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                 if (no_splits)
 595                         goto next;
 596
 597                 if (em->start < start) {
 598                         split->start = em->start;
 599                         split->len = start - em->start;
 600
 601                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 602                                 split->orig_start = em->orig_start;
 603                                 split->block_start = em->block_start;
 604
 605                                 if (compressed)
 606                                         split->block_len = em->block_len;
 607                                 else
 608                                         split->block_len = split->len;
 609                                 split->orig_block_len = max(split->block_len,
 610                                                 em->orig_block_len);
 611                                 split->ram_bytes = em->ram_bytes;
 612                         } else {
 613                                 split->orig_start = split->start;
 614                                 split->block_len = 0;
 615                                 split->block_start = em->block_start;
 616                                 split->orig_block_len = 0;
 617                                 split->ram_bytes = split->len;
 618                         }
 619
 620                         split->generation = gen;
 621                         split->bdev = em->bdev;
 622                         split->flags = flags;
 623                         split->compress_type = em->compress_type;
 624                         replace_extent_mapping(em_tree, em, split, modified);
 625                         free_extent_map(split);
 626                         split = split2;
 627                         split2 = NULL;
 628                 }
 629                 if (testend && em->start + em->len > start + len) {
 630                         u64 diff = start + len - em->start;
 631
 632                         split->start = start + len;
 633                         split->len = em->start + em->len - (start + len);
 634                         split->bdev = em->bdev;
 635                         split->flags = flags;
 636                         split->compress_type = em->compress_type;
 637                         split->generation = gen;
 638
 639                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 640                                 split->orig_block_len = max(em->block_len,
 641                                                     em->orig_block_len);
 642
 643                                 split->ram_bytes = em->ram_bytes;
 644                                 if (compressed) {
 645                                         split->block_len = em->block_len;
 646                                         split->block_start = em->block_start;
 647                                         split->orig_start = em->orig_start;
 648                                 } else {
 649                                         split->block_len = split->len;
 650                                         split->block_start = em->block_start
 651                                                 + diff;
 652                                         split->orig_start = em->orig_start;
 653                                 }
 654                         } else {
 655                                 split->ram_bytes = split->len;
 656                                 split->orig_start = split->start;
 657                                 split->block_len = 0;
 658                                 split->block_start = em->block_start;
 659                                 split->orig_block_len = 0;
 660                         }
 661
 662                         if (extent_map_in_tree(em)) {
 663                                 replace_extent_mapping(em_tree, em, split,
 664                                                        modified);
 665                         } else {
 666                                 ret = add_extent_mapping(em_tree, split,
 667                                                          modified);
 668                                 ASSERT(ret == 0); /* Logic error */
 669                         }
 670                         free_extent_map(split);
 671                         split = NULL;
 672                 }
 673 next:
 674                 if (extent_map_in_tree(em))
 675                         remove_extent_mapping(em_tree, em);
 676                 write_unlock(&em_tree->lock);
 677
 678                 /* once for us */
 679                 free_extent_map(em);
 680                 /* once for the tree*/
 681                 free_extent_map(em);
 682         }
 683         if (split)
 684                 free_extent_map(split);
 685         if (split2)
 686                 free_extent_map(split2);
 687 }
 688
 689 /*
 690  * this is very complex, but the basic idea is to drop all extents
 691  * in the range start - end.  hint_block is filled in with a block number
 692  * that would be a good hint to the block allocator for this file.
 693  *
 694  * If an extent intersects the range but is not entirely inside the range
 695  * it is either truncated or split.  Anything entirely inside the range
 696  * is deleted from the tree.
 697  */
 698 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 699                          struct btrfs_root *root, struct inode *inode,
 700                          struct btrfs_path *path, u64 start, u64 end,
 701                          u64 *drop_end, int drop_cache,
 702                          int replace_extent,
 703                          u32 extent_item_size,
 704                          int *key_inserted)
 705 {
 706         struct extent_buffer *leaf;
 707         struct btrfs_file_extent_item *fi;
 708         struct btrfs_key key;
 709         struct btrfs_key new_key;
 710         u64 ino = btrfs_ino(inode);
 711         u64 search_start = start;
 712         u64 disk_bytenr = 0;
 713         u64 num_bytes = 0;
 714         u64 extent_offset = 0;
 715         u64 extent_end = 0;
 716         int del_nr = 0;
 717         int del_slot = 0;
 718         int extent_type;
 719         int recow;
 720         int ret;
 721         int modify_tree = -1;
 722         int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
 723         int found = 0;
 724         int leafs_visited = 0;
 725
 726         if (drop_cache)
 727                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 728
 729         if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
 730                 modify_tree = 0;
 731
 732         while (1) {
 733                 recow = 0;
 734                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
 735                                                search_start, modify_tree);
 736                 if (ret < 0)
 737                         break;
 738                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 739                         leaf = path->nodes[0];
 740                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 741                         if (key.objectid == ino &&
 742                             key.type == BTRFS_EXTENT_DATA_KEY)
 743                                 path->slots[0]--;
 744                 }
 745                 ret = 0;
 746                 leafs_visited++;
 747 next_slot:
 748                 leaf = path->nodes[0];
 749                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 750                         BUG_ON(del_nr > 0);
 751                         ret = btrfs_next_leaf(root, path);
 752                         if (ret < 0)
 753                                 break;
 754                         if (ret > 0) {
 755                                 ret = 0;
 756                                 break;
 757                         }
 758                         leafs_visited++;
 759                         leaf = path->nodes[0];
 760                         recow = 1;
 761                 }
 762
 763                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 764                 if (key.objectid > ino ||
 765                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 766                         break;
 767
 768                 fi = btrfs_item_ptr(leaf, path->slots[0],
 769                                     struct btrfs_file_extent_item);
 770                 extent_type = btrfs_file_extent_type(leaf, fi);
 771
 772                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 773                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 774                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 775                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 776                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 777                         extent_end = key.offset +
 778                                 btrfs_file_extent_num_bytes(leaf, fi);
 779                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 780                         extent_end = key.offset +
 781                                 btrfs_file_extent_inline_len(leaf,
 782                                                      path->slots[0], fi);
 783                 } else {
 784                         WARN_ON(1);
 785                         extent_end = search_start;
 786                 }
 787
 788                 if (extent_end <= search_start) {
 789                         path->slots[0]++;
 790                         goto next_slot;
 791                 }
 792
 793                 found = 1;
 794                 search_start = max(key.offset, start);
 795                 if (recow || !modify_tree) {
 796                         modify_tree = -1;
 797                         btrfs_release_path(path);
 798                         continue;
 799                 }
 800
 801                 /*
 802                  *     | - range to drop - |
 803                  *  | -------- extent -------- |
 804                  */
 805                 if (start > key.offset && end < extent_end) {
 806                         BUG_ON(del_nr > 0);
 807                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 808
 809                         memcpy(&new_key, &key, sizeof(new_key));
 810                         new_key.offset = start;
 811                         ret = btrfs_duplicate_item(trans, root, path,
 812                                                    &new_key);
 813                         if (ret == -EAGAIN) {
 814                                 btrfs_release_path(path);
 815                                 continue;
 816                         }
 817                         if (ret < 0)
 818                                 break;
 819
 820                         leaf = path->nodes[0];
 821                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 822                                             struct btrfs_file_extent_item);
 823                         btrfs_set_file_extent_num_bytes(leaf, fi,
 824                                                         start - key.offset);
 825
 826                         fi = btrfs_item_ptr(leaf, path->slots[0],
 827                                             struct btrfs_file_extent_item);
 828
 829                         extent_offset += start - key.offset;
 830                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 831                         btrfs_set_file_extent_num_bytes(leaf, fi,
 832                                                         extent_end - start);
 833                         btrfs_mark_buffer_dirty(leaf);
 834
 835                         if (update_refs && disk_bytenr > 0) {
 836                                 ret = btrfs_inc_extent_ref(trans, root,
 837                                                 disk_bytenr, num_bytes, 0,
 838                                                 root->root_key.objectid,
 839                                                 new_key.objectid,
 840                                                 start - extent_offset, 0);
 841                                 BUG_ON(ret); /* -ENOMEM */
 842                         }
 843                         key.offset = start;
 844                 }
 845                 /*
 846                  *  | ---- range to drop ----- |
 847                  *      | -------- extent -------- |
 848                  */
 849                 if (start <= key.offset && end < extent_end) {
 850                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 851
 852                         memcpy(&new_key, &key, sizeof(new_key));
 853                         new_key.offset = end;
 854                         btrfs_set_item_key_safe(root, path, &new_key);
 855
 856                         extent_offset += end - key.offset;
 857                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 858                         btrfs_set_file_extent_num_bytes(leaf, fi,
 859                                                         extent_end - end);
 860                         btrfs_mark_buffer_dirty(leaf);
 861                         if (update_refs && disk_bytenr > 0)
 862                                 inode_sub_bytes(inode, end - key.offset);
 863                         break;
 864                 }
 865
 866                 search_start = extent_end;
 867                 /*
 868                  *       | ---- range to drop ----- |
 869                  *  | -------- extent -------- |
 870                  */
 871                 if (start > key.offset && end >= extent_end) {
 872                         BUG_ON(del_nr > 0);
 873                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 874
 875                         btrfs_set_file_extent_num_bytes(leaf, fi,
 876                                                         start - key.offset);
 877                         btrfs_mark_buffer_dirty(leaf);
 878                         if (update_refs && disk_bytenr > 0)
 879                                 inode_sub_bytes(inode, extent_end - start);
 880                         if (end == extent_end)
 881                                 break;
 882
 883                         path->slots[0]++;
 884                         goto next_slot;
 885                 }
 886
 887                 /*
 888                  *  | ---- range to drop ----- |
 889                  *    | ------ extent ------ |
 890                  */
 891                 if (start <= key.offset && end >= extent_end) {
 892                         if (del_nr == 0) {
 893                                 del_slot = path->slots[0];
 894                                 del_nr = 1;
 895                         } else {
 896                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 897                                 del_nr++;
 898                         }
 899
 900                         if (update_refs &&
 901                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 902                                 inode_sub_bytes(inode,
 903                                                 extent_end - key.offset);
 904                                 extent_end = ALIGN(extent_end,
 905                                                    root->sectorsize);
 906                         } else if (update_refs && disk_bytenr > 0) {
 907                                 ret = btrfs_free_extent(trans, root,
 908                                                 disk_bytenr, num_bytes, 0,
 909                                                 root->root_key.objectid,
 910                                                 key.objectid, key.offset -
 911                                                 extent_offset, 0);
 912                                 BUG_ON(ret); /* -ENOMEM */
 913                                 inode_sub_bytes(inode,
 914                                                 extent_end - key.offset);
 915                         }
 916
 917                         if (end == extent_end)
 918                                 break;
 919
 920                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 921                                 path->slots[0]++;
 922                                 goto next_slot;
 923                         }
 924
 925                         ret = btrfs_del_items(trans, root, path, del_slot,
 926                                               del_nr);
 927                         if (ret) {
 928                                 btrfs_abort_transaction(trans, root, ret);
 929                                 break;
 930                         }
 931
 932                         del_nr = 0;
 933                         del_slot = 0;
 934
 935                         btrfs_release_path(path);
 936                         continue;
 937                 }
 938
 939                 BUG_ON(1);
 940         }
 941
 942         if (!ret && del_nr > 0) {
 943                 /*
 944                  * Set path->slots[0] to first slot, so that after the delete
 945                  * if items are move off from our leaf to its immediate left or
 946                  * right neighbor leafs, we end up with a correct and adjusted
 947                  * path->slots[0] for our insertion (if replace_extent != 0).
 948                  */
 949                 path->slots[0] = del_slot;
 950                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 951                 if (ret)
 952                         btrfs_abort_transaction(trans, root, ret);
 953         }
 954
 955         leaf = path->nodes[0];
 956         /*
 957          * If btrfs_del_items() was called, it might have deleted a leaf, in
 958          * which case it unlocked our path, so check path->locks[0] matches a
 959          * write lock.
 960          */
 961         if (!ret && replace_extent && leafs_visited == 1 &&
 962             (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
 963              path->locks[0] == BTRFS_WRITE_LOCK) &&
 964             btrfs_leaf_free_space(root, leaf) >=
 965             sizeof(struct btrfs_item) + extent_item_size) {
 966
 967                 key.objectid = ino;
 968                 key.type = BTRFS_EXTENT_DATA_KEY;
 969                 key.offset = start;
 970                 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
 971                         struct btrfs_key slot_key;
 972
 973                         btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
 974                         if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
 975                                 path->slots[0]++;
 976                 }
 977                 setup_items_for_insert(root, path, &key,
 978                                        &extent_item_size,
 979                                        extent_item_size,
 980                                        sizeof(struct btrfs_item) +
 981                                        extent_item_size, 1);
 982                 *key_inserted = 1;
 983         }
 984
 985         if (!replace_extent || !(*key_inserted))
 986                 btrfs_release_path(path);
 987         if (drop_end)
 988                 *drop_end = found ? min(end, extent_end) : end;
 989         return ret;
 990 }
 991
 992 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 993                        struct btrfs_root *root, struct inode *inode, u64 start,
 994                        u64 end, int drop_cache)
 995 {
 996         struct btrfs_path *path;
 997         int ret;
 998
 999         path = btrfs_alloc_path();
1000         if (!path)
1001                 return -ENOMEM;
1002         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1003                                    drop_cache, 0, 0, NULL);
1004         btrfs_free_path(path);
1005         return ret;
1006 }
1007
1008 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1009                             u64 objectid, u64 bytenr, u64 orig_offset,
1010                             u64 *start, u64 *end)
1011 {
1012         struct btrfs_file_extent_item *fi;
1013         struct btrfs_key key;
1014         u64 extent_end;
1015
1016         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1017                 return 0;
1018
1019         btrfs_item_key_to_cpu(leaf, &key, slot);
1020         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1021                 return 0;
1022
1023         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1024         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1025             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1026             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1027             btrfs_file_extent_compression(leaf, fi) ||
1028             btrfs_file_extent_encryption(leaf, fi) ||
1029             btrfs_file_extent_other_encoding(leaf, fi))
1030                 return 0;
1031
1032         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1033         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1034                 return 0;
1035
1036         *start = key.offset;
1037         *end = extent_end;
1038         return 1;
1039 }
1040
1041 /*
1042  * Mark extent in the range start - end as written.
1043  *
1044  * This changes extent type from 'pre-allocated' to 'regular'. If only
1045  * part of extent is marked as written, the extent will be split into
1046  * two or three.
1047  */
1048 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1049                               struct inode *inode, u64 start, u64 end)
1050 {
1051         struct btrfs_root *root = BTRFS_I(inode)->root;
1052         struct extent_buffer *leaf;
1053         struct btrfs_path *path;
1054         struct btrfs_file_extent_item *fi;
1055         struct btrfs_key key;
1056         struct btrfs_key new_key;
1057         u64 bytenr;
1058         u64 num_bytes;
1059         u64 extent_end;
1060         u64 orig_offset;
1061         u64 other_start;
1062         u64 other_end;
1063         u64 split;
1064         int del_nr = 0;
1065         int del_slot = 0;
1066         int recow;
1067         int ret;
1068         u64 ino = btrfs_ino(inode);
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073 again:
1074         recow = 0;
1075         split = start;
1076         key.objectid = ino;
1077         key.type = BTRFS_EXTENT_DATA_KEY;
1078         key.offset = split;
1079
1080         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1081         if (ret < 0)
1082                 goto out;
1083         if (ret > 0 && path->slots[0] > 0)
1084                 path->slots[0]--;
1085
1086         leaf = path->nodes[0];
1087         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1088         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1089         fi = btrfs_item_ptr(leaf, path->slots[0],
1090                             struct btrfs_file_extent_item);
1091         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1092                BTRFS_FILE_EXTENT_PREALLOC);
1093         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1094         BUG_ON(key.offset > start || extent_end < end);
1095
1096         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1097         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1098         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1099         memcpy(&new_key, &key, sizeof(new_key));
1100
1101         if (start == key.offset && end < extent_end) {
1102                 other_start = 0;
1103                 other_end = start;
1104                 if (extent_mergeable(leaf, path->slots[0] - 1,
1105                                      ino, bytenr, orig_offset,
1106                                      &other_start, &other_end)) {
1107                         new_key.offset = end;
1108                         btrfs_set_item_key_safe(root, path, &new_key);
1109                         fi = btrfs_item_ptr(leaf, path->slots[0],
1110                                             struct btrfs_file_extent_item);
1111                         btrfs_set_file_extent_generation(leaf, fi,
1112                                                          trans->transid);
1113                         btrfs_set_file_extent_num_bytes(leaf, fi,
1114                                                         extent_end - end);
1115                         btrfs_set_file_extent_offset(leaf, fi,
1116                                                      end - orig_offset);
1117                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1118                                             struct btrfs_file_extent_item);
1119                         btrfs_set_file_extent_generation(leaf, fi,
1120                                                          trans->transid);
1121                         btrfs_set_file_extent_num_bytes(leaf, fi,
1122                                                         end - other_start);
1123                         btrfs_mark_buffer_dirty(leaf);
1124                         goto out;
1125                 }
1126         }
1127
1128         if (start > key.offset && end == extent_end) {
1129                 other_start = end;
1130                 other_end = 0;
1131                 if (extent_mergeable(leaf, path->slots[0] + 1,
1132                                      ino, bytenr, orig_offset,
1133                                      &other_start, &other_end)) {
1134                         fi = btrfs_item_ptr(leaf, path->slots[0],
1135                                             struct btrfs_file_extent_item);
1136                         btrfs_set_file_extent_num_bytes(leaf, fi,
1137                                                         start - key.offset);
1138                         btrfs_set_file_extent_generation(leaf, fi,
1139                                                          trans->transid);
1140                         path->slots[0]++;
1141                         new_key.offset = start;
1142                         btrfs_set_item_key_safe(root, path, &new_key);
1143
1144                         fi = btrfs_item_ptr(leaf, path->slots[0],
1145                                             struct btrfs_file_extent_item);
1146                         btrfs_set_file_extent_generation(leaf, fi,
1147                                                          trans->transid);
1148                         btrfs_set_file_extent_num_bytes(leaf, fi,
1149                                                         other_end - start);
1150                         btrfs_set_file_extent_offset(leaf, fi,
1151                                                      start - orig_offset);
1152                         btrfs_mark_buffer_dirty(leaf);
1153                         goto out;
1154                 }
1155         }
1156
1157         while (start > key.offset || end < extent_end) {
1158                 if (key.offset == start)
1159                         split = end;
1160
1161                 new_key.offset = split;
1162                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1163                 if (ret == -EAGAIN) {
1164                         btrfs_release_path(path);
1165                         goto again;
1166                 }
1167                 if (ret < 0) {
1168                         btrfs_abort_transaction(trans, root, ret);
1169                         goto out;
1170                 }
1171
1172                 leaf = path->nodes[0];
1173                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1174                                     struct btrfs_file_extent_item);
1175                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1176                 btrfs_set_file_extent_num_bytes(leaf, fi,
1177                                                 split - key.offset);
1178
1179                 fi = btrfs_item_ptr(leaf, path->slots[0],
1180                                     struct btrfs_file_extent_item);
1181
1182                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1183                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1184                 btrfs_set_file_extent_num_bytes(leaf, fi,
1185                                                 extent_end - split);
1186                 btrfs_mark_buffer_dirty(leaf);
1187
1188                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1189                                            root->root_key.objectid,
1190                                            ino, orig_offset, 0);
1191                 BUG_ON(ret); /* -ENOMEM */
1192
1193                 if (split == start) {
1194                         key.offset = start;
1195                 } else {
1196                         BUG_ON(start != key.offset);
1197                         path->slots[0]--;
1198                         extent_end = end;
1199                 }
1200                 recow = 1;
1201         }
1202
1203         other_start = end;
1204         other_end = 0;
1205         if (extent_mergeable(leaf, path->slots[0] + 1,
1206                              ino, bytenr, orig_offset,
1207                              &other_start, &other_end)) {
1208                 if (recow) {
1209                         btrfs_release_path(path);
1210                         goto again;
1211                 }
1212                 extent_end = other_end;
1213                 del_slot = path->slots[0] + 1;
1214                 del_nr++;
1215                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1216                                         0, root->root_key.objectid,
1217                                         ino, orig_offset, 0);
1218                 BUG_ON(ret); /* -ENOMEM */
1219         }
1220         other_start = 0;
1221         other_end = start;
1222         if (extent_mergeable(leaf, path->slots[0] - 1,
1223                              ino, bytenr, orig_offset,
1224                              &other_start, &other_end)) {
1225                 if (recow) {
1226                         btrfs_release_path(path);
1227                         goto again;
1228                 }
1229                 key.offset = other_start;
1230                 del_slot = path->slots[0];
1231                 del_nr++;
1232                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1233                                         0, root->root_key.objectid,
1234                                         ino, orig_offset, 0);
1235                 BUG_ON(ret); /* -ENOMEM */
1236         }
1237         if (del_nr == 0) {
1238                 fi = btrfs_item_ptr(leaf, path->slots[0],
1239                            struct btrfs_file_extent_item);
1240                 btrfs_set_file_extent_type(leaf, fi,
1241                                            BTRFS_FILE_EXTENT_REG);
1242                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1243                 btrfs_mark_buffer_dirty(leaf);
1244         } else {
1245                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1246                            struct btrfs_file_extent_item);
1247                 btrfs_set_file_extent_type(leaf, fi,
1248                                            BTRFS_FILE_EXTENT_REG);
1249                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1250                 btrfs_set_file_extent_num_bytes(leaf, fi,
1251                                                 extent_end - key.offset);
1252                 btrfs_mark_buffer_dirty(leaf);
1253
1254                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1255                 if (ret < 0) {
1256                         btrfs_abort_transaction(trans, root, ret);
1257                         goto out;
1258                 }
1259         }
1260 out:
1261         btrfs_free_path(path);
1262         return 0;
1263 }
1264
1265 /*
1266  * on error we return an unlocked page and the error value
1267  * on success we return a locked page and 0
1268  */
1269 static int prepare_uptodate_page(struct page *page, u64 pos,
1270                                  bool force_uptodate)
1271 {
1272         int ret = 0;
1273
1274         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1275             !PageUptodate(page)) {
1276                 ret = btrfs_readpage(NULL, page);
1277                 if (ret)
1278                         return ret;
1279                 lock_page(page);
1280                 if (!PageUptodate(page)) {
1281                         unlock_page(page);
1282                         return -EIO;
1283                 }
1284         }
1285         return 0;
1286 }
1287
1288 /*
1289  * this just gets pages into the page cache and locks them down.
1290  */
1291 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1292                                   size_t num_pages, loff_t pos,
1293                                   size_t write_bytes, bool force_uptodate)
1294 {
1295         int i;
1296         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1297         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1298         int err = 0;
1299         int faili;
1300
1301         for (i = 0; i < num_pages; i++) {
1302                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1303                                                mask | __GFP_WRITE);
1304                 if (!pages[i]) {
1305                         faili = i - 1;
1306                         err = -ENOMEM;
1307                         goto fail;
1308                 }
1309
1310                 if (i == 0)
1311                         err = prepare_uptodate_page(pages[i], pos,
1312                                                     force_uptodate);
1313                 if (i == num_pages - 1)
1314                         err = prepare_uptodate_page(pages[i],
1315                                                     pos + write_bytes, false);
1316                 if (err) {
1317                         page_cache_release(pages[i]);
1318                         faili = i - 1;
1319                         goto fail;
1320                 }
1321                 wait_on_page_writeback(pages[i]);
1322         }
1323
1324         return 0;
1325 fail:
1326         while (faili >= 0) {
1327                 unlock_page(pages[faili]);
1328                 page_cache_release(pages[faili]);
1329                 faili--;
1330         }
1331         return err;
1332
1333 }
1334
1335 /*
1336  * This function locks the extent and properly waits for data=ordered extents
1337  * to finish before allowing the pages to be modified if need.
1338  *
1339  * The return value:
1340  * 1 - the extent is locked
1341  * 0 - the extent is not locked, and everything is OK
1342  * -EAGAIN - need re-prepare the pages
1343  * the other < 0 number - Something wrong happens
1344  */
1345 static noinline int
1346 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1347                                 size_t num_pages, loff_t pos,
1348                                 u64 *lockstart, u64 *lockend,
1349                                 struct extent_state **cached_state)
1350 {
1351         u64 start_pos;
1352         u64 last_pos;
1353         int i;
1354         int ret = 0;
1355
1356         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1357         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1358
1359         if (start_pos < inode->i_size) {
1360                 struct btrfs_ordered_extent *ordered;
1361                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1362                                  start_pos, last_pos, 0, cached_state);
1363                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1364                 if (ordered &&
1365                     ordered->file_offset + ordered->len > start_pos &&
1366                     ordered->file_offset <= last_pos) {
1367                         btrfs_put_ordered_extent(ordered);
1368                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1369                                              start_pos, last_pos,
1370                                              cached_state, GFP_NOFS);
1371                         for (i = 0; i < num_pages; i++) {
1372                                 unlock_page(pages[i]);
1373                                 page_cache_release(pages[i]);
1374                         }
1375                         ret = btrfs_wait_ordered_range(inode, start_pos,
1376                                                 last_pos - start_pos + 1);
1377                         if (ret)
1378                                 return ret;
1379                         else
1380                                 return -EAGAIN;
1381                 }
1382                 if (ordered)
1383                         btrfs_put_ordered_extent(ordered);
1384
1385                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1386                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1387                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1388                                   0, 0, cached_state, GFP_NOFS);
1389                 *lockstart = start_pos;
1390                 *lockend = last_pos;
1391                 ret = 1;
1392         }
1393
1394         for (i = 0; i < num_pages; i++) {
1395                 if (clear_page_dirty_for_io(pages[i]))
1396                         account_page_redirty(pages[i]);
1397                 set_page_extent_mapped(pages[i]);
1398                 WARN_ON(!PageLocked(pages[i]));
1399         }
1400
1401         return ret;
1402 }
1403
1404 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1405                                     size_t *write_bytes)
1406 {
1407         struct btrfs_root *root = BTRFS_I(inode)->root;
1408         struct btrfs_ordered_extent *ordered;
1409         u64 lockstart, lockend;
1410         u64 num_bytes;
1411         int ret;
1412
1413         lockstart = round_down(pos, root->sectorsize);
1414         lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1415
1416         while (1) {
1417                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1418                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1419                                                      lockend - lockstart + 1);
1420                 if (!ordered) {
1421                         break;
1422                 }
1423                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1424                 btrfs_start_ordered_extent(inode, ordered, 1);
1425                 btrfs_put_ordered_extent(ordered);
1426         }
1427
1428         num_bytes = lockend - lockstart + 1;
1429         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1430         if (ret <= 0) {
1431                 ret = 0;
1432         } else {
1433                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1434                                  EXTENT_DIRTY | EXTENT_DELALLOC |
1435                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1436                                  NULL, GFP_NOFS);
1437                 *write_bytes = min_t(size_t, *write_bytes ,
1438                                      num_bytes - pos + lockstart);
1439         }
1440
1441         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1442
1443         return ret;
1444 }
1445
1446 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1447                                                struct iov_iter *i,
1448                                                loff_t pos)
1449 {
1450         struct inode *inode = file_inode(file);
1451         struct btrfs_root *root = BTRFS_I(inode)->root;
1452         struct page **pages = NULL;
1453         struct extent_state *cached_state = NULL;
1454         u64 release_bytes = 0;
1455         u64 lockstart;
1456         u64 lockend;
1457         unsigned long first_index;
1458         size_t num_written = 0;
1459         int nrptrs;
1460         int ret = 0;
1461         bool only_release_metadata = false;
1462         bool force_page_uptodate = false;
1463         bool need_unlock;
1464
1465         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1466                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1467                      (sizeof(struct page *)));
1468         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1469         nrptrs = max(nrptrs, 8);
1470         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1471         if (!pages)
1472                 return -ENOMEM;
1473
1474         first_index = pos >> PAGE_CACHE_SHIFT;
1475
1476         while (iov_iter_count(i) > 0) {
1477                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1478                 size_t write_bytes = min(iov_iter_count(i),
1479                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1480                                          offset);
1481                 size_t num_pages = (write_bytes + offset +
1482                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1483                 size_t reserve_bytes;
1484                 size_t dirty_pages;
1485                 size_t copied;
1486
1487                 WARN_ON(num_pages > nrptrs);
1488
1489                 /*
1490                  * Fault pages before locking them in prepare_pages
1491                  * to avoid recursive lock
1492                  */
1493                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1494                         ret = -EFAULT;
1495                         break;
1496                 }
1497
1498                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1499                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1500                 if (ret == -ENOSPC &&
1501                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1502                                               BTRFS_INODE_PREALLOC))) {
1503                         ret = check_can_nocow(inode, pos, &write_bytes);
1504                         if (ret > 0) {
1505                                 only_release_metadata = true;
1506                                 /*
1507                                  * our prealloc extent may be smaller than
1508                                  * write_bytes, so scale down.
1509                                  */
1510                                 num_pages = (write_bytes + offset +
1511                                              PAGE_CACHE_SIZE - 1) >>
1512                                         PAGE_CACHE_SHIFT;
1513                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1514                                 ret = 0;
1515                         } else {
1516                                 ret = -ENOSPC;
1517                         }
1518                 }
1519
1520                 if (ret)
1521                         break;
1522
1523                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1524                 if (ret) {
1525                         if (!only_release_metadata)
1526                                 btrfs_free_reserved_data_space(inode,
1527                                                                reserve_bytes);
1528                         break;
1529                 }
1530
1531                 release_bytes = reserve_bytes;
1532                 need_unlock = false;
1533 again:
1534                 /*
1535                  * This is going to setup the pages array with the number of
1536                  * pages we want, so we don't really need to worry about the
1537                  * contents of pages from loop to loop
1538                  */
1539                 ret = prepare_pages(inode, pages, num_pages,
1540                                     pos, write_bytes,
1541                                     force_page_uptodate);
1542                 if (ret)
1543                         break;
1544
1545                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1546                                                       pos, &lockstart, &lockend,
1547                                                       &cached_state);
1548                 if (ret < 0) {
1549                         if (ret == -EAGAIN)
1550                                 goto again;
1551                         break;
1552                 } else if (ret > 0) {
1553                         need_unlock = true;
1554                         ret = 0;
1555                 }
1556
1557                 copied = btrfs_copy_from_user(pos, num_pages,
1558                                            write_bytes, pages, i);
1559
1560                 /*
1561                  * if we have trouble faulting in the pages, fall
1562                  * back to one page at a time
1563                  */
1564                 if (copied < write_bytes)
1565                         nrptrs = 1;
1566
1567                 if (copied == 0) {
1568                         force_page_uptodate = true;
1569                         dirty_pages = 0;
1570                 } else {
1571                         force_page_uptodate = false;
1572                         dirty_pages = (copied + offset +
1573                                        PAGE_CACHE_SIZE - 1) >>
1574                                        PAGE_CACHE_SHIFT;
1575                 }
1576
1577                 /*
1578                  * If we had a short copy we need to release the excess delaloc
1579                  * bytes we reserved.  We need to increment outstanding_extents
1580                  * because btrfs_delalloc_release_space will decrement it, but
1581                  * we still have an outstanding extent for the chunk we actually
1582                  * managed to copy.
1583                  */
1584                 if (num_pages > dirty_pages) {
1585                         release_bytes = (num_pages - dirty_pages) <<
1586                                 PAGE_CACHE_SHIFT;
1587                         if (copied > 0) {
1588                                 spin_lock(&BTRFS_I(inode)->lock);
1589                                 BTRFS_I(inode)->outstanding_extents++;
1590                                 spin_unlock(&BTRFS_I(inode)->lock);
1591                         }
1592                         if (only_release_metadata)
1593                                 btrfs_delalloc_release_metadata(inode,
1594                                                                 release_bytes);
1595                         else
1596                                 btrfs_delalloc_release_space(inode,
1597                                                              release_bytes);
1598                 }
1599
1600                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1601
1602                 if (copied > 0)
1603                         ret = btrfs_dirty_pages(root, inode, pages,
1604                                                 dirty_pages, pos, copied,
1605                                                 NULL);
1606                 if (need_unlock)
1607                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1608                                              lockstart, lockend, &cached_state,
1609                                              GFP_NOFS);
1610                 if (ret) {
1611                         btrfs_drop_pages(pages, num_pages);
1612                         break;
1613                 }
1614
1615                 release_bytes = 0;
1616                 if (only_release_metadata && copied > 0) {
1617                         u64 lockstart = round_down(pos, root->sectorsize);
1618                         u64 lockend = lockstart +
1619                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1620
1621                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1622                                        lockend, EXTENT_NORESERVE, NULL,
1623                                        NULL, GFP_NOFS);
1624                         only_release_metadata = false;
1625                 }
1626
1627                 btrfs_drop_pages(pages, num_pages);
1628
1629                 cond_resched();
1630
1631                 balance_dirty_pages_ratelimited(inode->i_mapping);
1632                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1633                         btrfs_btree_balance_dirty(root);
1634
1635                 pos += copied;
1636                 num_written += copied;
1637         }
1638
1639         kfree(pages);
1640
1641         if (release_bytes) {
1642                 if (only_release_metadata)
1643                         btrfs_delalloc_release_metadata(inode, release_bytes);
1644                 else
1645                         btrfs_delalloc_release_space(inode, release_bytes);
1646         }
1647
1648         return num_written ? num_written : ret;
1649 }
1650
1651 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1652                                     const struct iovec *iov,
1653                                     unsigned long nr_segs, loff_t pos,
1654                                     loff_t *ppos, size_t count, size_t ocount)
1655 {
1656         struct file *file = iocb->ki_filp;
1657         struct iov_iter i;
1658         ssize_t written;
1659         ssize_t written_buffered;
1660         loff_t endbyte;
1661         int err;
1662
1663         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1664                                             count, ocount);
1665
1666         if (written < 0 || written == count)
1667                 return written;
1668
1669         pos += written;
1670         count -= written;
1671         iov_iter_init(&i, iov, nr_segs, count, written);
1672         written_buffered = __btrfs_buffered_write(file, &i, pos);
1673         if (written_buffered < 0) {
1674                 err = written_buffered;
1675                 goto out;
1676         }
1677         endbyte = pos + written_buffered - 1;
1678         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1679         if (err)
1680                 goto out;
1681         written += written_buffered;
1682         *ppos = pos + written_buffered;
1683         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1684                                  endbyte >> PAGE_CACHE_SHIFT);
1685 out:
1686         return written ? written : err;
1687 }
1688
1689 static void update_time_for_write(struct inode *inode)
1690 {
1691         struct timespec now;
1692
1693         if (IS_NOCMTIME(inode))
1694                 return;
1695
1696         now = current_fs_time(inode->i_sb);
1697         if (!timespec_equal(&inode->i_mtime, &now))
1698                 inode->i_mtime = now;
1699
1700         if (!timespec_equal(&inode->i_ctime, &now))
1701                 inode->i_ctime = now;
1702
1703         if (IS_I_VERSION(inode))
1704                 inode_inc_iversion(inode);
1705 }
1706
1707 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1708                                     const struct iovec *iov,
1709                                     unsigned long nr_segs, loff_t pos)
1710 {
1711         struct file *file = iocb->ki_filp;
1712         struct inode *inode = file_inode(file);
1713         struct btrfs_root *root = BTRFS_I(inode)->root;
1714         loff_t *ppos = &iocb->ki_pos;
1715         u64 start_pos;
1716         ssize_t num_written = 0;
1717         ssize_t err = 0;
1718         size_t count, ocount;
1719         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1720
1721         mutex_lock(&inode->i_mutex);
1722
1723         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1724         if (err) {
1725                 mutex_unlock(&inode->i_mutex);
1726                 goto out;
1727         }
1728         count = ocount;
1729
1730         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1731         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1732         if (err) {
1733                 mutex_unlock(&inode->i_mutex);
1734                 goto out;
1735         }
1736
1737         if (count == 0) {
1738                 mutex_unlock(&inode->i_mutex);
1739                 goto out;
1740         }
1741
1742         err = file_remove_suid(file);
1743         if (err) {
1744                 mutex_unlock(&inode->i_mutex);
1745                 goto out;
1746         }
1747
1748         /*
1749          * If BTRFS flips readonly due to some impossible error
1750          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1751          * although we have opened a file as writable, we have
1752          * to stop this write operation to ensure FS consistency.
1753          */
1754         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1755                 mutex_unlock(&inode->i_mutex);
1756                 err = -EROFS;
1757                 goto out;
1758         }
1759
1760         /*
1761          * We reserve space for updating the inode when we reserve space for the
1762          * extent we are going to write, so we will enospc out there.  We don't
1763          * need to start yet another transaction to update the inode as we will
1764          * update the inode when we finish writing whatever data we write.
1765          */
1766         update_time_for_write(inode);
1767
1768         start_pos = round_down(pos, root->sectorsize);
1769         if (start_pos > i_size_read(inode)) {
1770                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1771                 if (err) {
1772                         mutex_unlock(&inode->i_mutex);
1773                         goto out;
1774                 }
1775         }
1776
1777         if (sync)
1778                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1779
1780         if (unlikely(file->f_flags & O_DIRECT)) {
1781                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1782                                                    pos, ppos, count, ocount);
1783         } else {
1784                 struct iov_iter i;
1785
1786                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1787
1788                 num_written = __btrfs_buffered_write(file, &i, pos);
1789                 if (num_written > 0)
1790                         *ppos = pos + num_written;
1791         }
1792
1793         mutex_unlock(&inode->i_mutex);
1794
1795         /*
1796          * we want to make sure fsync finds this change
1797          * but we haven't joined a transaction running right now.
1798          *
1799          * Later on, someone is sure to update the inode and get the
1800          * real transid recorded.
1801          *
1802          * We set last_trans now to the fs_info generation + 1,
1803          * this will either be one more than the running transaction
1804          * or the generation used for the next transaction if there isn't
1805          * one running right now.
1806          *
1807          * We also have to set last_sub_trans to the current log transid,
1808          * otherwise subsequent syncs to a file that's been synced in this
1809          * transaction will appear to have already occured.
1810          */
1811         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1812         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1813         if (num_written > 0) {
1814                 err = generic_write_sync(file, pos, num_written);
1815                 if (err < 0 && num_written > 0)
1816                         num_written = err;
1817         }
1818
1819         if (sync)
1820                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1821 out:
1822         current->backing_dev_info = NULL;
1823         return num_written ? num_written : err;
1824 }
1825
1826 int btrfs_release_file(struct inode *inode, struct file *filp)
1827 {
1828         /*
1829          * ordered_data_close is set by settattr when we are about to truncate
1830          * a file from a non-zero size to a zero size.  This tries to
1831          * flush down new bytes that may have been written if the
1832          * application were using truncate to replace a file in place.
1833          */
1834         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1835                                &BTRFS_I(inode)->runtime_flags)) {
1836                 struct btrfs_trans_handle *trans;
1837                 struct btrfs_root *root = BTRFS_I(inode)->root;
1838
1839                 /*
1840                  * We need to block on a committing transaction to keep us from
1841                  * throwing a ordered operation on to the list and causing
1842                  * something like sync to deadlock trying to flush out this
1843                  * inode.
1844                  */
1845                 trans = btrfs_start_transaction(root, 0);
1846                 if (IS_ERR(trans))
1847                         return PTR_ERR(trans);
1848                 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1849                 btrfs_end_transaction(trans, root);
1850                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1851                         filemap_flush(inode->i_mapping);
1852         }
1853         if (filp->private_data)
1854                 btrfs_ioctl_trans_end(filp);
1855         return 0;
1856 }
1857
1858 /*
1859  * fsync call for both files and directories.  This logs the inode into
1860  * the tree log instead of forcing full commits whenever possible.
1861  *
1862  * It needs to call filemap_fdatawait so that all ordered extent updates are
1863  * in the metadata btree are up to date for copying to the log.
1864  *
1865  * It drops the inode mutex before doing the tree log commit.  This is an
1866  * important optimization for directories because holding the mutex prevents
1867  * new operations on the dir while we write to disk.
1868  */
1869 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1870 {
1871         struct dentry *dentry = file->f_path.dentry;
1872         struct inode *inode = dentry->d_inode;
1873         struct btrfs_root *root = BTRFS_I(inode)->root;
1874         struct btrfs_trans_handle *trans;
1875         struct btrfs_log_ctx ctx;
1876         int ret = 0;
1877         bool full_sync = 0;
1878
1879         trace_btrfs_sync_file(file, datasync);
1880
1881         /*
1882          * We write the dirty pages in the range and wait until they complete
1883          * out of the ->i_mutex. If so, we can flush the dirty pages by
1884          * multi-task, and make the performance up.  See
1885          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1886          */
1887         atomic_inc(&BTRFS_I(inode)->sync_writers);
1888         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1889         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1890                              &BTRFS_I(inode)->runtime_flags))
1891                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1892         atomic_dec(&BTRFS_I(inode)->sync_writers);
1893         if (ret)
1894                 return ret;
1895
1896         mutex_lock(&inode->i_mutex);
1897
1898         /*
1899          * We flush the dirty pages again to avoid some dirty pages in the
1900          * range being left.
1901          */
1902         atomic_inc(&root->log_batch);
1903         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1904                              &BTRFS_I(inode)->runtime_flags);
1905         if (full_sync) {
1906                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1907                 if (ret) {
1908                         mutex_unlock(&inode->i_mutex);
1909                         goto out;
1910                 }
1911         }
1912         atomic_inc(&root->log_batch);
1913
1914         /*
1915          * check the transaction that last modified this inode
1916          * and see if its already been committed
1917          */
1918         if (!BTRFS_I(inode)->last_trans) {
1919                 mutex_unlock(&inode->i_mutex);
1920                 goto out;
1921         }
1922
1923         /*
1924          * if the last transaction that changed this file was before
1925          * the current transaction, we can bail out now without any
1926          * syncing
1927          */
1928         smp_mb();
1929         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1930             BTRFS_I(inode)->last_trans <=
1931             root->fs_info->last_trans_committed) {
1932                 BTRFS_I(inode)->last_trans = 0;
1933
1934                 /*
1935                  * We'v had everything committed since the last time we were
1936                  * modified so clear this flag in case it was set for whatever
1937                  * reason, it's no longer relevant.
1938                  */
1939                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1940                           &BTRFS_I(inode)->runtime_flags);
1941                 mutex_unlock(&inode->i_mutex);
1942                 goto out;
1943         }
1944
1945         /*
1946          * ok we haven't committed the transaction yet, lets do a commit
1947          */
1948         if (file->private_data)
1949                 btrfs_ioctl_trans_end(file);
1950
1951         /*
1952          * We use start here because we will need to wait on the IO to complete
1953          * in btrfs_sync_log, which could require joining a transaction (for
1954          * example checking cross references in the nocow path).  If we use join
1955          * here we could get into a situation where we're waiting on IO to
1956          * happen that is blocked on a transaction trying to commit.  With start
1957          * we inc the extwriter counter, so we wait for all extwriters to exit
1958          * before we start blocking join'ers.  This comment is to keep somebody
1959          * from thinking they are super smart and changing this to
1960          * btrfs_join_transaction *cough*Josef*cough*.
1961          */
1962         trans = btrfs_start_transaction(root, 0);
1963         if (IS_ERR(trans)) {
1964                 ret = PTR_ERR(trans);
1965                 mutex_unlock(&inode->i_mutex);
1966                 goto out;
1967         }
1968         trans->sync = true;
1969
1970         btrfs_init_log_ctx(&ctx);
1971
1972         ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1973         if (ret < 0) {
1974                 /* Fallthrough and commit/free transaction. */
1975                 ret = 1;
1976         }
1977
1978         /* we've logged all the items and now have a consistent
1979          * version of the file in the log.  It is possible that
1980          * someone will come in and modify the file, but that's
1981          * fine because the log is consistent on disk, and we
1982          * have references to all of the file's extents
1983          *
1984          * It is possible that someone will come in and log the
1985          * file again, but that will end up using the synchronization
1986          * inside btrfs_sync_log to keep things safe.
1987          */
1988         mutex_unlock(&inode->i_mutex);
1989
1990         if (ret != BTRFS_NO_LOG_SYNC) {
1991                 if (!ret) {
1992                         ret = btrfs_sync_log(trans, root, &ctx);
1993                         if (!ret) {
1994                                 ret = btrfs_end_transaction(trans, root);
1995                                 goto out;
1996                         }
1997                 }
1998                 if (!full_sync) {
1999                         ret = btrfs_wait_ordered_range(inode, start,
2000                                                        end - start + 1);
2001                         if (ret)
2002                                 goto out;
2003                 }
2004                 ret = btrfs_commit_transaction(trans, root);
2005         } else {
2006                 ret = btrfs_end_transaction(trans, root);
2007         }
2008 out:
2009         return ret > 0 ? -EIO : ret;
2010 }
2011
2012 static const struct vm_operations_struct btrfs_file_vm_ops = {
2013         .fault          = filemap_fault,
2014         .page_mkwrite   = btrfs_page_mkwrite,
2015         .remap_pages    = generic_file_remap_pages,
2016 };
2017
2018 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
2019 {
2020         struct address_space *mapping = filp->f_mapping;
2021
2022         if (!mapping->a_ops->readpage)
2023                 return -ENOEXEC;
2024
2025         file_accessed(filp);
2026         vma->vm_ops = &btrfs_file_vm_ops;
2027
2028         return 0;
2029 }
2030
2031 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2032                           int slot, u64 start, u64 end)
2033 {
2034         struct btrfs_file_extent_item *fi;
2035         struct btrfs_key key;
2036
2037         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2038                 return 0;
2039
2040         btrfs_item_key_to_cpu(leaf, &key, slot);
2041         if (key.objectid != btrfs_ino(inode) ||
2042             key.type != BTRFS_EXTENT_DATA_KEY)
2043                 return 0;
2044
2045         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2046
2047         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2048                 return 0;
2049
2050         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2051                 return 0;
2052
2053         if (key.offset == end)
2054                 return 1;
2055         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2056                 return 1;
2057         return 0;
2058 }
2059
2060 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2061                       struct btrfs_path *path, u64 offset, u64 end)
2062 {
2063         struct btrfs_root *root = BTRFS_I(inode)->root;
2064         struct extent_buffer *leaf;
2065         struct btrfs_file_extent_item *fi;
2066         struct extent_map *hole_em;
2067         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2068         struct btrfs_key key;
2069         int ret;
2070
2071         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2072                 goto out;
2073
2074         key.objectid = btrfs_ino(inode);
2075         key.type = BTRFS_EXTENT_DATA_KEY;
2076         key.offset = offset;
2077
2078         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2079         if (ret < 0)
2080                 return ret;
2081         BUG_ON(!ret);
2082
2083         leaf = path->nodes[0];
2084         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2085                 u64 num_bytes;
2086
2087                 path->slots[0]--;
2088                 fi = btrfs_item_ptr(leaf, path->slots[0],
2089                                     struct btrfs_file_extent_item);
2090                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2091                         end - offset;
2092                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2093                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2094                 btrfs_set_file_extent_offset(leaf, fi, 0);
2095                 btrfs_mark_buffer_dirty(leaf);
2096                 goto out;
2097         }
2098
2099         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2100                 u64 num_bytes;
2101
2102                 path->slots[0]++;
2103                 key.offset = offset;
2104                 btrfs_set_item_key_safe(root, path, &key);
2105                 fi = btrfs_item_ptr(leaf, path->slots[0],
2106                                     struct btrfs_file_extent_item);
2107                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2108                         offset;
2109                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2110                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2111                 btrfs_set_file_extent_offset(leaf, fi, 0);
2112                 btrfs_mark_buffer_dirty(leaf);
2113                 goto out;
2114         }
2115         btrfs_release_path(path);
2116
2117         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2118                                        0, 0, end - offset, 0, end - offset,
2119                                        0, 0, 0);
2120         if (ret)
2121                 return ret;
2122
2123 out:
2124         btrfs_release_path(path);
2125
2126         hole_em = alloc_extent_map();
2127         if (!hole_em) {
2128                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2129                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2130                         &BTRFS_I(inode)->runtime_flags);
2131         } else {
2132                 hole_em->start = offset;
2133                 hole_em->len = end - offset;
2134                 hole_em->ram_bytes = hole_em->len;
2135                 hole_em->orig_start = offset;
2136
2137                 hole_em->block_start = EXTENT_MAP_HOLE;
2138                 hole_em->block_len = 0;
2139                 hole_em->orig_block_len = 0;
2140                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2141                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2142                 hole_em->generation = trans->transid;
2143
2144                 do {
2145                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2146                         write_lock(&em_tree->lock);
2147                         ret = add_extent_mapping(em_tree, hole_em, 1);
2148                         write_unlock(&em_tree->lock);
2149                 } while (ret == -EEXIST);
2150                 free_extent_map(hole_em);
2151                 if (ret)
2152                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2153                                 &BTRFS_I(inode)->runtime_flags);
2154         }
2155
2156         return 0;
2157 }
2158
2159 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2160 {
2161         struct btrfs_root *root = BTRFS_I(inode)->root;
2162         struct extent_state *cached_state = NULL;
2163         struct btrfs_path *path;
2164         struct btrfs_block_rsv *rsv;
2165         struct btrfs_trans_handle *trans;
2166         u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2167         u64 lockend = round_down(offset + len,
2168                                  BTRFS_I(inode)->root->sectorsize) - 1;
2169         u64 cur_offset = lockstart;
2170         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2171         u64 drop_end;
2172         int ret = 0;
2173         int err = 0;
2174         int rsv_count;
2175         bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2176                           ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2177         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2178         u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2179
2180         ret = btrfs_wait_ordered_range(inode, offset, len);
2181         if (ret)
2182                 return ret;
2183
2184         mutex_lock(&inode->i_mutex);
2185         /*
2186          * We needn't truncate any page which is beyond the end of the file
2187          * because we are sure there is no data there.
2188          */
2189         /*
2190          * Only do this if we are in the same page and we aren't doing the
2191          * entire page.
2192          */
2193         if (same_page && len < PAGE_CACHE_SIZE) {
2194                 if (offset < ino_size)
2195                         ret = btrfs_truncate_page(inode, offset, len, 0);
2196                 mutex_unlock(&inode->i_mutex);
2197                 return ret;
2198         }
2199
2200         /* zero back part of the first page */
2201         if (offset < ino_size) {
2202                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2203                 if (ret) {
2204                         mutex_unlock(&inode->i_mutex);
2205                         return ret;
2206                 }
2207         }
2208
2209         /* zero the front end of the last page */
2210         if (offset + len < ino_size) {
2211                 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2212                 if (ret) {
2213                         mutex_unlock(&inode->i_mutex);
2214                         return ret;
2215                 }
2216         }
2217
2218         if (lockend < lockstart) {
2219                 mutex_unlock(&inode->i_mutex);
2220                 return 0;
2221         }
2222
2223         while (1) {
2224                 struct btrfs_ordered_extent *ordered;
2225
2226                 truncate_pagecache_range(inode, lockstart, lockend);
2227
2228                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2229                                  0, &cached_state);
2230                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2231
2232                 /*
2233                  * We need to make sure we have no ordered extents in this range
2234                  * and nobody raced in and read a page in this range, if we did
2235                  * we need to try again.
2236                  */
2237                 if ((!ordered ||
2238                     (ordered->file_offset + ordered->len <= lockstart ||
2239                      ordered->file_offset > lockend)) &&
2240                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2241                                      lockend, EXTENT_UPTODATE, 0,
2242                                      cached_state)) {
2243                         if (ordered)
2244                                 btrfs_put_ordered_extent(ordered);
2245                         break;
2246                 }
2247                 if (ordered)
2248                         btrfs_put_ordered_extent(ordered);
2249                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2250                                      lockend, &cached_state, GFP_NOFS);
2251                 ret = btrfs_wait_ordered_range(inode, lockstart,
2252                                                lockend - lockstart + 1);
2253                 if (ret) {
2254                         mutex_unlock(&inode->i_mutex);
2255                         return ret;
2256                 }
2257         }
2258
2259         path = btrfs_alloc_path();
2260         if (!path) {
2261                 ret = -ENOMEM;
2262                 goto out;
2263         }
2264
2265         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2266         if (!rsv) {
2267                 ret = -ENOMEM;
2268                 goto out_free;
2269         }
2270         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2271         rsv->failfast = 1;
2272
2273         /*
2274          * 1 - update the inode
2275          * 1 - removing the extents in the range
2276          * 1 - adding the hole extent if no_holes isn't set
2277          */
2278         rsv_count = no_holes ? 2 : 3;
2279         trans = btrfs_start_transaction(root, rsv_count);
2280         if (IS_ERR(trans)) {
2281                 err = PTR_ERR(trans);
2282                 goto out_free;
2283         }
2284
2285         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2286                                       min_size);
2287         BUG_ON(ret);
2288         trans->block_rsv = rsv;
2289
2290         while (cur_offset < lockend) {
2291                 ret = __btrfs_drop_extents(trans, root, inode, path,
2292                                            cur_offset, lockend + 1,
2293                                            &drop_end, 1, 0, 0, NULL);
2294                 if (ret != -ENOSPC)
2295                         break;
2296
2297                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2298
2299                 if (cur_offset < ino_size) {
2300                         ret = fill_holes(trans, inode, path, cur_offset,
2301                                          drop_end);
2302                         if (ret) {
2303                                 err = ret;
2304                                 break;
2305                         }
2306                 }
2307
2308                 cur_offset = drop_end;
2309
2310                 ret = btrfs_update_inode(trans, root, inode);
2311                 if (ret) {
2312                         err = ret;
2313                         break;
2314                 }
2315
2316                 btrfs_end_transaction(trans, root);
2317                 btrfs_btree_balance_dirty(root);
2318
2319                 trans = btrfs_start_transaction(root, rsv_count);
2320                 if (IS_ERR(trans)) {
2321                         ret = PTR_ERR(trans);
2322                         trans = NULL;
2323                         break;
2324                 }
2325
2326                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2327                                               rsv, min_size);
2328                 BUG_ON(ret);    /* shouldn't happen */
2329                 trans->block_rsv = rsv;
2330         }
2331
2332         if (ret) {
2333                 err = ret;
2334                 goto out_trans;
2335         }
2336
2337         trans->block_rsv = &root->fs_info->trans_block_rsv;
2338         if (cur_offset < ino_size) {
2339                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2340                 if (ret) {
2341                         err = ret;
2342                         goto out_trans;
2343                 }
2344         }
2345
2346 out_trans:
2347         if (!trans)
2348                 goto out_free;
2349
2350         inode_inc_iversion(inode);
2351         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2352
2353         trans->block_rsv = &root->fs_info->trans_block_rsv;
2354         ret = btrfs_update_inode(trans, root, inode);
2355         btrfs_end_transaction(trans, root);
2356         btrfs_btree_balance_dirty(root);
2357 out_free:
2358         btrfs_free_path(path);
2359         btrfs_free_block_rsv(root, rsv);
2360 out:
2361         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2362                              &cached_state, GFP_NOFS);
2363         mutex_unlock(&inode->i_mutex);
2364         if (ret && !err)
2365                 err = ret;
2366         return err;
2367 }
2368
2369 static long btrfs_fallocate(struct file *file, int mode,
2370                             loff_t offset, loff_t len)
2371 {
2372         struct inode *inode = file_inode(file);
2373         struct extent_state *cached_state = NULL;
2374         struct btrfs_root *root = BTRFS_I(inode)->root;
2375         u64 cur_offset;
2376         u64 last_byte;
2377         u64 alloc_start;
2378         u64 alloc_end;
2379         u64 alloc_hint = 0;
2380         u64 locked_end;
2381         struct extent_map *em;
2382         int blocksize = BTRFS_I(inode)->root->sectorsize;
2383         int ret;
2384
2385         alloc_start = round_down(offset, blocksize);
2386         alloc_end = round_up(offset + len, blocksize);
2387
2388         /* Make sure we aren't being give some crap mode */
2389         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2390                 return -EOPNOTSUPP;
2391
2392         if (mode & FALLOC_FL_PUNCH_HOLE)
2393                 return btrfs_punch_hole(inode, offset, len);
2394
2395         /*
2396          * Make sure we have enough space before we do the
2397          * allocation.
2398          */
2399         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2400         if (ret)
2401                 return ret;
2402         if (root->fs_info->quota_enabled) {
2403                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2404                 if (ret)
2405                         goto out_reserve_fail;
2406         }
2407
2408         mutex_lock(&inode->i_mutex);
2409         ret = inode_newsize_ok(inode, alloc_end);
2410         if (ret)
2411                 goto out;
2412
2413         if (alloc_start > inode->i_size) {
2414                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2415                                         alloc_start);
2416                 if (ret)
2417                         goto out;
2418         } else {
2419                 /*
2420                  * If we are fallocating from the end of the file onward we
2421                  * need to zero out the end of the page if i_size lands in the
2422                  * middle of a page.
2423                  */
2424                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2425                 if (ret)
2426                         goto out;
2427         }
2428
2429         /*
2430          * wait for ordered IO before we have any locks.  We'll loop again
2431          * below with the locks held.
2432          */
2433         ret = btrfs_wait_ordered_range(inode, alloc_start,
2434                                        alloc_end - alloc_start);
2435         if (ret)
2436                 goto out;
2437
2438         locked_end = alloc_end - 1;
2439         while (1) {
2440                 struct btrfs_ordered_extent *ordered;
2441
2442                 /* the extent lock is ordered inside the running
2443                  * transaction
2444                  */
2445                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2446                                  locked_end, 0, &cached_state);
2447                 ordered = btrfs_lookup_first_ordered_extent(inode,
2448                                                             alloc_end - 1);
2449                 if (ordered &&
2450                     ordered->file_offset + ordered->len > alloc_start &&
2451                     ordered->file_offset < alloc_end) {
2452                         btrfs_put_ordered_extent(ordered);
2453                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2454                                              alloc_start, locked_end,
2455                                              &cached_state, GFP_NOFS);
2456                         /*
2457                          * we can't wait on the range with the transaction
2458                          * running or with the extent lock held
2459                          */
2460                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2461                                                        alloc_end - alloc_start);
2462                         if (ret)
2463                                 goto out;
2464                 } else {
2465                         if (ordered)
2466                                 btrfs_put_ordered_extent(ordered);
2467                         break;
2468                 }
2469         }
2470
2471         cur_offset = alloc_start;
2472         while (1) {
2473                 u64 actual_end;
2474
2475                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2476                                       alloc_end - cur_offset, 0);
2477                 if (IS_ERR_OR_NULL(em)) {
2478                         if (!em)
2479                                 ret = -ENOMEM;
2480                         else
2481                                 ret = PTR_ERR(em);
2482                         break;
2483                 }
2484                 last_byte = min(extent_map_end(em), alloc_end);
2485                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2486                 last_byte = ALIGN(last_byte, blocksize);
2487
2488                 if (em->block_start == EXTENT_MAP_HOLE ||
2489                     (cur_offset >= inode->i_size &&
2490                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2491                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2492                                                         last_byte - cur_offset,
2493                                                         1 << inode->i_blkbits,
2494                                                         offset + len,
2495                                                         &alloc_hint);
2496
2497                         if (ret < 0) {
2498                                 free_extent_map(em);
2499                                 break;
2500                         }
2501                 } else if (actual_end > inode->i_size &&
2502                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2503                         /*
2504                          * We didn't need to allocate any more space, but we
2505                          * still extended the size of the file so we need to
2506                          * update i_size.
2507                          */
2508                         inode->i_ctime = CURRENT_TIME;
2509                         i_size_write(inode, actual_end);
2510                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2511                 }
2512                 free_extent_map(em);
2513
2514                 cur_offset = last_byte;
2515                 if (cur_offset >= alloc_end) {
2516                         ret = 0;
2517                         break;
2518                 }
2519         }
2520         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2521                              &cached_state, GFP_NOFS);
2522 out:
2523         mutex_unlock(&inode->i_mutex);
2524         if (root->fs_info->quota_enabled)
2525                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2526 out_reserve_fail:
2527         /* Let go of our reservation. */
2528         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2529         return ret;
2530 }
2531
2532 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2533 {
2534         struct btrfs_root *root = BTRFS_I(inode)->root;
2535         struct extent_map *em = NULL;
2536         struct extent_state *cached_state = NULL;
2537         u64 lockstart = *offset;
2538         u64 lockend = i_size_read(inode);
2539         u64 start = *offset;
2540         u64 len = i_size_read(inode);
2541         int ret = 0;
2542
2543         lockend = max_t(u64, root->sectorsize, lockend);
2544         if (lockend <= lockstart)
2545                 lockend = lockstart + root->sectorsize;
2546
2547         lockend--;
2548         len = lockend - lockstart + 1;
2549
2550         len = max_t(u64, len, root->sectorsize);
2551         if (inode->i_size == 0)
2552                 return -ENXIO;
2553
2554         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2555                          &cached_state);
2556
2557         while (start < inode->i_size) {
2558                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2559                 if (IS_ERR(em)) {
2560                         ret = PTR_ERR(em);
2561                         em = NULL;
2562                         break;
2563                 }
2564
2565                 if (whence == SEEK_HOLE &&
2566                     (em->block_start == EXTENT_MAP_HOLE ||
2567                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2568                         break;
2569                 else if (whence == SEEK_DATA &&
2570                            (em->block_start != EXTENT_MAP_HOLE &&
2571                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2572                         break;
2573
2574                 start = em->start + em->len;
2575                 free_extent_map(em);
2576                 em = NULL;
2577                 cond_resched();
2578         }
2579         free_extent_map(em);
2580         if (!ret) {
2581                 if (whence == SEEK_DATA && start >= inode->i_size)
2582                         ret = -ENXIO;
2583                 else
2584                         *offset = min_t(loff_t, start, inode->i_size);
2585         }
2586         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2587                              &cached_state, GFP_NOFS);
2588         return ret;
2589 }
2590
2591 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2592 {
2593         struct inode *inode = file->f_mapping->host;
2594         int ret;
2595
2596         mutex_lock(&inode->i_mutex);
2597         switch (whence) {
2598         case SEEK_END:
2599         case SEEK_CUR:
2600                 offset = generic_file_llseek(file, offset, whence);
2601                 goto out;
2602         case SEEK_DATA:
2603         case SEEK_HOLE:
2604                 if (offset >= i_size_read(inode)) {
2605                         mutex_unlock(&inode->i_mutex);
2606                         return -ENXIO;
2607                 }
2608
2609                 ret = find_desired_extent(inode, &offset, whence);
2610                 if (ret) {
2611                         mutex_unlock(&inode->i_mutex);
2612                         return ret;
2613                 }
2614         }
2615
2616         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2617 out:
2618         mutex_unlock(&inode->i_mutex);
2619         return offset;
2620 }
2621
2622 const struct file_operations btrfs_file_operations = {
2623         .llseek         = btrfs_file_llseek,
2624         .read           = do_sync_read,
2625         .write          = do_sync_write,
2626         .aio_read       = generic_file_aio_read,
2627         .splice_read    = generic_file_splice_read,
2628         .aio_write      = btrfs_file_aio_write,
2629         .mmap           = btrfs_file_mmap,
2630         .open           = generic_file_open,
2631         .release        = btrfs_release_file,
2632         .fsync          = btrfs_sync_file,
2633         .fallocate      = btrfs_fallocate,
2634         .unlocked_ioctl = btrfs_ioctl,
2635 #ifdef CONFIG_COMPAT
2636         .compat_ioctl   = btrfs_ioctl,
2637 #endif
2638 };
2639
2640 void btrfs_auto_defrag_exit(void)
2641 {
2642         if (btrfs_inode_defrag_cachep)
2643                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2644 }
2645
2646 int btrfs_auto_defrag_init(void)
2647 {
2648         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2649                                         sizeof(struct inode_defrag), 0,
2650                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2651                                         NULL);
2652         if (!btrfs_inode_defrag_cachep)
2653                 return -ENOMEM;
2654
2655         return 0;
2656 }