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                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 808                                 ret = -EINVAL;
 809                                 break;
 810                         }
 811
 812                         memcpy(&new_key, &key, sizeof(new_key));
 813                         new_key.offset = start;
 814                         ret = btrfs_duplicate_item(trans, root, path,
 815                                                    &new_key);
 816                         if (ret == -EAGAIN) {
 817                                 btrfs_release_path(path);
 818                                 continue;
 819                         }
 820                         if (ret < 0)
 821                                 break;
 822
 823                         leaf = path->nodes[0];
 824                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 825                                             struct btrfs_file_extent_item);
 826                         btrfs_set_file_extent_num_bytes(leaf, fi,
 827                                                         start - key.offset);
 828
 829                         fi = btrfs_item_ptr(leaf, path->slots[0],
 830                                             struct btrfs_file_extent_item);
 831
 832                         extent_offset += start - key.offset;
 833                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 834                         btrfs_set_file_extent_num_bytes(leaf, fi,
 835                                                         extent_end - start);
 836                         btrfs_mark_buffer_dirty(leaf);
 837
 838                         if (update_refs && disk_bytenr > 0) {
 839                                 ret = btrfs_inc_extent_ref(trans, root,
 840                                                 disk_bytenr, num_bytes, 0,
 841                                                 root->root_key.objectid,
 842                                                 new_key.objectid,
 843                                                 start - extent_offset, 0);
 844                                 BUG_ON(ret); /* -ENOMEM */
 845                         }
 846                         key.offset = start;
 847                 }
 848                 /*
 849                  *  | ---- range to drop ----- |
 850                  *      | -------- extent -------- |
 851                  */
 852                 if (start <= key.offset && end < extent_end) {
 853                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 854                                 ret = -EINVAL;
 855                                 break;
 856                         }
 857
 858                         memcpy(&new_key, &key, sizeof(new_key));
 859                         new_key.offset = end;
 860                         btrfs_set_item_key_safe(root, path, &new_key);
 861
 862                         extent_offset += end - key.offset;
 863                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 864                         btrfs_set_file_extent_num_bytes(leaf, fi,
 865                                                         extent_end - end);
 866                         btrfs_mark_buffer_dirty(leaf);
 867                         if (update_refs && disk_bytenr > 0)
 868                                 inode_sub_bytes(inode, end - key.offset);
 869                         break;
 870                 }
 871
 872                 search_start = extent_end;
 873                 /*
 874                  *       | ---- range to drop ----- |
 875                  *  | -------- extent -------- |
 876                  */
 877                 if (start > key.offset && end >= extent_end) {
 878                         BUG_ON(del_nr > 0);
 879                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 880                                 ret = -EINVAL;
 881                                 break;
 882                         }
 883
 884                         btrfs_set_file_extent_num_bytes(leaf, fi,
 885                                                         start - key.offset);
 886                         btrfs_mark_buffer_dirty(leaf);
 887                         if (update_refs && disk_bytenr > 0)
 888                                 inode_sub_bytes(inode, extent_end - start);
 889                         if (end == extent_end)
 890                                 break;
 891
 892                         path->slots[0]++;
 893                         goto next_slot;
 894                 }
 895
 896                 /*
 897                  *  | ---- range to drop ----- |
 898                  *    | ------ extent ------ |
 899                  */
 900                 if (start <= key.offset && end >= extent_end) {
 901                         if (del_nr == 0) {
 902                                 del_slot = path->slots[0];
 903                                 del_nr = 1;
 904                         } else {
 905                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 906                                 del_nr++;
 907                         }
 908
 909                         if (update_refs &&
 910                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 911                                 inode_sub_bytes(inode,
 912                                                 extent_end - key.offset);
 913                                 extent_end = ALIGN(extent_end,
 914                                                    root->sectorsize);
 915                         } else if (update_refs && disk_bytenr > 0) {
 916                                 ret = btrfs_free_extent(trans, root,
 917                                                 disk_bytenr, num_bytes, 0,
 918                                                 root->root_key.objectid,
 919                                                 key.objectid, key.offset -
 920                                                 extent_offset, 0);
 921                                 BUG_ON(ret); /* -ENOMEM */
 922                                 inode_sub_bytes(inode,
 923                                                 extent_end - key.offset);
 924                         }
 925
 926                         if (end == extent_end)
 927                                 break;
 928
 929                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 930                                 path->slots[0]++;
 931                                 goto next_slot;
 932                         }
 933
 934                         ret = btrfs_del_items(trans, root, path, del_slot,
 935                                               del_nr);
 936                         if (ret) {
 937                                 btrfs_abort_transaction(trans, root, ret);
 938                                 break;
 939                         }
 940
 941                         del_nr = 0;
 942                         del_slot = 0;
 943
 944                         btrfs_release_path(path);
 945                         continue;
 946                 }
 947
 948                 BUG_ON(1);
 949         }
 950
 951         if (!ret && del_nr > 0) {
 952                 /*
 953                  * Set path->slots[0] to first slot, so that after the delete
 954                  * if items are move off from our leaf to its immediate left or
 955                  * right neighbor leafs, we end up with a correct and adjusted
 956                  * path->slots[0] for our insertion (if replace_extent != 0).
 957                  */
 958                 path->slots[0] = del_slot;
 959                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 960                 if (ret)
 961                         btrfs_abort_transaction(trans, root, ret);
 962         }
 963
 964         leaf = path->nodes[0];
 965         /*
 966          * If btrfs_del_items() was called, it might have deleted a leaf, in
 967          * which case it unlocked our path, so check path->locks[0] matches a
 968          * write lock.
 969          */
 970         if (!ret && replace_extent && leafs_visited == 1 &&
 971             (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
 972              path->locks[0] == BTRFS_WRITE_LOCK) &&
 973             btrfs_leaf_free_space(root, leaf) >=
 974             sizeof(struct btrfs_item) + extent_item_size) {
 975
 976                 key.objectid = ino;
 977                 key.type = BTRFS_EXTENT_DATA_KEY;
 978                 key.offset = start;
 979                 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
 980                         struct btrfs_key slot_key;
 981
 982                         btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
 983                         if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
 984                                 path->slots[0]++;
 985                 }
 986                 setup_items_for_insert(root, path, &key,
 987                                        &extent_item_size,
 988                                        extent_item_size,
 989                                        sizeof(struct btrfs_item) +
 990                                        extent_item_size, 1);
 991                 *key_inserted = 1;
 992         }
 993
 994         if (!replace_extent || !(*key_inserted))
 995                 btrfs_release_path(path);
 996         if (drop_end)
 997                 *drop_end = found ? min(end, extent_end) : end;
 998         return ret;
 999 }
1000
1001 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1002                        struct btrfs_root *root, struct inode *inode, u64 start,
1003                        u64 end, int drop_cache)
1004 {
1005         struct btrfs_path *path;
1006         int ret;
1007
1008         path = btrfs_alloc_path();
1009         if (!path)
1010                 return -ENOMEM;
1011         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1012                                    drop_cache, 0, 0, NULL);
1013         btrfs_free_path(path);
1014         return ret;
1015 }
1016
1017 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1018                             u64 objectid, u64 bytenr, u64 orig_offset,
1019                             u64 *start, u64 *end)
1020 {
1021         struct btrfs_file_extent_item *fi;
1022         struct btrfs_key key;
1023         u64 extent_end;
1024
1025         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1026                 return 0;
1027
1028         btrfs_item_key_to_cpu(leaf, &key, slot);
1029         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1030                 return 0;
1031
1032         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1033         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1034             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1035             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1036             btrfs_file_extent_compression(leaf, fi) ||
1037             btrfs_file_extent_encryption(leaf, fi) ||
1038             btrfs_file_extent_other_encoding(leaf, fi))
1039                 return 0;
1040
1041         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1042         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1043                 return 0;
1044
1045         *start = key.offset;
1046         *end = extent_end;
1047         return 1;
1048 }
1049
1050 /*
1051  * Mark extent in the range start - end as written.
1052  *
1053  * This changes extent type from 'pre-allocated' to 'regular'. If only
1054  * part of extent is marked as written, the extent will be split into
1055  * two or three.
1056  */
1057 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1058                               struct inode *inode, u64 start, u64 end)
1059 {
1060         struct btrfs_root *root = BTRFS_I(inode)->root;
1061         struct extent_buffer *leaf;
1062         struct btrfs_path *path;
1063         struct btrfs_file_extent_item *fi;
1064         struct btrfs_key key;
1065         struct btrfs_key new_key;
1066         u64 bytenr;
1067         u64 num_bytes;
1068         u64 extent_end;
1069         u64 orig_offset;
1070         u64 other_start;
1071         u64 other_end;
1072         u64 split;
1073         int del_nr = 0;
1074         int del_slot = 0;
1075         int recow;
1076         int ret;
1077         u64 ino = btrfs_ino(inode);
1078
1079         path = btrfs_alloc_path();
1080         if (!path)
1081                 return -ENOMEM;
1082 again:
1083         recow = 0;
1084         split = start;
1085         key.objectid = ino;
1086         key.type = BTRFS_EXTENT_DATA_KEY;
1087         key.offset = split;
1088
1089         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1090         if (ret < 0)
1091                 goto out;
1092         if (ret > 0 && path->slots[0] > 0)
1093                 path->slots[0]--;
1094
1095         leaf = path->nodes[0];
1096         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1097         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1098         fi = btrfs_item_ptr(leaf, path->slots[0],
1099                             struct btrfs_file_extent_item);
1100         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1101                BTRFS_FILE_EXTENT_PREALLOC);
1102         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1103         BUG_ON(key.offset > start || extent_end < end);
1104
1105         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1106         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1107         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1108         memcpy(&new_key, &key, sizeof(new_key));
1109
1110         if (start == key.offset && end < extent_end) {
1111                 other_start = 0;
1112                 other_end = start;
1113                 if (extent_mergeable(leaf, path->slots[0] - 1,
1114                                      ino, bytenr, orig_offset,
1115                                      &other_start, &other_end)) {
1116                         new_key.offset = end;
1117                         btrfs_set_item_key_safe(root, path, &new_key);
1118                         fi = btrfs_item_ptr(leaf, path->slots[0],
1119                                             struct btrfs_file_extent_item);
1120                         btrfs_set_file_extent_generation(leaf, fi,
1121                                                          trans->transid);
1122                         btrfs_set_file_extent_num_bytes(leaf, fi,
1123                                                         extent_end - end);
1124                         btrfs_set_file_extent_offset(leaf, fi,
1125                                                      end - orig_offset);
1126                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1127                                             struct btrfs_file_extent_item);
1128                         btrfs_set_file_extent_generation(leaf, fi,
1129                                                          trans->transid);
1130                         btrfs_set_file_extent_num_bytes(leaf, fi,
1131                                                         end - other_start);
1132                         btrfs_mark_buffer_dirty(leaf);
1133                         goto out;
1134                 }
1135         }
1136
1137         if (start > key.offset && end == extent_end) {
1138                 other_start = end;
1139                 other_end = 0;
1140                 if (extent_mergeable(leaf, path->slots[0] + 1,
1141                                      ino, bytenr, orig_offset,
1142                                      &other_start, &other_end)) {
1143                         fi = btrfs_item_ptr(leaf, path->slots[0],
1144                                             struct btrfs_file_extent_item);
1145                         btrfs_set_file_extent_num_bytes(leaf, fi,
1146                                                         start - key.offset);
1147                         btrfs_set_file_extent_generation(leaf, fi,
1148                                                          trans->transid);
1149                         path->slots[0]++;
1150                         new_key.offset = start;
1151                         btrfs_set_item_key_safe(root, path, &new_key);
1152
1153                         fi = btrfs_item_ptr(leaf, path->slots[0],
1154                                             struct btrfs_file_extent_item);
1155                         btrfs_set_file_extent_generation(leaf, fi,
1156                                                          trans->transid);
1157                         btrfs_set_file_extent_num_bytes(leaf, fi,
1158                                                         other_end - start);
1159                         btrfs_set_file_extent_offset(leaf, fi,
1160                                                      start - orig_offset);
1161                         btrfs_mark_buffer_dirty(leaf);
1162                         goto out;
1163                 }
1164         }
1165
1166         while (start > key.offset || end < extent_end) {
1167                 if (key.offset == start)
1168                         split = end;
1169
1170                 new_key.offset = split;
1171                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1172                 if (ret == -EAGAIN) {
1173                         btrfs_release_path(path);
1174                         goto again;
1175                 }
1176                 if (ret < 0) {
1177                         btrfs_abort_transaction(trans, root, ret);
1178                         goto out;
1179                 }
1180
1181                 leaf = path->nodes[0];
1182                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1183                                     struct btrfs_file_extent_item);
1184                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1185                 btrfs_set_file_extent_num_bytes(leaf, fi,
1186                                                 split - key.offset);
1187
1188                 fi = btrfs_item_ptr(leaf, path->slots[0],
1189                                     struct btrfs_file_extent_item);
1190
1191                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1192                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1193                 btrfs_set_file_extent_num_bytes(leaf, fi,
1194                                                 extent_end - split);
1195                 btrfs_mark_buffer_dirty(leaf);
1196
1197                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1198                                            root->root_key.objectid,
1199                                            ino, orig_offset, 0);
1200                 BUG_ON(ret); /* -ENOMEM */
1201
1202                 if (split == start) {
1203                         key.offset = start;
1204                 } else {
1205                         BUG_ON(start != key.offset);
1206                         path->slots[0]--;
1207                         extent_end = end;
1208                 }
1209                 recow = 1;
1210         }
1211
1212         other_start = end;
1213         other_end = 0;
1214         if (extent_mergeable(leaf, path->slots[0] + 1,
1215                              ino, bytenr, orig_offset,
1216                              &other_start, &other_end)) {
1217                 if (recow) {
1218                         btrfs_release_path(path);
1219                         goto again;
1220                 }
1221                 extent_end = other_end;
1222                 del_slot = path->slots[0] + 1;
1223                 del_nr++;
1224                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1225                                         0, root->root_key.objectid,
1226                                         ino, orig_offset, 0);
1227                 BUG_ON(ret); /* -ENOMEM */
1228         }
1229         other_start = 0;
1230         other_end = start;
1231         if (extent_mergeable(leaf, path->slots[0] - 1,
1232                              ino, bytenr, orig_offset,
1233                              &other_start, &other_end)) {
1234                 if (recow) {
1235                         btrfs_release_path(path);
1236                         goto again;
1237                 }
1238                 key.offset = other_start;
1239                 del_slot = path->slots[0];
1240                 del_nr++;
1241                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1242                                         0, root->root_key.objectid,
1243                                         ino, orig_offset, 0);
1244                 BUG_ON(ret); /* -ENOMEM */
1245         }
1246         if (del_nr == 0) {
1247                 fi = btrfs_item_ptr(leaf, path->slots[0],
1248                            struct btrfs_file_extent_item);
1249                 btrfs_set_file_extent_type(leaf, fi,
1250                                            BTRFS_FILE_EXTENT_REG);
1251                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1252                 btrfs_mark_buffer_dirty(leaf);
1253         } else {
1254                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1255                            struct btrfs_file_extent_item);
1256                 btrfs_set_file_extent_type(leaf, fi,
1257                                            BTRFS_FILE_EXTENT_REG);
1258                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1259                 btrfs_set_file_extent_num_bytes(leaf, fi,
1260                                                 extent_end - key.offset);
1261                 btrfs_mark_buffer_dirty(leaf);
1262
1263                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1264                 if (ret < 0) {
1265                         btrfs_abort_transaction(trans, root, ret);
1266                         goto out;
1267                 }
1268         }
1269 out:
1270         btrfs_free_path(path);
1271         return 0;
1272 }
1273
1274 /*
1275  * on error we return an unlocked page and the error value
1276  * on success we return a locked page and 0
1277  */
1278 static int prepare_uptodate_page(struct page *page, u64 pos,
1279                                  bool force_uptodate)
1280 {
1281         int ret = 0;
1282
1283         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1284             !PageUptodate(page)) {
1285                 ret = btrfs_readpage(NULL, page);
1286                 if (ret)
1287                         return ret;
1288                 lock_page(page);
1289                 if (!PageUptodate(page)) {
1290                         unlock_page(page);
1291                         return -EIO;
1292                 }
1293         }
1294         return 0;
1295 }
1296
1297 /*
1298  * this just gets pages into the page cache and locks them down.
1299  */
1300 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1301                                   size_t num_pages, loff_t pos,
1302                                   size_t write_bytes, bool force_uptodate)
1303 {
1304         int i;
1305         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1306         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1307         int err = 0;
1308         int faili;
1309
1310         for (i = 0; i < num_pages; i++) {
1311                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1312                                                mask | __GFP_WRITE);
1313                 if (!pages[i]) {
1314                         faili = i - 1;
1315                         err = -ENOMEM;
1316                         goto fail;
1317                 }
1318
1319                 if (i == 0)
1320                         err = prepare_uptodate_page(pages[i], pos,
1321                                                     force_uptodate);
1322                 if (i == num_pages - 1)
1323                         err = prepare_uptodate_page(pages[i],
1324                                                     pos + write_bytes, false);
1325                 if (err) {
1326                         page_cache_release(pages[i]);
1327                         faili = i - 1;
1328                         goto fail;
1329                 }
1330                 wait_on_page_writeback(pages[i]);
1331         }
1332
1333         return 0;
1334 fail:
1335         while (faili >= 0) {
1336                 unlock_page(pages[faili]);
1337                 page_cache_release(pages[faili]);
1338                 faili--;
1339         }
1340         return err;
1341
1342 }
1343
1344 /*
1345  * This function locks the extent and properly waits for data=ordered extents
1346  * to finish before allowing the pages to be modified if need.
1347  *
1348  * The return value:
1349  * 1 - the extent is locked
1350  * 0 - the extent is not locked, and everything is OK
1351  * -EAGAIN - need re-prepare the pages
1352  * the other < 0 number - Something wrong happens
1353  */
1354 static noinline int
1355 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1356                                 size_t num_pages, loff_t pos,
1357                                 u64 *lockstart, u64 *lockend,
1358                                 struct extent_state **cached_state)
1359 {
1360         u64 start_pos;
1361         u64 last_pos;
1362         int i;
1363         int ret = 0;
1364
1365         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1366         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1367
1368         if (start_pos < inode->i_size) {
1369                 struct btrfs_ordered_extent *ordered;
1370                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1371                                  start_pos, last_pos, 0, cached_state);
1372                 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1373                                                      last_pos - start_pos + 1);
1374                 if (ordered &&
1375                     ordered->file_offset + ordered->len > start_pos &&
1376                     ordered->file_offset <= last_pos) {
1377                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1378                                              start_pos, last_pos,
1379                                              cached_state, GFP_NOFS);
1380                         for (i = 0; i < num_pages; i++) {
1381                                 unlock_page(pages[i]);
1382                                 page_cache_release(pages[i]);
1383                         }
1384                         btrfs_start_ordered_extent(inode, ordered, 1);
1385                         btrfs_put_ordered_extent(ordered);
1386                         return -EAGAIN;
1387                 }
1388                 if (ordered)
1389                         btrfs_put_ordered_extent(ordered);
1390
1391                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1392                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1393                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1394                                   0, 0, cached_state, GFP_NOFS);
1395                 *lockstart = start_pos;
1396                 *lockend = last_pos;
1397                 ret = 1;
1398         }
1399
1400         for (i = 0; i < num_pages; i++) {
1401                 if (clear_page_dirty_for_io(pages[i]))
1402                         account_page_redirty(pages[i]);
1403                 set_page_extent_mapped(pages[i]);
1404                 WARN_ON(!PageLocked(pages[i]));
1405         }
1406
1407         return ret;
1408 }
1409
1410 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1411                                     size_t *write_bytes)
1412 {
1413         struct btrfs_root *root = BTRFS_I(inode)->root;
1414         struct btrfs_ordered_extent *ordered;
1415         u64 lockstart, lockend;
1416         u64 num_bytes;
1417         int ret;
1418
1419         ret = btrfs_start_nocow_write(root);
1420         if (!ret)
1421                 return -ENOSPC;
1422
1423         lockstart = round_down(pos, root->sectorsize);
1424         lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1425
1426         while (1) {
1427                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1428                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1429                                                      lockend - lockstart + 1);
1430                 if (!ordered) {
1431                         break;
1432                 }
1433                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1434                 btrfs_start_ordered_extent(inode, ordered, 1);
1435                 btrfs_put_ordered_extent(ordered);
1436         }
1437
1438         num_bytes = lockend - lockstart + 1;
1439         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1440         if (ret <= 0) {
1441                 ret = 0;
1442                 btrfs_end_nocow_write(root);
1443         } else {
1444                 *write_bytes = min_t(size_t, *write_bytes ,
1445                                      num_bytes - pos + lockstart);
1446         }
1447
1448         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1449
1450         return ret;
1451 }
1452
1453 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1454                                                struct iov_iter *i,
1455                                                loff_t pos)
1456 {
1457         struct inode *inode = file_inode(file);
1458         struct btrfs_root *root = BTRFS_I(inode)->root;
1459         struct page **pages = NULL;
1460         struct extent_state *cached_state = NULL;
1461         u64 release_bytes = 0;
1462         u64 lockstart;
1463         u64 lockend;
1464         unsigned long first_index;
1465         size_t num_written = 0;
1466         int nrptrs;
1467         int ret = 0;
1468         bool only_release_metadata = false;
1469         bool force_page_uptodate = false;
1470         bool need_unlock;
1471
1472         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1473                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1474                      (sizeof(struct page *)));
1475         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1476         nrptrs = max(nrptrs, 8);
1477         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1478         if (!pages)
1479                 return -ENOMEM;
1480
1481         first_index = pos >> PAGE_CACHE_SHIFT;
1482
1483         while (iov_iter_count(i) > 0) {
1484                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1485                 size_t write_bytes = min(iov_iter_count(i),
1486                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1487                                          offset);
1488                 size_t num_pages = (write_bytes + offset +
1489                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1490                 size_t reserve_bytes;
1491                 size_t dirty_pages;
1492                 size_t copied;
1493
1494                 WARN_ON(num_pages > nrptrs);
1495
1496                 /*
1497                  * Fault pages before locking them in prepare_pages
1498                  * to avoid recursive lock
1499                  */
1500                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1501                         ret = -EFAULT;
1502                         break;
1503                 }
1504
1505                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1506                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1507                 if (ret == -ENOSPC &&
1508                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1509                                               BTRFS_INODE_PREALLOC))) {
1510                         ret = check_can_nocow(inode, pos, &write_bytes);
1511                         if (ret > 0) {
1512                                 only_release_metadata = true;
1513                                 /*
1514                                  * our prealloc extent may be smaller than
1515                                  * write_bytes, so scale down.
1516                                  */
1517                                 num_pages = (write_bytes + offset +
1518                                              PAGE_CACHE_SIZE - 1) >>
1519                                         PAGE_CACHE_SHIFT;
1520                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1521                                 ret = 0;
1522                         } else {
1523                                 ret = -ENOSPC;
1524                         }
1525                 }
1526
1527                 if (ret)
1528                         break;
1529
1530                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1531                 if (ret) {
1532                         if (!only_release_metadata)
1533                                 btrfs_free_reserved_data_space(inode,
1534                                                                reserve_bytes);
1535                         else
1536                                 btrfs_end_nocow_write(root);
1537                         break;
1538                 }
1539
1540                 release_bytes = reserve_bytes;
1541                 need_unlock = false;
1542 again:
1543                 /*
1544                  * This is going to setup the pages array with the number of
1545                  * pages we want, so we don't really need to worry about the
1546                  * contents of pages from loop to loop
1547                  */
1548                 ret = prepare_pages(inode, pages, num_pages,
1549                                     pos, write_bytes,
1550                                     force_page_uptodate);
1551                 if (ret)
1552                         break;
1553
1554                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1555                                                       pos, &lockstart, &lockend,
1556                                                       &cached_state);
1557                 if (ret < 0) {
1558                         if (ret == -EAGAIN)
1559                                 goto again;
1560                         break;
1561                 } else if (ret > 0) {
1562                         need_unlock = true;
1563                         ret = 0;
1564                 }
1565
1566                 copied = btrfs_copy_from_user(pos, num_pages,
1567                                            write_bytes, pages, i);
1568
1569                 /*
1570                  * if we have trouble faulting in the pages, fall
1571                  * back to one page at a time
1572                  */
1573                 if (copied < write_bytes)
1574                         nrptrs = 1;
1575
1576                 if (copied == 0) {
1577                         force_page_uptodate = true;
1578                         dirty_pages = 0;
1579                 } else {
1580                         force_page_uptodate = false;
1581                         dirty_pages = (copied + offset +
1582                                        PAGE_CACHE_SIZE - 1) >>
1583                                        PAGE_CACHE_SHIFT;
1584                 }
1585
1586                 /*
1587                  * If we had a short copy we need to release the excess delaloc
1588                  * bytes we reserved.  We need to increment outstanding_extents
1589                  * because btrfs_delalloc_release_space will decrement it, but
1590                  * we still have an outstanding extent for the chunk we actually
1591                  * managed to copy.
1592                  */
1593                 if (num_pages > dirty_pages) {
1594                         release_bytes = (num_pages - dirty_pages) <<
1595                                 PAGE_CACHE_SHIFT;
1596                         if (copied > 0) {
1597                                 spin_lock(&BTRFS_I(inode)->lock);
1598                                 BTRFS_I(inode)->outstanding_extents++;
1599                                 spin_unlock(&BTRFS_I(inode)->lock);
1600                         }
1601                         if (only_release_metadata)
1602                                 btrfs_delalloc_release_metadata(inode,
1603                                                                 release_bytes);
1604                         else
1605                                 btrfs_delalloc_release_space(inode,
1606                                                              release_bytes);
1607                 }
1608
1609                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1610
1611                 if (copied > 0)
1612                         ret = btrfs_dirty_pages(root, inode, pages,
1613                                                 dirty_pages, pos, copied,
1614                                                 NULL);
1615                 if (need_unlock)
1616                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1617                                              lockstart, lockend, &cached_state,
1618                                              GFP_NOFS);
1619                 if (ret) {
1620                         btrfs_drop_pages(pages, num_pages);
1621                         break;
1622                 }
1623
1624                 release_bytes = 0;
1625                 if (only_release_metadata)
1626                         btrfs_end_nocow_write(root);
1627
1628                 if (only_release_metadata && copied > 0) {
1629                         u64 lockstart = round_down(pos, root->sectorsize);
1630                         u64 lockend = lockstart +
1631                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1632
1633                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1634                                        lockend, EXTENT_NORESERVE, NULL,
1635                                        NULL, GFP_NOFS);
1636                         only_release_metadata = false;
1637                 }
1638
1639                 btrfs_drop_pages(pages, num_pages);
1640
1641                 cond_resched();
1642
1643                 balance_dirty_pages_ratelimited(inode->i_mapping);
1644                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1645                         btrfs_btree_balance_dirty(root);
1646
1647                 pos += copied;
1648                 num_written += copied;
1649         }
1650
1651         kfree(pages);
1652
1653         if (release_bytes) {
1654                 if (only_release_metadata) {
1655                         btrfs_end_nocow_write(root);
1656                         btrfs_delalloc_release_metadata(inode, release_bytes);
1657                 } else {
1658                         btrfs_delalloc_release_space(inode, release_bytes);
1659                 }
1660         }
1661
1662         return num_written ? num_written : ret;
1663 }
1664
1665 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1666                                     const struct iovec *iov,
1667                                     unsigned long nr_segs, loff_t pos,
1668                                     loff_t *ppos, size_t count, size_t ocount)
1669 {
1670         struct file *file = iocb->ki_filp;
1671         struct iov_iter i;
1672         ssize_t written;
1673         ssize_t written_buffered;
1674         loff_t endbyte;
1675         int err;
1676
1677         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1678                                             count, ocount);
1679
1680         if (written < 0 || written == count)
1681                 return written;
1682
1683         pos += written;
1684         count -= written;
1685         iov_iter_init(&i, iov, nr_segs, count, written);
1686         written_buffered = __btrfs_buffered_write(file, &i, pos);
1687         if (written_buffered < 0) {
1688                 err = written_buffered;
1689                 goto out;
1690         }
1691         endbyte = pos + written_buffered - 1;
1692         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1693         if (err)
1694                 goto out;
1695         written += written_buffered;
1696         *ppos = pos + written_buffered;
1697         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1698                                  endbyte >> PAGE_CACHE_SHIFT);
1699 out:
1700         return written ? written : err;
1701 }
1702
1703 static void update_time_for_write(struct inode *inode)
1704 {
1705         struct timespec now;
1706
1707         if (IS_NOCMTIME(inode))
1708                 return;
1709
1710         now = current_fs_time(inode->i_sb);
1711         if (!timespec_equal(&inode->i_mtime, &now))
1712                 inode->i_mtime = now;
1713
1714         if (!timespec_equal(&inode->i_ctime, &now))
1715                 inode->i_ctime = now;
1716
1717         if (IS_I_VERSION(inode))
1718                 inode_inc_iversion(inode);
1719 }
1720
1721 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1722                                     const struct iovec *iov,
1723                                     unsigned long nr_segs, loff_t pos)
1724 {
1725         struct file *file = iocb->ki_filp;
1726         struct inode *inode = file_inode(file);
1727         struct btrfs_root *root = BTRFS_I(inode)->root;
1728         loff_t *ppos = &iocb->ki_pos;
1729         u64 start_pos;
1730         u64 end_pos;
1731         ssize_t num_written = 0;
1732         ssize_t err = 0;
1733         size_t count, ocount;
1734         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1735
1736         mutex_lock(&inode->i_mutex);
1737
1738         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1739         if (err) {
1740                 mutex_unlock(&inode->i_mutex);
1741                 goto out;
1742         }
1743         count = ocount;
1744
1745         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1746         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1747         if (err) {
1748                 mutex_unlock(&inode->i_mutex);
1749                 goto out;
1750         }
1751
1752         if (count == 0) {
1753                 mutex_unlock(&inode->i_mutex);
1754                 goto out;
1755         }
1756
1757         err = file_remove_suid(file);
1758         if (err) {
1759                 mutex_unlock(&inode->i_mutex);
1760                 goto out;
1761         }
1762
1763         /*
1764          * If BTRFS flips readonly due to some impossible error
1765          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1766          * although we have opened a file as writable, we have
1767          * to stop this write operation to ensure FS consistency.
1768          */
1769         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1770                 mutex_unlock(&inode->i_mutex);
1771                 err = -EROFS;
1772                 goto out;
1773         }
1774
1775         /*
1776          * We reserve space for updating the inode when we reserve space for the
1777          * extent we are going to write, so we will enospc out there.  We don't
1778          * need to start yet another transaction to update the inode as we will
1779          * update the inode when we finish writing whatever data we write.
1780          */
1781         update_time_for_write(inode);
1782
1783         start_pos = round_down(pos, root->sectorsize);
1784         if (start_pos > i_size_read(inode)) {
1785                 /* Expand hole size to cover write data, preventing empty gap */
1786                 end_pos = round_up(pos + iov->iov_len, root->sectorsize);
1787                 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1788                 if (err) {
1789                         mutex_unlock(&inode->i_mutex);
1790                         goto out;
1791                 }
1792         }
1793
1794         if (sync)
1795                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1796
1797         if (unlikely(file->f_flags & O_DIRECT)) {
1798                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1799                                                    pos, ppos, count, ocount);
1800         } else {
1801                 struct iov_iter i;
1802
1803                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1804
1805                 num_written = __btrfs_buffered_write(file, &i, pos);
1806                 if (num_written > 0)
1807                         *ppos = pos + num_written;
1808         }
1809
1810         mutex_unlock(&inode->i_mutex);
1811
1812         /*
1813          * we want to make sure fsync finds this change
1814          * but we haven't joined a transaction running right now.
1815          *
1816          * Later on, someone is sure to update the inode and get the
1817          * real transid recorded.
1818          *
1819          * We set last_trans now to the fs_info generation + 1,
1820          * this will either be one more than the running transaction
1821          * or the generation used for the next transaction if there isn't
1822          * one running right now.
1823          *
1824          * We also have to set last_sub_trans to the current log transid,
1825          * otherwise subsequent syncs to a file that's been synced in this
1826          * transaction will appear to have already occured.
1827          */
1828         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1829         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1830         if (num_written > 0) {
1831                 err = generic_write_sync(file, pos, num_written);
1832                 if (err < 0)
1833                         num_written = err;
1834         }
1835
1836         if (sync)
1837                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1838 out:
1839         current->backing_dev_info = NULL;
1840         return num_written ? num_written : err;
1841 }
1842
1843 int btrfs_release_file(struct inode *inode, struct file *filp)
1844 {
1845         /*
1846          * ordered_data_close is set by settattr when we are about to truncate
1847          * a file from a non-zero size to a zero size.  This tries to
1848          * flush down new bytes that may have been written if the
1849          * application were using truncate to replace a file in place.
1850          */
1851         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1852                                &BTRFS_I(inode)->runtime_flags)) {
1853                 struct btrfs_trans_handle *trans;
1854                 struct btrfs_root *root = BTRFS_I(inode)->root;
1855
1856                 /*
1857                  * We need to block on a committing transaction to keep us from
1858                  * throwing a ordered operation on to the list and causing
1859                  * something like sync to deadlock trying to flush out this
1860                  * inode.
1861                  */
1862                 trans = btrfs_start_transaction(root, 0);
1863                 if (IS_ERR(trans))
1864                         return PTR_ERR(trans);
1865                 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1866                 btrfs_end_transaction(trans, root);
1867                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1868                         filemap_flush(inode->i_mapping);
1869         }
1870         if (filp->private_data)
1871                 btrfs_ioctl_trans_end(filp);
1872         return 0;
1873 }
1874
1875 /*
1876  * fsync call for both files and directories.  This logs the inode into
1877  * the tree log instead of forcing full commits whenever possible.
1878  *
1879  * It needs to call filemap_fdatawait so that all ordered extent updates are
1880  * in the metadata btree are up to date for copying to the log.
1881  *
1882  * It drops the inode mutex before doing the tree log commit.  This is an
1883  * important optimization for directories because holding the mutex prevents
1884  * new operations on the dir while we write to disk.
1885  */
1886 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1887 {
1888         struct dentry *dentry = file->f_path.dentry;
1889         struct inode *inode = dentry->d_inode;
1890         struct btrfs_root *root = BTRFS_I(inode)->root;
1891         struct btrfs_trans_handle *trans;
1892         struct btrfs_log_ctx ctx;
1893         int ret = 0;
1894         bool full_sync = 0;
1895
1896         trace_btrfs_sync_file(file, datasync);
1897
1898         /*
1899          * We write the dirty pages in the range and wait until they complete
1900          * out of the ->i_mutex. If so, we can flush the dirty pages by
1901          * multi-task, and make the performance up.  See
1902          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1903          */
1904         atomic_inc(&BTRFS_I(inode)->sync_writers);
1905         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1906         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1907                              &BTRFS_I(inode)->runtime_flags))
1908                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1909         atomic_dec(&BTRFS_I(inode)->sync_writers);
1910         if (ret)
1911                 return ret;
1912
1913         mutex_lock(&inode->i_mutex);
1914
1915         /*
1916          * We flush the dirty pages again to avoid some dirty pages in the
1917          * range being left.
1918          */
1919         atomic_inc(&root->log_batch);
1920         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1921                              &BTRFS_I(inode)->runtime_flags);
1922         if (full_sync) {
1923                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1924                 if (ret) {
1925                         mutex_unlock(&inode->i_mutex);
1926                         goto out;
1927                 }
1928         }
1929         atomic_inc(&root->log_batch);
1930
1931         /*
1932          * check the transaction that last modified this inode
1933          * and see if its already been committed
1934          */
1935         if (!BTRFS_I(inode)->last_trans) {
1936                 mutex_unlock(&inode->i_mutex);
1937                 goto out;
1938         }
1939
1940         /*
1941          * if the last transaction that changed this file was before
1942          * the current transaction, we can bail out now without any
1943          * syncing
1944          */
1945         smp_mb();
1946         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1947             BTRFS_I(inode)->last_trans <=
1948             root->fs_info->last_trans_committed) {
1949                 BTRFS_I(inode)->last_trans = 0;
1950
1951                 /*
1952                  * We'v had everything committed since the last time we were
1953                  * modified so clear this flag in case it was set for whatever
1954                  * reason, it's no longer relevant.
1955                  */
1956                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1957                           &BTRFS_I(inode)->runtime_flags);
1958                 mutex_unlock(&inode->i_mutex);
1959                 goto out;
1960         }
1961
1962         /*
1963          * ok we haven't committed the transaction yet, lets do a commit
1964          */
1965         if (file->private_data)
1966                 btrfs_ioctl_trans_end(file);
1967
1968         /*
1969          * We use start here because we will need to wait on the IO to complete
1970          * in btrfs_sync_log, which could require joining a transaction (for
1971          * example checking cross references in the nocow path).  If we use join
1972          * here we could get into a situation where we're waiting on IO to
1973          * happen that is blocked on a transaction trying to commit.  With start
1974          * we inc the extwriter counter, so we wait for all extwriters to exit
1975          * before we start blocking join'ers.  This comment is to keep somebody
1976          * from thinking they are super smart and changing this to
1977          * btrfs_join_transaction *cough*Josef*cough*.
1978          */
1979         trans = btrfs_start_transaction(root, 0);
1980         if (IS_ERR(trans)) {
1981                 ret = PTR_ERR(trans);
1982                 mutex_unlock(&inode->i_mutex);
1983                 goto out;
1984         }
1985         trans->sync = true;
1986
1987         btrfs_init_log_ctx(&ctx);
1988
1989         ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1990         if (ret < 0) {
1991                 /* Fallthrough and commit/free transaction. */
1992                 ret = 1;
1993         }
1994
1995         /* we've logged all the items and now have a consistent
1996          * version of the file in the log.  It is possible that
1997          * someone will come in and modify the file, but that's
1998          * fine because the log is consistent on disk, and we
1999          * have references to all of the file's extents
2000          *
2001          * It is possible that someone will come in and log the
2002          * file again, but that will end up using the synchronization
2003          * inside btrfs_sync_log to keep things safe.
2004          */
2005         mutex_unlock(&inode->i_mutex);
2006
2007         if (ret != BTRFS_NO_LOG_SYNC) {
2008                 if (!ret) {
2009                         ret = btrfs_sync_log(trans, root, &ctx);
2010                         if (!ret) {
2011                                 ret = btrfs_end_transaction(trans, root);
2012                                 goto out;
2013                         }
2014                 }
2015                 if (!full_sync) {
2016                         ret = btrfs_wait_ordered_range(inode, start,
2017                                                        end - start + 1);
2018                         if (ret)
2019                                 goto out;
2020                 }
2021                 ret = btrfs_commit_transaction(trans, root);
2022         } else {
2023                 ret = btrfs_end_transaction(trans, root);
2024         }
2025 out:
2026         return ret > 0 ? -EIO : ret;
2027 }
2028
2029 static const struct vm_operations_struct btrfs_file_vm_ops = {
2030         .fault          = filemap_fault,
2031         .map_pages      = filemap_map_pages,
2032         .page_mkwrite   = btrfs_page_mkwrite,
2033         .remap_pages    = generic_file_remap_pages,
2034 };
2035
2036 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
2037 {
2038         struct address_space *mapping = filp->f_mapping;
2039
2040         if (!mapping->a_ops->readpage)
2041                 return -ENOEXEC;
2042
2043         file_accessed(filp);
2044         vma->vm_ops = &btrfs_file_vm_ops;
2045
2046         return 0;
2047 }
2048
2049 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2050                           int slot, u64 start, u64 end)
2051 {
2052         struct btrfs_file_extent_item *fi;
2053         struct btrfs_key key;
2054
2055         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2056                 return 0;
2057
2058         btrfs_item_key_to_cpu(leaf, &key, slot);
2059         if (key.objectid != btrfs_ino(inode) ||
2060             key.type != BTRFS_EXTENT_DATA_KEY)
2061                 return 0;
2062
2063         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2064
2065         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2066                 return 0;
2067
2068         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2069                 return 0;
2070
2071         if (key.offset == end)
2072                 return 1;
2073         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2074                 return 1;
2075         return 0;
2076 }
2077
2078 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2079                       struct btrfs_path *path, u64 offset, u64 end)
2080 {
2081         struct btrfs_root *root = BTRFS_I(inode)->root;
2082         struct extent_buffer *leaf;
2083         struct btrfs_file_extent_item *fi;
2084         struct extent_map *hole_em;
2085         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2086         struct btrfs_key key;
2087         int ret;
2088
2089         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2090                 goto out;
2091
2092         key.objectid = btrfs_ino(inode);
2093         key.type = BTRFS_EXTENT_DATA_KEY;
2094         key.offset = offset;
2095
2096         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2097         if (ret < 0)
2098                 return ret;
2099         BUG_ON(!ret);
2100
2101         leaf = path->nodes[0];
2102         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2103                 u64 num_bytes;
2104
2105                 path->slots[0]--;
2106                 fi = btrfs_item_ptr(leaf, path->slots[0],
2107                                     struct btrfs_file_extent_item);
2108                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2109                         end - offset;
2110                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2111                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2112                 btrfs_set_file_extent_offset(leaf, fi, 0);
2113                 btrfs_mark_buffer_dirty(leaf);
2114                 goto out;
2115         }
2116
2117         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2118                 u64 num_bytes;
2119
2120                 path->slots[0]++;
2121                 key.offset = offset;
2122                 btrfs_set_item_key_safe(root, path, &key);
2123                 fi = btrfs_item_ptr(leaf, path->slots[0],
2124                                     struct btrfs_file_extent_item);
2125                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2126                         offset;
2127                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2128                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2129                 btrfs_set_file_extent_offset(leaf, fi, 0);
2130                 btrfs_mark_buffer_dirty(leaf);
2131                 goto out;
2132         }
2133         btrfs_release_path(path);
2134
2135         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2136                                        0, 0, end - offset, 0, end - offset,
2137                                        0, 0, 0);
2138         if (ret)
2139                 return ret;
2140
2141 out:
2142         btrfs_release_path(path);
2143
2144         hole_em = alloc_extent_map();
2145         if (!hole_em) {
2146                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2147                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2148                         &BTRFS_I(inode)->runtime_flags);
2149         } else {
2150                 hole_em->start = offset;
2151                 hole_em->len = end - offset;
2152                 hole_em->ram_bytes = hole_em->len;
2153                 hole_em->orig_start = offset;
2154
2155                 hole_em->block_start = EXTENT_MAP_HOLE;
2156                 hole_em->block_len = 0;
2157                 hole_em->orig_block_len = 0;
2158                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2159                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2160                 hole_em->generation = trans->transid;
2161
2162                 do {
2163                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2164                         write_lock(&em_tree->lock);
2165                         ret = add_extent_mapping(em_tree, hole_em, 1);
2166                         write_unlock(&em_tree->lock);
2167                 } while (ret == -EEXIST);
2168                 free_extent_map(hole_em);
2169                 if (ret)
2170                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2171                                 &BTRFS_I(inode)->runtime_flags);
2172         }
2173
2174         return 0;
2175 }
2176
2177 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2178 {
2179         struct btrfs_root *root = BTRFS_I(inode)->root;
2180         struct extent_state *cached_state = NULL;
2181         struct btrfs_path *path;
2182         struct btrfs_block_rsv *rsv;
2183         struct btrfs_trans_handle *trans;
2184         u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2185         u64 lockend = round_down(offset + len,
2186                                  BTRFS_I(inode)->root->sectorsize) - 1;
2187         u64 cur_offset = lockstart;
2188         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2189         u64 drop_end;
2190         int ret = 0;
2191         int err = 0;
2192         int rsv_count;
2193         bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2194                           ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2195         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2196         u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2197
2198         ret = btrfs_wait_ordered_range(inode, offset, len);
2199         if (ret)
2200                 return ret;
2201
2202         mutex_lock(&inode->i_mutex);
2203         /*
2204          * We needn't truncate any page which is beyond the end of the file
2205          * because we are sure there is no data there.
2206          */
2207         /*
2208          * Only do this if we are in the same page and we aren't doing the
2209          * entire page.
2210          */
2211         if (same_page && len < PAGE_CACHE_SIZE) {
2212                 if (offset < ino_size)
2213                         ret = btrfs_truncate_page(inode, offset, len, 0);
2214                 mutex_unlock(&inode->i_mutex);
2215                 return ret;
2216         }
2217
2218         /* zero back part of the first page */
2219         if (offset < ino_size) {
2220                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2221                 if (ret) {
2222                         mutex_unlock(&inode->i_mutex);
2223                         return ret;
2224                 }
2225         }
2226
2227         /* zero the front end of the last page */
2228         if (offset + len < ino_size) {
2229                 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2230                 if (ret) {
2231                         mutex_unlock(&inode->i_mutex);
2232                         return ret;
2233                 }
2234         }
2235
2236         if (lockend < lockstart) {
2237                 mutex_unlock(&inode->i_mutex);
2238                 return 0;
2239         }
2240
2241         while (1) {
2242                 struct btrfs_ordered_extent *ordered;
2243
2244                 truncate_pagecache_range(inode, lockstart, lockend);
2245
2246                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2247                                  0, &cached_state);
2248                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2249
2250                 /*
2251                  * We need to make sure we have no ordered extents in this range
2252                  * and nobody raced in and read a page in this range, if we did
2253                  * we need to try again.
2254                  */
2255                 if ((!ordered ||
2256                     (ordered->file_offset + ordered->len <= lockstart ||
2257                      ordered->file_offset > lockend)) &&
2258                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2259                                      lockend, EXTENT_UPTODATE, 0,
2260                                      cached_state)) {
2261                         if (ordered)
2262                                 btrfs_put_ordered_extent(ordered);
2263                         break;
2264                 }
2265                 if (ordered)
2266                         btrfs_put_ordered_extent(ordered);
2267                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2268                                      lockend, &cached_state, GFP_NOFS);
2269                 ret = btrfs_wait_ordered_range(inode, lockstart,
2270                                                lockend - lockstart + 1);
2271                 if (ret) {
2272                         mutex_unlock(&inode->i_mutex);
2273                         return ret;
2274                 }
2275         }
2276
2277         path = btrfs_alloc_path();
2278         if (!path) {
2279                 ret = -ENOMEM;
2280                 goto out;
2281         }
2282
2283         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2284         if (!rsv) {
2285                 ret = -ENOMEM;
2286                 goto out_free;
2287         }
2288         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2289         rsv->failfast = 1;
2290
2291         /*
2292          * 1 - update the inode
2293          * 1 - removing the extents in the range
2294          * 1 - adding the hole extent if no_holes isn't set
2295          */
2296         rsv_count = no_holes ? 2 : 3;
2297         trans = btrfs_start_transaction(root, rsv_count);
2298         if (IS_ERR(trans)) {
2299                 err = PTR_ERR(trans);
2300                 goto out_free;
2301         }
2302
2303         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2304                                       min_size);
2305         BUG_ON(ret);
2306         trans->block_rsv = rsv;
2307
2308         while (cur_offset < lockend) {
2309                 ret = __btrfs_drop_extents(trans, root, inode, path,
2310                                            cur_offset, lockend + 1,
2311                                            &drop_end, 1, 0, 0, NULL);
2312                 if (ret != -ENOSPC)
2313                         break;
2314
2315                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2316
2317                 if (cur_offset < ino_size) {
2318                         ret = fill_holes(trans, inode, path, cur_offset,
2319                                          drop_end);
2320                         if (ret) {
2321                                 err = ret;
2322                                 break;
2323                         }
2324                 }
2325
2326                 cur_offset = drop_end;
2327
2328                 ret = btrfs_update_inode(trans, root, inode);
2329                 if (ret) {
2330                         err = ret;
2331                         break;
2332                 }
2333
2334                 btrfs_end_transaction(trans, root);
2335                 btrfs_btree_balance_dirty(root);
2336
2337                 trans = btrfs_start_transaction(root, rsv_count);
2338                 if (IS_ERR(trans)) {
2339                         ret = PTR_ERR(trans);
2340                         trans = NULL;
2341                         break;
2342                 }
2343
2344                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2345                                               rsv, min_size);
2346                 BUG_ON(ret);    /* shouldn't happen */
2347                 trans->block_rsv = rsv;
2348         }
2349
2350         if (ret) {
2351                 err = ret;
2352                 goto out_trans;
2353         }
2354
2355         trans->block_rsv = &root->fs_info->trans_block_rsv;
2356         if (cur_offset < ino_size) {
2357                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2358                 if (ret) {
2359                         err = ret;
2360                         goto out_trans;
2361                 }
2362         }
2363
2364 out_trans:
2365         if (!trans)
2366                 goto out_free;
2367
2368         inode_inc_iversion(inode);
2369         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2370
2371         trans->block_rsv = &root->fs_info->trans_block_rsv;
2372         ret = btrfs_update_inode(trans, root, inode);
2373         btrfs_end_transaction(trans, root);
2374         btrfs_btree_balance_dirty(root);
2375 out_free:
2376         btrfs_free_path(path);
2377         btrfs_free_block_rsv(root, rsv);
2378 out:
2379         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2380                              &cached_state, GFP_NOFS);
2381         mutex_unlock(&inode->i_mutex);
2382         if (ret && !err)
2383                 err = ret;
2384         return err;
2385 }
2386
2387 static long btrfs_fallocate(struct file *file, int mode,
2388                             loff_t offset, loff_t len)
2389 {
2390         struct inode *inode = file_inode(file);
2391         struct extent_state *cached_state = NULL;
2392         struct btrfs_root *root = BTRFS_I(inode)->root;
2393         u64 cur_offset;
2394         u64 last_byte;
2395         u64 alloc_start;
2396         u64 alloc_end;
2397         u64 alloc_hint = 0;
2398         u64 locked_end;
2399         struct extent_map *em;
2400         int blocksize = BTRFS_I(inode)->root->sectorsize;
2401         int ret;
2402
2403         alloc_start = round_down(offset, blocksize);
2404         alloc_end = round_up(offset + len, blocksize);
2405
2406         /* Make sure we aren't being give some crap mode */
2407         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2408                 return -EOPNOTSUPP;
2409
2410         if (mode & FALLOC_FL_PUNCH_HOLE)
2411                 return btrfs_punch_hole(inode, offset, len);
2412
2413         /*
2414          * Make sure we have enough space before we do the
2415          * allocation.
2416          */
2417         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2418         if (ret)
2419                 return ret;
2420         if (root->fs_info->quota_enabled) {
2421                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2422                 if (ret)
2423                         goto out_reserve_fail;
2424         }
2425
2426         mutex_lock(&inode->i_mutex);
2427         ret = inode_newsize_ok(inode, alloc_end);
2428         if (ret)
2429                 goto out;
2430
2431         if (alloc_start > inode->i_size) {
2432                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2433                                         alloc_start);
2434                 if (ret)
2435                         goto out;
2436         } else {
2437                 /*
2438                  * If we are fallocating from the end of the file onward we
2439                  * need to zero out the end of the page if i_size lands in the
2440                  * middle of a page.
2441                  */
2442                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2443                 if (ret)
2444                         goto out;
2445         }
2446
2447         /*
2448          * wait for ordered IO before we have any locks.  We'll loop again
2449          * below with the locks held.
2450          */
2451         ret = btrfs_wait_ordered_range(inode, alloc_start,
2452                                        alloc_end - alloc_start);
2453         if (ret)
2454                 goto out;
2455
2456         locked_end = alloc_end - 1;
2457         while (1) {
2458                 struct btrfs_ordered_extent *ordered;
2459
2460                 /* the extent lock is ordered inside the running
2461                  * transaction
2462                  */
2463                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2464                                  locked_end, 0, &cached_state);
2465                 ordered = btrfs_lookup_first_ordered_extent(inode,
2466                                                             alloc_end - 1);
2467                 if (ordered &&
2468                     ordered->file_offset + ordered->len > alloc_start &&
2469                     ordered->file_offset < alloc_end) {
2470                         btrfs_put_ordered_extent(ordered);
2471                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2472                                              alloc_start, locked_end,
2473                                              &cached_state, GFP_NOFS);
2474                         /*
2475                          * we can't wait on the range with the transaction
2476                          * running or with the extent lock held
2477                          */
2478                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2479                                                        alloc_end - alloc_start);
2480                         if (ret)
2481                                 goto out;
2482                 } else {
2483                         if (ordered)
2484                                 btrfs_put_ordered_extent(ordered);
2485                         break;
2486                 }
2487         }
2488
2489         cur_offset = alloc_start;
2490         while (1) {
2491                 u64 actual_end;
2492
2493                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2494                                       alloc_end - cur_offset, 0);
2495                 if (IS_ERR_OR_NULL(em)) {
2496                         if (!em)
2497                                 ret = -ENOMEM;
2498                         else
2499                                 ret = PTR_ERR(em);
2500                         break;
2501                 }
2502                 last_byte = min(extent_map_end(em), alloc_end);
2503                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2504                 last_byte = ALIGN(last_byte, blocksize);
2505
2506                 if (em->block_start == EXTENT_MAP_HOLE ||
2507                     (cur_offset >= inode->i_size &&
2508                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2509                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2510                                                         last_byte - cur_offset,
2511                                                         1 << inode->i_blkbits,
2512                                                         offset + len,
2513                                                         &alloc_hint);
2514
2515                         if (ret < 0) {
2516                                 free_extent_map(em);
2517                                 break;
2518                         }
2519                 } else if (actual_end > inode->i_size &&
2520                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2521                         /*
2522                          * We didn't need to allocate any more space, but we
2523                          * still extended the size of the file so we need to
2524                          * update i_size.
2525                          */
2526                         inode->i_ctime = CURRENT_TIME;
2527                         i_size_write(inode, actual_end);
2528                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2529                 }
2530                 free_extent_map(em);
2531
2532                 cur_offset = last_byte;
2533                 if (cur_offset >= alloc_end) {
2534                         ret = 0;
2535                         break;
2536                 }
2537         }
2538         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2539                              &cached_state, GFP_NOFS);
2540 out:
2541         mutex_unlock(&inode->i_mutex);
2542         if (root->fs_info->quota_enabled)
2543                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2544 out_reserve_fail:
2545         /* Let go of our reservation. */
2546         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2547         return ret;
2548 }
2549
2550 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2551 {
2552         struct btrfs_root *root = BTRFS_I(inode)->root;
2553         struct extent_map *em = NULL;
2554         struct extent_state *cached_state = NULL;
2555         u64 lockstart = *offset;
2556         u64 lockend = i_size_read(inode);
2557         u64 start = *offset;
2558         u64 len = i_size_read(inode);
2559         int ret = 0;
2560
2561         lockend = max_t(u64, root->sectorsize, lockend);
2562         if (lockend <= lockstart)
2563                 lockend = lockstart + root->sectorsize;
2564
2565         lockend--;
2566         len = lockend - lockstart + 1;
2567
2568         len = max_t(u64, len, root->sectorsize);
2569         if (inode->i_size == 0)
2570                 return -ENXIO;
2571
2572         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2573                          &cached_state);
2574
2575         while (start < inode->i_size) {
2576                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2577                 if (IS_ERR(em)) {
2578                         ret = PTR_ERR(em);
2579                         em = NULL;
2580                         break;
2581                 }
2582
2583                 if (whence == SEEK_HOLE &&
2584                     (em->block_start == EXTENT_MAP_HOLE ||
2585                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2586                         break;
2587                 else if (whence == SEEK_DATA &&
2588                            (em->block_start != EXTENT_MAP_HOLE &&
2589                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2590                         break;
2591
2592                 start = em->start + em->len;
2593                 free_extent_map(em);
2594                 em = NULL;
2595                 cond_resched();
2596         }
2597         free_extent_map(em);
2598         if (!ret) {
2599                 if (whence == SEEK_DATA && start >= inode->i_size)
2600                         ret = -ENXIO;
2601                 else
2602                         *offset = min_t(loff_t, start, inode->i_size);
2603         }
2604         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2605                              &cached_state, GFP_NOFS);
2606         return ret;
2607 }
2608
2609 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2610 {
2611         struct inode *inode = file->f_mapping->host;
2612         int ret;
2613
2614         mutex_lock(&inode->i_mutex);
2615         switch (whence) {
2616         case SEEK_END:
2617         case SEEK_CUR:
2618                 offset = generic_file_llseek(file, offset, whence);
2619                 goto out;
2620         case SEEK_DATA:
2621         case SEEK_HOLE:
2622                 if (offset >= i_size_read(inode)) {
2623                         mutex_unlock(&inode->i_mutex);
2624                         return -ENXIO;
2625                 }
2626
2627                 ret = find_desired_extent(inode, &offset, whence);
2628                 if (ret) {
2629                         mutex_unlock(&inode->i_mutex);
2630                         return ret;
2631                 }
2632         }
2633
2634         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2635 out:
2636         mutex_unlock(&inode->i_mutex);
2637         return offset;
2638 }
2639
2640 const struct file_operations btrfs_file_operations = {
2641         .llseek         = btrfs_file_llseek,
2642         .read           = do_sync_read,
2643         .write          = do_sync_write,
2644         .aio_read       = generic_file_aio_read,
2645         .splice_read    = generic_file_splice_read,
2646         .aio_write      = btrfs_file_aio_write,
2647         .mmap           = btrfs_file_mmap,
2648         .open           = generic_file_open,
2649         .release        = btrfs_release_file,
2650         .fsync          = btrfs_sync_file,
2651         .fallocate      = btrfs_fallocate,
2652         .unlocked_ioctl = btrfs_ioctl,
2653 #ifdef CONFIG_COMPAT
2654         .compat_ioctl   = btrfs_ioctl,
2655 #endif
2656 };
2657
2658 void btrfs_auto_defrag_exit(void)
2659 {
2660         if (btrfs_inode_defrag_cachep)
2661                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2662 }
2663
2664 int btrfs_auto_defrag_init(void)
2665 {
2666         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2667                                         sizeof(struct inode_defrag), 0,
2668                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2669                                         NULL);
2670         if (!btrfs_inode_defrag_cachep)
2671                 return -ENOMEM;
2672
2673         return 0;
2674 }