 if (!em) {
 572                         write_unlock(&em_tree->lock);
 573                         break;
 574                 }
 575                 flags = em->flags;
 576                 gen = em->generation;
 577                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 578                         if (testend && em->start + em->len >= start + len) {
 579                                 free_extent_map(em);
 580                                 write_unlock(&em_tree->lock);
 581                                 break;
 582                         }
 583                         start = em->start + em->len;
 584                         if (testend)
 585                                 len = start + len - (em->start + em->len);
 586                         free_extent_map(em);
 587                         write_unlock(&em_tree->lock);
 588                         continue;
 589                 }
 590                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 591                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 592                 clear_bit(EXTENT_FLAG_LOGGING, &flags);
 593                 modified = !list_empty(&em->list);
 594                 remove_extent_mapping(em_tree, em);
 595                 if (no_splits)
 596                         goto next;
 597
 598                 if (em->start < start) {
 599                         split->start = em->start;
 600                         split->len = start - em->start;
 601
 602                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 603                                 split->orig_start = em->orig_start;
 604                                 split->block_start = em->block_start;
 605
 606                                 if (compressed)
 607                                         split->block_len = em->block_len;
 608                                 else
 609                                         split->block_len = split->len;
 610                                 split->orig_block_len = max(split->block_len,
 611                                                 em->orig_block_len);
 612                                 split->ram_bytes = em->ram_bytes;
 613                         } else {
 614                                 split->orig_start = split->start;
 615                                 split->block_len = 0;
 616                                 split->block_start = em->block_start;
 617                                 split->orig_block_len = 0;
 618                                 split->ram_bytes = split->len;
 619                         }
 620
 621                         split->generation = gen;
 622                         split->bdev = em->bdev;
 623                         split->flags = flags;
 624                         split->compress_type = em->compress_type;
 625                         ret = add_extent_mapping(em_tree, split, modified);
 626                         BUG_ON(ret); /* Logic error */
 627                         free_extent_map(split);
 628                         split = split2;
 629                         split2 = NULL;
 630                 }
 631                 if (testend && em->start + em->len > start + len) {
 632                         u64 diff = start + len - em->start;
 633
 634                         split->start = start + len;
 635                         split->len = em->start + em->len - (start + len);
 636                         split->bdev = em->bdev;
 637                         split->flags = flags;
 638                         split->compress_type = em->compress_type;
 639                         split->generation = gen;
 640
 641                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 642                                 split->orig_block_len = max(em->block_len,
 643                                                     em->orig_block_len);
 644
 645                                 split->ram_bytes = em->ram_bytes;
 646                                 if (compressed) {
 647                                         split->block_len = em->block_len;
 648                                         split->block_start = em->block_start;
 649                                         split->orig_start = em->orig_start;
 650                                 } else {
 651                                         split->block_len = split->len;
 652                                         split->block_start = em->block_start
 653                                                 + diff;
 654                                         split->orig_start = em->orig_start;
 655                                 }
 656                         } else {
 657                                 split->ram_bytes = split->len;
 658                                 split->orig_start = split->start;
 659                                 split->block_len = 0;
 660                                 split->block_start = em->block_start;
 661                                 split->orig_block_len = 0;
 662                         }
 663
 664                         ret = add_extent_mapping(em_tree, split, modified);
 665                         BUG_ON(ret); /* Logic error */
 666                         free_extent_map(split);
 667                         split = NULL;
 668                 }
 669 next:
 670                 write_unlock(&em_tree->lock);
 671
 672                 /* once for us */
 673                 free_extent_map(em);
 674                 /* once for the tree*/
 675                 free_extent_map(em);
 676         }
 677         if (split)
 678                 free_extent_map(split);
 679         if (split2)
 680                 free_extent_map(split2);
 681 }
 682
 683 /*
 684  * this is very complex, but the basic idea is to drop all extents
 685  * in the range start - end.  hint_block is filled in with a block number
 686  * that would be a good hint to the block allocator for this file.
 687  *
 688  * If an extent intersects the range but is not entirely inside the range
 689  * it is either truncated or split.  Anything entirely inside the range
 690  * is deleted from the tree.
 691  */
 692 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 693                          struct btrfs_root *root, struct inode *inode,
 694                          struct btrfs_path *path, u64 start, u64 end,
 695                          u64 *drop_end, int drop_cache,
 696                          int replace_extent,
 697                          u32 extent_item_size,
 698                          int *key_inserted)
 699 {
 700         struct extent_buffer *leaf;
 701         struct btrfs_file_extent_item *fi;
 702         struct btrfs_key key;
 703         struct btrfs_key new_key;
 704         u64 ino = btrfs_ino(inode);
 705         u64 search_start = start;
 706         u64 disk_bytenr = 0;
 707         u64 num_bytes = 0;
 708         u64 extent_offset = 0;
 709         u64 extent_end = 0;
 710         int del_nr = 0;
 711         int del_slot = 0;
 712         int extent_type;
 713         int recow;
 714         int ret;
 715         int modify_tree = -1;
 716         int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
 717         int found = 0;
 718         int leafs_visited = 0;
 719
 720         if (drop_cache)
 721                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 722
 723         if (start >= BTRFS_I(inode)->disk_i_size)
 724                 modify_tree = 0;
 725
 726         while (1) {
 727                 recow = 0;
 728                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
 729                                                search_start, modify_tree);
 730                 if (ret < 0)
 731                         break;
 732                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 733                         leaf = path->nodes[0];
 734                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 735                         if (key.objectid == ino &&
 736                             key.type == BTRFS_EXTENT_DATA_KEY)
 737                                 path->slots[0]--;
 738                 }
 739                 ret = 0;
 740                 leafs_visited++;
 741 next_slot:
 742                 leaf = path->nodes[0];
 743                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 744                         BUG_ON(del_nr > 0);
 745                         ret = btrfs_next_leaf(root, path);
 746                         if (ret < 0)
 747                                 break;
 748                         if (ret > 0) {
 749                                 ret = 0;
 750                                 break;
 751                         }
 752                         leafs_visited++;
 753                         leaf = path->nodes[0];
 754                         recow = 1;
 755                 }
 756
 757                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 758                 if (key.objectid > ino ||
 759                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 760                         break;
 761
 762                 fi = btrfs_item_ptr(leaf, path->slots[0],
 763                                     struct btrfs_file_extent_item);
 764                 extent_type = btrfs_file_extent_type(leaf, fi);
 765
 766                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 767                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 768                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 769                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 770                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 771                         extent_end = key.offset +
 772                                 btrfs_file_extent_num_bytes(leaf, fi);
 773                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 774                         extent_end = key.offset +
 775                                 btrfs_file_extent_inline_len(leaf, fi);
 776                 } else {
 777                         WARN_ON(1);
 778                         extent_end = search_start;
 779                 }
 780
 781                 if (extent_end <= search_start) {
 782                         path->slots[0]++;
 783                         goto next_slot;
 784                 }
 785
 786                 found = 1;
 787                 search_start = max(key.offset, start);
 788                 if (recow || !modify_tree) {
 789                         modify_tree = -1;
 790                         btrfs_release_path(path);
 791                         continue;
 792                 }
 793
 794                 /*
 795                  *     | - range to drop - |
 796                  *  | -------- extent -------- |
 797                  */
 798                 if (start > key.offset && end < extent_end) {
 799                         BUG_ON(del_nr > 0);
 800                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 801
 802                         memcpy(&new_key, &key, sizeof(new_key));
 803                         new_key.offset = start;
 804                         ret = btrfs_duplicate_item(trans, root, path,
 805                                                    &new_key);
 806                         if (ret == -EAGAIN) {
 807                                 btrfs_release_path(path);
 808                                 continue;
 809                         }
 810                         if (ret < 0)
 811                                 break;
 812
 813                         leaf = path->nodes[0];
 814                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 815                                             struct btrfs_file_extent_item);
 816                         btrfs_set_file_extent_num_bytes(leaf, fi,
 817                                                         start - key.offset);
 818
 819                         fi = btrfs_item_ptr(leaf, path->slots[0],
 820                                             struct btrfs_file_extent_item);
 821
 822                         extent_offset += start - key.offset;
 823                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 824                         btrfs_set_file_extent_num_bytes(leaf, fi,
 825                                                         extent_end - start);
 826                         btrfs_mark_buffer_dirty(leaf);
 827
 828                         if (update_refs && disk_bytenr > 0) {
 829                                 ret = btrfs_inc_extent_ref(trans, root,
 830                                                 disk_bytenr, num_bytes, 0,
 831                                                 root->root_key.objectid,
 832                                                 new_key.objectid,
 833                                                 start - extent_offset, 0);
 834                                 BUG_ON(ret); /* -ENOMEM */
 835                         }
 836                         key.offset = start;
 837                 }
 838                 /*
 839                  *  | ---- range to drop ----- |
 840                  *      | -------- extent -------- |
 841                  */
 842                 if (start <= key.offset && end < extent_end) {
 843                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 844
 845                         memcpy(&new_key, &key, sizeof(new_key));
 846                         new_key.offset = end;
 847                         btrfs_set_item_key_safe(root, path, &new_key);
 848
 849                         extent_offset += end - key.offset;
 850                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 851                         btrfs_set_file_extent_num_bytes(leaf, fi,
 852                                                         extent_end - end);
 853                         btrfs_mark_buffer_dirty(leaf);
 854                         if (update_refs && disk_bytenr > 0)
 855                                 inode_sub_bytes(inode, end - key.offset);
 856                         break;
 857                 }
 858
 859                 search_start = extent_end;
 860                 /*
 861                  *       | ---- range to drop ----- |
 862                  *  | -------- extent -------- |
 863                  */
 864                 if (start > key.offset && end >= extent_end) {
 865                         BUG_ON(del_nr > 0);
 866                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 867
 868                         btrfs_set_file_extent_num_bytes(leaf, fi,
 869                                                         start - key.offset);
 870                         btrfs_mark_buffer_dirty(leaf);
 871                         if (update_refs && disk_bytenr > 0)
 872                                 inode_sub_bytes(inode, extent_end - start);
 873                         if (end == extent_end)
 874                                 break;
 875
 876                         path->slots[0]++;
 877                         goto next_slot;
 878                 }
 879
 880                 /*
 881                  *  | ---- range to drop ----- |
 882                  *    | ------ extent ------ |
 883                  */
 884                 if (start <= key.offset && end >= extent_end) {
 885                         if (del_nr == 0) {
 886                                 del_slot = path->slots[0];
 887                                 del_nr = 1;
 888                         } else {
 889                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 890                                 del_nr++;
 891                         }
 892
 893                         if (update_refs &&
 894                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 895                                 inode_sub_bytes(inode,
 896                                                 extent_end - key.offset);
 897                                 extent_end = ALIGN(extent_end,
 898                                                    root->sectorsize);
 899                         } else if (update_refs && disk_bytenr > 0) {
 900                                 ret = btrfs_free_extent(trans, root,
 901                                                 disk_bytenr, num_bytes, 0,
 902                                                 root->root_key.objectid,
 903                                                 key.objectid, key.offset -
 904                                                 extent_offset, 0);
 905                                 BUG_ON(ret); /* -ENOMEM */
 906                                 inode_sub_bytes(inode,
 907                                                 extent_end - key.offset);
 908                         }
 909
 910                         if (end == extent_end)
 911                                 break;
 912
 913                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 914                                 path->slots[0]++;
 915                                 goto next_slot;
 916                         }
 917
 918                         ret = btrfs_del_items(trans, root, path, del_slot,
 919                                               del_nr);
 920                         if (ret) {
 921                                 btrfs_abort_transaction(trans, root, ret);
 922                                 break;
 923                         }
 924
 925                         del_nr = 0;
 926                         del_slot = 0;
 927
 928                         btrfs_release_path(path);
 929                         continue;
 930                 }
 931
 932                 BUG_ON(1);
 933         }
 934
 935         if (!ret && del_nr > 0) {
 936                 /*
 937                  * Set path->slots[0] to first slot, so that after the delete
 938                  * if items are move off from our leaf to its immediate left or
 939                  * right neighbor leafs, we end up with a correct and adjusted
 940                  * path->slots[0] for our insertion.
 941                  */
 942                 path->slots[0] = del_slot;
 943                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 944                 if (ret)
 945                         btrfs_abort_transaction(trans, root, ret);
 946
 947                 leaf = path->nodes[0];
 948                 /*
 949                  * leaf eb has flag EXTENT_BUFFER_STALE if it was deleted (that
 950                  * is, its contents got pushed to its neighbors), in which case
 951                  * it means path->locks[0] == 0
 952                  */
 953                 if (!ret && replace_extent && leafs_visited == 1 &&
 954                     path->locks[0] &&
 955                     btrfs_leaf_free_space(root, leaf) >=
 956                     sizeof(struct btrfs_item) + extent_item_size) {
 957
 958                         key.objectid = ino;
 959                         key.type = BTRFS_EXTENT_DATA_KEY;
 960                         key.offset = start;
 961                         setup_items_for_insert(root, path, &key,
 962                                                &extent_item_size,
 963                                                extent_item_size,
 964                                                sizeof(struct btrfs_item) +
 965                                                extent_item_size, 1);
 966                         *key_inserted = 1;
 967                 }
 968         }
 969
 970         if (!replace_extent || !(*key_inserted))
 971                 btrfs_release_path(path);
 972         if (drop_end)
 973                 *drop_end = found ? min(end, extent_end) : end;
 974         return ret;
 975 }
 976
 977 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 978                        struct btrfs_root *root, struct inode *inode, u64 start,
 979                        u64 end, int drop_cache)
 980 {
 981         struct btrfs_path *path;
 982         int ret;
 983
 984         path = btrfs_alloc_path();
 985         if (!path)
 986                 return -ENOMEM;
 987         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
 988                                    drop_cache, 0, 0, NULL);
 989         btrfs_free_path(path);
 990         return ret;
 991 }
 992
 993 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 994                             u64 objectid, u64 bytenr, u64 orig_offset,
 995                             u64 *start, u64 *end)
 996 {
 997         struct btrfs_file_extent_item *fi;
 998         struct btrfs_key key;
 999         u64 extent_end;
1000
1001         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1002                 return 0;
1003
1004         btrfs_item_key_to_cpu(leaf, &key, slot);
1005         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1006                 return 0;
1007
1008         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1009         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1010             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1011             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1012             btrfs_file_extent_compression(leaf, fi) ||
1013             btrfs_file_extent_encryption(leaf, fi) ||
1014             btrfs_file_extent_other_encoding(leaf, fi))
1015                 return 0;
1016
1017         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1018         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1019                 return 0;
1020
1021         *start = key.offset;
1022         *end = extent_end;
1023         return 1;
1024 }
1025
1026 /*
1027  * Mark extent in the range start - end as written.
1028  *
1029  * This changes extent type from 'pre-allocated' to 'regular'. If only
1030  * part of extent is marked as written, the extent will be split into
1031  * two or three.
1032  */
1033 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1034                               struct inode *inode, u64 start, u64 end)
1035 {
1036         struct btrfs_root *root = BTRFS_I(inode)->root;
1037         struct extent_buffer *leaf;
1038         struct btrfs_path *path;
1039         struct btrfs_file_extent_item *fi;
1040         struct btrfs_key key;
1041         struct btrfs_key new_key;
1042         u64 bytenr;
1043         u64 num_bytes;
1044         u64 extent_end;
1045         u64 orig_offset;
1046         u64 other_start;
1047         u64 other_end;
1048         u64 split;
1049         int del_nr = 0;
1050         int del_slot = 0;
1051         int recow;
1052         int ret;
1053         u64 ino = btrfs_ino(inode);
1054
1055         path = btrfs_alloc_path();
1056         if (!path)
1057                 return -ENOMEM;
1058 again:
1059         recow = 0;
1060         split = start;
1061         key.objectid = ino;
1062         key.type = BTRFS_EXTENT_DATA_KEY;
1063         key.offset = split;
1064
1065         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1066         if (ret < 0)
1067                 goto out;
1068         if (ret > 0 && path->slots[0] > 0)
1069                 path->slots[0]--;
1070
1071         leaf = path->nodes[0];
1072         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1073         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1074         fi = btrfs_item_ptr(leaf, path->slots[0],
1075                             struct btrfs_file_extent_item);
1076         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1077                BTRFS_FILE_EXTENT_PREALLOC);
1078         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1079         BUG_ON(key.offset > start || extent_end < end);
1080
1081         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1082         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1083         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1084         memcpy(&new_key, &key, sizeof(new_key));
1085
1086         if (start == key.offset && end < extent_end) {
1087                 other_start = 0;
1088                 other_end = start;
1089                 if (extent_mergeable(leaf, path->slots[0] - 1,
1090                                      ino, bytenr, orig_offset,
1091                                      &other_start, &other_end)) {
1092                         new_key.offset = end;
1093                         btrfs_set_item_key_safe(root, path, &new_key);
1094                         fi = btrfs_item_ptr(leaf, path->slots[0],
1095                                             struct btrfs_file_extent_item);
1096                         btrfs_set_file_extent_generation(leaf, fi,
1097                                                          trans->transid);
1098                         btrfs_set_file_extent_num_bytes(leaf, fi,
1099                                                         extent_end - end);
1100                         btrfs_set_file_extent_offset(leaf, fi,
1101                                                      end - orig_offset);
1102                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1103                                             struct btrfs_file_extent_item);
1104                         btrfs_set_file_extent_generation(leaf, fi,
1105                                                          trans->transid);
1106                         btrfs_set_file_extent_num_bytes(leaf, fi,
1107                                                         end - other_start);
1108                         btrfs_mark_buffer_dirty(leaf);
1109                         goto out;
1110                 }
1111         }
1112
1113         if (start > key.offset && end == extent_end) {
1114                 other_start = end;
1115                 other_end = 0;
1116                 if (extent_mergeable(leaf, path->slots[0] + 1,
1117                                      ino, bytenr, orig_offset,
1118                                      &other_start, &other_end)) {
1119                         fi = btrfs_item_ptr(leaf, path->slots[0],
1120                                             struct btrfs_file_extent_item);
1121                         btrfs_set_file_extent_num_bytes(leaf, fi,
1122                                                         start - key.offset);
1123                         btrfs_set_file_extent_generation(leaf, fi,
1124                                                          trans->transid);
1125                         path->slots[0]++;
1126                         new_key.offset = start;
1127                         btrfs_set_item_key_safe(root, path, &new_key);
1128
1129                         fi = btrfs_item_ptr(leaf, path->slots[0],
1130                                             struct btrfs_file_extent_item);
1131                         btrfs_set_file_extent_generation(leaf, fi,
1132                                                          trans->transid);
1133                         btrfs_set_file_extent_num_bytes(leaf, fi,
1134                                                         other_end - start);
1135                         btrfs_set_file_extent_offset(leaf, fi,
1136                                                      start - orig_offset);
1137                         btrfs_mark_buffer_dirty(leaf);
1138                         goto out;
1139                 }
1140         }
1141
1142         while (start > key.offset || end < extent_end) {
1143                 if (key.offset == start)
1144                         split = end;
1145
1146                 new_key.offset = split;
1147                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1148                 if (ret == -EAGAIN) {
1149                         btrfs_release_path(path);
1150                         goto again;
1151                 }
1152                 if (ret < 0) {
1153                         btrfs_abort_transaction(trans, root, ret);
1154                         goto out;
1155                 }
1156
1157                 leaf = path->nodes[0];
1158                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1159                                     struct btrfs_file_extent_item);
1160                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1161                 btrfs_set_file_extent_num_bytes(leaf, fi,
1162                                                 split - key.offset);
1163
1164                 fi = btrfs_item_ptr(leaf, path->slots[0],
1165                                     struct btrfs_file_extent_item);
1166
1167                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1168                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1169                 btrfs_set_file_extent_num_bytes(leaf, fi,
1170                                                 extent_end - split);
1171                 btrfs_mark_buffer_dirty(leaf);
1172
1173                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1174                                            root->root_key.objectid,
1175                                            ino, orig_offset, 0);
1176                 BUG_ON(ret); /* -ENOMEM */
1177
1178                 if (split == start) {
1179                         key.offset = start;
1180                 } else {
1181                         BUG_ON(start != key.offset);
1182                         path->slots[0]--;
1183                         extent_end = end;
1184                 }
1185                 recow = 1;
1186         }
1187
1188         other_start = end;
1189         other_end = 0;
1190         if (extent_mergeable(leaf, path->slots[0] + 1,
1191                              ino, bytenr, orig_offset,
1192                              &other_start, &other_end)) {
1193                 if (recow) {
1194                         btrfs_release_path(path);
1195                         goto again;
1196                 }
1197                 extent_end = other_end;
1198                 del_slot = path->slots[0] + 1;
1199                 del_nr++;
1200                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1201                                         0, root->root_key.objectid,
1202                                         ino, orig_offset, 0);
1203                 BUG_ON(ret); /* -ENOMEM */
1204         }
1205         other_start = 0;
1206         other_end = start;
1207         if (extent_mergeable(leaf, path->slots[0] - 1,
1208                              ino, bytenr, orig_offset,
1209                              &other_start, &other_end)) {
1210                 if (recow) {
1211                         btrfs_release_path(path);
1212                         goto again;
1213                 }
1214                 key.offset = other_start;
1215                 del_slot = path->slots[0];
1216                 del_nr++;
1217                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1218                                         0, root->root_key.objectid,
1219                                         ino, orig_offset, 0);
1220                 BUG_ON(ret); /* -ENOMEM */
1221         }
1222         if (del_nr == 0) {
1223                 fi = btrfs_item_ptr(leaf, path->slots[0],
1224                            struct btrfs_file_extent_item);
1225                 btrfs_set_file_extent_type(leaf, fi,
1226                                            BTRFS_FILE_EXTENT_REG);
1227                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1228                 btrfs_mark_buffer_dirty(leaf);
1229         } else {
1230                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1231                            struct btrfs_file_extent_item);
1232                 btrfs_set_file_extent_type(leaf, fi,
1233                                            BTRFS_FILE_EXTENT_REG);
1234                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1235                 btrfs_set_file_extent_num_bytes(leaf, fi,
1236                                                 extent_end - key.offset);
1237                 btrfs_mark_buffer_dirty(leaf);
1238
1239                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1240                 if (ret < 0) {
1241                         btrfs_abort_transaction(trans, root, ret);
1242                         goto out;
1243                 }
1244         }
1245 out:
1246         btrfs_free_path(path);
1247         return 0;
1248 }
1249
1250 /*
1251  * on error we return an unlocked page and the error value
1252  * on success we return a locked page and 0
1253  */
1254 static int prepare_uptodate_page(struct page *page, u64 pos,
1255                                  bool force_uptodate)
1256 {
1257         int ret = 0;
1258
1259         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1260             !PageUptodate(page)) {
1261                 ret = btrfs_readpage(NULL, page);
1262                 if (ret)
1263                         return ret;
1264                 lock_page(page);
1265                 if (!PageUptodate(page)) {
1266                         unlock_page(page);
1267                         return -EIO;
1268                 }
1269         }
1270         return 0;
1271 }
1272
1273 /*
1274  * this just gets pages into the page cache and locks them down.
1275  */
1276 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1277                                   size_t num_pages, loff_t pos,
1278                                   size_t write_bytes, bool force_uptodate)
1279 {
1280         int i;
1281         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1282         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1283         int err = 0;
1284         int faili;
1285
1286         for (i = 0; i < num_pages; i++) {
1287                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1288                                                mask | __GFP_WRITE);
1289                 if (!pages[i]) {
1290                         faili = i - 1;
1291                         err = -ENOMEM;
1292                         goto fail;
1293                 }
1294
1295                 if (i == 0)
1296                         err = prepare_uptodate_page(pages[i], pos,
1297                                                     force_uptodate);
1298                 if (i == num_pages - 1)
1299                         err = prepare_uptodate_page(pages[i],
1300                                                     pos + write_bytes, false);
1301                 if (err) {
1302                         page_cache_release(pages[i]);
1303                         faili = i - 1;
1304                         goto fail;
1305                 }
1306                 wait_on_page_writeback(pages[i]);
1307         }
1308
1309         return 0;
1310 fail:
1311         while (faili >= 0) {
1312                 unlock_page(pages[faili]);
1313                 page_cache_release(pages[faili]);
1314                 faili--;
1315         }
1316         return err;
1317
1318 }
1319
1320 /*
1321  * This function locks the extent and properly waits for data=ordered extents
1322  * to finish before allowing the pages to be modified if need.
1323  *
1324  * The return value:
1325  * 1 - the extent is locked
1326  * 0 - the extent is not locked, and everything is OK
1327  * -EAGAIN - need re-prepare the pages
1328  * the other < 0 number - Something wrong happens
1329  */
1330 static noinline int
1331 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1332                                 size_t num_pages, loff_t pos,
1333                                 u64 *lockstart, u64 *lockend,
1334                                 struct extent_state **cached_state)
1335 {
1336         u64 start_pos;
1337         u64 last_pos;
1338         int i;
1339         int ret = 0;
1340
1341         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1342         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1343
1344         if (start_pos < inode->i_size) {
1345                 struct btrfs_ordered_extent *ordered;
1346                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1347                                  start_pos, last_pos, 0, cached_state);
1348                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1349                 if (ordered &&
1350                     ordered->file_offset + ordered->len > start_pos &&
1351                     ordered->file_offset <= last_pos) {
1352                         btrfs_put_ordered_extent(ordered);
1353                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1354                                              start_pos, last_pos,
1355                                              cached_state, GFP_NOFS);
1356                         for (i = 0; i < num_pages; i++) {
1357                                 unlock_page(pages[i]);
1358                                 page_cache_release(pages[i]);
1359                         }
1360                         ret = btrfs_wait_ordered_range(inode, start_pos,
1361                                                 last_pos - start_pos + 1);
1362                         if (ret)
1363                                 return ret;
1364                         else
1365                                 return -EAGAIN;
1366                 }
1367                 if (ordered)
1368                         btrfs_put_ordered_extent(ordered);
1369
1370                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1371                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1372                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1373                                   0, 0, cached_state, GFP_NOFS);
1374                 *lockstart = start_pos;
1375                 *lockend = last_pos;
1376                 ret = 1;
1377         }
1378
1379         for (i = 0; i < num_pages; i++) {
1380                 if (clear_page_dirty_for_io(pages[i]))
1381                         account_page_redirty(pages[i]);
1382                 set_page_extent_mapped(pages[i]);
1383                 WARN_ON(!PageLocked(pages[i]));
1384         }
1385
1386         return ret;
1387 }
1388
1389 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1390                                     size_t *write_bytes)
1391 {
1392         struct btrfs_root *root = BTRFS_I(inode)->root;
1393         struct btrfs_ordered_extent *ordered;
1394         u64 lockstart, lockend;
1395         u64 num_bytes;
1396         int ret;
1397
1398         lockstart = round_down(pos, root->sectorsize);
1399         lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1400
1401         while (1) {
1402                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1403                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1404                                                      lockend - lockstart + 1);
1405                 if (!ordered) {
1406                         break;
1407                 }
1408                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1409                 btrfs_start_ordered_extent(inode, ordered, 1);
1410                 btrfs_put_ordered_extent(ordered);
1411         }
1412
1413         num_bytes = lockend - lockstart + 1;
1414         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1415         if (ret <= 0) {
1416                 ret = 0;
1417         } else {
1418                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1419                                  EXTENT_DIRTY | EXTENT_DELALLOC |
1420                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1421                                  NULL, GFP_NOFS);
1422                 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1423         }
1424
1425         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1426
1427         return ret;
1428 }
1429
1430 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1431                                                struct iov_iter *i,
1432                                                loff_t pos)
1433 {
1434         struct inode *inode = file_inode(file);
1435         struct btrfs_root *root = BTRFS_I(inode)->root;
1436         struct page **pages = NULL;
1437         struct extent_state *cached_state = NULL;
1438         u64 release_bytes = 0;
1439         u64 lockstart;
1440         u64 lockend;
1441         unsigned long first_index;
1442         size_t num_written = 0;
1443         int nrptrs;
1444         int ret = 0;
1445         bool only_release_metadata = false;
1446         bool force_page_uptodate = false;
1447         bool need_unlock;
1448
1449         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1450                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1451                      (sizeof(struct page *)));
1452         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1453         nrptrs = max(nrptrs, 8);
1454         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1455         if (!pages)
1456                 return -ENOMEM;
1457
1458         first_index = pos >> PAGE_CACHE_SHIFT;
1459
1460         while (iov_iter_count(i) > 0) {
1461                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1462                 size_t write_bytes = min(iov_iter_count(i),
1463                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1464                                          offset);
1465                 size_t num_pages = (write_bytes + offset +
1466                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1467                 size_t reserve_bytes;
1468                 size_t dirty_pages;
1469                 size_t copied;
1470
1471                 WARN_ON(num_pages > nrptrs);
1472
1473                 /*
1474                  * Fault pages before locking them in prepare_pages
1475                  * to avoid recursive lock
1476                  */
1477                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1478                         ret = -EFAULT;
1479                         break;
1480                 }
1481
1482                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1483                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1484                 if (ret == -ENOSPC &&
1485                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1486                                               BTRFS_INODE_PREALLOC))) {
1487                         ret = check_can_nocow(inode, pos, &write_bytes);
1488                         if (ret > 0) {
1489                                 only_release_metadata = true;
1490                                 /*
1491                                  * our prealloc extent may be smaller than
1492                                  * write_bytes, so scale down.
1493                                  */
1494                                 num_pages = (write_bytes + offset +
1495                                              PAGE_CACHE_SIZE - 1) >>
1496                                         PAGE_CACHE_SHIFT;
1497                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1498                                 ret = 0;
1499                         } else {
1500                                 ret = -ENOSPC;
1501                         }
1502                 }
1503
1504                 if (ret)
1505                         break;
1506
1507                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1508                 if (ret) {
1509                         if (!only_release_metadata)
1510                                 btrfs_free_reserved_data_space(inode,
1511                                                                reserve_bytes);
1512                         break;
1513                 }
1514
1515                 release_bytes = reserve_bytes;
1516                 need_unlock = false;
1517 again:
1518                 /*
1519                  * This is going to setup the pages array with the number of
1520                  * pages we want, so we don't really need to worry about the
1521                  * contents of pages from loop to loop
1522                  */
1523                 ret = prepare_pages(inode, pages, num_pages,
1524                                     pos, write_bytes,
1525                                     force_page_uptodate);
1526                 if (ret)
1527                         break;
1528
1529                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1530                                                       pos, &lockstart, &lockend,
1531                                                       &cached_state);
1532                 if (ret < 0) {
1533                         if (ret == -EAGAIN)
1534                                 goto again;
1535                         break;
1536                 } else if (ret > 0) {
1537                         need_unlock = true;
1538                         ret = 0;
1539                 }
1540
1541                 copied = btrfs_copy_from_user(pos, num_pages,
1542                                            write_bytes, pages, i);
1543
1544                 /*
1545                  * if we have trouble faulting in the pages, fall
1546                  * back to one page at a time
1547                  */
1548                 if (copied < write_bytes)
1549                         nrptrs = 1;
1550
1551                 if (copied == 0) {
1552                         force_page_uptodate = true;
1553                         dirty_pages = 0;
1554                 } else {
1555                         force_page_uptodate = false;
1556                         dirty_pages = (copied + offset +
1557                                        PAGE_CACHE_SIZE - 1) >>
1558                                        PAGE_CACHE_SHIFT;
1559                 }
1560
1561                 /*
1562                  * If we had a short copy we need to release the excess delaloc
1563                  * bytes we reserved.  We need to increment outstanding_extents
1564                  * because btrfs_delalloc_release_space will decrement it, but
1565                  * we still have an outstanding extent for the chunk we actually
1566                  * managed to copy.
1567                  */
1568                 if (num_pages > dirty_pages) {
1569                         release_bytes = (num_pages - dirty_pages) <<
1570                                 PAGE_CACHE_SHIFT;
1571                         if (copied > 0) {
1572                                 spin_lock(&BTRFS_I(inode)->lock);
1573                                 BTRFS_I(inode)->outstanding_extents++;
1574                                 spin_unlock(&BTRFS_I(inode)->lock);
1575                         }
1576                         if (only_release_metadata)
1577                                 btrfs_delalloc_release_metadata(inode,
1578                                                                 release_bytes);
1579                         else
1580                                 btrfs_delalloc_release_space(inode,
1581                                                              release_bytes);
1582                 }
1583
1584                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1585
1586                 if (copied > 0)
1587                         ret = btrfs_dirty_pages(root, inode, pages,
1588                                                 dirty_pages, pos, copied,
1589                                                 NULL);
1590                 if (need_unlock)
1591                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1592                                              lockstart, lockend, &cached_state,
1593                                              GFP_NOFS);
1594                 btrfs_drop_pages(pages, num_pages);
1595                 if (ret)
1596                         break;
1597
1598                 release_bytes = 0;
1599                 if (only_release_metadata && copied > 0) {
1600                         u64 lockstart = round_down(pos, root->sectorsize);
1601                         u64 lockend = lockstart +
1602                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1603
1604                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1605                                        lockend, EXTENT_NORESERVE, NULL,
1606                                        NULL, GFP_NOFS);
1607                         only_release_metadata = false;
1608                 }
1609
1610                 cond_resched();
1611
1612                 balance_dirty_pages_ratelimited(inode->i_mapping);
1613                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1614                         btrfs_btree_balance_dirty(root);
1615
1616                 pos += copied;
1617                 num_written += copied;
1618         }
1619
1620         kfree(pages);
1621
1622         if (release_bytes) {
1623                 if (only_release_metadata)
1624                         btrfs_delalloc_release_metadata(inode, release_bytes);
1625                 else
1626                         btrfs_delalloc_release_space(inode, release_bytes);
1627         }
1628
1629         return num_written ? num_written : ret;
1630 }
1631
1632 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1633                                     const struct iovec *iov,
1634                                     unsigned long nr_segs, loff_t pos,
1635                                     loff_t *ppos, size_t count, size_t ocount)
1636 {
1637         struct file *file = iocb->ki_filp;
1638         struct iov_iter i;
1639         ssize_t written;
1640         ssize_t written_buffered;
1641         loff_t endbyte;
1642         int err;
1643
1644         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1645                                             count, ocount);
1646
1647         if (written < 0 || written == count)
1648                 return written;
1649
1650         pos += written;
1651         count -= written;
1652         iov_iter_init(&i, iov, nr_segs, count, written);
1653         written_buffered = __btrfs_buffered_write(file, &i, pos);
1654         if (written_buffered < 0) {
1655                 err = written_buffered;
1656                 goto out;
1657         }
1658         endbyte = pos + written_buffered - 1;
1659         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1660         if (err)
1661                 goto out;
1662         written += written_buffered;
1663         *ppos = pos + written_buffered;
1664         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1665                                  endbyte >> PAGE_CACHE_SHIFT);
1666 out:
1667         return written ? written : err;
1668 }
1669
1670 static void update_time_for_write(struct inode *inode)
1671 {
1672         struct timespec now;
1673
1674         if (IS_NOCMTIME(inode))
1675                 return;
1676
1677         now = current_fs_time(inode->i_sb);
1678         if (!timespec_equal(&inode->i_mtime, &now))
1679                 inode->i_mtime = now;
1680
1681         if (!timespec_equal(&inode->i_ctime, &now))
1682                 inode->i_ctime = now;
1683
1684         if (IS_I_VERSION(inode))
1685                 inode_inc_iversion(inode);
1686 }
1687
1688 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1689                                     const struct iovec *iov,
1690                                     unsigned long nr_segs, loff_t pos)
1691 {
1692         struct file *file = iocb->ki_filp;
1693         struct inode *inode = file_inode(file);
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         loff_t *ppos = &iocb->ki_pos;
1696         u64 start_pos;
1697         ssize_t num_written = 0;
1698         ssize_t err = 0;
1699         size_t count, ocount;
1700         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1701
1702         mutex_lock(&inode->i_mutex);
1703
1704         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1705         if (err) {
1706                 mutex_unlock(&inode->i_mutex);
1707                 goto out;
1708         }
1709         count = ocount;
1710
1711         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1712         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1713         if (err) {
1714                 mutex_unlock(&inode->i_mutex);
1715                 goto out;
1716         }
1717
1718         if (count == 0) {
1719                 mutex_unlock(&inode->i_mutex);
1720                 goto out;
1721         }
1722
1723         err = file_remove_suid(file);
1724         if (err) {
1725                 mutex_unlock(&inode->i_mutex);
1726                 goto out;
1727         }
1728
1729         /*
1730          * If BTRFS flips readonly due to some impossible error
1731          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1732          * although we have opened a file as writable, we have
1733          * to stop this write operation to ensure FS consistency.
1734          */
1735         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1736                 mutex_unlock(&inode->i_mutex);
1737                 err = -EROFS;
1738                 goto out;
1739         }
1740
1741         /*
1742          * We reserve space for updating the inode when we reserve space for the
1743          * extent we are going to write, so we will enospc out there.  We don't
1744          * need to start yet another transaction to update the inode as we will
1745          * update the inode when we finish writing whatever data we write.
1746          */
1747         update_time_for_write(inode);
1748
1749         start_pos = round_down(pos, root->sectorsize);
1750         if (start_pos > i_size_read(inode)) {
1751                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1752                 if (err) {
1753                         mutex_unlock(&inode->i_mutex);
1754                         goto out;
1755                 }
1756         }
1757
1758         if (sync)
1759                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1760
1761         if (unlikely(file->f_flags & O_DIRECT)) {
1762                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1763                                                    pos, ppos, count, ocount);
1764         } else {
1765                 struct iov_iter i;
1766
1767                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1768
1769                 num_written = __btrfs_buffered_write(file, &i, pos);
1770                 if (num_written > 0)
1771                         *ppos = pos + num_written;
1772         }
1773
1774         mutex_unlock(&inode->i_mutex);
1775
1776         /*
1777          * we want to make sure fsync finds this change
1778          * but we haven't joined a transaction running right now.
1779          *
1780          * Later on, someone is sure to update the inode and get the
1781          * real transid recorded.
1782          *
1783          * We set last_trans now to the fs_info generation + 1,
1784          * this will either be one more than the running transaction
1785          * or the generation used for the next transaction if there isn't
1786          * one running right now.
1787          *
1788          * We also have to set last_sub_trans to the current log transid,
1789          * otherwise subsequent syncs to a file that's been synced in this
1790          * transaction will appear to have already occured.
1791          */
1792         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1793         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1794         if (num_written > 0) {
1795                 err = generic_write_sync(file, pos, num_written);
1796                 if (err < 0 && num_written > 0)
1797                         num_written = err;
1798         }
1799
1800         if (sync)
1801                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1802 out:
1803         current->backing_dev_info = NULL;
1804         return num_written ? num_written : err;
1805 }
1806
1807 int btrfs_release_file(struct inode *inode, struct file *filp)
1808 {
1809         /*
1810          * ordered_data_close is set by settattr when we are about to truncate
1811          * a file from a non-zero size to a zero size.  This tries to
1812          * flush down new bytes that may have been written if the
1813          * application were using truncate to replace a file in place.
1814          */
1815         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1816                                &BTRFS_I(inode)->runtime_flags)) {
1817                 struct btrfs_trans_handle *trans;
1818                 struct btrfs_root *root = BTRFS_I(inode)->root;
1819
1820                 /*
1821                  * We need to block on a committing transaction to keep us from
1822                  * throwing a ordered operation on to the list and causing
1823                  * something like sync to deadlock trying to flush out this
1824                  * inode.
1825                  */
1826                 trans = btrfs_start_transaction(root, 0);
1827                 if (IS_ERR(trans))
1828                         return PTR_ERR(trans);
1829                 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1830                 btrfs_end_transaction(trans, root);
1831                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1832                         filemap_flush(inode->i_mapping);
1833         }
1834         if (filp->private_data)
1835                 btrfs_ioctl_trans_end(filp);
1836         return 0;
1837 }
1838
1839 /*
1840  * fsync call for both files and directories.  This logs the inode into
1841  * the tree log instead of forcing full commits whenever possible.
1842  *
1843  * It needs to call filemap_fdatawait so that all ordered extent updates are
1844  * in the metadata btree are up to date for copying to the log.
1845  *
1846  * It drops the inode mutex before doing the tree log commit.  This is an
1847  * important optimization for directories because holding the mutex prevents
1848  * new operations on the dir while we write to disk.
1849  */
1850 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1851 {
1852         struct dentry *dentry = file->f_path.dentry;
1853         struct inode *inode = dentry->d_inode;
1854         struct btrfs_root *root = BTRFS_I(inode)->root;
1855         int ret = 0;
1856         struct btrfs_trans_handle *trans;
1857         bool full_sync = 0;
1858
1859         trace_btrfs_sync_file(file, datasync);
1860
1861         /*
1862          * We write the dirty pages in the range and wait until they complete
1863          * out of the ->i_mutex. If so, we can flush the dirty pages by
1864          * multi-task, and make the performance up.  See
1865          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1866          */
1867         atomic_inc(&BTRFS_I(inode)->sync_writers);
1868         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1869         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1870                              &BTRFS_I(inode)->runtime_flags))
1871                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1872         atomic_dec(&BTRFS_I(inode)->sync_writers);
1873         if (ret)
1874                 return ret;
1875
1876         mutex_lock(&inode->i_mutex);
1877
1878         /*
1879          * We flush the dirty pages again to avoid some dirty pages in the
1880          * range being left.
1881          */
1882         atomic_inc(&root->log_batch);
1883         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1884                              &BTRFS_I(inode)->runtime_flags);
1885         if (full_sync) {
1886                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1887                 if (ret) {
1888                         mutex_unlock(&inode->i_mutex);
1889                         goto out;
1890                 }
1891         }
1892         atomic_inc(&root->log_batch);
1893
1894         /*
1895          * check the transaction that last modified this inode
1896          * and see if its already been committed
1897          */
1898         if (!BTRFS_I(inode)->last_trans) {
1899                 mutex_unlock(&inode->i_mutex);
1900                 goto out;
1901         }
1902
1903         /*
1904          * if the last transaction that changed this file was before
1905          * the current transaction, we can bail out now without any
1906          * syncing
1907          */
1908         smp_mb();
1909         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1910             BTRFS_I(inode)->last_trans <=
1911             root->fs_info->last_trans_committed) {
1912                 BTRFS_I(inode)->last_trans = 0;
1913
1914                 /*
1915                  * We'v had everything committed since the last time we were
1916                  * modified so clear this flag in case it was set for whatever
1917                  * reason, it's no longer relevant.
1918                  */
1919                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1920                           &BTRFS_I(inode)->runtime_flags);
1921                 mutex_unlock(&inode->i_mutex);
1922                 goto out;
1923         }
1924
1925         /*
1926          * ok we haven't committed the transaction yet, lets do a commit
1927          */
1928         if (file->private_data)
1929                 btrfs_ioctl_trans_end(file);
1930
1931         trans = btrfs_start_transaction(root, 0);
1932         if (IS_ERR(trans)) {
1933                 ret = PTR_ERR(trans);
1934                 mutex_unlock(&inode->i_mutex);
1935                 goto out;
1936         }
1937
1938         ret = btrfs_log_dentry_safe(trans, root, dentry);
1939         if (ret < 0) {
1940                 /* Fallthrough and commit/free transaction. */
1941                 ret = 1;
1942         }
1943
1944         /* we've logged all the items and now have a consistent
1945          * version of the file in the log.  It is possible that
1946          * someone will come in and modify the file, but that's
1947          * fine because the log is consistent on disk, and we
1948          * have references to all of the file's extents
1949          *
1950          * It is possible that someone will come in and log the
1951          * file again, but that will end up using the synchronization
1952          * inside btrfs_sync_log to keep things safe.
1953          */
1954         mutex_unlock(&inode->i_mutex);
1955
1956         if (ret != BTRFS_NO_LOG_SYNC) {
1957                 if (!ret) {
1958                         ret = btrfs_sync_log(trans, root);
1959                         if (!ret) {
1960                                 ret = btrfs_end_transaction(trans, root);
1961                                 goto out;
1962                         }
1963                 }
1964                 if (!full_sync) {
1965                         ret = btrfs_wait_ordered_range(inode, start,
1966                                                        end - start + 1);
1967                         if (ret)
1968                                 goto out;
1969                 }
1970                 ret = btrfs_commit_transaction(trans, root);
1971         } else {
1972                 ret = btrfs_end_transaction(trans, root);
1973         }
1974 out:
1975         return ret > 0 ? -EIO : ret;
1976 }
1977
1978 static const struct vm_operations_struct btrfs_file_vm_ops = {
1979         .fault          = filemap_fault,
1980         .page_mkwrite   = btrfs_page_mkwrite,
1981         .remap_pages    = generic_file_remap_pages,
1982 };
1983
1984 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1985 {
1986         struct address_space *mapping = filp->f_mapping;
1987
1988         if (!mapping->a_ops->readpage)
1989                 return -ENOEXEC;
1990
1991         file_accessed(filp);
1992         vma->vm_ops = &btrfs_file_vm_ops;
1993
1994         return 0;
1995 }
1996
1997 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1998                           int slot, u64 start, u64 end)
1999 {
2000         struct btrfs_file_extent_item *fi;
2001         struct btrfs_key key;
2002
2003         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2004                 return 0;
2005
2006         btrfs_item_key_to_cpu(leaf, &key, slot);
2007         if (key.objectid != btrfs_ino(inode) ||
2008             key.type != BTRFS_EXTENT_DATA_KEY)
2009                 return 0;
2010
2011         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2012
2013         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2014                 return 0;
2015
2016         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2017                 return 0;
2018
2019         if (key.offset == end)
2020                 return 1;
2021         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2022                 return 1;
2023         return 0;
2024 }
2025
2026 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2027                       struct btrfs_path *path, u64 offset, u64 end)
2028 {
2029         struct btrfs_root *root = BTRFS_I(inode)->root;
2030         struct extent_buffer *leaf;
2031         struct btrfs_file_extent_item *fi;
2032         struct extent_map *hole_em;
2033         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2034         struct btrfs_key key;
2035         int ret;
2036
2037         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2038                 goto out;
2039
2040         key.objectid = btrfs_ino(inode);
2041         key.type = BTRFS_EXTENT_DATA_KEY;
2042         key.offset = offset;
2043
2044         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2045         if (ret < 0)
2046                 return ret;
2047         BUG_ON(!ret);
2048
2049         leaf = path->nodes[0];
2050         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2051                 u64 num_bytes;
2052
2053                 path->slots[0]--;
2054                 fi = btrfs_item_ptr(leaf, path->slots[0],
2055                                     struct btrfs_file_extent_item);
2056                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2057                         end - offset;
2058                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2059                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2060                 btrfs_set_file_extent_offset(leaf, fi, 0);
2061                 btrfs_mark_buffer_dirty(leaf);
2062                 goto out;
2063         }
2064
2065         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2066                 u64 num_bytes;
2067
2068                 path->slots[0]++;
2069                 key.offset = offset;
2070                 btrfs_set_item_key_safe(root, path, &key);
2071                 fi = btrfs_item_ptr(leaf, path->slots[0],
2072                                     struct btrfs_file_extent_item);
2073                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2074                         offset;
2075                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2076                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2077                 btrfs_set_file_extent_offset(leaf, fi, 0);
2078                 btrfs_mark_buffer_dirty(leaf);
2079                 goto out;
2080         }
2081         btrfs_release_path(path);
2082
2083         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2084                                        0, 0, end - offset, 0, end - offset,
2085                                        0, 0, 0);
2086         if (ret)
2087                 return ret;
2088
2089 out:
2090         btrfs_release_path(path);
2091
2092         hole_em = alloc_extent_map();
2093         if (!hole_em) {
2094                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2095                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2096                         &BTRFS_I(inode)->runtime_flags);
2097         } else {
2098                 hole_em->start = offset;
2099                 hole_em->len = end - offset;
2100                 hole_em->ram_bytes = hole_em->len;
2101                 hole_em->orig_start = offset;
2102
2103                 hole_em->block_start = EXTENT_MAP_HOLE;
2104                 hole_em->block_len = 0;
2105                 hole_em->orig_block_len = 0;
2106                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2107                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2108                 hole_em->generation = trans->transid;
2109
2110                 do {
2111                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2112                         write_lock(&em_tree->lock);
2113                         ret = add_extent_mapping(em_tree, hole_em, 1);
2114                         write_unlock(&em_tree->lock);
2115                 } while (ret == -EEXIST);
2116                 free_extent_map(hole_em);
2117                 if (ret)
2118                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2119                                 &BTRFS_I(inode)->runtime_flags);
2120         }
2121
2122         return 0;
2123 }
2124
2125 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2126 {
2127         struct btrfs_root *root = BTRFS_I(inode)->root;
2128         struct extent_state *cached_state = NULL;
2129         struct btrfs_path *path;
2130         struct btrfs_block_rsv *rsv;
2131         struct btrfs_trans_handle *trans;
2132         u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2133         u64 lockend = round_down(offset + len,
2134                                  BTRFS_I(inode)->root->sectorsize) - 1;
2135         u64 cur_offset = lockstart;
2136         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2137         u64 drop_end;
2138         int ret = 0;
2139         int err = 0;
2140         int rsv_count;
2141         bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2142                           ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2143         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2144
2145         ret = btrfs_wait_ordered_range(inode, offset, len);
2146         if (ret)
2147                 return ret;
2148
2149         mutex_lock(&inode->i_mutex);
2150         /*
2151          * We needn't truncate any page which is beyond the end of the file
2152          * because we are sure there is no data there.
2153          */
2154         /*
2155          * Only do this if we are in the same page and we aren't doing the
2156          * entire page.
2157          */
2158         if (same_page && len < PAGE_CACHE_SIZE) {
2159                 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2160                         ret = btrfs_truncate_page(inode, offset, len, 0);
2161                 mutex_unlock(&inode->i_mutex);
2162                 return ret;
2163         }
2164
2165         /* zero back part of the first page */
2166         if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2167                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2168                 if (ret) {
2169                         mutex_unlock(&inode->i_mutex);
2170                         return ret;
2171                 }
2172         }
2173
2174         /* zero the front end of the last page */
2175         if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2176                 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2177                 if (ret) {
2178                         mutex_unlock(&inode->i_mutex);
2179                         return ret;
2180                 }
2181         }
2182
2183         if (lockend < lockstart) {
2184                 mutex_unlock(&inode->i_mutex);
2185                 return 0;
2186         }
2187
2188         while (1) {
2189                 struct btrfs_ordered_extent *ordered;
2190
2191                 truncate_pagecache_range(inode, lockstart, lockend);
2192
2193                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2194                                  0, &cached_state);
2195                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2196
2197                 /*
2198                  * We need to make sure we have no ordered extents in this range
2199                  * and nobody raced in and read a page in this range, if we did
2200                  * we need to try again.
2201                  */
2202                 if ((!ordered ||
2203                     (ordered->file_offset + ordered->len <= lockstart ||
2204                      ordered->file_offset > lockend)) &&
2205                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2206                                      lockend, EXTENT_UPTODATE, 0,
2207                                      cached_state)) {
2208                         if (ordered)
2209                                 btrfs_put_ordered_extent(ordered);
2210                         break;
2211                 }
2212                 if (ordered)
2213                         btrfs_put_ordered_extent(ordered);
2214                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2215                                      lockend, &cached_state, GFP_NOFS);
2216                 ret = btrfs_wait_ordered_range(inode, lockstart,
2217                                                lockend - lockstart + 1);
2218                 if (ret) {
2219                         mutex_unlock(&inode->i_mutex);
2220                         return ret;
2221                 }
2222         }
2223
2224         path = btrfs_alloc_path();
2225         if (!path) {
2226                 ret = -ENOMEM;
2227                 goto out;
2228         }
2229
2230         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2231         if (!rsv) {
2232                 ret = -ENOMEM;
2233                 goto out_free;
2234         }
2235         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2236         rsv->failfast = 1;
2237
2238         /*
2239          * 1 - update the inode
2240          * 1 - removing the extents in the range
2241          * 1 - adding the hole extent if no_holes isn't set
2242          */
2243         rsv_count = no_holes ? 2 : 3;
2244         trans = btrfs_start_transaction(root, rsv_count);
2245         if (IS_ERR(trans)) {
2246                 err = PTR_ERR(trans);
2247                 goto out_free;
2248         }
2249
2250         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2251                                       min_size);
2252         BUG_ON(ret);
2253         trans->block_rsv = rsv;
2254
2255         while (cur_offset < lockend) {
2256                 ret = __btrfs_drop_extents(trans, root, inode, path,
2257                                            cur_offset, lockend + 1,
2258                                            &drop_end, 1, 0, 0, NULL);
2259                 if (ret != -ENOSPC)
2260                         break;
2261
2262                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2263
2264                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2265                 if (ret) {
2266                         err = ret;
2267                         break;
2268                 }
2269
2270                 cur_offset = drop_end;
2271
2272                 ret = btrfs_update_inode(trans, root, inode);
2273                 if (ret) {
2274                         err = ret;
2275                         break;
2276                 }
2277
2278                 btrfs_end_transaction(trans, root);
2279                 btrfs_btree_balance_dirty(root);
2280
2281                 trans = btrfs_start_transaction(root, rsv_count);
2282                 if (IS_ERR(trans)) {
2283                         ret = PTR_ERR(trans);
2284                         trans = NULL;
2285                         break;
2286                 }
2287
2288                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2289                                               rsv, min_size);
2290                 BUG_ON(ret);    /* shouldn't happen */
2291                 trans->block_rsv = rsv;
2292         }
2293
2294         if (ret) {
2295                 err = ret;
2296                 goto out_trans;
2297         }
2298
2299         trans->block_rsv = &root->fs_info->trans_block_rsv;
2300         ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2301         if (ret) {
2302                 err = ret;
2303                 goto out_trans;
2304         }
2305
2306 out_trans:
2307         if (!trans)
2308                 goto out_free;
2309
2310         inode_inc_iversion(inode);
2311         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2312
2313         trans->block_rsv = &root->fs_info->trans_block_rsv;
2314         ret = btrfs_update_inode(trans, root, inode);
2315         btrfs_end_transaction(trans, root);
2316         btrfs_btree_balance_dirty(root);
2317 out_free:
2318         btrfs_free_path(path);
2319         btrfs_free_block_rsv(root, rsv);
2320 out:
2321         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2322                              &cached_state, GFP_NOFS);
2323         mutex_unlock(&inode->i_mutex);
2324         if (ret && !err)
2325                 err = ret;
2326         return err;
2327 }
2328
2329 static long btrfs_fallocate(struct file *file, int mode,
2330                             loff_t offset, loff_t len)
2331 {
2332         struct inode *inode = file_inode(file);
2333         struct extent_state *cached_state = NULL;
2334         struct btrfs_root *root = BTRFS_I(inode)->root;
2335         u64 cur_offset;
2336         u64 last_byte;
2337         u64 alloc_start;
2338         u64 alloc_end;
2339         u64 alloc_hint = 0;
2340         u64 locked_end;
2341         struct extent_map *em;
2342         int blocksize = BTRFS_I(inode)->root->sectorsize;
2343         int ret;
2344
2345         alloc_start = round_down(offset, blocksize);
2346         alloc_end = round_up(offset + len, blocksize);
2347
2348         /* Make sure we aren't being give some crap mode */
2349         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2350                 return -EOPNOTSUPP;
2351
2352         if (mode & FALLOC_FL_PUNCH_HOLE)
2353                 return btrfs_punch_hole(inode, offset, len);
2354
2355         /*
2356          * Make sure we have enough space before we do the
2357          * allocation.
2358          */
2359         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2360         if (ret)
2361                 return ret;
2362         if (root->fs_info->quota_enabled) {
2363                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2364                 if (ret)
2365                         goto out_reserve_fail;
2366         }
2367
2368         mutex_lock(&inode->i_mutex);
2369         ret = inode_newsize_ok(inode, alloc_end);
2370         if (ret)
2371                 goto out;
2372
2373         if (alloc_start > inode->i_size) {
2374                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2375                                         alloc_start);
2376                 if (ret)
2377                         goto out;
2378         } else {
2379                 /*
2380                  * If we are fallocating from the end of the file onward we
2381                  * need to zero out the end of the page if i_size lands in the
2382                  * middle of a page.
2383                  */
2384                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2385                 if (ret)
2386                         goto out;
2387         }
2388
2389         /*
2390          * wait for ordered IO before we have any locks.  We'll loop again
2391          * below with the locks held.
2392          */
2393         ret = btrfs_wait_ordered_range(inode, alloc_start,
2394                                        alloc_end - alloc_start);
2395         if (ret)
2396                 goto out;
2397
2398         locked_end = alloc_end - 1;
2399         while (1) {
2400                 struct btrfs_ordered_extent *ordered;
2401
2402                 /* the extent lock is ordered inside the running
2403                  * transaction
2404                  */
2405                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2406                                  locked_end, 0, &cached_state);
2407                 ordered = btrfs_lookup_first_ordered_extent(inode,
2408                                                             alloc_end - 1);
2409                 if (ordered &&
2410                     ordered->file_offset + ordered->len > alloc_start &&
2411                     ordered->file_offset < alloc_end) {
2412                         btrfs_put_ordered_extent(ordered);
2413                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2414                                              alloc_start, locked_end,
2415                                              &cached_state, GFP_NOFS);
2416                         /*
2417                          * we can't wait on the range with the transaction
2418                          * running or with the extent lock held
2419                          */
2420                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2421                                                        alloc_end - alloc_start);
2422                         if (ret)
2423                                 goto out;
2424                 } else {
2425                         if (ordered)
2426                                 btrfs_put_ordered_extent(ordered);
2427                         break;
2428                 }
2429         }
2430
2431         cur_offset = alloc_start;
2432         while (1) {
2433                 u64 actual_end;
2434
2435                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2436                                       alloc_end - cur_offset, 0);
2437                 if (IS_ERR_OR_NULL(em)) {
2438                         if (!em)
2439                                 ret = -ENOMEM;
2440                         else
2441                                 ret = PTR_ERR(em);
2442                         break;
2443                 }
2444                 last_byte = min(extent_map_end(em), alloc_end);
2445                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2446                 last_byte = ALIGN(last_byte, blocksize);
2447
2448                 if (em->block_start == EXTENT_MAP_HOLE ||
2449                     (cur_offset >= inode->i_size &&
2450                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2451                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2452                                                         last_byte - cur_offset,
2453                                                         1 << inode->i_blkbits,
2454                                                         offset + len,
2455                                                         &alloc_hint);
2456
2457                         if (ret < 0) {
2458                                 free_extent_map(em);
2459                                 break;
2460                         }
2461                 } else if (actual_end > inode->i_size &&
2462                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2463                         /*
2464                          * We didn't need to allocate any more space, but we
2465                          * still extended the size of the file so we need to
2466                          * update i_size.
2467                          */
2468                         inode->i_ctime = CURRENT_TIME;
2469                         i_size_write(inode, actual_end);
2470                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2471                 }
2472                 free_extent_map(em);
2473
2474                 cur_offset = last_byte;
2475                 if (cur_offset >= alloc_end) {
2476                         ret = 0;
2477                         break;
2478                 }
2479         }
2480         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2481                              &cached_state, GFP_NOFS);
2482 out:
2483         mutex_unlock(&inode->i_mutex);
2484         if (root->fs_info->quota_enabled)
2485                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2486 out_reserve_fail:
2487         /* Let go of our reservation. */
2488         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2489         return ret;
2490 }
2491
2492 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2493 {
2494         struct btrfs_root *root = BTRFS_I(inode)->root;
2495         struct extent_map *em = NULL;
2496         struct extent_state *cached_state = NULL;
2497         u64 lockstart = *offset;
2498         u64 lockend = i_size_read(inode);
2499         u64 start = *offset;
2500         u64 len = i_size_read(inode);
2501         int ret = 0;
2502
2503         lockend = max_t(u64, root->sectorsize, lockend);
2504         if (lockend <= lockstart)
2505                 lockend = lockstart + root->sectorsize;
2506
2507         lockend--;
2508         len = lockend - lockstart + 1;
2509
2510         len = max_t(u64, len, root->sectorsize);
2511         if (inode->i_size == 0)
2512                 return -ENXIO;
2513
2514         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2515                          &cached_state);
2516
2517         while (start < inode->i_size) {
2518                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2519                 if (IS_ERR(em)) {
2520                         ret = PTR_ERR(em);
2521                         em = NULL;
2522                         break;
2523                 }
2524
2525                 if (whence == SEEK_HOLE &&
2526                     (em->block_start == EXTENT_MAP_HOLE ||
2527                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2528                         break;
2529                 else if (whence == SEEK_DATA &&
2530                            (em->block_start != EXTENT_MAP_HOLE &&
2531                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2532                         break;
2533
2534                 start = em->start + em->len;
2535                 free_extent_map(em);
2536                 em = NULL;
2537                 cond_resched();
2538         }
2539         free_extent_map(em);
2540         if (!ret) {
2541                 if (whence == SEEK_DATA && start >= inode->i_size)
2542                         ret = -ENXIO;
2543                 else
2544                         *offset = min_t(loff_t, start, inode->i_size);
2545         }
2546         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2547                              &cached_state, GFP_NOFS);
2548         return ret;
2549 }
2550
2551 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2552 {
2553         struct inode *inode = file->f_mapping->host;
2554         int ret;
2555
2556         mutex_lock(&inode->i_mutex);
2557         switch (whence) {
2558         case SEEK_END:
2559         case SEEK_CUR:
2560                 offset = generic_file_llseek(file, offset, whence);
2561                 goto out;
2562         case SEEK_DATA:
2563         case SEEK_HOLE:
2564                 if (offset >= i_size_read(inode)) {
2565                         mutex_unlock(&inode->i_mutex);
2566                         return -ENXIO;
2567                 }
2568
2569                 ret = find_desired_extent(inode, &offset, whence);
2570                 if (ret) {
2571                         mutex_unlock(&inode->i_mutex);
2572                         return ret;
2573                 }
2574         }
2575
2576         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2577 out:
2578         mutex_unlock(&inode->i_mutex);
2579         return offset;
2580 }
2581
2582 const struct file_operations btrfs_file_operations = {
2583         .llseek         = btrfs_file_llseek,
2584         .read           = do_sync_read,
2585         .write          = do_sync_write,
2586         .aio_read       = generic_file_aio_read,
2587         .splice_read    = generic_file_splice_read,
2588         .aio_write      = btrfs_file_aio_write,
2589         .mmap           = btrfs_file_mmap,
2590         .open           = generic_file_open,
2591         .release        = btrfs_release_file,
2592         .fsync          = btrfs_sync_file,
2593         .fallocate      = btrfs_fallocate,
2594         .unlocked_ioctl = btrfs_ioctl,
2595 #ifdef CONFIG_COMPAT
2596         .compat_ioctl   = btrfs_ioctl,
2597 #endif
2598 };
2599
2600 void btrfs_auto_defrag_exit(void)
2601 {
2602         if (btrfs_inode_defrag_cachep)
2603                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2604 }
2605
2606 int btrfs_auto_defrag_init(void)
2607 {
2608         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2609                                         sizeof(struct inode_defrag), 0,
2610                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2611                                         NULL);
2612         if (!btrfs_inode_defrag_cachep)
2613                 return -ENOMEM;
2614
2615         return 0;
2616 }