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                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
 430
 431                 /* Flush processor's dcache for this page */
 432                 flush_dcache_page(page);
 433
 434                 /*
 435                  * if we get a partial write, we can end up with
 436                  * partially up to date pages.  These add
 437                  * a lot of complexity, so make sure they don't
 438                  * happen by forcing this copy to be retried.
 439                  *
 440                  * The rest of the btrfs_file_write code will fall
 441                  * back to page at a time copies after we return 0.
 442                  */
 443                 if (!PageUptodate(page) && copied < count)
 444                         copied = 0;
 445
 446                 iov_iter_advance(i, copied);
 447                 write_bytes -= copied;
 448                 total_copied += copied;
 449
 450                 /* Return to btrfs_file_aio_write to fault page */
 451                 if (unlikely(copied == 0))
 452                         break;
 453
 454                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
 455                         offset += copied;
 456                 } else {
 457                         pg++;
 458                         offset = 0;
 459                 }
 460         }
 461         return total_copied;
 462 }
 463
 464 /*
 465  * unlocks pages after btrfs_file_write is done with them
 466  */
 467 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
 468 {
 469         size_t i;
 470         for (i = 0; i < num_pages; i++) {
 471                 /* page checked is some magic around finding pages that
 472                  * have been modified without going through btrfs_set_page_dirty
 473                  * clear it here
 474                  */
 475                 ClearPageChecked(pages[i]);
 476                 unlock_page(pages[i]);
 477                 mark_page_accessed(pages[i]);
 478                 page_cache_release(pages[i]);
 479         }
 480 }
 481
 482 /*
 483  * after copy_from_user, pages need to be dirtied and we need to make
 484  * sure holes are created between the current EOF and the start of
 485  * any next extents (if required).
 486  *
 487  * this also makes the decision about creating an inline extent vs
 488  * doing real data extents, marking pages dirty and delalloc as required.
 489  */
 490 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
 491                              struct page **pages, size_t num_pages,
 492                              loff_t pos, size_t write_bytes,
 493                              struct extent_state **cached)
 494 {
 495         int err = 0;
 496         int i;
 497         u64 num_bytes;
 498         u64 start_pos;
 499         u64 end_of_last_block;
 500         u64 end_pos = pos + write_bytes;
 501         loff_t isize = i_size_read(inode);
 502
 503         start_pos = pos & ~((u64)root->sectorsize - 1);
 504         num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
 505
 506         end_of_last_block = start_pos + num_bytes - 1;
 507         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 508                                         cached);
 509         if (err)
 510                 return err;
 511
 512         for (i = 0; i < num_pages; i++) {
 513                 struct page *p = pages[i];
 514                 SetPageUptodate(p);
 515                 ClearPageChecked(p);
 516                 set_page_dirty(p);
 517         }
 518
 519         /*
 520          * we've only changed i_size in ram, and we haven't updated
 521          * the disk i_size.  There is no need to log the inode
 522          * at this time.
 523          */
 524         if (end_pos > isize)
 525                 i_size_write(inode, end_pos);
 526         return 0;
 527 }
 528
 529 /*
 530  * this drops all the extents in the cache that intersect the range
 531  * [start, end].  Existing extents are split as required.
 532  */
 533 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 534                              int skip_pinned)
 535 {
 536         struct extent_map *em;
 537         struct extent_map *split = NULL;
 538         struct extent_map *split2 = NULL;
 539         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 540         u64 len = end - start + 1;
 541         u64 gen;
 542         int ret;
 543         int testend = 1;
 544         unsigned long flags;
 545         int compressed = 0;
 546         bool modified;
 547
 548         WARN_ON(end < start);
 549         if (end == (u64)-1) {
 550                 len = (u64)-1;
 551                 testend = 0;
 552         }
 553         while (1) {
 554                 int no_splits = 0;
 555
 556                 modified = false;
 557                 if (!split)
 558                         split = alloc_extent_map();
 559                 if (!split2)
 560                         split2 = alloc_extent_map();
 561                 if (!split || !split2)
 562                         no_splits = 1;
 563
 564                 write_lock(&em_tree->lock);
 565                 em = lookup_extent_mapping(em_tree, start, len);
 566                 if (!em) {
 567                         write_unlock(&em_tree->lock);
 568                         break;
 569                 }
 570                 flags = em->flags;
 571                 gen = em->generation;
 572                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 573                         if (testend && em->start + em->len >= start + len) {
 574                                 free_extent_map(em);
 575                                 write_unlock(&em_tree->lock);
 576                                 break;
 577                         }
 578                         start = em->start + em->len;
 579                         if (testend)
 580                                 len = start + len - (em->start + em->len);
 581                         free_extent_map(em);
 582                         write_unlock(&em_tree->lock);
 583                         continue;
 584                 }
 585                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 586                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 587                 clear_bit(EXTENT_FLAG_LOGGING, &flags);
 588                 modified = !list_empty(&em->list);
 589                 remove_extent_mapping(em_tree, em);
 590                 if (no_splits)
 591                         goto next;
 592
 593                 if (em->start < start) {
 594                         split->start = em->start;
 595                         split->len = start - em->start;
 596
 597                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 598                                 split->orig_start = em->orig_start;
 599                                 split->block_start = em->block_start;
 600
 601                                 if (compressed)
 602                                         split->block_len = em->block_len;
 603                                 else
 604                                         split->block_len = split->len;
 605                                 split->orig_block_len = max(split->block_len,
 606                                                 em->orig_block_len);
 607                                 split->ram_bytes = em->ram_bytes;
 608                         } else {
 609                                 split->orig_start = split->start;
 610                                 split->block_len = 0;
 611                                 split->block_start = em->block_start;
 612                                 split->orig_block_len = 0;
 613                                 split->ram_bytes = split->len;
 614                         }
 615
 616                         split->generation = gen;
 617                         split->bdev = em->bdev;
 618                         split->flags = flags;
 619                         split->compress_type = em->compress_type;
 620                         ret = add_extent_mapping(em_tree, split, modified);
 621                         BUG_ON(ret); /* Logic error */
 622                         free_extent_map(split);
 623                         split = split2;
 624                         split2 = NULL;
 625                 }
 626                 if (testend && em->start + em->len > start + len) {
 627                         u64 diff = start + len - em->start;
 628
 629                         split->start = start + len;
 630                         split->len = em->start + em->len - (start + len);
 631                         split->bdev = em->bdev;
 632                         split->flags = flags;
 633                         split->compress_type = em->compress_type;
 634                         split->generation = gen;
 635
 636                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 637                                 split->orig_block_len = max(em->block_len,
 638                                                     em->orig_block_len);
 639
 640                                 split->ram_bytes = em->ram_bytes;
 641                                 if (compressed) {
 642                                         split->block_len = em->block_len;
 643                                         split->block_start = em->block_start;
 644                                         split->orig_start = em->orig_start;
 645                                 } else {
 646                                         split->block_len = split->len;
 647                                         split->block_start = em->block_start
 648                                                 + diff;
 649                                         split->orig_start = em->orig_start;
 650                                 }
 651                         } else {
 652                                 split->ram_bytes = split->len;
 653                                 split->orig_start = split->start;
 654                                 split->block_len = 0;
 655                                 split->block_start = em->block_start;
 656                                 split->orig_block_len = 0;
 657                         }
 658
 659                         ret = add_extent_mapping(em_tree, split, modified);
 660                         BUG_ON(ret); /* Logic error */
 661                         free_extent_map(split);
 662                         split = NULL;
 663                 }
 664 next:
 665                 write_unlock(&em_tree->lock);
 666
 667                 /* once for us */
 668                 free_extent_map(em);
 669                 /* once for the tree*/
 670                 free_extent_map(em);
 671         }
 672         if (split)
 673                 free_extent_map(split);
 674         if (split2)
 675                 free_extent_map(split2);
 676 }
 677
 678 /*
 679  * this is very complex, but the basic idea is to drop all extents
 680  * in the range start - end.  hint_block is filled in with a block number
 681  * that would be a good hint to the block allocator for this file.
 682  *
 683  * If an extent intersects the range but is not entirely inside the range
 684  * it is either truncated or split.  Anything entirely inside the range
 685  * is deleted from the tree.
 686  */
 687 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 688                          struct btrfs_root *root, struct inode *inode,
 689                          struct btrfs_path *path, u64 start, u64 end,
 690                          u64 *drop_end, int drop_cache,
 691                          int replace_extent,
 692                          u32 extent_item_size,
 693                          int *key_inserted)
 694 {
 695         struct extent_buffer *leaf;
 696         struct btrfs_file_extent_item *fi;
 697         struct btrfs_key key;
 698         struct btrfs_key new_key;
 699         u64 ino = btrfs_ino(inode);
 700         u64 search_start = start;
 701         u64 disk_bytenr = 0;
 702         u64 num_bytes = 0;
 703         u64 extent_offset = 0;
 704         u64 extent_end = 0;
 705         int del_nr = 0;
 706         int del_slot = 0;
 707         int extent_type;
 708         int recow;
 709         int ret;
 710         int modify_tree = -1;
 711         int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
 712         int found = 0;
 713         int leafs_visited = 0;
 714
 715         if (drop_cache)
 716                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 717
 718         if (start >= BTRFS_I(inode)->disk_i_size)
 719                 modify_tree = 0;
 720
 721         while (1) {
 722                 recow = 0;
 723                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
 724                                                search_start, modify_tree);
 725                 if (ret < 0)
 726                         break;
 727                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 728                         leaf = path->nodes[0];
 729                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 730                         if (key.objectid == ino &&
 731                             key.type == BTRFS_EXTENT_DATA_KEY)
 732                                 path->slots[0]--;
 733                 }
 734                 ret = 0;
 735                 leafs_visited++;
 736 next_slot:
 737                 leaf = path->nodes[0];
 738                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 739                         BUG_ON(del_nr > 0);
 740                         ret = btrfs_next_leaf(root, path);
 741                         if (ret < 0)
 742                                 break;
 743                         if (ret > 0) {
 744                                 ret = 0;
 745                                 break;
 746                         }
 747                         leafs_visited++;
 748                         leaf = path->nodes[0];
 749                         recow = 1;
 750                 }
 751
 752                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 753                 if (key.objectid > ino ||
 754                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 755                         break;
 756
 757                 fi = btrfs_item_ptr(leaf, path->slots[0],
 758                                     struct btrfs_file_extent_item);
 759                 extent_type = btrfs_file_extent_type(leaf, fi);
 760
 761                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 762                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 763                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 764                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 765                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 766                         extent_end = key.offset +
 767                                 btrfs_file_extent_num_bytes(leaf, fi);
 768                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 769                         extent_end = key.offset +
 770                                 btrfs_file_extent_inline_len(leaf,
 771                                                      path->slots[0], fi);
 772                 } else {
 773                         WARN_ON(1);
 774                         extent_end = search_start;
 775                 }
 776
 777                 if (extent_end <= search_start) {
 778                         path->slots[0]++;
 779                         goto next_slot;
 780                 }
 781
 782                 found = 1;
 783                 search_start = max(key.offset, start);
 784                 if (recow || !modify_tree) {
 785                         modify_tree = -1;
 786                         btrfs_release_path(path);
 787                         continue;
 788                 }
 789
 790                 /*
 791                  *     | - range to drop - |
 792                  *  | -------- extent -------- |
 793                  */
 794                 if (start > key.offset && end < extent_end) {
 795                         BUG_ON(del_nr > 0);
 796                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 797
 798                         memcpy(&new_key, &key, sizeof(new_key));
 799                         new_key.offset = start;
 800                         ret = btrfs_duplicate_item(trans, root, path,
 801                                                    &new_key);
 802                         if (ret == -EAGAIN) {
 803                                 btrfs_release_path(path);
 804                                 continue;
 805                         }
 806                         if (ret < 0)
 807                                 break;
 808
 809                         leaf = path->nodes[0];
 810                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 811                                             struct btrfs_file_extent_item);
 812                         btrfs_set_file_extent_num_bytes(leaf, fi,
 813                                                         start - key.offset);
 814
 815                         fi = btrfs_item_ptr(leaf, path->slots[0],
 816                                             struct btrfs_file_extent_item);
 817
 818                         extent_offset += start - key.offset;
 819                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 820                         btrfs_set_file_extent_num_bytes(leaf, fi,
 821                                                         extent_end - start);
 822                         btrfs_mark_buffer_dirty(leaf);
 823
 824                         if (update_refs && disk_bytenr > 0) {
 825                                 ret = btrfs_inc_extent_ref(trans, root,
 826                                                 disk_bytenr, num_bytes, 0,
 827                                                 root->root_key.objectid,
 828                                                 new_key.objectid,
 829                                                 start - extent_offset, 0);
 830                                 BUG_ON(ret); /* -ENOMEM */
 831                         }
 832                         key.offset = start;
 833                 }
 834                 /*
 835                  *  | ---- range to drop ----- |
 836                  *      | -------- extent -------- |
 837                  */
 838                 if (start <= key.offset && end < extent_end) {
 839                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 840
 841                         memcpy(&new_key, &key, sizeof(new_key));
 842                         new_key.offset = end;
 843                         btrfs_set_item_key_safe(root, path, &new_key);
 844
 845                         extent_offset += end - key.offset;
 846                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 847                         btrfs_set_file_extent_num_bytes(leaf, fi,
 848                                                         extent_end - end);
 849                         btrfs_mark_buffer_dirty(leaf);
 850                         if (update_refs && disk_bytenr > 0)
 851                                 inode_sub_bytes(inode, end - key.offset);
 852                         break;
 853                 }
 854
 855                 search_start = extent_end;
 856                 /*
 857                  *       | ---- range to drop ----- |
 858                  *  | -------- extent -------- |
 859                  */
 860                 if (start > key.offset && end >= extent_end) {
 861                         BUG_ON(del_nr > 0);
 862                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 863
 864                         btrfs_set_file_extent_num_bytes(leaf, fi,
 865                                                         start - key.offset);
 866                         btrfs_mark_buffer_dirty(leaf);
 867                         if (update_refs && disk_bytenr > 0)
 868                                 inode_sub_bytes(inode, extent_end - start);
 869                         if (end == extent_end)
 870                                 break;
 871
 872                         path->slots[0]++;
 873                         goto next_slot;
 874                 }
 875
 876                 /*
 877                  *  | ---- range to drop ----- |
 878                  *    | ------ extent ------ |
 879                  */
 880                 if (start <= key.offset && end >= extent_end) {
 881                         if (del_nr == 0) {
 882                                 del_slot = path->slots[0];
 883                                 del_nr = 1;
 884                         } else {
 885                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 886                                 del_nr++;
 887                         }
 888
 889                         if (update_refs &&
 890                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 891                                 inode_sub_bytes(inode,
 892                                                 extent_end - key.offset);
 893                                 extent_end = ALIGN(extent_end,
 894                                                    root->sectorsize);
 895                         } else if (update_refs && disk_bytenr > 0) {
 896                                 ret = btrfs_free_extent(trans, root,
 897                                                 disk_bytenr, num_bytes, 0,
 898                                                 root->root_key.objectid,
 899                                                 key.objectid, key.offset -
 900                                                 extent_offset, 0);
 901                                 BUG_ON(ret); /* -ENOMEM */
 902                                 inode_sub_bytes(inode,
 903                                                 extent_end - key.offset);
 904                         }
 905
 906                         if (end == extent_end)
 907                                 break;
 908
 909                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 910                                 path->slots[0]++;
 911                                 goto next_slot;
 912                         }
 913
 914                         ret = btrfs_del_items(trans, root, path, del_slot,
 915                                               del_nr);
 916                         if (ret) {
 917                                 btrfs_abort_transaction(trans, root, ret);
 918                                 break;
 919                         }
 920
 921                         del_nr = 0;
 922                         del_slot = 0;
 923
 924                         btrfs_release_path(path);
 925                         continue;
 926                 }
 927
 928                 BUG_ON(1);
 929         }
 930
 931         if (!ret && del_nr > 0) {
 932                 /*
 933                  * Set path->slots[0] to first slot, so that after the delete
 934                  * if items are move off from our leaf to its immediate left or
 935                  * right neighbor leafs, we end up with a correct and adjusted
 936                  * path->slots[0] for our insertion.
 937                  */
 938                 path->slots[0] = del_slot;
 939                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 940                 if (ret)
 941                         btrfs_abort_transaction(trans, root, ret);
 942
 943                 leaf = path->nodes[0];
 944                 /*
 945                  * leaf eb has flag EXTENT_BUFFER_STALE if it was deleted (that
 946                  * is, its contents got pushed to its neighbors), in which case
 947                  * it means path->locks[0] == 0
 948                  */
 949                 if (!ret && replace_extent && leafs_visited == 1 &&
 950                     path->locks[0] &&
 951                     btrfs_leaf_free_space(root, leaf) >=
 952                     sizeof(struct btrfs_item) + extent_item_size) {
 953
 954                         key.objectid = ino;
 955                         key.type = BTRFS_EXTENT_DATA_KEY;
 956                         key.offset = start;
 957                         setup_items_for_insert(root, path, &key,
 958                                                &extent_item_size,
 959                                                extent_item_size,
 960                                                sizeof(struct btrfs_item) +
 961                                                extent_item_size, 1);
 962                         *key_inserted = 1;
 963                 }
 964         }
 965
 966         if (!replace_extent || !(*key_inserted))
 967                 btrfs_release_path(path);
 968         if (drop_end)
 969                 *drop_end = found ? min(end, extent_end) : end;
 970         return ret;
 971 }
 972
 973 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 974                        struct btrfs_root *root, struct inode *inode, u64 start,
 975                        u64 end, int drop_cache)
 976 {
 977         struct btrfs_path *path;
 978         int ret;
 979
 980         path = btrfs_alloc_path();
 981         if (!path)
 982                 return -ENOMEM;
 983         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
 984                                    drop_cache, 0, 0, NULL);
 985         btrfs_free_path(path);
 986         return ret;
 987 }
 988
 989 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 990                             u64 objectid, u64 bytenr, u64 orig_offset,
 991                             u64 *start, u64 *end)
 992 {
 993         struct btrfs_file_extent_item *fi;
 994         struct btrfs_key key;
 995         u64 extent_end;
 996
 997         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 998                 return 0;
 999
1000         btrfs_item_key_to_cpu(leaf, &key, slot);
1001         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1002                 return 0;
1003
1004         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1005         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1006             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1007             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1008             btrfs_file_extent_compression(leaf, fi) ||
1009             btrfs_file_extent_encryption(leaf, fi) ||
1010             btrfs_file_extent_other_encoding(leaf, fi))
1011                 return 0;
1012
1013         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1014         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1015                 return 0;
1016
1017         *start = key.offset;
1018         *end = extent_end;
1019         return 1;
1020 }
1021
1022 /*
1023  * Mark extent in the range start - end as written.
1024  *
1025  * This changes extent type from 'pre-allocated' to 'regular'. If only
1026  * part of extent is marked as written, the extent will be split into
1027  * two or three.
1028  */
1029 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1030                               struct inode *inode, u64 start, u64 end)
1031 {
1032         struct btrfs_root *root = BTRFS_I(inode)->root;
1033         struct extent_buffer *leaf;
1034         struct btrfs_path *path;
1035         struct btrfs_file_extent_item *fi;
1036         struct btrfs_key key;
1037         struct btrfs_key new_key;
1038         u64 bytenr;
1039         u64 num_bytes;
1040         u64 extent_end;
1041         u64 orig_offset;
1042         u64 other_start;
1043         u64 other_end;
1044         u64 split;
1045         int del_nr = 0;
1046         int del_slot = 0;
1047         int recow;
1048         int ret;
1049         u64 ino = btrfs_ino(inode);
1050
1051         path = btrfs_alloc_path();
1052         if (!path)
1053                 return -ENOMEM;
1054 again:
1055         recow = 0;
1056         split = start;
1057         key.objectid = ino;
1058         key.type = BTRFS_EXTENT_DATA_KEY;
1059         key.offset = split;
1060
1061         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1062         if (ret < 0)
1063                 goto out;
1064         if (ret > 0 && path->slots[0] > 0)
1065                 path->slots[0]--;
1066
1067         leaf = path->nodes[0];
1068         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1069         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1070         fi = btrfs_item_ptr(leaf, path->slots[0],
1071                             struct btrfs_file_extent_item);
1072         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1073                BTRFS_FILE_EXTENT_PREALLOC);
1074         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1075         BUG_ON(key.offset > start || extent_end < end);
1076
1077         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1078         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1079         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1080         memcpy(&new_key, &key, sizeof(new_key));
1081
1082         if (start == key.offset && end < extent_end) {
1083                 other_start = 0;
1084                 other_end = start;
1085                 if (extent_mergeable(leaf, path->slots[0] - 1,
1086                                      ino, bytenr, orig_offset,
1087                                      &other_start, &other_end)) {
1088                         new_key.offset = end;
1089                         btrfs_set_item_key_safe(root, path, &new_key);
1090                         fi = btrfs_item_ptr(leaf, path->slots[0],
1091                                             struct btrfs_file_extent_item);
1092                         btrfs_set_file_extent_generation(leaf, fi,
1093                                                          trans->transid);
1094                         btrfs_set_file_extent_num_bytes(leaf, fi,
1095                                                         extent_end - end);
1096                         btrfs_set_file_extent_offset(leaf, fi,
1097                                                      end - orig_offset);
1098                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1099                                             struct btrfs_file_extent_item);
1100                         btrfs_set_file_extent_generation(leaf, fi,
1101                                                          trans->transid);
1102                         btrfs_set_file_extent_num_bytes(leaf, fi,
1103                                                         end - other_start);
1104                         btrfs_mark_buffer_dirty(leaf);
1105                         goto out;
1106                 }
1107         }
1108
1109         if (start > key.offset && end == extent_end) {
1110                 other_start = end;
1111                 other_end = 0;
1112                 if (extent_mergeable(leaf, path->slots[0] + 1,
1113                                      ino, bytenr, orig_offset,
1114                                      &other_start, &other_end)) {
1115                         fi = btrfs_item_ptr(leaf, path->slots[0],
1116                                             struct btrfs_file_extent_item);
1117                         btrfs_set_file_extent_num_bytes(leaf, fi,
1118                                                         start - key.offset);
1119                         btrfs_set_file_extent_generation(leaf, fi,
1120                                                          trans->transid);
1121                         path->slots[0]++;
1122                         new_key.offset = start;
1123                         btrfs_set_item_key_safe(root, path, &new_key);
1124
1125                         fi = btrfs_item_ptr(leaf, path->slots[0],
1126                                             struct btrfs_file_extent_item);
1127                         btrfs_set_file_extent_generation(leaf, fi,
1128                                                          trans->transid);
1129                         btrfs_set_file_extent_num_bytes(leaf, fi,
1130                                                         other_end - start);
1131                         btrfs_set_file_extent_offset(leaf, fi,
1132                                                      start - orig_offset);
1133                         btrfs_mark_buffer_dirty(leaf);
1134                         goto out;
1135                 }
1136         }
1137
1138         while (start > key.offset || end < extent_end) {
1139                 if (key.offset == start)
1140                         split = end;
1141
1142                 new_key.offset = split;
1143                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1144                 if (ret == -EAGAIN) {
1145                         btrfs_release_path(path);
1146                         goto again;
1147                 }
1148                 if (ret < 0) {
1149                         btrfs_abort_transaction(trans, root, ret);
1150                         goto out;
1151                 }
1152
1153                 leaf = path->nodes[0];
1154                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1155                                     struct btrfs_file_extent_item);
1156                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1157                 btrfs_set_file_extent_num_bytes(leaf, fi,
1158                                                 split - key.offset);
1159
1160                 fi = btrfs_item_ptr(leaf, path->slots[0],
1161                                     struct btrfs_file_extent_item);
1162
1163                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1164                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1165                 btrfs_set_file_extent_num_bytes(leaf, fi,
1166                                                 extent_end - split);
1167                 btrfs_mark_buffer_dirty(leaf);
1168
1169                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1170                                            root->root_key.objectid,
1171                                            ino, orig_offset, 0);
1172                 BUG_ON(ret); /* -ENOMEM */
1173
1174                 if (split == start) {
1175                         key.offset = start;
1176                 } else {
1177                         BUG_ON(start != key.offset);
1178                         path->slots[0]--;
1179                         extent_end = end;
1180                 }
1181                 recow = 1;
1182         }
1183
1184         other_start = end;
1185         other_end = 0;
1186         if (extent_mergeable(leaf, path->slots[0] + 1,
1187                              ino, bytenr, orig_offset,
1188                              &other_start, &other_end)) {
1189                 if (recow) {
1190                         btrfs_release_path(path);
1191                         goto again;
1192                 }
1193                 extent_end = other_end;
1194                 del_slot = path->slots[0] + 1;
1195                 del_nr++;
1196                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1197                                         0, root->root_key.objectid,
1198                                         ino, orig_offset, 0);
1199                 BUG_ON(ret); /* -ENOMEM */
1200         }
1201         other_start = 0;
1202         other_end = start;
1203         if (extent_mergeable(leaf, path->slots[0] - 1,
1204                              ino, bytenr, orig_offset,
1205                              &other_start, &other_end)) {
1206                 if (recow) {
1207                         btrfs_release_path(path);
1208                         goto again;
1209                 }
1210                 key.offset = other_start;
1211                 del_slot = path->slots[0];
1212                 del_nr++;
1213                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1214                                         0, root->root_key.objectid,
1215                                         ino, orig_offset, 0);
1216                 BUG_ON(ret); /* -ENOMEM */
1217         }
1218         if (del_nr == 0) {
1219                 fi = btrfs_item_ptr(leaf, path->slots[0],
1220                            struct btrfs_file_extent_item);
1221                 btrfs_set_file_extent_type(leaf, fi,
1222                                            BTRFS_FILE_EXTENT_REG);
1223                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1224                 btrfs_mark_buffer_dirty(leaf);
1225         } else {
1226                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1227                            struct btrfs_file_extent_item);
1228                 btrfs_set_file_extent_type(leaf, fi,
1229                                            BTRFS_FILE_EXTENT_REG);
1230                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1231                 btrfs_set_file_extent_num_bytes(leaf, fi,
1232                                                 extent_end - key.offset);
1233                 btrfs_mark_buffer_dirty(leaf);
1234
1235                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1236                 if (ret < 0) {
1237                         btrfs_abort_transaction(trans, root, ret);
1238                         goto out;
1239                 }
1240         }
1241 out:
1242         btrfs_free_path(path);
1243         return 0;
1244 }
1245
1246 /*
1247  * on error we return an unlocked page and the error value
1248  * on success we return a locked page and 0
1249  */
1250 static int prepare_uptodate_page(struct page *page, u64 pos,
1251                                  bool force_uptodate)
1252 {
1253         int ret = 0;
1254
1255         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1256             !PageUptodate(page)) {
1257                 ret = btrfs_readpage(NULL, page);
1258                 if (ret)
1259                         return ret;
1260                 lock_page(page);
1261                 if (!PageUptodate(page)) {
1262                         unlock_page(page);
1263                         return -EIO;
1264                 }
1265         }
1266         return 0;
1267 }
1268
1269 /*
1270  * this just gets pages into the page cache and locks them down.
1271  */
1272 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1273                                   size_t num_pages, loff_t pos,
1274                                   size_t write_bytes, bool force_uptodate)
1275 {
1276         int i;
1277         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1278         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1279         int err = 0;
1280         int faili;
1281
1282         for (i = 0; i < num_pages; i++) {
1283                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1284                                                mask | __GFP_WRITE);
1285                 if (!pages[i]) {
1286                         faili = i - 1;
1287                         err = -ENOMEM;
1288                         goto fail;
1289                 }
1290
1291                 if (i == 0)
1292                         err = prepare_uptodate_page(pages[i], pos,
1293                                                     force_uptodate);
1294                 if (i == num_pages - 1)
1295                         err = prepare_uptodate_page(pages[i],
1296                                                     pos + write_bytes, false);
1297                 if (err) {
1298                         page_cache_release(pages[i]);
1299                         faili = i - 1;
1300                         goto fail;
1301                 }
1302                 wait_on_page_writeback(pages[i]);
1303         }
1304
1305         return 0;
1306 fail:
1307         while (faili >= 0) {
1308                 unlock_page(pages[faili]);
1309                 page_cache_release(pages[faili]);
1310                 faili--;
1311         }
1312         return err;
1313
1314 }
1315
1316 /*
1317  * This function locks the extent and properly waits for data=ordered extents
1318  * to finish before allowing the pages to be modified if need.
1319  *
1320  * The return value:
1321  * 1 - the extent is locked
1322  * 0 - the extent is not locked, and everything is OK
1323  * -EAGAIN - need re-prepare the pages
1324  * the other < 0 number - Something wrong happens
1325  */
1326 static noinline int
1327 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1328                                 size_t num_pages, loff_t pos,
1329                                 u64 *lockstart, u64 *lockend,
1330                                 struct extent_state **cached_state)
1331 {
1332         u64 start_pos;
1333         u64 last_pos;
1334         int i;
1335         int ret = 0;
1336
1337         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1338         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1339
1340         if (start_pos < inode->i_size) {
1341                 struct btrfs_ordered_extent *ordered;
1342                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1343                                  start_pos, last_pos, 0, cached_state);
1344                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1345                 if (ordered &&
1346                     ordered->file_offset + ordered->len > start_pos &&
1347                     ordered->file_offset <= last_pos) {
1348                         btrfs_put_ordered_extent(ordered);
1349                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1350                                              start_pos, last_pos,
1351                                              cached_state, GFP_NOFS);
1352                         for (i = 0; i < num_pages; i++) {
1353                                 unlock_page(pages[i]);
1354                                 page_cache_release(pages[i]);
1355                         }
1356                         ret = btrfs_wait_ordered_range(inode, start_pos,
1357                                                 last_pos - start_pos + 1);
1358                         if (ret)
1359                                 return ret;
1360                         else
1361                                 return -EAGAIN;
1362                 }
1363                 if (ordered)
1364                         btrfs_put_ordered_extent(ordered);
1365
1366                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1367                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1368                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1369                                   0, 0, cached_state, GFP_NOFS);
1370                 *lockstart = start_pos;
1371                 *lockend = last_pos;
1372                 ret = 1;
1373         }
1374
1375         for (i = 0; i < num_pages; i++) {
1376                 if (clear_page_dirty_for_io(pages[i]))
1377                         account_page_redirty(pages[i]);
1378                 set_page_extent_mapped(pages[i]);
1379                 WARN_ON(!PageLocked(pages[i]));
1380         }
1381
1382         return ret;
1383 }
1384
1385 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1386                                     size_t *write_bytes)
1387 {
1388         struct btrfs_root *root = BTRFS_I(inode)->root;
1389         struct btrfs_ordered_extent *ordered;
1390         u64 lockstart, lockend;
1391         u64 num_bytes;
1392         int ret;
1393
1394         lockstart = round_down(pos, root->sectorsize);
1395         lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1396
1397         while (1) {
1398                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1399                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1400                                                      lockend - lockstart + 1);
1401                 if (!ordered) {
1402                         break;
1403                 }
1404                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1405                 btrfs_start_ordered_extent(inode, ordered, 1);
1406                 btrfs_put_ordered_extent(ordered);
1407         }
1408
1409         num_bytes = lockend - lockstart + 1;
1410         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1411         if (ret <= 0) {
1412                 ret = 0;
1413         } else {
1414                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1415                                  EXTENT_DIRTY | EXTENT_DELALLOC |
1416                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1417                                  NULL, GFP_NOFS);
1418                 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1419         }
1420
1421         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1422
1423         return ret;
1424 }
1425
1426 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1427                                                struct iov_iter *i,
1428                                                loff_t pos)
1429 {
1430         struct inode *inode = file_inode(file);
1431         struct btrfs_root *root = BTRFS_I(inode)->root;
1432         struct page **pages = NULL;
1433         struct extent_state *cached_state = NULL;
1434         u64 release_bytes = 0;
1435         u64 lockstart;
1436         u64 lockend;
1437         unsigned long first_index;
1438         size_t num_written = 0;
1439         int nrptrs;
1440         int ret = 0;
1441         bool only_release_metadata = false;
1442         bool force_page_uptodate = false;
1443         bool need_unlock;
1444
1445         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1446                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1447                      (sizeof(struct page *)));
1448         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1449         nrptrs = max(nrptrs, 8);
1450         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1451         if (!pages)
1452                 return -ENOMEM;
1453
1454         first_index = pos >> PAGE_CACHE_SHIFT;
1455
1456         while (iov_iter_count(i) > 0) {
1457                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1458                 size_t write_bytes = min(iov_iter_count(i),
1459                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1460                                          offset);
1461                 size_t num_pages = (write_bytes + offset +
1462                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1463                 size_t reserve_bytes;
1464                 size_t dirty_pages;
1465                 size_t copied;
1466
1467                 WARN_ON(num_pages > nrptrs);
1468
1469                 /*
1470                  * Fault pages before locking them in prepare_pages
1471                  * to avoid recursive lock
1472                  */
1473                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1474                         ret = -EFAULT;
1475                         break;
1476                 }
1477
1478                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1479                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1480                 if (ret == -ENOSPC &&
1481                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1482                                               BTRFS_INODE_PREALLOC))) {
1483                         ret = check_can_nocow(inode, pos, &write_bytes);
1484                         if (ret > 0) {
1485                                 only_release_metadata = true;
1486                                 /*
1487                                  * our prealloc extent may be smaller than
1488                                  * write_bytes, so scale down.
1489                                  */
1490                                 num_pages = (write_bytes + offset +
1491                                              PAGE_CACHE_SIZE - 1) >>
1492                                         PAGE_CACHE_SHIFT;
1493                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1494                                 ret = 0;
1495                         } else {
1496                                 ret = -ENOSPC;
1497                         }
1498                 }
1499
1500                 if (ret)
1501                         break;
1502
1503                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1504                 if (ret) {
1505                         if (!only_release_metadata)
1506                                 btrfs_free_reserved_data_space(inode,
1507                                                                reserve_bytes);
1508                         break;
1509                 }
1510
1511                 release_bytes = reserve_bytes;
1512                 need_unlock = false;
1513 again:
1514                 /*
1515                  * This is going to setup the pages array with the number of
1516                  * pages we want, so we don't really need to worry about the
1517                  * contents of pages from loop to loop
1518                  */
1519                 ret = prepare_pages(inode, pages, num_pages,
1520                                     pos, write_bytes,
1521                                     force_page_uptodate);
1522                 if (ret)
1523                         break;
1524
1525                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1526                                                       pos, &lockstart, &lockend,
1527                                                       &cached_state);
1528                 if (ret < 0) {
1529                         if (ret == -EAGAIN)
1530                                 goto again;
1531                         break;
1532                 } else if (ret > 0) {
1533                         need_unlock = true;
1534                         ret = 0;
1535                 }
1536
1537                 copied = btrfs_copy_from_user(pos, num_pages,
1538                                            write_bytes, pages, i);
1539
1540                 /*
1541                  * if we have trouble faulting in the pages, fall
1542                  * back to one page at a time
1543                  */
1544                 if (copied < write_bytes)
1545                         nrptrs = 1;
1546
1547                 if (copied == 0) {
1548                         force_page_uptodate = true;
1549                         dirty_pages = 0;
1550                 } else {
1551                         force_page_uptodate = false;
1552                         dirty_pages = (copied + offset +
1553                                        PAGE_CACHE_SIZE - 1) >>
1554                                        PAGE_CACHE_SHIFT;
1555                 }
1556
1557                 /*
1558                  * If we had a short copy we need to release the excess delaloc
1559                  * bytes we reserved.  We need to increment outstanding_extents
1560                  * because btrfs_delalloc_release_space will decrement it, but
1561                  * we still have an outstanding extent for the chunk we actually
1562                  * managed to copy.
1563                  */
1564                 if (num_pages > dirty_pages) {
1565                         release_bytes = (num_pages - dirty_pages) <<
1566                                 PAGE_CACHE_SHIFT;
1567                         if (copied > 0) {
1568                                 spin_lock(&BTRFS_I(inode)->lock);
1569                                 BTRFS_I(inode)->outstanding_extents++;
1570                                 spin_unlock(&BTRFS_I(inode)->lock);
1571                         }
1572                         if (only_release_metadata)
1573                                 btrfs_delalloc_release_metadata(inode,
1574                                                                 release_bytes);
1575                         else
1576                                 btrfs_delalloc_release_space(inode,
1577                                                              release_bytes);
1578                 }
1579
1580                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1581
1582                 if (copied > 0)
1583                         ret = btrfs_dirty_pages(root, inode, pages,
1584                                                 dirty_pages, pos, copied,
1585                                                 NULL);
1586                 if (need_unlock)
1587                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1588                                              lockstart, lockend, &cached_state,
1589                                              GFP_NOFS);
1590                 if (ret) {
1591                         btrfs_drop_pages(pages, num_pages);
1592                         break;
1593                 }
1594
1595                 release_bytes = 0;
1596                 if (only_release_metadata && copied > 0) {
1597                         u64 lockstart = round_down(pos, root->sectorsize);
1598                         u64 lockend = lockstart +
1599                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1600
1601                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1602                                        lockend, EXTENT_NORESERVE, NULL,
1603                                        NULL, GFP_NOFS);
1604                         only_release_metadata = false;
1605                 }
1606
1607                 btrfs_drop_pages(pages, num_pages);
1608
1609                 cond_resched();
1610
1611                 balance_dirty_pages_ratelimited(inode->i_mapping);
1612                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1613                         btrfs_btree_balance_dirty(root);
1614
1615                 pos += copied;
1616                 num_written += copied;
1617         }
1618
1619         kfree(pages);
1620
1621         if (release_bytes) {
1622                 if (only_release_metadata)
1623                         btrfs_delalloc_release_metadata(inode, release_bytes);
1624                 else
1625                         btrfs_delalloc_release_space(inode, release_bytes);
1626         }
1627
1628         return num_written ? num_written : ret;
1629 }
1630
1631 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1632                                     const struct iovec *iov,
1633                                     unsigned long nr_segs, loff_t pos,
1634                                     loff_t *ppos, size_t count, size_t ocount)
1635 {
1636         struct file *file = iocb->ki_filp;
1637         struct iov_iter i;
1638         ssize_t written;
1639         ssize_t written_buffered;
1640         loff_t endbyte;
1641         int err;
1642
1643         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1644                                             count, ocount);
1645
1646         if (written < 0 || written == count)
1647                 return written;
1648
1649         pos += written;
1650         count -= written;
1651         iov_iter_init(&i, iov, nr_segs, count, written);
1652         written_buffered = __btrfs_buffered_write(file, &i, pos);
1653         if (written_buffered < 0) {
1654                 err = written_buffered;
1655                 goto out;
1656         }
1657         endbyte = pos + written_buffered - 1;
1658         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1659         if (err)
1660                 goto out;
1661         written += written_buffered;
1662         *ppos = pos + written_buffered;
1663         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1664                                  endbyte >> PAGE_CACHE_SHIFT);
1665 out:
1666         return written ? written : err;
1667 }
1668
1669 static void update_time_for_write(struct inode *inode)
1670 {
1671         struct timespec now;
1672
1673         if (IS_NOCMTIME(inode))
1674                 return;
1675
1676         now = current_fs_time(inode->i_sb);
1677         if (!timespec_equal(&inode->i_mtime, &now))
1678                 inode->i_mtime = now;
1679
1680         if (!timespec_equal(&inode->i_ctime, &now))
1681                 inode->i_ctime = now;
1682
1683         if (IS_I_VERSION(inode))
1684                 inode_inc_iversion(inode);
1685 }
1686
1687 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1688                                     const struct iovec *iov,
1689                                     unsigned long nr_segs, loff_t pos)
1690 {
1691         struct file *file = iocb->ki_filp;
1692         struct inode *inode = file_inode(file);
1693         struct btrfs_root *root = BTRFS_I(inode)->root;
1694         loff_t *ppos = &iocb->ki_pos;
1695         u64 start_pos;
1696         ssize_t num_written = 0;
1697         ssize_t err = 0;
1698         size_t count, ocount;
1699         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1700
1701         mutex_lock(&inode->i_mutex);
1702
1703         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1704         if (err) {
1705                 mutex_unlock(&inode->i_mutex);
1706                 goto out;
1707         }
1708         count = ocount;
1709
1710         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1711         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1712         if (err) {
1713                 mutex_unlock(&inode->i_mutex);
1714                 goto out;
1715         }
1716
1717         if (count == 0) {
1718                 mutex_unlock(&inode->i_mutex);
1719                 goto out;
1720         }
1721
1722         err = file_remove_suid(file);
1723         if (err) {
1724                 mutex_unlock(&inode->i_mutex);
1725                 goto out;
1726         }
1727
1728         /*
1729          * If BTRFS flips readonly due to some impossible error
1730          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1731          * although we have opened a file as writable, we have
1732          * to stop this write operation to ensure FS consistency.
1733          */
1734         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1735                 mutex_unlock(&inode->i_mutex);
1736                 err = -EROFS;
1737                 goto out;
1738         }
1739
1740         /*
1741          * We reserve space for updating the inode when we reserve space for the
1742          * extent we are going to write, so we will enospc out there.  We don't
1743          * need to start yet another transaction to update the inode as we will
1744          * update the inode when we finish writing whatever data we write.
1745          */
1746         update_time_for_write(inode);
1747
1748         start_pos = round_down(pos, root->sectorsize);
1749         if (start_pos > i_size_read(inode)) {
1750                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1751                 if (err) {
1752                         mutex_unlock(&inode->i_mutex);
1753                         goto out;
1754                 }
1755         }
1756
1757         if (sync)
1758                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1759
1760         if (unlikely(file->f_flags & O_DIRECT)) {
1761                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1762                                                    pos, ppos, count, ocount);
1763         } else {
1764                 struct iov_iter i;
1765
1766                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1767
1768                 num_written = __btrfs_buffered_write(file, &i, pos);
1769                 if (num_written > 0)
1770                         *ppos = pos + num_written;
1771         }
1772
1773         mutex_unlock(&inode->i_mutex);
1774
1775         /*
1776          * we want to make sure fsync finds this change
1777          * but we haven't joined a transaction running right now.
1778          *
1779          * Later on, someone is sure to update the inode and get the
1780          * real transid recorded.
1781          *
1782          * We set last_trans now to the fs_info generation + 1,
1783          * this will either be one more than the running transaction
1784          * or the generation used for the next transaction if there isn't
1785          * one running right now.
1786          *
1787          * We also have to set last_sub_trans to the current log transid,
1788          * otherwise subsequent syncs to a file that's been synced in this
1789          * transaction will appear to have already occured.
1790          */
1791         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1792         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1793         if (num_written > 0) {
1794                 err = generic_write_sync(file, pos, num_written);
1795                 if (err < 0 && num_written > 0)
1796                         num_written = err;
1797         }
1798
1799         if (sync)
1800                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1801 out:
1802         current->backing_dev_info = NULL;
1803         return num_written ? num_written : err;
1804 }
1805
1806 int btrfs_release_file(struct inode *inode, struct file *filp)
1807 {
1808         /*
1809          * ordered_data_close is set by settattr when we are about to truncate
1810          * a file from a non-zero size to a zero size.  This tries to
1811          * flush down new bytes that may have been written if the
1812          * application were using truncate to replace a file in place.
1813          */
1814         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1815                                &BTRFS_I(inode)->runtime_flags)) {
1816                 struct btrfs_trans_handle *trans;
1817                 struct btrfs_root *root = BTRFS_I(inode)->root;
1818
1819                 /*
1820                  * We need to block on a committing transaction to keep us from
1821                  * throwing a ordered operation on to the list and causing
1822                  * something like sync to deadlock trying to flush out this
1823                  * inode.
1824                  */
1825                 trans = btrfs_start_transaction(root, 0);
1826                 if (IS_ERR(trans))
1827                         return PTR_ERR(trans);
1828                 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1829                 btrfs_end_transaction(trans, root);
1830                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1831                         filemap_flush(inode->i_mapping);
1832         }
1833         if (filp->private_data)
1834                 btrfs_ioctl_trans_end(filp);
1835         return 0;
1836 }
1837
1838 /*
1839  * fsync call for both files and directories.  This logs the inode into
1840  * the tree log instead of forcing full commits whenever possible.
1841  *
1842  * It needs to call filemap_fdatawait so that all ordered extent updates are
1843  * in the metadata btree are up to date for copying to the log.
1844  *
1845  * It drops the inode mutex before doing the tree log commit.  This is an
1846  * important optimization for directories because holding the mutex prevents
1847  * new operations on the dir while we write to disk.
1848  */
1849 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1850 {
1851         struct dentry *dentry = file->f_path.dentry;
1852         struct inode *inode = dentry->d_inode;
1853         struct btrfs_root *root = BTRFS_I(inode)->root;
1854         int ret = 0;
1855         struct btrfs_trans_handle *trans;
1856         bool full_sync = 0;
1857
1858         trace_btrfs_sync_file(file, datasync);
1859
1860         /*
1861          * We write the dirty pages in the range and wait until they complete
1862          * out of the ->i_mutex. If so, we can flush the dirty pages by
1863          * multi-task, and make the performance up.  See
1864          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1865          */
1866         atomic_inc(&BTRFS_I(inode)->sync_writers);
1867         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1868         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1869                              &BTRFS_I(inode)->runtime_flags))
1870                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1871         atomic_dec(&BTRFS_I(inode)->sync_writers);
1872         if (ret)
1873                 return ret;
1874
1875         mutex_lock(&inode->i_mutex);
1876
1877         /*
1878          * We flush the dirty pages again to avoid some dirty pages in the
1879          * range being left.
1880          */
1881         atomic_inc(&root->log_batch);
1882         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1883                              &BTRFS_I(inode)->runtime_flags);
1884         if (full_sync) {
1885                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1886                 if (ret) {
1887                         mutex_unlock(&inode->i_mutex);
1888                         goto out;
1889                 }
1890         }
1891         atomic_inc(&root->log_batch);
1892
1893         /*
1894          * check the transaction that last modified this inode
1895          * and see if its already been committed
1896          */
1897         if (!BTRFS_I(inode)->last_trans) {
1898                 mutex_unlock(&inode->i_mutex);
1899                 goto out;
1900         }
1901
1902         /*
1903          * if the last transaction that changed this file was before
1904          * the current transaction, we can bail out now without any
1905          * syncing
1906          */
1907         smp_mb();
1908         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1909             BTRFS_I(inode)->last_trans <=
1910             root->fs_info->last_trans_committed) {
1911                 BTRFS_I(inode)->last_trans = 0;
1912
1913                 /*
1914                  * We'v had everything committed since the last time we were
1915                  * modified so clear this flag in case it was set for whatever
1916                  * reason, it's no longer relevant.
1917                  */
1918                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1919                           &BTRFS_I(inode)->runtime_flags);
1920                 mutex_unlock(&inode->i_mutex);
1921                 goto out;
1922         }
1923
1924         /*
1925          * ok we haven't committed the transaction yet, lets do a commit
1926          */
1927         if (file->private_data)
1928                 btrfs_ioctl_trans_end(file);
1929
1930         /*
1931          * We use start here because we will need to wait on the IO to complete
1932          * in btrfs_sync_log, which could require joining a transaction (for
1933          * example checking cross references in the nocow path).  If we use join
1934          * here we could get into a situation where we're waiting on IO to
1935          * happen that is blocked on a transaction trying to commit.  With start
1936          * we inc the extwriter counter, so we wait for all extwriters to exit
1937          * before we start blocking join'ers.  This comment is to keep somebody
1938          * from thinking they are super smart and changing this to
1939          * btrfs_join_transaction *cough*Josef*cough*.
1940          */
1941         trans = btrfs_start_transaction(root, 0);
1942         if (IS_ERR(trans)) {
1943                 ret = PTR_ERR(trans);
1944                 mutex_unlock(&inode->i_mutex);
1945                 goto out;
1946         }
1947         trans->sync = true;
1948
1949         ret = btrfs_log_dentry_safe(trans, root, dentry);
1950         if (ret < 0) {
1951                 /* Fallthrough and commit/free transaction. */
1952                 ret = 1;
1953         }
1954
1955         /* we've logged all the items and now have a consistent
1956          * version of the file in the log.  It is possible that
1957          * someone will come in and modify the file, but that's
1958          * fine because the log is consistent on disk, and we
1959          * have references to all of the file's extents
1960          *
1961          * It is possible that someone will come in and log the
1962          * file again, but that will end up using the synchronization
1963          * inside btrfs_sync_log to keep things safe.
1964          */
1965         mutex_unlock(&inode->i_mutex);
1966
1967         if (ret != BTRFS_NO_LOG_SYNC) {
1968                 if (!ret) {
1969                         ret = btrfs_sync_log(trans, root);
1970                         if (!ret) {
1971                                 ret = btrfs_end_transaction(trans, root);
1972                                 goto out;
1973                         }
1974                 }
1975                 if (!full_sync) {
1976                         ret = btrfs_wait_ordered_range(inode, start,
1977                                                        end - start + 1);
1978                         if (ret)
1979                                 goto out;
1980                 }
1981                 ret = btrfs_commit_transaction(trans, root);
1982         } else {
1983                 ret = btrfs_end_transaction(trans, root);
1984         }
1985 out:
1986         return ret > 0 ? -EIO : ret;
1987 }
1988
1989 static const struct vm_operations_struct btrfs_file_vm_ops = {
1990         .fault          = filemap_fault,
1991         .page_mkwrite   = btrfs_page_mkwrite,
1992         .remap_pages    = generic_file_remap_pages,
1993 };
1994
1995 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1996 {
1997         struct address_space *mapping = filp->f_mapping;
1998
1999         if (!mapping->a_ops->readpage)
2000                 return -ENOEXEC;
2001
2002         file_accessed(filp);
2003         vma->vm_ops = &btrfs_file_vm_ops;
2004
2005         return 0;
2006 }
2007
2008 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2009                           int slot, u64 start, u64 end)
2010 {
2011         struct btrfs_file_extent_item *fi;
2012         struct btrfs_key key;
2013
2014         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2015                 return 0;
2016
2017         btrfs_item_key_to_cpu(leaf, &key, slot);
2018         if (key.objectid != btrfs_ino(inode) ||
2019             key.type != BTRFS_EXTENT_DATA_KEY)
2020                 return 0;
2021
2022         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2023
2024         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2025                 return 0;
2026
2027         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2028                 return 0;
2029
2030         if (key.offset == end)
2031                 return 1;
2032         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2033                 return 1;
2034         return 0;
2035 }
2036
2037 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2038                       struct btrfs_path *path, u64 offset, u64 end)
2039 {
2040         struct btrfs_root *root = BTRFS_I(inode)->root;
2041         struct extent_buffer *leaf;
2042         struct btrfs_file_extent_item *fi;
2043         struct extent_map *hole_em;
2044         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2045         struct btrfs_key key;
2046         int ret;
2047
2048         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2049                 goto out;
2050
2051         key.objectid = btrfs_ino(inode);
2052         key.type = BTRFS_EXTENT_DATA_KEY;
2053         key.offset = offset;
2054
2055         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2056         if (ret < 0)
2057                 return ret;
2058         BUG_ON(!ret);
2059
2060         leaf = path->nodes[0];
2061         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2062                 u64 num_bytes;
2063
2064                 path->slots[0]--;
2065                 fi = btrfs_item_ptr(leaf, path->slots[0],
2066                                     struct btrfs_file_extent_item);
2067                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2068                         end - offset;
2069                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2070                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2071                 btrfs_set_file_extent_offset(leaf, fi, 0);
2072                 btrfs_mark_buffer_dirty(leaf);
2073                 goto out;
2074         }
2075
2076         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2077                 u64 num_bytes;
2078
2079                 path->slots[0]++;
2080                 key.offset = offset;
2081                 btrfs_set_item_key_safe(root, path, &key);
2082                 fi = btrfs_item_ptr(leaf, path->slots[0],
2083                                     struct btrfs_file_extent_item);
2084                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2085                         offset;
2086                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2087                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2088                 btrfs_set_file_extent_offset(leaf, fi, 0);
2089                 btrfs_mark_buffer_dirty(leaf);
2090                 goto out;
2091         }
2092         btrfs_release_path(path);
2093
2094         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2095                                        0, 0, end - offset, 0, end - offset,
2096                                        0, 0, 0);
2097         if (ret)
2098                 return ret;
2099
2100 out:
2101         btrfs_release_path(path);
2102
2103         hole_em = alloc_extent_map();
2104         if (!hole_em) {
2105                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2106                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2107                         &BTRFS_I(inode)->runtime_flags);
2108         } else {
2109                 hole_em->start = offset;
2110                 hole_em->len = end - offset;
2111                 hole_em->ram_bytes = hole_em->len;
2112                 hole_em->orig_start = offset;
2113
2114                 hole_em->block_start = EXTENT_MAP_HOLE;
2115                 hole_em->block_len = 0;
2116                 hole_em->orig_block_len = 0;
2117                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2118                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2119                 hole_em->generation = trans->transid;
2120
2121                 do {
2122                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2123                         write_lock(&em_tree->lock);
2124                         ret = add_extent_mapping(em_tree, hole_em, 1);
2125                         write_unlock(&em_tree->lock);
2126                 } while (ret == -EEXIST);
2127                 free_extent_map(hole_em);
2128                 if (ret)
2129                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2130                                 &BTRFS_I(inode)->runtime_flags);
2131         }
2132
2133         return 0;
2134 }
2135
2136 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2137 {
2138         struct btrfs_root *root = BTRFS_I(inode)->root;
2139         struct extent_state *cached_state = NULL;
2140         struct btrfs_path *path;
2141         struct btrfs_block_rsv *rsv;
2142         struct btrfs_trans_handle *trans;
2143         u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2144         u64 lockend = round_down(offset + len,
2145                                  BTRFS_I(inode)->root->sectorsize) - 1;
2146         u64 cur_offset = lockstart;
2147         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2148         u64 drop_end;
2149         int ret = 0;
2150         int err = 0;
2151         int rsv_count;
2152         bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2153                           ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2154         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2155
2156         ret = btrfs_wait_ordered_range(inode, offset, len);
2157         if (ret)
2158                 return ret;
2159
2160         mutex_lock(&inode->i_mutex);
2161         /*
2162          * We needn't truncate any page which is beyond the end of the file
2163          * because we are sure there is no data there.
2164          */
2165         /*
2166          * Only do this if we are in the same page and we aren't doing the
2167          * entire page.
2168          */
2169         if (same_page && len < PAGE_CACHE_SIZE) {
2170                 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2171                         ret = btrfs_truncate_page(inode, offset, len, 0);
2172                 mutex_unlock(&inode->i_mutex);
2173                 return ret;
2174         }
2175
2176         /* zero back part of the first page */
2177         if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2178                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2179                 if (ret) {
2180                         mutex_unlock(&inode->i_mutex);
2181                         return ret;
2182                 }
2183         }
2184
2185         /* zero the front end of the last page */
2186         if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2187                 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2188                 if (ret) {
2189                         mutex_unlock(&inode->i_mutex);
2190                         return ret;
2191                 }
2192         }
2193
2194         if (lockend < lockstart) {
2195                 mutex_unlock(&inode->i_mutex);
2196                 return 0;
2197         }
2198
2199         while (1) {
2200                 struct btrfs_ordered_extent *ordered;
2201
2202                 truncate_pagecache_range(inode, lockstart, lockend);
2203
2204                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2205                                  0, &cached_state);
2206                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2207
2208                 /*
2209                  * We need to make sure we have no ordered extents in this range
2210                  * and nobody raced in and read a page in this range, if we did
2211                  * we need to try again.
2212                  */
2213                 if ((!ordered ||
2214                     (ordered->file_offset + ordered->len <= lockstart ||
2215                      ordered->file_offset > lockend)) &&
2216                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2217                                      lockend, EXTENT_UPTODATE, 0,
2218                                      cached_state)) {
2219                         if (ordered)
2220                                 btrfs_put_ordered_extent(ordered);
2221                         break;
2222                 }
2223                 if (ordered)
2224                         btrfs_put_ordered_extent(ordered);
2225                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2226                                      lockend, &cached_state, GFP_NOFS);
2227                 ret = btrfs_wait_ordered_range(inode, lockstart,
2228                                                lockend - lockstart + 1);
2229                 if (ret) {
2230                         mutex_unlock(&inode->i_mutex);
2231                         return ret;
2232                 }
2233         }
2234
2235         path = btrfs_alloc_path();
2236         if (!path) {
2237                 ret = -ENOMEM;
2238                 goto out;
2239         }
2240
2241         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2242         if (!rsv) {
2243                 ret = -ENOMEM;
2244                 goto out_free;
2245         }
2246         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2247         rsv->failfast = 1;
2248
2249         /*
2250          * 1 - update the inode
2251          * 1 - removing the extents in the range
2252          * 1 - adding the hole extent if no_holes isn't set
2253          */
2254         rsv_count = no_holes ? 2 : 3;
2255         trans = btrfs_start_transaction(root, rsv_count);
2256         if (IS_ERR(trans)) {
2257                 err = PTR_ERR(trans);
2258                 goto out_free;
2259         }
2260
2261         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2262                                       min_size);
2263         BUG_ON(ret);
2264         trans->block_rsv = rsv;
2265
2266         while (cur_offset < lockend) {
2267                 ret = __btrfs_drop_extents(trans, root, inode, path,
2268                                            cur_offset, lockend + 1,
2269                                            &drop_end, 1, 0, 0, NULL);
2270                 if (ret != -ENOSPC)
2271                         break;
2272
2273                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2274
2275                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2276                 if (ret) {
2277                         err = ret;
2278                         break;
2279                 }
2280
2281                 cur_offset = drop_end;
2282
2283                 ret = btrfs_update_inode(trans, root, inode);
2284                 if (ret) {
2285                         err = ret;
2286                         break;
2287                 }
2288
2289                 btrfs_end_transaction(trans, root);
2290                 btrfs_btree_balance_dirty(root);
2291
2292                 trans = btrfs_start_transaction(root, rsv_count);
2293                 if (IS_ERR(trans)) {
2294                         ret = PTR_ERR(trans);
2295                         trans = NULL;
2296                         break;
2297                 }
2298
2299                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2300                                               rsv, min_size);
2301                 BUG_ON(ret);    /* shouldn't happen */
2302                 trans->block_rsv = rsv;
2303         }
2304
2305         if (ret) {
2306                 err = ret;
2307                 goto out_trans;
2308         }
2309
2310         trans->block_rsv = &root->fs_info->trans_block_rsv;
2311         ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2312         if (ret) {
2313                 err = ret;
2314                 goto out_trans;
2315         }
2316
2317 out_trans:
2318         if (!trans)
2319                 goto out_free;
2320
2321         inode_inc_iversion(inode);
2322         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2323
2324         trans->block_rsv = &root->fs_info->trans_block_rsv;
2325         ret = btrfs_update_inode(trans, root, inode);
2326         btrfs_end_transaction(trans, root);
2327         btrfs_btree_balance_dirty(root);
2328 out_free:
2329         btrfs_free_path(path);
2330         btrfs_free_block_rsv(root, rsv);
2331 out:
2332         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2333                              &cached_state, GFP_NOFS);
2334         mutex_unlock(&inode->i_mutex);
2335         if (ret && !err)
2336                 err = ret;
2337         return err;
2338 }
2339
2340 static long btrfs_fallocate(struct file *file, int mode,
2341                             loff_t offset, loff_t len)
2342 {
2343         struct inode *inode = file_inode(file);
2344         struct extent_state *cached_state = NULL;
2345         struct btrfs_root *root = BTRFS_I(inode)->root;
2346         u64 cur_offset;
2347         u64 last_byte;
2348         u64 alloc_start;
2349         u64 alloc_end;
2350         u64 alloc_hint = 0;
2351         u64 locked_end;
2352         struct extent_map *em;
2353         int blocksize = BTRFS_I(inode)->root->sectorsize;
2354         int ret;
2355
2356         alloc_start = round_down(offset, blocksize);
2357         alloc_end = round_up(offset + len, blocksize);
2358
2359         /* Make sure we aren't being give some crap mode */
2360         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2361                 return -EOPNOTSUPP;
2362
2363         if (mode & FALLOC_FL_PUNCH_HOLE)
2364                 return btrfs_punch_hole(inode, offset, len);
2365
2366         /*
2367          * Make sure we have enough space before we do the
2368          * allocation.
2369          */
2370         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2371         if (ret)
2372                 return ret;
2373         if (root->fs_info->quota_enabled) {
2374                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2375                 if (ret)
2376                         goto out_reserve_fail;
2377         }
2378
2379         mutex_lock(&inode->i_mutex);
2380         ret = inode_newsize_ok(inode, alloc_end);
2381         if (ret)
2382                 goto out;
2383
2384         if (alloc_start > inode->i_size) {
2385                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2386                                         alloc_start);
2387                 if (ret)
2388                         goto out;
2389         } else {
2390                 /*
2391                  * If we are fallocating from the end of the file onward we
2392                  * need to zero out the end of the page if i_size lands in the
2393                  * middle of a page.
2394                  */
2395                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2396                 if (ret)
2397                         goto out;
2398         }
2399
2400         /*
2401          * wait for ordered IO before we have any locks.  We'll loop again
2402          * below with the locks held.
2403          */
2404         ret = btrfs_wait_ordered_range(inode, alloc_start,
2405                                        alloc_end - alloc_start);
2406         if (ret)
2407                 goto out;
2408
2409         locked_end = alloc_end - 1;
2410         while (1) {
2411                 struct btrfs_ordered_extent *ordered;
2412
2413                 /* the extent lock is ordered inside the running
2414                  * transaction
2415                  */
2416                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2417                                  locked_end, 0, &cached_state);
2418                 ordered = btrfs_lookup_first_ordered_extent(inode,
2419                                                             alloc_end - 1);
2420                 if (ordered &&
2421                     ordered->file_offset + ordered->len > alloc_start &&
2422                     ordered->file_offset < alloc_end) {
2423                         btrfs_put_ordered_extent(ordered);
2424                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2425                                              alloc_start, locked_end,
2426                                              &cached_state, GFP_NOFS);
2427                         /*
2428                          * we can't wait on the range with the transaction
2429                          * running or with the extent lock held
2430                          */
2431                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2432                                                        alloc_end - alloc_start);
2433                         if (ret)
2434                                 goto out;
2435                 } else {
2436                         if (ordered)
2437                                 btrfs_put_ordered_extent(ordered);
2438                         break;
2439                 }
2440         }
2441
2442         cur_offset = alloc_start;
2443         while (1) {
2444                 u64 actual_end;
2445
2446                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2447                                       alloc_end - cur_offset, 0);
2448                 if (IS_ERR_OR_NULL(em)) {
2449                         if (!em)
2450                                 ret = -ENOMEM;
2451                         else
2452                                 ret = PTR_ERR(em);
2453                         break;
2454                 }
2455                 last_byte = min(extent_map_end(em), alloc_end);
2456                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2457                 last_byte = ALIGN(last_byte, blocksize);
2458
2459                 if (em->block_start == EXTENT_MAP_HOLE ||
2460                     (cur_offset >= inode->i_size &&
2461                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2462                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2463                                                         last_byte - cur_offset,
2464                                                         1 << inode->i_blkbits,
2465                                                         offset + len,
2466                                                         &alloc_hint);
2467
2468                         if (ret < 0) {
2469                                 free_extent_map(em);
2470                                 break;
2471                         }
2472                 } else if (actual_end > inode->i_size &&
2473                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2474                         /*
2475                          * We didn't need to allocate any more space, but we
2476                          * still extended the size of the file so we need to
2477                          * update i_size.
2478                          */
2479                         inode->i_ctime = CURRENT_TIME;
2480                         i_size_write(inode, actual_end);
2481                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2482                 }
2483                 free_extent_map(em);
2484
2485                 cur_offset = last_byte;
2486                 if (cur_offset >= alloc_end) {
2487                         ret = 0;
2488                         break;
2489                 }
2490         }
2491         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2492                              &cached_state, GFP_NOFS);
2493 out:
2494         mutex_unlock(&inode->i_mutex);
2495         if (root->fs_info->quota_enabled)
2496                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2497 out_reserve_fail:
2498         /* Let go of our reservation. */
2499         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2500         return ret;
2501 }
2502
2503 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2504 {
2505         struct btrfs_root *root = BTRFS_I(inode)->root;
2506         struct extent_map *em = NULL;
2507         struct extent_state *cached_state = NULL;
2508         u64 lockstart;
2509         u64 lockend;
2510         u64 start;
2511         u64 len;
2512         int ret = 0;
2513
2514         if (inode->i_size == 0)
2515                 return -ENXIO;
2516
2517         /*
2518          * *offset can be negative, in this case we start finding DATA/HOLE from
2519          * the very start of the file.
2520          */
2521         start = max_t(loff_t, 0, *offset);
2522
2523         lockstart = round_down(start, root->sectorsize);
2524         lockend = round_up(i_size_read(inode), root->sectorsize);
2525         if (lockend <= lockstart)
2526                 lockend = lockstart + root->sectorsize;
2527         lockend--;
2528         len = lockend - lockstart + 1;
2529
2530         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2531                          &cached_state);
2532
2533         while (start < inode->i_size) {
2534                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2535                 if (IS_ERR(em)) {
2536                         ret = PTR_ERR(em);
2537                         em = NULL;
2538                         break;
2539                 }
2540
2541                 if (whence == SEEK_HOLE &&
2542                     (em->block_start == EXTENT_MAP_HOLE ||
2543                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2544                         break;
2545                 else if (whence == SEEK_DATA &&
2546                            (em->block_start != EXTENT_MAP_HOLE &&
2547                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2548                         break;
2549
2550                 start = em->start + em->len;
2551                 free_extent_map(em);
2552                 em = NULL;
2553                 cond_resched();
2554         }
2555         free_extent_map(em);
2556         if (!ret) {
2557                 if (whence == SEEK_DATA && start >= inode->i_size)
2558                         ret = -ENXIO;
2559                 else
2560                         *offset = min_t(loff_t, start, inode->i_size);
2561         }
2562         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2563                              &cached_state, GFP_NOFS);
2564         return ret;
2565 }
2566
2567 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2568 {
2569         struct inode *inode = file->f_mapping->host;
2570         int ret;
2571
2572         mutex_lock(&inode->i_mutex);
2573         switch (whence) {
2574         case SEEK_END:
2575         case SEEK_CUR:
2576                 offset = generic_file_llseek(file, offset, whence);
2577                 goto out;
2578         case SEEK_DATA:
2579         case SEEK_HOLE:
2580                 if (offset >= i_size_read(inode)) {
2581                         mutex_unlock(&inode->i_mutex);
2582                         return -ENXIO;
2583                 }
2584
2585                 ret = find_desired_extent(inode, &offset, whence);
2586                 if (ret) {
2587                         mutex_unlock(&inode->i_mutex);
2588                         return ret;
2589                 }
2590         }
2591
2592         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2593 out:
2594         mutex_unlock(&inode->i_mutex);
2595         return offset;
2596 }
2597
2598 const struct file_operations btrfs_file_operations = {
2599         .llseek         = btrfs_file_llseek,
2600         .read           = do_sync_read,
2601         .write          = do_sync_write,
2602         .aio_read       = generic_file_aio_read,
2603         .splice_read    = generic_file_splice_read,
2604         .aio_write      = btrfs_file_aio_write,
2605         .mmap           = btrfs_file_mmap,
2606         .open           = generic_file_open,
2607         .release        = btrfs_release_file,
2608         .fsync          = btrfs_sync_file,
2609         .fallocate      = btrfs_fallocate,
2610         .unlocked_ioctl = btrfs_ioctl,
2611 #ifdef CONFIG_COMPAT
2612         .compat_ioctl   = btrfs_ioctl,
2613 #endif
2614 };
2615
2616 void btrfs_auto_defrag_exit(void)
2617 {
2618         if (btrfs_inode_defrag_cachep)
2619                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2620 }
2621
2622 int btrfs_auto_defrag_init(void)
2623 {
2624         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2625                                         sizeof(struct inode_defrag), 0,
2626                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2627                                         NULL);
2628         if (!btrfs_inode_defrag_cachep)
2629                 return -ENOMEM;
2630
2631         return 0;
2632 }