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,
 776                                                      path->slots[0], fi);
 777                 } else {
 778                         WARN_ON(1);
 779                         extent_end = search_start;
 780                 }
 781
 782                 if (extent_end <= search_start) {
 783                         path->slots[0]++;
 784                         goto next_slot;
 785                 }
 786
 787                 found = 1;
 788                 search_start = max(key.offset, start);
 789                 if (recow || !modify_tree) {
 790                         modify_tree = -1;
 791                         btrfs_release_path(path);
 792                         continue;
 793                 }
 794
 795                 /*
 796                  *     | - range to drop - |
 797                  *  | -------- extent -------- |
 798                  */
 799                 if (start > key.offset && end < extent_end) {
 800                         BUG_ON(del_nr > 0);
 801                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 802
 803                         memcpy(&new_key, &key, sizeof(new_key));
 804                         new_key.offset = start;
 805                         ret = btrfs_duplicate_item(trans, root, path,
 806                                                    &new_key);
 807                         if (ret == -EAGAIN) {
 808                                 btrfs_release_path(path);
 809                                 continue;
 810                         }
 811                         if (ret < 0)
 812                                 break;
 813
 814                         leaf = path->nodes[0];
 815                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 816                                             struct btrfs_file_extent_item);
 817                         btrfs_set_file_extent_num_bytes(leaf, fi,
 818                                                         start - key.offset);
 819
 820                         fi = btrfs_item_ptr(leaf, path->slots[0],
 821                                             struct btrfs_file_extent_item);
 822
 823                         extent_offset += start - key.offset;
 824                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 825                         btrfs_set_file_extent_num_bytes(leaf, fi,
 826                                                         extent_end - start);
 827                         btrfs_mark_buffer_dirty(leaf);
 828
 829                         if (update_refs && disk_bytenr > 0) {
 830                                 ret = btrfs_inc_extent_ref(trans, root,
 831                                                 disk_bytenr, num_bytes, 0,
 832                                                 root->root_key.objectid,
 833                                                 new_key.objectid,
 834                                                 start - extent_offset, 0);
 835                                 BUG_ON(ret); /* -ENOMEM */
 836                         }
 837                         key.offset = start;
 838                 }
 839                 /*
 840                  *  | ---- range to drop ----- |
 841                  *      | -------- extent -------- |
 842                  */
 843                 if (start <= key.offset && end < extent_end) {
 844                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 845
 846                         memcpy(&new_key, &key, sizeof(new_key));
 847                         new_key.offset = end;
 848                         btrfs_set_item_key_safe(root, path, &new_key);
 849
 850                         extent_offset += end - key.offset;
 851                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 852                         btrfs_set_file_extent_num_bytes(leaf, fi,
 853                                                         extent_end - end);
 854                         btrfs_mark_buffer_dirty(leaf);
 855                         if (update_refs && disk_bytenr > 0)
 856                                 inode_sub_bytes(inode, end - key.offset);
 857                         break;
 858                 }
 859
 860                 search_start = extent_end;
 861                 /*
 862                  *       | ---- range to drop ----- |
 863                  *  | -------- extent -------- |
 864                  */
 865                 if (start > key.offset && end >= extent_end) {
 866                         BUG_ON(del_nr > 0);
 867                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 868
 869                         btrfs_set_file_extent_num_bytes(leaf, fi,
 870                                                         start - key.offset);
 871                         btrfs_mark_buffer_dirty(leaf);
 872                         if (update_refs && disk_bytenr > 0)
 873                                 inode_sub_bytes(inode, extent_end - start);
 874                         if (end == extent_end)
 875                                 break;
 876
 877                         path->slots[0]++;
 878                         goto next_slot;
 879                 }
 880
 881                 /*
 882                  *  | ---- range to drop ----- |
 883                  *    | ------ extent ------ |
 884                  */
 885                 if (start <= key.offset && end >= extent_end) {
 886                         if (del_nr == 0) {
 887                                 del_slot = path->slots[0];
 888                                 del_nr = 1;
 889                         } else {
 890                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 891                                 del_nr++;
 892                         }
 893
 894                         if (update_refs &&
 895                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 896                                 inode_sub_bytes(inode,
 897                                                 extent_end - key.offset);
 898                                 extent_end = ALIGN(extent_end,
 899                                                    root->sectorsize);
 900                         } else if (update_refs && disk_bytenr > 0) {
 901                                 ret = btrfs_free_extent(trans, root,
 902                                                 disk_bytenr, num_bytes, 0,
 903                                                 root->root_key.objectid,
 904                                                 key.objectid, key.offset -
 905                                                 extent_offset, 0);
 906                                 BUG_ON(ret); /* -ENOMEM */
 907                                 inode_sub_bytes(inode,
 908                                                 extent_end - key.offset);
 909                         }
 910
 911                         if (end == extent_end)
 912                                 break;
 913
 914                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 915                                 path->slots[0]++;
 916                                 goto next_slot;
 917                         }
 918
 919                         ret = btrfs_del_items(trans, root, path, del_slot,
 920                                               del_nr);
 921                         if (ret) {
 922                                 btrfs_abort_transaction(trans, root, ret);
 923                                 break;
 924                         }
 925
 926                         del_nr = 0;
 927                         del_slot = 0;
 928
 929                         btrfs_release_path(path);
 930                         continue;
 931                 }
 932
 933                 BUG_ON(1);
 934         }
 935
 936         if (!ret && del_nr > 0) {
 937                 /*
 938                  * Set path->slots[0] to first slot, so that after the delete
 939                  * if items are move off from our leaf to its immediate left or
 940                  * right neighbor leafs, we end up with a correct and adjusted
 941                  * path->slots[0] for our insertion.
 942                  */
 943                 path->slots[0] = del_slot;
 944                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 945                 if (ret)
 946                         btrfs_abort_transaction(trans, root, ret);
 947
 948                 leaf = path->nodes[0];
 949                 /*
 950                  * leaf eb has flag EXTENT_BUFFER_STALE if it was deleted (that
 951                  * is, its contents got pushed to its neighbors), in which case
 952                  * it means path->locks[0] == 0
 953                  */
 954                 if (!ret && replace_extent && leafs_visited == 1 &&
 955                     path->locks[0] &&
 956                     btrfs_leaf_free_space(root, leaf) >=
 957                     sizeof(struct btrfs_item) + extent_item_size) {
 958
 959                         key.objectid = ino;
 960                         key.type = BTRFS_EXTENT_DATA_KEY;
 961                         key.offset = start;
 962                         setup_items_for_insert(root, path, &key,
 963                                                &extent_item_size,
 964                                                extent_item_size,
 965                                                sizeof(struct btrfs_item) +
 966                                                extent_item_size, 1);
 967                         *key_inserted = 1;
 968                 }
 969         }
 970
 971         if (!replace_extent || !(*key_inserted))
 972                 btrfs_release_path(path);
 973         if (drop_end)
 974                 *drop_end = found ? min(end, extent_end) : end;
 975         return ret;
 976 }
 977
 978 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 979                        struct btrfs_root *root, struct inode *inode, u64 start,
 980                        u64 end, int drop_cache)
 981 {
 982         struct btrfs_path *path;
 983         int ret;
 984
 985         path = btrfs_alloc_path();
 986         if (!path)
 987                 return -ENOMEM;
 988         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
 989                                    drop_cache, 0, 0, NULL);
 990         btrfs_free_path(path);
 991         return ret;
 992 }
 993
 994 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 995                             u64 objectid, u64 bytenr, u64 orig_offset,
 996                             u64 *start, u64 *end)
 997 {
 998         struct btrfs_file_extent_item *fi;
 999         struct btrfs_key key;
1000         u64 extent_end;
1001
1002         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1003                 return 0;
1004
1005         btrfs_item_key_to_cpu(leaf, &key, slot);
1006         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1007                 return 0;
1008
1009         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1010         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1011             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1012             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1013             btrfs_file_extent_compression(leaf, fi) ||
1014             btrfs_file_extent_encryption(leaf, fi) ||
1015             btrfs_file_extent_other_encoding(leaf, fi))
1016                 return 0;
1017
1018         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1019         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1020                 return 0;
1021
1022         *start = key.offset;
1023         *end = extent_end;
1024         return 1;
1025 }
1026
1027 /*
1028  * Mark extent in the range start - end as written.
1029  *
1030  * This changes extent type from 'pre-allocated' to 'regular'. If only
1031  * part of extent is marked as written, the extent will be split into
1032  * two or three.
1033  */
1034 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1035                               struct inode *inode, u64 start, u64 end)
1036 {
1037         struct btrfs_root *root = BTRFS_I(inode)->root;
1038         struct extent_buffer *leaf;
1039         struct btrfs_path *path;
1040         struct btrfs_file_extent_item *fi;
1041         struct btrfs_key key;
1042         struct btrfs_key new_key;
1043         u64 bytenr;
1044         u64 num_bytes;
1045         u64 extent_end;
1046         u64 orig_offset;
1047         u64 other_start;
1048         u64 other_end;
1049         u64 split;
1050         int del_nr = 0;
1051         int del_slot = 0;
1052         int recow;
1053         int ret;
1054         u64 ino = btrfs_ino(inode);
1055
1056         path = btrfs_alloc_path();
1057         if (!path)
1058                 return -ENOMEM;
1059 again:
1060         recow = 0;
1061         split = start;
1062         key.objectid = ino;
1063         key.type = BTRFS_EXTENT_DATA_KEY;
1064         key.offset = split;
1065
1066         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1067         if (ret < 0)
1068                 goto out;
1069         if (ret > 0 && path->slots[0] > 0)
1070                 path->slots[0]--;
1071
1072         leaf = path->nodes[0];
1073         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1074         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1075         fi = btrfs_item_ptr(leaf, path->slots[0],
1076                             struct btrfs_file_extent_item);
1077         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1078                BTRFS_FILE_EXTENT_PREALLOC);
1079         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1080         BUG_ON(key.offset > start || extent_end < end);
1081
1082         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1083         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1084         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1085         memcpy(&new_key, &key, sizeof(new_key));
1086
1087         if (start == key.offset && end < extent_end) {
1088                 other_start = 0;
1089                 other_end = start;
1090                 if (extent_mergeable(leaf, path->slots[0] - 1,
1091                                      ino, bytenr, orig_offset,
1092                                      &other_start, &other_end)) {
1093                         new_key.offset = end;
1094                         btrfs_set_item_key_safe(root, path, &new_key);
1095                         fi = btrfs_item_ptr(leaf, path->slots[0],
1096                                             struct btrfs_file_extent_item);
1097                         btrfs_set_file_extent_generation(leaf, fi,
1098                                                          trans->transid);
1099                         btrfs_set_file_extent_num_bytes(leaf, fi,
1100                                                         extent_end - end);
1101                         btrfs_set_file_extent_offset(leaf, fi,
1102                                                      end - orig_offset);
1103                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1104                                             struct btrfs_file_extent_item);
1105                         btrfs_set_file_extent_generation(leaf, fi,
1106                                                          trans->transid);
1107                         btrfs_set_file_extent_num_bytes(leaf, fi,
1108                                                         end - other_start);
1109                         btrfs_mark_buffer_dirty(leaf);
1110                         goto out;
1111                 }
1112         }
1113
1114         if (start > key.offset && end == extent_end) {
1115                 other_start = end;
1116                 other_end = 0;
1117                 if (extent_mergeable(leaf, path->slots[0] + 1,
1118                                      ino, bytenr, orig_offset,
1119                                      &other_start, &other_end)) {
1120                         fi = btrfs_item_ptr(leaf, path->slots[0],
1121                                             struct btrfs_file_extent_item);
1122                         btrfs_set_file_extent_num_bytes(leaf, fi,
1123                                                         start - key.offset);
1124                         btrfs_set_file_extent_generation(leaf, fi,
1125                                                          trans->transid);
1126                         path->slots[0]++;
1127                         new_key.offset = start;
1128                         btrfs_set_item_key_safe(root, path, &new_key);
1129
1130                         fi = btrfs_item_ptr(leaf, path->slots[0],
1131                                             struct btrfs_file_extent_item);
1132                         btrfs_set_file_extent_generation(leaf, fi,
1133                                                          trans->transid);
1134                         btrfs_set_file_extent_num_bytes(leaf, fi,
1135                                                         other_end - start);
1136                         btrfs_set_file_extent_offset(leaf, fi,
1137                                                      start - orig_offset);
1138                         btrfs_mark_buffer_dirty(leaf);
1139                         goto out;
1140                 }
1141         }
1142
1143         while (start > key.offset || end < extent_end) {
1144                 if (key.offset == start)
1145                         split = end;
1146
1147                 new_key.offset = split;
1148                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1149                 if (ret == -EAGAIN) {
1150                         btrfs_release_path(path);
1151                         goto again;
1152                 }
1153                 if (ret < 0) {
1154                         btrfs_abort_transaction(trans, root, ret);
1155                         goto out;
1156                 }
1157
1158                 leaf = path->nodes[0];
1159                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1160                                     struct btrfs_file_extent_item);
1161                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1162                 btrfs_set_file_extent_num_bytes(leaf, fi,
1163                                                 split - key.offset);
1164
1165                 fi = btrfs_item_ptr(leaf, path->slots[0],
1166                                     struct btrfs_file_extent_item);
1167
1168                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1169                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1170                 btrfs_set_file_extent_num_bytes(leaf, fi,
1171                                                 extent_end - split);
1172                 btrfs_mark_buffer_dirty(leaf);
1173
1174                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1175                                            root->root_key.objectid,
1176                                            ino, orig_offset, 0);
1177                 BUG_ON(ret); /* -ENOMEM */
1178
1179                 if (split == start) {
1180                         key.offset = start;
1181                 } else {
1182                         BUG_ON(start != key.offset);
1183                         path->slots[0]--;
1184                         extent_end = end;
1185                 }
1186                 recow = 1;
1187         }
1188
1189         other_start = end;
1190         other_end = 0;
1191         if (extent_mergeable(leaf, path->slots[0] + 1,
1192                              ino, bytenr, orig_offset,
1193                              &other_start, &other_end)) {
1194                 if (recow) {
1195                         btrfs_release_path(path);
1196                         goto again;
1197                 }
1198                 extent_end = other_end;
1199                 del_slot = path->slots[0] + 1;
1200                 del_nr++;
1201                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1202                                         0, root->root_key.objectid,
1203                                         ino, orig_offset, 0);
1204                 BUG_ON(ret); /* -ENOMEM */
1205         }
1206         other_start = 0;
1207         other_end = start;
1208         if (extent_mergeable(leaf, path->slots[0] - 1,
1209                              ino, bytenr, orig_offset,
1210                              &other_start, &other_end)) {
1211                 if (recow) {
1212                         btrfs_release_path(path);
1213                         goto again;
1214                 }
1215                 key.offset = other_start;
1216                 del_slot = path->slots[0];
1217                 del_nr++;
1218                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1219                                         0, root->root_key.objectid,
1220                                         ino, orig_offset, 0);
1221                 BUG_ON(ret); /* -ENOMEM */
1222         }
1223         if (del_nr == 0) {
1224                 fi = btrfs_item_ptr(leaf, path->slots[0],
1225                            struct btrfs_file_extent_item);
1226                 btrfs_set_file_extent_type(leaf, fi,
1227                                            BTRFS_FILE_EXTENT_REG);
1228                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1229                 btrfs_mark_buffer_dirty(leaf);
1230         } else {
1231                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1232                            struct btrfs_file_extent_item);
1233                 btrfs_set_file_extent_type(leaf, fi,
1234                                            BTRFS_FILE_EXTENT_REG);
1235                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1236                 btrfs_set_file_extent_num_bytes(leaf, fi,
1237                                                 extent_end - key.offset);
1238                 btrfs_mark_buffer_dirty(leaf);
1239
1240                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1241                 if (ret < 0) {
1242                         btrfs_abort_transaction(trans, root, ret);
1243                         goto out;
1244                 }
1245         }
1246 out:
1247         btrfs_free_path(path);
1248         return 0;
1249 }
1250
1251 /*
1252  * on error we return an unlocked page and the error value
1253  * on success we return a locked page and 0
1254  */
1255 static int prepare_uptodate_page(struct page *page, u64 pos,
1256                                  bool force_uptodate)
1257 {
1258         int ret = 0;
1259
1260         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1261             !PageUptodate(page)) {
1262                 ret = btrfs_readpage(NULL, page);
1263                 if (ret)
1264                         return ret;
1265                 lock_page(page);
1266                 if (!PageUptodate(page)) {
1267                         unlock_page(page);
1268                         return -EIO;
1269                 }
1270         }
1271         return 0;
1272 }
1273
1274 /*
1275  * this just gets pages into the page cache and locks them down.
1276  */
1277 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1278                                   size_t num_pages, loff_t pos,
1279                                   size_t write_bytes, bool force_uptodate)
1280 {
1281         int i;
1282         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1283         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1284         int err = 0;
1285         int faili;
1286
1287         for (i = 0; i < num_pages; i++) {
1288                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1289                                                mask | __GFP_WRITE);
1290                 if (!pages[i]) {
1291                         faili = i - 1;
1292                         err = -ENOMEM;
1293                         goto fail;
1294                 }
1295
1296                 if (i == 0)
1297                         err = prepare_uptodate_page(pages[i], pos,
1298                                                     force_uptodate);
1299                 if (i == num_pages - 1)
1300                         err = prepare_uptodate_page(pages[i],
1301                                                     pos + write_bytes, false);
1302                 if (err) {
1303                         page_cache_release(pages[i]);
1304                         faili = i - 1;
1305                         goto fail;
1306                 }
1307                 wait_on_page_writeback(pages[i]);
1308         }
1309
1310         return 0;
1311 fail:
1312         while (faili >= 0) {
1313                 unlock_page(pages[faili]);
1314                 page_cache_release(pages[faili]);
1315                 faili--;
1316         }
1317         return err;
1318
1319 }
1320
1321 /*
1322  * This function locks the extent and properly waits for data=ordered extents
1323  * to finish before allowing the pages to be modified if need.
1324  *
1325  * The return value:
1326  * 1 - the extent is locked
1327  * 0 - the extent is not locked, and everything is OK
1328  * -EAGAIN - need re-prepare the pages
1329  * the other < 0 number - Something wrong happens
1330  */
1331 static noinline int
1332 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1333                                 size_t num_pages, loff_t pos,
1334                                 u64 *lockstart, u64 *lockend,
1335                                 struct extent_state **cached_state)
1336 {
1337         u64 start_pos;
1338         u64 last_pos;
1339         int i;
1340         int ret = 0;
1341
1342         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1343         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1344
1345         if (start_pos < inode->i_size) {
1346                 struct btrfs_ordered_extent *ordered;
1347                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1348                                  start_pos, last_pos, 0, cached_state);
1349                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1350                 if (ordered &&
1351                     ordered->file_offset + ordered->len > start_pos &&
1352                     ordered->file_offset <= last_pos) {
1353                         btrfs_put_ordered_extent(ordered);
1354                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1355                                              start_pos, last_pos,
1356                                              cached_state, GFP_NOFS);
1357                         for (i = 0; i < num_pages; i++) {
1358                                 unlock_page(pages[i]);
1359                                 page_cache_release(pages[i]);
1360                         }
1361                         ret = btrfs_wait_ordered_range(inode, start_pos,
1362                                                 last_pos - start_pos + 1);
1363                         if (ret)
1364                                 return ret;
1365                         else
1366                                 return -EAGAIN;
1367                 }
1368                 if (ordered)
1369                         btrfs_put_ordered_extent(ordered);
1370
1371                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1372                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1373                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1374                                   0, 0, cached_state, GFP_NOFS);
1375                 *lockstart = start_pos;
1376                 *lockend = last_pos;
1377                 ret = 1;
1378         }
1379
1380         for (i = 0; i < num_pages; i++) {
1381                 if (clear_page_dirty_for_io(pages[i]))
1382                         account_page_redirty(pages[i]);
1383                 set_page_extent_mapped(pages[i]);
1384                 WARN_ON(!PageLocked(pages[i]));
1385         }
1386
1387         return ret;
1388 }
1389
1390 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1391                                     size_t *write_bytes)
1392 {
1393         struct btrfs_root *root = BTRFS_I(inode)->root;
1394         struct btrfs_ordered_extent *ordered;
1395         u64 lockstart, lockend;
1396         u64 num_bytes;
1397         int ret;
1398
1399         lockstart = round_down(pos, root->sectorsize);
1400         lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1401
1402         while (1) {
1403                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1404                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1405                                                      lockend - lockstart + 1);
1406                 if (!ordered) {
1407                         break;
1408                 }
1409                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1410                 btrfs_start_ordered_extent(inode, ordered, 1);
1411                 btrfs_put_ordered_extent(ordered);
1412         }
1413
1414         num_bytes = lockend - lockstart + 1;
1415         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1416         if (ret <= 0) {
1417                 ret = 0;
1418         } else {
1419                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1420                                  EXTENT_DIRTY | EXTENT_DELALLOC |
1421                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1422                                  NULL, GFP_NOFS);
1423                 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1424         }
1425
1426         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1427
1428         return ret;
1429 }
1430
1431 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1432                                                struct iov_iter *i,
1433                                                loff_t pos)
1434 {
1435         struct inode *inode = file_inode(file);
1436         struct btrfs_root *root = BTRFS_I(inode)->root;
1437         struct page **pages = NULL;
1438         struct extent_state *cached_state = NULL;
1439         u64 release_bytes = 0;
1440         u64 lockstart;
1441         u64 lockend;
1442         unsigned long first_index;
1443         size_t num_written = 0;
1444         int nrptrs;
1445         int ret = 0;
1446         bool only_release_metadata = false;
1447         bool force_page_uptodate = false;
1448         bool need_unlock;
1449
1450         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1451                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1452                      (sizeof(struct page *)));
1453         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1454         nrptrs = max(nrptrs, 8);
1455         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1456         if (!pages)
1457                 return -ENOMEM;
1458
1459         first_index = pos >> PAGE_CACHE_SHIFT;
1460
1461         while (iov_iter_count(i) > 0) {
1462                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1463                 size_t write_bytes = min(iov_iter_count(i),
1464                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1465                                          offset);
1466                 size_t num_pages = (write_bytes + offset +
1467                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1468                 size_t reserve_bytes;
1469                 size_t dirty_pages;
1470                 size_t copied;
1471
1472                 WARN_ON(num_pages > nrptrs);
1473
1474                 /*
1475                  * Fault pages before locking them in prepare_pages
1476                  * to avoid recursive lock
1477                  */
1478                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1479                         ret = -EFAULT;
1480                         break;
1481                 }
1482
1483                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1484                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1485                 if (ret == -ENOSPC &&
1486                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1487                                               BTRFS_INODE_PREALLOC))) {
1488                         ret = check_can_nocow(inode, pos, &write_bytes);
1489                         if (ret > 0) {
1490                                 only_release_metadata = true;
1491                                 /*
1492                                  * our prealloc extent may be smaller than
1493                                  * write_bytes, so scale down.
1494                                  */
1495                                 num_pages = (write_bytes + offset +
1496                                              PAGE_CACHE_SIZE - 1) >>
1497                                         PAGE_CACHE_SHIFT;
1498                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1499                                 ret = 0;
1500                         } else {
1501                                 ret = -ENOSPC;
1502                         }
1503                 }
1504
1505                 if (ret)
1506                         break;
1507
1508                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1509                 if (ret) {
1510                         if (!only_release_metadata)
1511                                 btrfs_free_reserved_data_space(inode,
1512                                                                reserve_bytes);
1513                         break;
1514                 }
1515
1516                 release_bytes = reserve_bytes;
1517                 need_unlock = false;
1518 again:
1519                 /*
1520                  * This is going to setup the pages array with the number of
1521                  * pages we want, so we don't really need to worry about the
1522                  * contents of pages from loop to loop
1523                  */
1524                 ret = prepare_pages(inode, pages, num_pages,
1525                                     pos, write_bytes,
1526                                     force_page_uptodate);
1527                 if (ret)
1528                         break;
1529
1530                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1531                                                       pos, &lockstart, &lockend,
1532                                                       &cached_state);
1533                 if (ret < 0) {
1534                         if (ret == -EAGAIN)
1535                                 goto again;
1536                         break;
1537                 } else if (ret > 0) {
1538                         need_unlock = true;
1539                         ret = 0;
1540                 }
1541
1542                 copied = btrfs_copy_from_user(pos, num_pages,
1543                                            write_bytes, pages, i);
1544
1545                 /*
1546                  * if we have trouble faulting in the pages, fall
1547                  * back to one page at a time
1548                  */
1549                 if (copied < write_bytes)
1550                         nrptrs = 1;
1551
1552                 if (copied == 0) {
1553                         force_page_uptodate = true;
1554                         dirty_pages = 0;
1555                 } else {
1556                         force_page_uptodate = false;
1557                         dirty_pages = (copied + offset +
1558                                        PAGE_CACHE_SIZE - 1) >>
1559                                        PAGE_CACHE_SHIFT;
1560                 }
1561
1562                 /*
1563                  * If we had a short copy we need to release the excess delaloc
1564                  * bytes we reserved.  We need to increment outstanding_extents
1565                  * because btrfs_delalloc_release_space will decrement it, but
1566                  * we still have an outstanding extent for the chunk we actually
1567                  * managed to copy.
1568                  */
1569                 if (num_pages > dirty_pages) {
1570                         release_bytes = (num_pages - dirty_pages) <<
1571                                 PAGE_CACHE_SHIFT;
1572                         if (copied > 0) {
1573                                 spin_lock(&BTRFS_I(inode)->lock);
1574                                 BTRFS_I(inode)->outstanding_extents++;
1575                                 spin_unlock(&BTRFS_I(inode)->lock);
1576                         }
1577                         if (only_release_metadata)
1578                                 btrfs_delalloc_release_metadata(inode,
1579                                                                 release_bytes);
1580                         else
1581                                 btrfs_delalloc_release_space(inode,
1582                                                              release_bytes);
1583                 }
1584
1585                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1586
1587                 if (copied > 0)
1588                         ret = btrfs_dirty_pages(root, inode, pages,
1589                                                 dirty_pages, pos, copied,
1590                                                 NULL);
1591                 if (need_unlock)
1592                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1593                                              lockstart, lockend, &cached_state,
1594                                              GFP_NOFS);
1595                 if (ret) {
1596                         btrfs_drop_pages(pages, num_pages);
1597                         break;
1598                 }
1599
1600                 release_bytes = 0;
1601                 if (only_release_metadata && copied > 0) {
1602                         u64 lockstart = round_down(pos, root->sectorsize);
1603                         u64 lockend = lockstart +
1604                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1605
1606                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1607                                        lockend, EXTENT_NORESERVE, NULL,
1608                                        NULL, GFP_NOFS);
1609                         only_release_metadata = false;
1610                 }
1611
1612                 btrfs_drop_pages(pages, num_pages);
1613
1614                 cond_resched();
1615
1616                 balance_dirty_pages_ratelimited(inode->i_mapping);
1617                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1618                         btrfs_btree_balance_dirty(root);
1619
1620                 pos += copied;
1621                 num_written += copied;
1622         }
1623
1624         kfree(pages);
1625
1626         if (release_bytes) {
1627                 if (only_release_metadata)
1628                         btrfs_delalloc_release_metadata(inode, release_bytes);
1629                 else
1630                         btrfs_delalloc_release_space(inode, release_bytes);
1631         }
1632
1633         return num_written ? num_written : ret;
1634 }
1635
1636 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1637                                     const struct iovec *iov,
1638                                     unsigned long nr_segs, loff_t pos,
1639                                     loff_t *ppos, size_t count, size_t ocount)
1640 {
1641         struct file *file = iocb->ki_filp;
1642         struct iov_iter i;
1643         ssize_t written;
1644         ssize_t written_buffered;
1645         loff_t endbyte;
1646         int err;
1647
1648         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1649                                             count, ocount);
1650
1651         if (written < 0 || written == count)
1652                 return written;
1653
1654         pos += written;
1655         count -= written;
1656         iov_iter_init(&i, iov, nr_segs, count, written);
1657         written_buffered = __btrfs_buffered_write(file, &i, pos);
1658         if (written_buffered < 0) {
1659                 err = written_buffered;
1660                 goto out;
1661         }
1662         endbyte = pos + written_buffered - 1;
1663         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1664         if (err)
1665                 goto out;
1666         written += written_buffered;
1667         *ppos = pos + written_buffered;
1668         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1669                                  endbyte >> PAGE_CACHE_SHIFT);
1670 out:
1671         return written ? written : err;
1672 }
1673
1674 static void update_time_for_write(struct inode *inode)
1675 {
1676         struct timespec now;
1677
1678         if (IS_NOCMTIME(inode))
1679                 return;
1680
1681         now = current_fs_time(inode->i_sb);
1682         if (!timespec_equal(&inode->i_mtime, &now))
1683                 inode->i_mtime = now;
1684
1685         if (!timespec_equal(&inode->i_ctime, &now))
1686                 inode->i_ctime = now;
1687
1688         if (IS_I_VERSION(inode))
1689                 inode_inc_iversion(inode);
1690 }
1691
1692 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1693                                     const struct iovec *iov,
1694                                     unsigned long nr_segs, loff_t pos)
1695 {
1696         struct file *file = iocb->ki_filp;
1697         struct inode *inode = file_inode(file);
1698         struct btrfs_root *root = BTRFS_I(inode)->root;
1699         loff_t *ppos = &iocb->ki_pos;
1700         u64 start_pos;
1701         ssize_t num_written = 0;
1702         ssize_t err = 0;
1703         size_t count, ocount;
1704         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1705
1706         mutex_lock(&inode->i_mutex);
1707
1708         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1709         if (err) {
1710                 mutex_unlock(&inode->i_mutex);
1711                 goto out;
1712         }
1713         count = ocount;
1714
1715         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1716         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1717         if (err) {
1718                 mutex_unlock(&inode->i_mutex);
1719                 goto out;
1720         }
1721
1722         if (count == 0) {
1723                 mutex_unlock(&inode->i_mutex);
1724                 goto out;
1725         }
1726
1727         err = file_remove_suid(file);
1728         if (err) {
1729                 mutex_unlock(&inode->i_mutex);
1730                 goto out;
1731         }
1732
1733         /*
1734          * If BTRFS flips readonly due to some impossible error
1735          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1736          * although we have opened a file as writable, we have
1737          * to stop this write operation to ensure FS consistency.
1738          */
1739         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1740                 mutex_unlock(&inode->i_mutex);
1741                 err = -EROFS;
1742                 goto out;
1743         }
1744
1745         /*
1746          * We reserve space for updating the inode when we reserve space for the
1747          * extent we are going to write, so we will enospc out there.  We don't
1748          * need to start yet another transaction to update the inode as we will
1749          * update the inode when we finish writing whatever data we write.
1750          */
1751         update_time_for_write(inode);
1752
1753         start_pos = round_down(pos, root->sectorsize);
1754         if (start_pos > i_size_read(inode)) {
1755                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1756                 if (err) {
1757                         mutex_unlock(&inode->i_mutex);
1758                         goto out;
1759                 }
1760         }
1761
1762         if (sync)
1763                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1764
1765         if (unlikely(file->f_flags & O_DIRECT)) {
1766                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1767                                                    pos, ppos, count, ocount);
1768         } else {
1769                 struct iov_iter i;
1770
1771                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1772
1773                 num_written = __btrfs_buffered_write(file, &i, pos);
1774                 if (num_written > 0)
1775                         *ppos = pos + num_written;
1776         }
1777
1778         mutex_unlock(&inode->i_mutex);
1779
1780         /*
1781          * we want to make sure fsync finds this change
1782          * but we haven't joined a transaction running right now.
1783          *
1784          * Later on, someone is sure to update the inode and get the
1785          * real transid recorded.
1786          *
1787          * We set last_trans now to the fs_info generation + 1,
1788          * this will either be one more than the running transaction
1789          * or the generation used for the next transaction if there isn't
1790          * one running right now.
1791          *
1792          * We also have to set last_sub_trans to the current log transid,
1793          * otherwise subsequent syncs to a file that's been synced in this
1794          * transaction will appear to have already occured.
1795          */
1796         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1797         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1798         if (num_written > 0) {
1799                 err = generic_write_sync(file, pos, num_written);
1800                 if (err < 0 && num_written > 0)
1801                         num_written = err;
1802         }
1803
1804         if (sync)
1805                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1806 out:
1807         current->backing_dev_info = NULL;
1808         return num_written ? num_written : err;
1809 }
1810
1811 int btrfs_release_file(struct inode *inode, struct file *filp)
1812 {
1813         /*
1814          * ordered_data_close is set by settattr when we are about to truncate
1815          * a file from a non-zero size to a zero size.  This tries to
1816          * flush down new bytes that may have been written if the
1817          * application were using truncate to replace a file in place.
1818          */
1819         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1820                                &BTRFS_I(inode)->runtime_flags)) {
1821                 struct btrfs_trans_handle *trans;
1822                 struct btrfs_root *root = BTRFS_I(inode)->root;
1823
1824                 /*
1825                  * We need to block on a committing transaction to keep us from
1826                  * throwing a ordered operation on to the list and causing
1827                  * something like sync to deadlock trying to flush out this
1828                  * inode.
1829                  */
1830                 trans = btrfs_start_transaction(root, 0);
1831                 if (IS_ERR(trans))
1832                         return PTR_ERR(trans);
1833                 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1834                 btrfs_end_transaction(trans, root);
1835                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1836                         filemap_flush(inode->i_mapping);
1837         }
1838         if (filp->private_data)
1839                 btrfs_ioctl_trans_end(filp);
1840         return 0;
1841 }
1842
1843 /*
1844  * fsync call for both files and directories.  This logs the inode into
1845  * the tree log instead of forcing full commits whenever possible.
1846  *
1847  * It needs to call filemap_fdatawait so that all ordered extent updates are
1848  * in the metadata btree are up to date for copying to the log.
1849  *
1850  * It drops the inode mutex before doing the tree log commit.  This is an
1851  * important optimization for directories because holding the mutex prevents
1852  * new operations on the dir while we write to disk.
1853  */
1854 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1855 {
1856         struct dentry *dentry = file->f_path.dentry;
1857         struct inode *inode = dentry->d_inode;
1858         struct btrfs_root *root = BTRFS_I(inode)->root;
1859         int ret = 0;
1860         struct btrfs_trans_handle *trans;
1861         bool full_sync = 0;
1862
1863         trace_btrfs_sync_file(file, datasync);
1864
1865         /*
1866          * We write the dirty pages in the range and wait until they complete
1867          * out of the ->i_mutex. If so, we can flush the dirty pages by
1868          * multi-task, and make the performance up.  See
1869          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1870          */
1871         atomic_inc(&BTRFS_I(inode)->sync_writers);
1872         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1873         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1874                              &BTRFS_I(inode)->runtime_flags))
1875                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1876         atomic_dec(&BTRFS_I(inode)->sync_writers);
1877         if (ret)
1878                 return ret;
1879
1880         mutex_lock(&inode->i_mutex);
1881
1882         /*
1883          * We flush the dirty pages again to avoid some dirty pages in the
1884          * range being left.
1885          */
1886         atomic_inc(&root->log_batch);
1887         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1888                              &BTRFS_I(inode)->runtime_flags);
1889         if (full_sync) {
1890                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1891                 if (ret) {
1892                         mutex_unlock(&inode->i_mutex);
1893                         goto out;
1894                 }
1895         }
1896         atomic_inc(&root->log_batch);
1897
1898         /*
1899          * check the transaction that last modified this inode
1900          * and see if its already been committed
1901          */
1902         if (!BTRFS_I(inode)->last_trans) {
1903                 mutex_unlock(&inode->i_mutex);
1904                 goto out;
1905         }
1906
1907         /*
1908          * if the last transaction that changed this file was before
1909          * the current transaction, we can bail out now without any
1910          * syncing
1911          */
1912         smp_mb();
1913         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1914             BTRFS_I(inode)->last_trans <=
1915             root->fs_info->last_trans_committed) {
1916                 BTRFS_I(inode)->last_trans = 0;
1917
1918                 /*
1919                  * We'v had everything committed since the last time we were
1920                  * modified so clear this flag in case it was set for whatever
1921                  * reason, it's no longer relevant.
1922                  */
1923                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1924                           &BTRFS_I(inode)->runtime_flags);
1925                 mutex_unlock(&inode->i_mutex);
1926                 goto out;
1927         }
1928
1929         /*
1930          * ok we haven't committed the transaction yet, lets do a commit
1931          */
1932         if (file->private_data)
1933                 btrfs_ioctl_trans_end(file);
1934
1935         /*
1936          * We use start here because we will need to wait on the IO to complete
1937          * in btrfs_sync_log, which could require joining a transaction (for
1938          * example checking cross references in the nocow path).  If we use join
1939          * here we could get into a situation where we're waiting on IO to
1940          * happen that is blocked on a transaction trying to commit.  With start
1941          * we inc the extwriter counter, so we wait for all extwriters to exit
1942          * before we start blocking join'ers.  This comment is to keep somebody
1943          * from thinking they are super smart and changing this to
1944          * btrfs_join_transaction *cough*Josef*cough*.
1945          */
1946         trans = btrfs_start_transaction(root, 0);
1947         if (IS_ERR(trans)) {
1948                 ret = PTR_ERR(trans);
1949                 mutex_unlock(&inode->i_mutex);
1950                 goto out;
1951         }
1952         trans->sync = true;
1953
1954         ret = btrfs_log_dentry_safe(trans, root, dentry);
1955         if (ret < 0) {
1956                 /* Fallthrough and commit/free transaction. */
1957                 ret = 1;
1958         }
1959
1960         /* we've logged all the items and now have a consistent
1961          * version of the file in the log.  It is possible that
1962          * someone will come in and modify the file, but that's
1963          * fine because the log is consistent on disk, and we
1964          * have references to all of the file's extents
1965          *
1966          * It is possible that someone will come in and log the
1967          * file again, but that will end up using the synchronization
1968          * inside btrfs_sync_log to keep things safe.
1969          */
1970         mutex_unlock(&inode->i_mutex);
1971
1972         if (ret != BTRFS_NO_LOG_SYNC) {
1973                 if (!ret) {
1974                         ret = btrfs_sync_log(trans, root);
1975                         if (!ret) {
1976                                 ret = btrfs_end_transaction(trans, root);
1977                                 goto out;
1978                         }
1979                 }
1980                 if (!full_sync) {
1981                         ret = btrfs_wait_ordered_range(inode, start,
1982                                                        end - start + 1);
1983                         if (ret)
1984                                 goto out;
1985                 }
1986                 ret = btrfs_commit_transaction(trans, root);
1987         } else {
1988                 ret = btrfs_end_transaction(trans, root);
1989         }
1990 out:
1991         return ret > 0 ? -EIO : ret;
1992 }
1993
1994 static const struct vm_operations_struct btrfs_file_vm_ops = {
1995         .fault          = filemap_fault,
1996         .page_mkwrite   = btrfs_page_mkwrite,
1997         .remap_pages    = generic_file_remap_pages,
1998 };
1999
2000 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
2001 {
2002         struct address_space *mapping = filp->f_mapping;
2003
2004         if (!mapping->a_ops->readpage)
2005                 return -ENOEXEC;
2006
2007         file_accessed(filp);
2008         vma->vm_ops = &btrfs_file_vm_ops;
2009
2010         return 0;
2011 }
2012
2013 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2014                           int slot, u64 start, u64 end)
2015 {
2016         struct btrfs_file_extent_item *fi;
2017         struct btrfs_key key;
2018
2019         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2020                 return 0;
2021
2022         btrfs_item_key_to_cpu(leaf, &key, slot);
2023         if (key.objectid != btrfs_ino(inode) ||
2024             key.type != BTRFS_EXTENT_DATA_KEY)
2025                 return 0;
2026
2027         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2028
2029         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2030                 return 0;
2031
2032         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2033                 return 0;
2034
2035         if (key.offset == end)
2036                 return 1;
2037         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2038                 return 1;
2039         return 0;
2040 }
2041
2042 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2043                       struct btrfs_path *path, u64 offset, u64 end)
2044 {
2045         struct btrfs_root *root = BTRFS_I(inode)->root;
2046         struct extent_buffer *leaf;
2047         struct btrfs_file_extent_item *fi;
2048         struct extent_map *hole_em;
2049         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2050         struct btrfs_key key;
2051         int ret;
2052
2053         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2054                 goto out;
2055
2056         key.objectid = btrfs_ino(inode);
2057         key.type = BTRFS_EXTENT_DATA_KEY;
2058         key.offset = offset;
2059
2060         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2061         if (ret < 0)
2062                 return ret;
2063         BUG_ON(!ret);
2064
2065         leaf = path->nodes[0];
2066         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2067                 u64 num_bytes;
2068
2069                 path->slots[0]--;
2070                 fi = btrfs_item_ptr(leaf, path->slots[0],
2071                                     struct btrfs_file_extent_item);
2072                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2073                         end - offset;
2074                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2075                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2076                 btrfs_set_file_extent_offset(leaf, fi, 0);
2077                 btrfs_mark_buffer_dirty(leaf);
2078                 goto out;
2079         }
2080
2081         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2082                 u64 num_bytes;
2083
2084                 path->slots[0]++;
2085                 key.offset = offset;
2086                 btrfs_set_item_key_safe(root, path, &key);
2087                 fi = btrfs_item_ptr(leaf, path->slots[0],
2088                                     struct btrfs_file_extent_item);
2089                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2090                         offset;
2091                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2092                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2093                 btrfs_set_file_extent_offset(leaf, fi, 0);
2094                 btrfs_mark_buffer_dirty(leaf);
2095                 goto out;
2096         }
2097         btrfs_release_path(path);
2098
2099         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2100                                        0, 0, end - offset, 0, end - offset,
2101                                        0, 0, 0);
2102         if (ret)
2103                 return ret;
2104
2105 out:
2106         btrfs_release_path(path);
2107
2108         hole_em = alloc_extent_map();
2109         if (!hole_em) {
2110                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2111                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2112                         &BTRFS_I(inode)->runtime_flags);
2113         } else {
2114                 hole_em->start = offset;
2115                 hole_em->len = end - offset;
2116                 hole_em->ram_bytes = hole_em->len;
2117                 hole_em->orig_start = offset;
2118
2119                 hole_em->block_start = EXTENT_MAP_HOLE;
2120                 hole_em->block_len = 0;
2121                 hole_em->orig_block_len = 0;
2122                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2123                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2124                 hole_em->generation = trans->transid;
2125
2126                 do {
2127                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2128                         write_lock(&em_tree->lock);
2129                         ret = add_extent_mapping(em_tree, hole_em, 1);
2130                         write_unlock(&em_tree->lock);
2131                 } while (ret == -EEXIST);
2132                 free_extent_map(hole_em);
2133                 if (ret)
2134                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2135                                 &BTRFS_I(inode)->runtime_flags);
2136         }
2137
2138         return 0;
2139 }
2140
2141 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2142 {
2143         struct btrfs_root *root = BTRFS_I(inode)->root;
2144         struct extent_state *cached_state = NULL;
2145         struct btrfs_path *path;
2146         struct btrfs_block_rsv *rsv;
2147         struct btrfs_trans_handle *trans;
2148         u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2149         u64 lockend = round_down(offset + len,
2150                                  BTRFS_I(inode)->root->sectorsize) - 1;
2151         u64 cur_offset = lockstart;
2152         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2153         u64 drop_end;
2154         int ret = 0;
2155         int err = 0;
2156         int rsv_count;
2157         bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2158                           ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2159         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2160
2161         ret = btrfs_wait_ordered_range(inode, offset, len);
2162         if (ret)
2163                 return ret;
2164
2165         mutex_lock(&inode->i_mutex);
2166         /*
2167          * We needn't truncate any page which is beyond the end of the file
2168          * because we are sure there is no data there.
2169          */
2170         /*
2171          * Only do this if we are in the same page and we aren't doing the
2172          * entire page.
2173          */
2174         if (same_page && len < PAGE_CACHE_SIZE) {
2175                 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2176                         ret = btrfs_truncate_page(inode, offset, len, 0);
2177                 mutex_unlock(&inode->i_mutex);
2178                 return ret;
2179         }
2180
2181         /* zero back part of the first page */
2182         if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2183                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2184                 if (ret) {
2185                         mutex_unlock(&inode->i_mutex);
2186                         return ret;
2187                 }
2188         }
2189
2190         /* zero the front end of the last page */
2191         if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2192                 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2193                 if (ret) {
2194                         mutex_unlock(&inode->i_mutex);
2195                         return ret;
2196                 }
2197         }
2198
2199         if (lockend < lockstart) {
2200                 mutex_unlock(&inode->i_mutex);
2201                 return 0;
2202         }
2203
2204         while (1) {
2205                 struct btrfs_ordered_extent *ordered;
2206
2207                 truncate_pagecache_range(inode, lockstart, lockend);
2208
2209                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2210                                  0, &cached_state);
2211                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2212
2213                 /*
2214                  * We need to make sure we have no ordered extents in this range
2215                  * and nobody raced in and read a page in this range, if we did
2216                  * we need to try again.
2217                  */
2218                 if ((!ordered ||
2219                     (ordered->file_offset + ordered->len <= lockstart ||
2220                      ordered->file_offset > lockend)) &&
2221                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2222                                      lockend, EXTENT_UPTODATE, 0,
2223                                      cached_state)) {
2224                         if (ordered)
2225                                 btrfs_put_ordered_extent(ordered);
2226                         break;
2227                 }
2228                 if (ordered)
2229                         btrfs_put_ordered_extent(ordered);
2230                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2231                                      lockend, &cached_state, GFP_NOFS);
2232                 ret = btrfs_wait_ordered_range(inode, lockstart,
2233                                                lockend - lockstart + 1);
2234                 if (ret) {
2235                         mutex_unlock(&inode->i_mutex);
2236                         return ret;
2237                 }
2238         }
2239
2240         path = btrfs_alloc_path();
2241         if (!path) {
2242                 ret = -ENOMEM;
2243                 goto out;
2244         }
2245
2246         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2247         if (!rsv) {
2248                 ret = -ENOMEM;
2249                 goto out_free;
2250         }
2251         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2252         rsv->failfast = 1;
2253
2254         /*
2255          * 1 - update the inode
2256          * 1 - removing the extents in the range
2257          * 1 - adding the hole extent if no_holes isn't set
2258          */
2259         rsv_count = no_holes ? 2 : 3;
2260         trans = btrfs_start_transaction(root, rsv_count);
2261         if (IS_ERR(trans)) {
2262                 err = PTR_ERR(trans);
2263                 goto out_free;
2264         }
2265
2266         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2267                                       min_size);
2268         BUG_ON(ret);
2269         trans->block_rsv = rsv;
2270
2271         while (cur_offset < lockend) {
2272                 ret = __btrfs_drop_extents(trans, root, inode, path,
2273                                            cur_offset, lockend + 1,
2274                                            &drop_end, 1, 0, 0, NULL);
2275                 if (ret != -ENOSPC)
2276                         break;
2277
2278                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2279
2280                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2281                 if (ret) {
2282                         err = ret;
2283                         break;
2284                 }
2285
2286                 cur_offset = drop_end;
2287
2288                 ret = btrfs_update_inode(trans, root, inode);
2289                 if (ret) {
2290                         err = ret;
2291                         break;
2292                 }
2293
2294                 btrfs_end_transaction(trans, root);
2295                 btrfs_btree_balance_dirty(root);
2296
2297                 trans = btrfs_start_transaction(root, rsv_count);
2298                 if (IS_ERR(trans)) {
2299                         ret = PTR_ERR(trans);
2300                         trans = NULL;
2301                         break;
2302                 }
2303
2304                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2305                                               rsv, min_size);
2306                 BUG_ON(ret);    /* shouldn't happen */
2307                 trans->block_rsv = rsv;
2308         }
2309
2310         if (ret) {
2311                 err = ret;
2312                 goto out_trans;
2313         }
2314
2315         trans->block_rsv = &root->fs_info->trans_block_rsv;
2316         ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2317         if (ret) {
2318                 err = ret;
2319                 goto out_trans;
2320         }
2321
2322 out_trans:
2323         if (!trans)
2324                 goto out_free;
2325
2326         inode_inc_iversion(inode);
2327         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2328
2329         trans->block_rsv = &root->fs_info->trans_block_rsv;
2330         ret = btrfs_update_inode(trans, root, inode);
2331         btrfs_end_transaction(trans, root);
2332         btrfs_btree_balance_dirty(root);
2333 out_free:
2334         btrfs_free_path(path);
2335         btrfs_free_block_rsv(root, rsv);
2336 out:
2337         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2338                              &cached_state, GFP_NOFS);
2339         mutex_unlock(&inode->i_mutex);
2340         if (ret && !err)
2341                 err = ret;
2342         return err;
2343 }
2344
2345 static long btrfs_fallocate(struct file *file, int mode,
2346                             loff_t offset, loff_t len)
2347 {
2348         struct inode *inode = file_inode(file);
2349         struct extent_state *cached_state = NULL;
2350         struct btrfs_root *root = BTRFS_I(inode)->root;
2351         u64 cur_offset;
2352         u64 last_byte;
2353         u64 alloc_start;
2354         u64 alloc_end;
2355         u64 alloc_hint = 0;
2356         u64 locked_end;
2357         struct extent_map *em;
2358         int blocksize = BTRFS_I(inode)->root->sectorsize;
2359         int ret;
2360
2361         alloc_start = round_down(offset, blocksize);
2362         alloc_end = round_up(offset + len, blocksize);
2363
2364         /* Make sure we aren't being give some crap mode */
2365         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2366                 return -EOPNOTSUPP;
2367
2368         if (mode & FALLOC_FL_PUNCH_HOLE)
2369                 return btrfs_punch_hole(inode, offset, len);
2370
2371         /*
2372          * Make sure we have enough space before we do the
2373          * allocation.
2374          */
2375         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2376         if (ret)
2377                 return ret;
2378         if (root->fs_info->quota_enabled) {
2379                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2380                 if (ret)
2381                         goto out_reserve_fail;
2382         }
2383
2384         mutex_lock(&inode->i_mutex);
2385         ret = inode_newsize_ok(inode, alloc_end);
2386         if (ret)
2387                 goto out;
2388
2389         if (alloc_start > inode->i_size) {
2390                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2391                                         alloc_start);
2392                 if (ret)
2393                         goto out;
2394         } else {
2395                 /*
2396                  * If we are fallocating from the end of the file onward we
2397                  * need to zero out the end of the page if i_size lands in the
2398                  * middle of a page.
2399                  */
2400                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2401                 if (ret)
2402                         goto out;
2403         }
2404
2405         /*
2406          * wait for ordered IO before we have any locks.  We'll loop again
2407          * below with the locks held.
2408          */
2409         ret = btrfs_wait_ordered_range(inode, alloc_start,
2410                                        alloc_end - alloc_start);
2411         if (ret)
2412                 goto out;
2413
2414         locked_end = alloc_end - 1;
2415         while (1) {
2416                 struct btrfs_ordered_extent *ordered;
2417
2418                 /* the extent lock is ordered inside the running
2419                  * transaction
2420                  */
2421                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2422                                  locked_end, 0, &cached_state);
2423                 ordered = btrfs_lookup_first_ordered_extent(inode,
2424                                                             alloc_end - 1);
2425                 if (ordered &&
2426                     ordered->file_offset + ordered->len > alloc_start &&
2427                     ordered->file_offset < alloc_end) {
2428                         btrfs_put_ordered_extent(ordered);
2429                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2430                                              alloc_start, locked_end,
2431                                              &cached_state, GFP_NOFS);
2432                         /*
2433                          * we can't wait on the range with the transaction
2434                          * running or with the extent lock held
2435                          */
2436                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2437                                                        alloc_end - alloc_start);
2438                         if (ret)
2439                                 goto out;
2440                 } else {
2441                         if (ordered)
2442                                 btrfs_put_ordered_extent(ordered);
2443                         break;
2444                 }
2445         }
2446
2447         cur_offset = alloc_start;
2448         while (1) {
2449                 u64 actual_end;
2450
2451                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2452                                       alloc_end - cur_offset, 0);
2453                 if (IS_ERR_OR_NULL(em)) {
2454                         if (!em)
2455                                 ret = -ENOMEM;
2456                         else
2457                                 ret = PTR_ERR(em);
2458                         break;
2459                 }
2460                 last_byte = min(extent_map_end(em), alloc_end);
2461                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2462                 last_byte = ALIGN(last_byte, blocksize);
2463
2464                 if (em->block_start == EXTENT_MAP_HOLE ||
2465                     (cur_offset >= inode->i_size &&
2466                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2467                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2468                                                         last_byte - cur_offset,
2469                                                         1 << inode->i_blkbits,
2470                                                         offset + len,
2471                                                         &alloc_hint);
2472
2473                         if (ret < 0) {
2474                                 free_extent_map(em);
2475                                 break;
2476                         }
2477                 } else if (actual_end > inode->i_size &&
2478                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2479                         /*
2480                          * We didn't need to allocate any more space, but we
2481                          * still extended the size of the file so we need to
2482                          * update i_size.
2483                          */
2484                         inode->i_ctime = CURRENT_TIME;
2485                         i_size_write(inode, actual_end);
2486                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2487                 }
2488                 free_extent_map(em);
2489
2490                 cur_offset = last_byte;
2491                 if (cur_offset >= alloc_end) {
2492                         ret = 0;
2493                         break;
2494                 }
2495         }
2496         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2497                              &cached_state, GFP_NOFS);
2498 out:
2499         mutex_unlock(&inode->i_mutex);
2500         if (root->fs_info->quota_enabled)
2501                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2502 out_reserve_fail:
2503         /* Let go of our reservation. */
2504         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2505         return ret;
2506 }
2507
2508 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2509 {
2510         struct btrfs_root *root = BTRFS_I(inode)->root;
2511         struct extent_map *em = NULL;
2512         struct extent_state *cached_state = NULL;
2513         u64 lockstart;
2514         u64 lockend;
2515         u64 start;
2516         u64 len;
2517         int ret = 0;
2518
2519         if (inode->i_size == 0)
2520                 return -ENXIO;
2521
2522         /*
2523          * *offset can be negative, in this case we start finding DATA/HOLE from
2524          * the very start of the file.
2525          */
2526         start = max_t(loff_t, 0, *offset);
2527
2528         lockstart = round_down(start, root->sectorsize);
2529         lockend = round_up(i_size_read(inode), root->sectorsize);
2530         if (lockend <= lockstart)
2531                 lockend = lockstart + root->sectorsize;
2532         lockend--;
2533         len = lockend - lockstart + 1;
2534
2535         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2536                          &cached_state);
2537
2538         while (start < inode->i_size) {
2539                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2540                 if (IS_ERR(em)) {
2541                         ret = PTR_ERR(em);
2542                         em = NULL;
2543                         break;
2544                 }
2545
2546                 if (whence == SEEK_HOLE &&
2547                     (em->block_start == EXTENT_MAP_HOLE ||
2548                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2549                         break;
2550                 else if (whence == SEEK_DATA &&
2551                            (em->block_start != EXTENT_MAP_HOLE &&
2552                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2553                         break;
2554
2555                 start = em->start + em->len;
2556                 free_extent_map(em);
2557                 em = NULL;
2558                 cond_resched();
2559         }
2560         free_extent_map(em);
2561         if (!ret) {
2562                 if (whence == SEEK_DATA && start >= inode->i_size)
2563                         ret = -ENXIO;
2564                 else
2565                         *offset = min_t(loff_t, start, inode->i_size);
2566         }
2567         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2568                              &cached_state, GFP_NOFS);
2569         return ret;
2570 }
2571
2572 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2573 {
2574         struct inode *inode = file->f_mapping->host;
2575         int ret;
2576
2577         mutex_lock(&inode->i_mutex);
2578         switch (whence) {
2579         case SEEK_END:
2580         case SEEK_CUR:
2581                 offset = generic_file_llseek(file, offset, whence);
2582                 goto out;
2583         case SEEK_DATA:
2584         case SEEK_HOLE:
2585                 if (offset >= i_size_read(inode)) {
2586                         mutex_unlock(&inode->i_mutex);
2587                         return -ENXIO;
2588                 }
2589
2590                 ret = find_desired_extent(inode, &offset, whence);
2591                 if (ret) {
2592                         mutex_unlock(&inode->i_mutex);
2593                         return ret;
2594                 }
2595         }
2596
2597         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2598 out:
2599         mutex_unlock(&inode->i_mutex);
2600         return offset;
2601 }
2602
2603 const struct file_operations btrfs_file_operations = {
2604         .llseek         = btrfs_file_llseek,
2605         .read           = do_sync_read,
2606         .write          = do_sync_write,
2607         .aio_read       = generic_file_aio_read,
2608         .splice_read    = generic_file_splice_read,
2609         .aio_write      = btrfs_file_aio_write,
2610         .mmap           = btrfs_file_mmap,
2611         .open           = generic_file_open,
2612         .release        = btrfs_release_file,
2613         .fsync          = btrfs_sync_file,
2614         .fallocate      = btrfs_fallocate,
2615         .unlocked_ioctl = btrfs_ioctl,
2616 #ifdef CONFIG_COMPAT
2617         .compat_ioctl   = btrfs_ioctl,
2618 #endif
2619 };
2620
2621 void btrfs_auto_defrag_exit(void)
2622 {
2623         if (btrfs_inode_defrag_cachep)
2624                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2625 }
2626
2627 int btrfs_auto_defrag_init(void)
2628 {
2629         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2630                                         sizeof(struct inode_defrag), 0,
2631                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2632                                         NULL);
2633         if (!btrfs_inode_defrag_cachep)
2634                 return -ENOMEM;
2635
2636         return 0;
2637 }