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. There should be no need to mark the pages
 474                  * accessed as prepare_pages should have marked them accessed
 475                  * in prepare_pages via find_or_create_page()
 476                  */
 477                 ClearPageChecked(pages[i]);
 478                 unlock_page(pages[i]);
 479                 page_cache_release(pages[i]);
 480         }
 481 }
 482
 483 /*
 484  * after copy_from_user, pages need to be dirtied and we need to make
 485  * sure holes are created between the current EOF and the start of
 486  * any next extents (if required).
 487  *
 488  * this also makes the decision about creating an inline extent vs
 489  * doing real data extents, marking pages dirty and delalloc as required.
 490  */
 491 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
 492                              struct page **pages, size_t num_pages,
 493                              loff_t pos, size_t write_bytes,
 494                              struct extent_state **cached)
 495 {
 496         int err = 0;
 497         int i;
 498         u64 num_bytes;
 499         u64 start_pos;
 500         u64 end_of_last_block;
 501         u64 end_pos = pos + write_bytes;
 502         loff_t isize = i_size_read(inode);
 503
 504         start_pos = pos & ~((u64)root->sectorsize - 1);
 505         num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
 506
 507         end_of_last_block = start_pos + num_bytes - 1;
 508         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 509                                         cached);
 510         if (err)
 511                 return err;
 512
 513         for (i = 0; i < num_pages; i++) {
 514                 struct page *p = pages[i];
 515                 SetPageUptodate(p);
 516                 ClearPageChecked(p);
 517                 set_page_dirty(p);
 518         }
 519
 520         /*
 521          * we've only changed i_size in ram, and we haven't updated
 522          * the disk i_size.  There is no need to log the inode
 523          * at this time.
 524          */
 525         if (end_pos > isize)
 526                 i_size_write(inode, end_pos);
 527         return 0;
 528 }
 529
 530 /*
 531  * this drops all the extents in the cache that intersect the range
 532  * [start, end].  Existing extents are split as required.
 533  */
 534 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 535                              int skip_pinned)
 536 {
 537         struct extent_map *em;
 538         struct extent_map *split = NULL;
 539         struct extent_map *split2 = NULL;
 540         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 541         u64 len = end - start + 1;
 542         u64 gen;
 543         int ret;
 544         int testend = 1;
 545         unsigned long flags;
 546         int compressed = 0;
 547         bool modified;
 548
 549         WARN_ON(end < start);
 550         if (end == (u64)-1) {
 551                 len = (u64)-1;
 552                 testend = 0;
 553         }
 554         while (1) {
 555                 int no_splits = 0;
 556
 557                 modified = false;
 558                 if (!split)
 559                         split = alloc_extent_map();
 560                 if (!split2)
 561                         split2 = alloc_extent_map();
 562                 if (!split || !split2)
 563                         no_splits = 1;
 564
 565                 write_lock(&em_tree->lock);
 566                 em = lookup_extent_mapping(em_tree, start, len);
 567                 if (!em) {
 568                         write_unlock(&em_tree->lock);
 569                         break;
 570                 }
 571                 flags = em->flags;
 572                 gen = em->generation;
 573                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 574                         if (testend && em->start + em->len >= start + len) {
 575                                 free_extent_map(em);
 576                                 write_unlock(&em_tree->lock);
 577                                 break;
 578                         }
 579                         start = em->start + em->len;
 580                         if (testend)
 581                                 len = start + len - (em->start + em->len);
 582                         free_extent_map(em);
 583                         write_unlock(&em_tree->lock);
 584                         continue;
 585                 }
 586                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 587                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 588                 clear_bit(EXTENT_FLAG_LOGGING, &flags);
 589                 modified = !list_empty(&em->list);
 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                         replace_extent_mapping(em_tree, em, split, modified);
 621                         free_extent_map(split);
 622                         split = split2;
 623                         split2 = NULL;
 624                 }
 625                 if (testend && em->start + em->len > start + len) {
 626                         u64 diff = start + len - em->start;
 627
 628                         split->start = start + len;
 629                         split->len = em->start + em->len - (start + len);
 630                         split->bdev = em->bdev;
 631                         split->flags = flags;
 632                         split->compress_type = em->compress_type;
 633                         split->generation = gen;
 634
 635                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 636                                 split->orig_block_len = max(em->block_len,
 637                                                     em->orig_block_len);
 638
 639                                 split->ram_bytes = em->ram_bytes;
 640                                 if (compressed) {
 641                                         split->block_len = em->block_len;
 642                                         split->block_start = em->block_start;
 643                                         split->orig_start = em->orig_start;
 644                                 } else {
 645                                         split->block_len = split->len;
 646                                         split->block_start = em->block_start
 647                                                 + diff;
 648                                         split->orig_start = em->orig_start;
 649                                 }
 650                         } else {
 651                                 split->ram_bytes = split->len;
 652                                 split->orig_start = split->start;
 653                                 split->block_len = 0;
 654                                 split->block_start = em->block_start;
 655                                 split->orig_block_len = 0;
 656                         }
 657
 658                         if (extent_map_in_tree(em)) {
 659                                 replace_extent_mapping(em_tree, em, split,
 660                                                        modified);
 661                         } else {
 662                                 ret = add_extent_mapping(em_tree, split,
 663                                                          modified);
 664                                 ASSERT(ret == 0); /* Logic error */
 665                         }
 666                         free_extent_map(split);
 667                         split = NULL;
 668                 }
 669 next:
 670                 if (extent_map_in_tree(em))
 671                         remove_extent_mapping(em_tree, em);
 672                 write_unlock(&em_tree->lock);
 673
 674                 /* once for us */
 675                 free_extent_map(em);
 676                 /* once for the tree*/
 677                 free_extent_map(em);
 678         }
 679         if (split)
 680                 free_extent_map(split);
 681         if (split2)
 682                 free_extent_map(split2);
 683 }
 684
 685 /*
 686  * this is very complex, but the basic idea is to drop all extents
 687  * in the range start - end.  hint_block is filled in with a block number
 688  * that would be a good hint to the block allocator for this file.
 689  *
 690  * If an extent intersects the range but is not entirely inside the range
 691  * it is either truncated or split.  Anything entirely inside the range
 692  * is deleted from the tree.
 693  */
 694 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 695                          struct btrfs_root *root, struct inode *inode,
 696                          struct btrfs_path *path, u64 start, u64 end,
 697                          u64 *drop_end, int drop_cache,
 698                          int replace_extent,
 699                          u32 extent_item_size,
 700                          int *key_inserted)
 701 {
 702         struct extent_buffer *leaf;
 703         struct btrfs_file_extent_item *fi;
 704         struct btrfs_key key;
 705         struct btrfs_key new_key;
 706         u64 ino = btrfs_ino(inode);
 707         u64 search_start = start;
 708         u64 disk_bytenr = 0;
 709         u64 num_bytes = 0;
 710         u64 extent_offset = 0;
 711         u64 extent_end = 0;
 712         int del_nr = 0;
 713         int del_slot = 0;
 714         int extent_type;
 715         int recow;
 716         int ret;
 717         int modify_tree = -1;
 718         int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
 719         int found = 0;
 720         int leafs_visited = 0;
 721
 722         if (drop_cache)
 723                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 724
 725         if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
 726                 modify_tree = 0;
 727
 728         while (1) {
 729                 recow = 0;
 730                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
 731                                                search_start, modify_tree);
 732                 if (ret < 0)
 733                         break;
 734                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 735                         leaf = path->nodes[0];
 736                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 737                         if (key.objectid == ino &&
 738                             key.type == BTRFS_EXTENT_DATA_KEY)
 739                                 path->slots[0]--;
 740                 }
 741                 ret = 0;
 742                 leafs_visited++;
 743 next_slot:
 744                 leaf = path->nodes[0];
 745                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 746                         BUG_ON(del_nr > 0);
 747                         ret = btrfs_next_leaf(root, path);
 748                         if (ret < 0)
 749                                 break;
 750                         if (ret > 0) {
 751                                 ret = 0;
 752                                 break;
 753                         }
 754                         leafs_visited++;
 755                         leaf = path->nodes[0];
 756                         recow = 1;
 757                 }
 758
 759                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 760                 if (key.objectid > ino ||
 761                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 762                         break;
 763
 764                 fi = btrfs_item_ptr(leaf, path->slots[0],
 765                                     struct btrfs_file_extent_item);
 766                 extent_type = btrfs_file_extent_type(leaf, fi);
 767
 768                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 769                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 770                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 771                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 772                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 773                         extent_end = key.offset +
 774                                 btrfs_file_extent_num_bytes(leaf, fi);
 775                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 776                         extent_end = key.offset +
 777                                 btrfs_file_extent_inline_len(leaf,
 778                                                      path->slots[0], fi);
 779                 } else {
 780                         WARN_ON(1);
 781                         extent_end = search_start;
 782                 }
 783
 784                 if (extent_end <= search_start) {
 785                         path->slots[0]++;
 786                         goto next_slot;
 787                 }
 788
 789                 found = 1;
 790                 search_start = max(key.offset, start);
 791                 if (recow || !modify_tree) {
 792                         modify_tree = -1;
 793                         btrfs_release_path(path);
 794                         continue;
 795                 }
 796
 797                 /*
 798                  *     | - range to drop - |
 799                  *  | -------- extent -------- |
 800                  */
 801                 if (start > key.offset && end < extent_end) {
 802                         BUG_ON(del_nr > 0);
 803                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 804                                 ret = -EOPNOTSUPP;
 805                                 break;
 806                         }
 807
 808                         memcpy(&new_key, &key, sizeof(new_key));
 809                         new_key.offset = start;
 810                         ret = btrfs_duplicate_item(trans, root, path,
 811                                                    &new_key);
 812                         if (ret == -EAGAIN) {
 813                                 btrfs_release_path(path);
 814                                 continue;
 815                         }
 816                         if (ret < 0)
 817                                 break;
 818
 819                         leaf = path->nodes[0];
 820                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 821                                             struct btrfs_file_extent_item);
 822                         btrfs_set_file_extent_num_bytes(leaf, fi,
 823                                                         start - key.offset);
 824
 825                         fi = btrfs_item_ptr(leaf, path->slots[0],
 826                                             struct btrfs_file_extent_item);
 827
 828                         extent_offset += start - key.offset;
 829                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 830                         btrfs_set_file_extent_num_bytes(leaf, fi,
 831                                                         extent_end - start);
 832                         btrfs_mark_buffer_dirty(leaf);
 833
 834                         if (update_refs && disk_bytenr > 0) {
 835                                 ret = btrfs_inc_extent_ref(trans, root,
 836                                                 disk_bytenr, num_bytes, 0,
 837                                                 root->root_key.objectid,
 838                                                 new_key.objectid,
 839                                                 start - extent_offset, 0);
 840                                 BUG_ON(ret); /* -ENOMEM */
 841                         }
 842                         key.offset = start;
 843                 }
 844                 /*
 845                  *  | ---- range to drop ----- |
 846                  *      | -------- extent -------- |
 847                  */
 848                 if (start <= key.offset && end < extent_end) {
 849                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 850                                 ret = -EOPNOTSUPP;
 851                                 break;
 852                         }
 853
 854                         memcpy(&new_key, &key, sizeof(new_key));
 855                         new_key.offset = end;
 856                         btrfs_set_item_key_safe(root, path, &new_key);
 857
 858                         extent_offset += end - key.offset;
 859                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 860                         btrfs_set_file_extent_num_bytes(leaf, fi,
 861                                                         extent_end - end);
 862                         btrfs_mark_buffer_dirty(leaf);
 863                         if (update_refs && disk_bytenr > 0)
 864                                 inode_sub_bytes(inode, end - key.offset);
 865                         break;
 866                 }
 867
 868                 search_start = extent_end;
 869                 /*
 870                  *       | ---- range to drop ----- |
 871                  *  | -------- extent -------- |
 872                  */
 873                 if (start > key.offset && end >= extent_end) {
 874                         BUG_ON(del_nr > 0);
 875                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 876                                 ret = -EOPNOTSUPP;
 877                                 break;
 878                         }
 879
 880                         btrfs_set_file_extent_num_bytes(leaf, fi,
 881                                                         start - key.offset);
 882                         btrfs_mark_buffer_dirty(leaf);
 883                         if (update_refs && disk_bytenr > 0)
 884                                 inode_sub_bytes(inode, extent_end - start);
 885                         if (end == extent_end)
 886                                 break;
 887
 888                         path->slots[0]++;
 889                         goto next_slot;
 890                 }
 891
 892                 /*
 893                  *  | ---- range to drop ----- |
 894                  *    | ------ extent ------ |
 895                  */
 896                 if (start <= key.offset && end >= extent_end) {
 897                         if (del_nr == 0) {
 898                                 del_slot = path->slots[0];
 899                                 del_nr = 1;
 900                         } else {
 901                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 902                                 del_nr++;
 903                         }
 904
 905                         if (update_refs &&
 906                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 907                                 inode_sub_bytes(inode,
 908                                                 extent_end - key.offset);
 909                                 extent_end = ALIGN(extent_end,
 910                                                    root->sectorsize);
 911                         } else if (update_refs && disk_bytenr > 0) {
 912                                 ret = btrfs_free_extent(trans, root,
 913                                                 disk_bytenr, num_bytes, 0,
 914                                                 root->root_key.objectid,
 915                                                 key.objectid, key.offset -
 916                                                 extent_offset, 0);
 917                                 BUG_ON(ret); /* -ENOMEM */
 918                                 inode_sub_bytes(inode,
 919                                                 extent_end - key.offset);
 920                         }
 921
 922                         if (end == extent_end)
 923                                 break;
 924
 925                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 926                                 path->slots[0]++;
 927                                 goto next_slot;
 928                         }
 929
 930                         ret = btrfs_del_items(trans, root, path, del_slot,
 931                                               del_nr);
 932                         if (ret) {
 933                                 btrfs_abort_transaction(trans, root, ret);
 934                                 break;
 935                         }
 936
 937                         del_nr = 0;
 938                         del_slot = 0;
 939
 940                         btrfs_release_path(path);
 941                         continue;
 942                 }
 943
 944                 BUG_ON(1);
 945         }
 946
 947         if (!ret && del_nr > 0) {
 948                 /*
 949                  * Set path->slots[0] to first slot, so that after the delete
 950                  * if items are move off from our leaf to its immediate left or
 951                  * right neighbor leafs, we end up with a correct and adjusted
 952                  * path->slots[0] for our insertion (if replace_extent != 0).
 953                  */
 954                 path->slots[0] = del_slot;
 955                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 956                 if (ret)
 957                         btrfs_abort_transaction(trans, root, ret);
 958         }
 959
 960         leaf = path->nodes[0];
 961         /*
 962          * If btrfs_del_items() was called, it might have deleted a leaf, in
 963          * which case it unlocked our path, so check path->locks[0] matches a
 964          * write lock.
 965          */
 966         if (!ret && replace_extent && leafs_visited == 1 &&
 967             (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
 968              path->locks[0] == BTRFS_WRITE_LOCK) &&
 969             btrfs_leaf_free_space(root, leaf) >=
 970             sizeof(struct btrfs_item) + extent_item_size) {
 971
 972                 key.objectid = ino;
 973                 key.type = BTRFS_EXTENT_DATA_KEY;
 974                 key.offset = start;
 975                 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
 976                         struct btrfs_key slot_key;
 977
 978                         btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
 979                         if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
 980                                 path->slots[0]++;
 981                 }
 982                 setup_items_for_insert(root, path, &key,
 983                                        &extent_item_size,
 984                                        extent_item_size,
 985                                        sizeof(struct btrfs_item) +
 986                                        extent_item_size, 1);
 987                 *key_inserted = 1;
 988         }
 989
 990         if (!replace_extent || !(*key_inserted))
 991                 btrfs_release_path(path);
 992         if (drop_end)
 993                 *drop_end = found ? min(end, extent_end) : end;
 994         return ret;
 995 }
 996
 997 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 998                        struct btrfs_root *root, struct inode *inode, u64 start,
 999                        u64 end, int drop_cache)
1000 {
1001         struct btrfs_path *path;
1002         int ret;
1003
1004         path = btrfs_alloc_path();
1005         if (!path)
1006                 return -ENOMEM;
1007         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1008                                    drop_cache, 0, 0, NULL);
1009         btrfs_free_path(path);
1010         return ret;
1011 }
1012
1013 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1014                             u64 objectid, u64 bytenr, u64 orig_offset,
1015                             u64 *start, u64 *end)
1016 {
1017         struct btrfs_file_extent_item *fi;
1018         struct btrfs_key key;
1019         u64 extent_end;
1020
1021         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1022                 return 0;
1023
1024         btrfs_item_key_to_cpu(leaf, &key, slot);
1025         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1026                 return 0;
1027
1028         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1029         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1030             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1031             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1032             btrfs_file_extent_compression(leaf, fi) ||
1033             btrfs_file_extent_encryption(leaf, fi) ||
1034             btrfs_file_extent_other_encoding(leaf, fi))
1035                 return 0;
1036
1037         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1038         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1039                 return 0;
1040
1041         *start = key.offset;
1042         *end = extent_end;
1043         return 1;
1044 }
1045
1046 /*
1047  * Mark extent in the range start - end as written.
1048  *
1049  * This changes extent type from 'pre-allocated' to 'regular'. If only
1050  * part of extent is marked as written, the extent will be split into
1051  * two or three.
1052  */
1053 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1054                               struct inode *inode, u64 start, u64 end)
1055 {
1056         struct btrfs_root *root = BTRFS_I(inode)->root;
1057         struct extent_buffer *leaf;
1058         struct btrfs_path *path;
1059         struct btrfs_file_extent_item *fi;
1060         struct btrfs_key key;
1061         struct btrfs_key new_key;
1062         u64 bytenr;
1063         u64 num_bytes;
1064         u64 extent_end;
1065         u64 orig_offset;
1066         u64 other_start;
1067         u64 other_end;
1068         u64 split;
1069         int del_nr = 0;
1070         int del_slot = 0;
1071         int recow;
1072         int ret;
1073         u64 ino = btrfs_ino(inode);
1074
1075         path = btrfs_alloc_path();
1076         if (!path)
1077                 return -ENOMEM;
1078 again:
1079         recow = 0;
1080         split = start;
1081         key.objectid = ino;
1082         key.type = BTRFS_EXTENT_DATA_KEY;
1083         key.offset = split;
1084
1085         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1086         if (ret < 0)
1087                 goto out;
1088         if (ret > 0 && path->slots[0] > 0)
1089                 path->slots[0]--;
1090
1091         leaf = path->nodes[0];
1092         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1093         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1094         fi = btrfs_item_ptr(leaf, path->slots[0],
1095                             struct btrfs_file_extent_item);
1096         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1097                BTRFS_FILE_EXTENT_PREALLOC);
1098         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1099         BUG_ON(key.offset > start || extent_end < end);
1100
1101         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1102         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1103         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1104         memcpy(&new_key, &key, sizeof(new_key));
1105
1106         if (start == key.offset && end < extent_end) {
1107                 other_start = 0;
1108                 other_end = start;
1109                 if (extent_mergeable(leaf, path->slots[0] - 1,
1110                                      ino, bytenr, orig_offset,
1111                                      &other_start, &other_end)) {
1112                         new_key.offset = end;
1113                         btrfs_set_item_key_safe(root, path, &new_key);
1114                         fi = btrfs_item_ptr(leaf, path->slots[0],
1115                                             struct btrfs_file_extent_item);
1116                         btrfs_set_file_extent_generation(leaf, fi,
1117                                                          trans->transid);
1118                         btrfs_set_file_extent_num_bytes(leaf, fi,
1119                                                         extent_end - end);
1120                         btrfs_set_file_extent_offset(leaf, fi,
1121                                                      end - orig_offset);
1122                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1123                                             struct btrfs_file_extent_item);
1124                         btrfs_set_file_extent_generation(leaf, fi,
1125                                                          trans->transid);
1126                         btrfs_set_file_extent_num_bytes(leaf, fi,
1127                                                         end - other_start);
1128                         btrfs_mark_buffer_dirty(leaf);
1129                         goto out;
1130                 }
1131         }
1132
1133         if (start > key.offset && end == extent_end) {
1134                 other_start = end;
1135                 other_end = 0;
1136                 if (extent_mergeable(leaf, path->slots[0] + 1,
1137                                      ino, bytenr, orig_offset,
1138                                      &other_start, &other_end)) {
1139                         fi = btrfs_item_ptr(leaf, path->slots[0],
1140                                             struct btrfs_file_extent_item);
1141                         btrfs_set_file_extent_num_bytes(leaf, fi,
1142                                                         start - key.offset);
1143                         btrfs_set_file_extent_generation(leaf, fi,
1144                                                          trans->transid);
1145                         path->slots[0]++;
1146                         new_key.offset = start;
1147                         btrfs_set_item_key_safe(root, path, &new_key);
1148
1149                         fi = btrfs_item_ptr(leaf, path->slots[0],
1150                                             struct btrfs_file_extent_item);
1151                         btrfs_set_file_extent_generation(leaf, fi,
1152                                                          trans->transid);
1153                         btrfs_set_file_extent_num_bytes(leaf, fi,
1154                                                         other_end - start);
1155                         btrfs_set_file_extent_offset(leaf, fi,
1156                                                      start - orig_offset);
1157                         btrfs_mark_buffer_dirty(leaf);
1158                         goto out;
1159                 }
1160         }
1161
1162         while (start > key.offset || end < extent_end) {
1163                 if (key.offset == start)
1164                         split = end;
1165
1166                 new_key.offset = split;
1167                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1168                 if (ret == -EAGAIN) {
1169                         btrfs_release_path(path);
1170                         goto again;
1171                 }
1172                 if (ret < 0) {
1173                         btrfs_abort_transaction(trans, root, ret);
1174                         goto out;
1175                 }
1176
1177                 leaf = path->nodes[0];
1178                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1179                                     struct btrfs_file_extent_item);
1180                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1181                 btrfs_set_file_extent_num_bytes(leaf, fi,
1182                                                 split - key.offset);
1183
1184                 fi = btrfs_item_ptr(leaf, path->slots[0],
1185                                     struct btrfs_file_extent_item);
1186
1187                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1188                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1189                 btrfs_set_file_extent_num_bytes(leaf, fi,
1190                                                 extent_end - split);
1191                 btrfs_mark_buffer_dirty(leaf);
1192
1193                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1194                                            root->root_key.objectid,
1195                                            ino, orig_offset, 0);
1196                 BUG_ON(ret); /* -ENOMEM */
1197
1198                 if (split == start) {
1199                         key.offset = start;
1200                 } else {
1201                         BUG_ON(start != key.offset);
1202                         path->slots[0]--;
1203                         extent_end = end;
1204                 }
1205                 recow = 1;
1206         }
1207
1208         other_start = end;
1209         other_end = 0;
1210         if (extent_mergeable(leaf, path->slots[0] + 1,
1211                              ino, bytenr, orig_offset,
1212                              &other_start, &other_end)) {
1213                 if (recow) {
1214                         btrfs_release_path(path);
1215                         goto again;
1216                 }
1217                 extent_end = other_end;
1218                 del_slot = path->slots[0] + 1;
1219                 del_nr++;
1220                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1221                                         0, root->root_key.objectid,
1222                                         ino, orig_offset, 0);
1223                 BUG_ON(ret); /* -ENOMEM */
1224         }
1225         other_start = 0;
1226         other_end = start;
1227         if (extent_mergeable(leaf, path->slots[0] - 1,
1228                              ino, bytenr, orig_offset,
1229                              &other_start, &other_end)) {
1230                 if (recow) {
1231                         btrfs_release_path(path);
1232                         goto again;
1233                 }
1234                 key.offset = other_start;
1235                 del_slot = path->slots[0];
1236                 del_nr++;
1237                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1238                                         0, root->root_key.objectid,
1239                                         ino, orig_offset, 0);
1240                 BUG_ON(ret); /* -ENOMEM */
1241         }
1242         if (del_nr == 0) {
1243                 fi = btrfs_item_ptr(leaf, path->slots[0],
1244                            struct btrfs_file_extent_item);
1245                 btrfs_set_file_extent_type(leaf, fi,
1246                                            BTRFS_FILE_EXTENT_REG);
1247                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1248                 btrfs_mark_buffer_dirty(leaf);
1249         } else {
1250                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1251                            struct btrfs_file_extent_item);
1252                 btrfs_set_file_extent_type(leaf, fi,
1253                                            BTRFS_FILE_EXTENT_REG);
1254                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1255                 btrfs_set_file_extent_num_bytes(leaf, fi,
1256                                                 extent_end - key.offset);
1257                 btrfs_mark_buffer_dirty(leaf);
1258
1259                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1260                 if (ret < 0) {
1261                         btrfs_abort_transaction(trans, root, ret);
1262                         goto out;
1263                 }
1264         }
1265 out:
1266         btrfs_free_path(path);
1267         return 0;
1268 }
1269
1270 /*
1271  * on error we return an unlocked page and the error value
1272  * on success we return a locked page and 0
1273  */
1274 static int prepare_uptodate_page(struct page *page, u64 pos,
1275                                  bool force_uptodate)
1276 {
1277         int ret = 0;
1278
1279         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1280             !PageUptodate(page)) {
1281                 ret = btrfs_readpage(NULL, page);
1282                 if (ret)
1283                         return ret;
1284                 lock_page(page);
1285                 if (!PageUptodate(page)) {
1286                         unlock_page(page);
1287                         return -EIO;
1288                 }
1289         }
1290         return 0;
1291 }
1292
1293 /*
1294  * this just gets pages into the page cache and locks them down.
1295  */
1296 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1297                                   size_t num_pages, loff_t pos,
1298                                   size_t write_bytes, bool force_uptodate)
1299 {
1300         int i;
1301         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1302         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1303         int err = 0;
1304         int faili;
1305
1306         for (i = 0; i < num_pages; i++) {
1307                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1308                                                mask | __GFP_WRITE);
1309                 if (!pages[i]) {
1310                         faili = i - 1;
1311                         err = -ENOMEM;
1312                         goto fail;
1313                 }
1314
1315                 if (i == 0)
1316                         err = prepare_uptodate_page(pages[i], pos,
1317                                                     force_uptodate);
1318                 if (i == num_pages - 1)
1319                         err = prepare_uptodate_page(pages[i],
1320                                                     pos + write_bytes, false);
1321                 if (err) {
1322                         page_cache_release(pages[i]);
1323                         faili = i - 1;
1324                         goto fail;
1325                 }
1326                 wait_on_page_writeback(pages[i]);
1327         }
1328
1329         return 0;
1330 fail:
1331         while (faili >= 0) {
1332                 unlock_page(pages[faili]);
1333                 page_cache_release(pages[faili]);
1334                 faili--;
1335         }
1336         return err;
1337
1338 }
1339
1340 /*
1341  * This function locks the extent and properly waits for data=ordered extents
1342  * to finish before allowing the pages to be modified if need.
1343  *
1344  * The return value:
1345  * 1 - the extent is locked
1346  * 0 - the extent is not locked, and everything is OK
1347  * -EAGAIN - need re-prepare the pages
1348  * the other < 0 number - Something wrong happens
1349  */
1350 static noinline int
1351 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1352                                 size_t num_pages, loff_t pos,
1353                                 u64 *lockstart, u64 *lockend,
1354                                 struct extent_state **cached_state)
1355 {
1356         u64 start_pos;
1357         u64 last_pos;
1358         int i;
1359         int ret = 0;
1360
1361         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1362         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1363
1364         if (start_pos < inode->i_size) {
1365                 struct btrfs_ordered_extent *ordered;
1366                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1367                                  start_pos, last_pos, 0, cached_state);
1368                 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1369                                                      last_pos - start_pos + 1);
1370                 if (ordered &&
1371                     ordered->file_offset + ordered->len > start_pos &&
1372                     ordered->file_offset <= last_pos) {
1373                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1374                                              start_pos, last_pos,
1375                                              cached_state, GFP_NOFS);
1376                         for (i = 0; i < num_pages; i++) {
1377                                 unlock_page(pages[i]);
1378                                 page_cache_release(pages[i]);
1379                         }
1380                         btrfs_start_ordered_extent(inode, ordered, 1);
1381                         btrfs_put_ordered_extent(ordered);
1382                         return -EAGAIN;
1383                 }
1384                 if (ordered)
1385                         btrfs_put_ordered_extent(ordered);
1386
1387                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1388                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1389                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1390                                   0, 0, cached_state, GFP_NOFS);
1391                 *lockstart = start_pos;
1392                 *lockend = last_pos;
1393                 ret = 1;
1394         }
1395
1396         for (i = 0; i < num_pages; i++) {
1397                 if (clear_page_dirty_for_io(pages[i]))
1398                         account_page_redirty(pages[i]);
1399                 set_page_extent_mapped(pages[i]);
1400                 WARN_ON(!PageLocked(pages[i]));
1401         }
1402
1403         return ret;
1404 }
1405
1406 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1407                                     size_t *write_bytes)
1408 {
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410         struct btrfs_ordered_extent *ordered;
1411         u64 lockstart, lockend;
1412         u64 num_bytes;
1413         int ret;
1414
1415         ret = btrfs_start_nocow_write(root);
1416         if (!ret)
1417                 return -ENOSPC;
1418
1419         lockstart = round_down(pos, root->sectorsize);
1420         lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1421
1422         while (1) {
1423                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1424                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1425                                                      lockend - lockstart + 1);
1426                 if (!ordered) {
1427                         break;
1428                 }
1429                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1430                 btrfs_start_ordered_extent(inode, ordered, 1);
1431                 btrfs_put_ordered_extent(ordered);
1432         }
1433
1434         num_bytes = lockend - lockstart + 1;
1435         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1436         if (ret <= 0) {
1437                 ret = 0;
1438                 btrfs_end_nocow_write(root);
1439         } else {
1440                 *write_bytes = min_t(size_t, *write_bytes ,
1441                                      num_bytes - pos + lockstart);
1442         }
1443
1444         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1445
1446         return ret;
1447 }
1448
1449 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1450                                                struct iov_iter *i,
1451                                                loff_t pos)
1452 {
1453         struct inode *inode = file_inode(file);
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         struct page **pages = NULL;
1456         struct extent_state *cached_state = NULL;
1457         u64 release_bytes = 0;
1458         u64 lockstart;
1459         u64 lockend;
1460         unsigned long first_index;
1461         size_t num_written = 0;
1462         int nrptrs;
1463         int ret = 0;
1464         bool only_release_metadata = false;
1465         bool force_page_uptodate = false;
1466         bool need_unlock;
1467
1468         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1469                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1470                      (sizeof(struct page *)));
1471         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1472         nrptrs = max(nrptrs, 8);
1473         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1474         if (!pages)
1475                 return -ENOMEM;
1476
1477         first_index = pos >> PAGE_CACHE_SHIFT;
1478
1479         while (iov_iter_count(i) > 0) {
1480                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1481                 size_t write_bytes = min(iov_iter_count(i),
1482                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1483                                          offset);
1484                 size_t num_pages = (write_bytes + offset +
1485                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1486                 size_t reserve_bytes;
1487                 size_t dirty_pages;
1488                 size_t copied;
1489
1490                 WARN_ON(num_pages > nrptrs);
1491
1492                 /*
1493                  * Fault pages before locking them in prepare_pages
1494                  * to avoid recursive lock
1495                  */
1496                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1497                         ret = -EFAULT;
1498                         break;
1499                 }
1500
1501                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1502                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1503                 if (ret == -ENOSPC &&
1504                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1505                                               BTRFS_INODE_PREALLOC))) {
1506                         ret = check_can_nocow(inode, pos, &write_bytes);
1507                         if (ret > 0) {
1508                                 only_release_metadata = true;
1509                                 /*
1510                                  * our prealloc extent may be smaller than
1511                                  * write_bytes, so scale down.
1512                                  */
1513                                 num_pages = (write_bytes + offset +
1514                                              PAGE_CACHE_SIZE - 1) >>
1515                                         PAGE_CACHE_SHIFT;
1516                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1517                                 ret = 0;
1518                         } else {
1519                                 ret = -ENOSPC;
1520                         }
1521                 }
1522
1523                 if (ret)
1524                         break;
1525
1526                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1527                 if (ret) {
1528                         if (!only_release_metadata)
1529                                 btrfs_free_reserved_data_space(inode,
1530                                                                reserve_bytes);
1531                         else
1532                                 btrfs_end_nocow_write(root);
1533                         break;
1534                 }
1535
1536                 release_bytes = reserve_bytes;
1537                 need_unlock = false;
1538 again:
1539                 /*
1540                  * This is going to setup the pages array with the number of
1541                  * pages we want, so we don't really need to worry about the
1542                  * contents of pages from loop to loop
1543                  */
1544                 ret = prepare_pages(inode, pages, num_pages,
1545                                     pos, write_bytes,
1546                                     force_page_uptodate);
1547                 if (ret)
1548                         break;
1549
1550                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1551                                                       pos, &lockstart, &lockend,
1552                                                       &cached_state);
1553                 if (ret < 0) {
1554                         if (ret == -EAGAIN)
1555                                 goto again;
1556                         break;
1557                 } else if (ret > 0) {
1558                         need_unlock = true;
1559                         ret = 0;
1560                 }
1561
1562                 copied = btrfs_copy_from_user(pos, num_pages,
1563                                            write_bytes, pages, i);
1564
1565                 /*
1566                  * if we have trouble faulting in the pages, fall
1567                  * back to one page at a time
1568                  */
1569                 if (copied < write_bytes)
1570                         nrptrs = 1;
1571
1572                 if (copied == 0) {
1573                         force_page_uptodate = true;
1574                         dirty_pages = 0;
1575                 } else {
1576                         force_page_uptodate = false;
1577                         dirty_pages = (copied + offset +
1578                                        PAGE_CACHE_SIZE - 1) >>
1579                                        PAGE_CACHE_SHIFT;
1580                 }
1581
1582                 /*
1583                  * If we had a short copy we need to release the excess delaloc
1584                  * bytes we reserved.  We need to increment outstanding_extents
1585                  * because btrfs_delalloc_release_space will decrement it, but
1586                  * we still have an outstanding extent for the chunk we actually
1587                  * managed to copy.
1588                  */
1589                 if (num_pages > dirty_pages) {
1590                         release_bytes = (num_pages - dirty_pages) <<
1591                                 PAGE_CACHE_SHIFT;
1592                         if (copied > 0) {
1593                                 spin_lock(&BTRFS_I(inode)->lock);
1594                                 BTRFS_I(inode)->outstanding_extents++;
1595                                 spin_unlock(&BTRFS_I(inode)->lock);
1596                         }
1597                         if (only_release_metadata)
1598                                 btrfs_delalloc_release_metadata(inode,
1599                                                                 release_bytes);
1600                         else
1601                                 btrfs_delalloc_release_space(inode,
1602                                                              release_bytes);
1603                 }
1604
1605                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1606
1607                 if (copied > 0)
1608                         ret = btrfs_dirty_pages(root, inode, pages,
1609                                                 dirty_pages, pos, copied,
1610                                                 NULL);
1611                 if (need_unlock)
1612                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1613                                              lockstart, lockend, &cached_state,
1614                                              GFP_NOFS);
1615                 if (ret) {
1616                         btrfs_drop_pages(pages, num_pages);
1617                         break;
1618                 }
1619
1620                 release_bytes = 0;
1621                 if (only_release_metadata)
1622                         btrfs_end_nocow_write(root);
1623
1624                 if (only_release_metadata && copied > 0) {
1625                         u64 lockstart = round_down(pos, root->sectorsize);
1626                         u64 lockend = lockstart +
1627                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1628
1629                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1630                                        lockend, EXTENT_NORESERVE, NULL,
1631                                        NULL, GFP_NOFS);
1632                         only_release_metadata = false;
1633                 }
1634
1635                 btrfs_drop_pages(pages, num_pages);
1636
1637                 cond_resched();
1638
1639                 balance_dirty_pages_ratelimited(inode->i_mapping);
1640                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1641                         btrfs_btree_balance_dirty(root);
1642
1643                 pos += copied;
1644                 num_written += copied;
1645         }
1646
1647         kfree(pages);
1648
1649         if (release_bytes) {
1650                 if (only_release_metadata) {
1651                         btrfs_end_nocow_write(root);
1652                         btrfs_delalloc_release_metadata(inode, release_bytes);
1653                 } else {
1654                         btrfs_delalloc_release_space(inode, release_bytes);
1655                 }
1656         }
1657
1658         return num_written ? num_written : ret;
1659 }
1660
1661 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1662                                     const struct iovec *iov,
1663                                     unsigned long nr_segs, loff_t pos,
1664                                     size_t count, size_t ocount)
1665 {
1666         struct file *file = iocb->ki_filp;
1667         struct iov_iter i;
1668         ssize_t written;
1669         ssize_t written_buffered;
1670         loff_t endbyte;
1671         int err;
1672
1673         written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
1674                                             count, ocount);
1675
1676         if (written < 0 || written == count)
1677                 return written;
1678
1679         pos += written;
1680         count -= written;
1681         iov_iter_init(&i, iov, nr_segs, count, written);
1682         written_buffered = __btrfs_buffered_write(file, &i, pos);
1683         if (written_buffered < 0) {
1684                 err = written_buffered;
1685                 goto out;
1686         }
1687         endbyte = pos + written_buffered - 1;
1688         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1689         if (err)
1690                 goto out;
1691         written += written_buffered;
1692         iocb->ki_pos = pos + written_buffered;
1693         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1694                                  endbyte >> PAGE_CACHE_SHIFT);
1695 out:
1696         return written ? written : err;
1697 }
1698
1699 static void update_time_for_write(struct inode *inode)
1700 {
1701         struct timespec now;
1702
1703         if (IS_NOCMTIME(inode))
1704                 return;
1705
1706         now = current_fs_time(inode->i_sb);
1707         if (!timespec_equal(&inode->i_mtime, &now))
1708                 inode->i_mtime = now;
1709
1710         if (!timespec_equal(&inode->i_ctime, &now))
1711                 inode->i_ctime = now;
1712
1713         if (IS_I_VERSION(inode))
1714                 inode_inc_iversion(inode);
1715 }
1716
1717 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1718                                     const struct iovec *iov,
1719                                     unsigned long nr_segs, loff_t pos)
1720 {
1721         struct file *file = iocb->ki_filp;
1722         struct inode *inode = file_inode(file);
1723         struct btrfs_root *root = BTRFS_I(inode)->root;
1724         u64 start_pos;
1725         u64 end_pos;
1726         ssize_t num_written = 0;
1727         ssize_t err = 0;
1728         size_t count, ocount;
1729         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1730
1731         mutex_lock(&inode->i_mutex);
1732
1733         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1734         if (err) {
1735                 mutex_unlock(&inode->i_mutex);
1736                 goto out;
1737         }
1738         count = ocount;
1739
1740         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1741         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1742         if (err) {
1743                 mutex_unlock(&inode->i_mutex);
1744                 goto out;
1745         }
1746
1747         if (count == 0) {
1748                 mutex_unlock(&inode->i_mutex);
1749                 goto out;
1750         }
1751
1752         err = file_remove_suid(file);
1753         if (err) {
1754                 mutex_unlock(&inode->i_mutex);
1755                 goto out;
1756         }
1757
1758         /*
1759          * If BTRFS flips readonly due to some impossible error
1760          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1761          * although we have opened a file as writable, we have
1762          * to stop this write operation to ensure FS consistency.
1763          */
1764         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1765                 mutex_unlock(&inode->i_mutex);
1766                 err = -EROFS;
1767                 goto out;
1768         }
1769
1770         /*
1771          * We reserve space for updating the inode when we reserve space for the
1772          * extent we are going to write, so we will enospc out there.  We don't
1773          * need to start yet another transaction to update the inode as we will
1774          * update the inode when we finish writing whatever data we write.
1775          */
1776         update_time_for_write(inode);
1777
1778         start_pos = round_down(pos, root->sectorsize);
1779         if (start_pos > i_size_read(inode)) {
1780                 /* Expand hole size to cover write data, preventing empty gap */
1781                 end_pos = round_up(pos + count, root->sectorsize);
1782                 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1783                 if (err) {
1784                         mutex_unlock(&inode->i_mutex);
1785                         goto out;
1786                 }
1787         }
1788
1789         if (sync)
1790                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1791
1792         if (unlikely(file->f_flags & O_DIRECT)) {
1793                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1794                                                    pos, count, ocount);
1795         } else {
1796                 struct iov_iter i;
1797
1798                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1799
1800                 num_written = __btrfs_buffered_write(file, &i, pos);
1801                 if (num_written > 0)
1802                         iocb->ki_pos = pos + num_written;
1803         }
1804
1805         mutex_unlock(&inode->i_mutex);
1806
1807         /*
1808          * we want to make sure fsync finds this change
1809          * but we haven't joined a transaction running right now.
1810          *
1811          * Later on, someone is sure to update the inode and get the
1812          * real transid recorded.
1813          *
1814          * We set last_trans now to the fs_info generation + 1,
1815          * this will either be one more than the running transaction
1816          * or the generation used for the next transaction if there isn't
1817          * one running right now.
1818          *
1819          * We also have to set last_sub_trans to the current log transid,
1820          * otherwise subsequent syncs to a file that's been synced in this
1821          * transaction will appear to have already occured.
1822          */
1823         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1824         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1825         if (num_written > 0) {
1826                 err = generic_write_sync(file, pos, num_written);
1827                 if (err < 0)
1828                         num_written = err;
1829         }
1830
1831         if (sync)
1832                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1833 out:
1834         current->backing_dev_info = NULL;
1835         return num_written ? num_written : err;
1836 }
1837
1838 int btrfs_release_file(struct inode *inode, struct file *filp)
1839 {
1840         /*
1841          * ordered_data_close is set by settattr when we are about to truncate
1842          * a file from a non-zero size to a zero size.  This tries to
1843          * flush down new bytes that may have been written if the
1844          * application were using truncate to replace a file in place.
1845          */
1846         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1847                                &BTRFS_I(inode)->runtime_flags)) {
1848                 struct btrfs_trans_handle *trans;
1849                 struct btrfs_root *root = BTRFS_I(inode)->root;
1850
1851                 /*
1852                  * We need to block on a committing transaction to keep us from
1853                  * throwing a ordered operation on to the list and causing
1854                  * something like sync to deadlock trying to flush out this
1855                  * inode.
1856                  */
1857                 trans = btrfs_start_transaction(root, 0);
1858                 if (IS_ERR(trans))
1859                         return PTR_ERR(trans);
1860                 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1861                 btrfs_end_transaction(trans, root);
1862                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1863                         filemap_flush(inode->i_mapping);
1864         }
1865         if (filp->private_data)
1866                 btrfs_ioctl_trans_end(filp);
1867         return 0;
1868 }
1869
1870 /*
1871  * fsync call for both files and directories.  This logs the inode into
1872  * the tree log instead of forcing full commits whenever possible.
1873  *
1874  * It needs to call filemap_fdatawait so that all ordered extent updates are
1875  * in the metadata btree are up to date for copying to the log.
1876  *
1877  * It drops the inode mutex before doing the tree log commit.  This is an
1878  * important optimization for directories because holding the mutex prevents
1879  * new operations on the dir while we write to disk.
1880  */
1881 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1882 {
1883         struct dentry *dentry = file->f_path.dentry;
1884         struct inode *inode = dentry->d_inode;
1885         struct btrfs_root *root = BTRFS_I(inode)->root;
1886         struct btrfs_trans_handle *trans;
1887         struct btrfs_log_ctx ctx;
1888         int ret = 0;
1889         bool full_sync = 0;
1890
1891         trace_btrfs_sync_file(file, datasync);
1892
1893         /*
1894          * We write the dirty pages in the range and wait until they complete
1895          * out of the ->i_mutex. If so, we can flush the dirty pages by
1896          * multi-task, and make the performance up.  See
1897          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1898          */
1899         atomic_inc(&BTRFS_I(inode)->sync_writers);
1900         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1901         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1902                              &BTRFS_I(inode)->runtime_flags))
1903                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1904         atomic_dec(&BTRFS_I(inode)->sync_writers);
1905         if (ret)
1906                 return ret;
1907
1908         mutex_lock(&inode->i_mutex);
1909
1910         /*
1911          * We flush the dirty pages again to avoid some dirty pages in the
1912          * range being left.
1913          */
1914         atomic_inc(&root->log_batch);
1915         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1916                              &BTRFS_I(inode)->runtime_flags);
1917         if (full_sync) {
1918                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1919                 if (ret) {
1920                         mutex_unlock(&inode->i_mutex);
1921                         goto out;
1922                 }
1923         }
1924         atomic_inc(&root->log_batch);
1925
1926         /*
1927          * check the transaction that last modified this inode
1928          * and see if its already been committed
1929          */
1930         if (!BTRFS_I(inode)->last_trans) {
1931                 mutex_unlock(&inode->i_mutex);
1932                 goto out;
1933         }
1934
1935         /*
1936          * if the last transaction that changed this file was before
1937          * the current transaction, we can bail out now without any
1938          * syncing
1939          */
1940         smp_mb();
1941         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1942             BTRFS_I(inode)->last_trans <=
1943             root->fs_info->last_trans_committed) {
1944                 BTRFS_I(inode)->last_trans = 0;
1945
1946                 /*
1947                  * We'v had everything committed since the last time we were
1948                  * modified so clear this flag in case it was set for whatever
1949                  * reason, it's no longer relevant.
1950                  */
1951                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1952                           &BTRFS_I(inode)->runtime_flags);
1953                 mutex_unlock(&inode->i_mutex);
1954                 goto out;
1955         }
1956
1957         /*
1958          * ok we haven't committed the transaction yet, lets do a commit
1959          */
1960         if (file->private_data)
1961                 btrfs_ioctl_trans_end(file);
1962
1963         /*
1964          * We use start here because we will need to wait on the IO to complete
1965          * in btrfs_sync_log, which could require joining a transaction (for
1966          * example checking cross references in the nocow path).  If we use join
1967          * here we could get into a situation where we're waiting on IO to
1968          * happen that is blocked on a transaction trying to commit.  With start
1969          * we inc the extwriter counter, so we wait for all extwriters to exit
1970          * before we start blocking join'ers.  This comment is to keep somebody
1971          * from thinking they are super smart and changing this to
1972          * btrfs_join_transaction *cough*Josef*cough*.
1973          */
1974         trans = btrfs_start_transaction(root, 0);
1975         if (IS_ERR(trans)) {
1976                 ret = PTR_ERR(trans);
1977                 mutex_unlock(&inode->i_mutex);
1978                 goto out;
1979         }
1980         trans->sync = true;
1981
1982         btrfs_init_log_ctx(&ctx);
1983
1984         ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1985         if (ret < 0) {
1986                 /* Fallthrough and commit/free transaction. */
1987                 ret = 1;
1988         }
1989
1990         /* we've logged all the items and now have a consistent
1991          * version of the file in the log.  It is possible that
1992          * someone will come in and modify the file, but that's
1993          * fine because the log is consistent on disk, and we
1994          * have references to all of the file's extents
1995          *
1996          * It is possible that someone will come in and log the
1997          * file again, but that will end up using the synchronization
1998          * inside btrfs_sync_log to keep things safe.
1999          */
2000         mutex_unlock(&inode->i_mutex);
2001
2002         if (ret != BTRFS_NO_LOG_SYNC) {
2003                 if (!ret) {
2004                         ret = btrfs_sync_log(trans, root, &ctx);
2005                         if (!ret) {
2006                                 ret = btrfs_end_transaction(trans, root);
2007                                 goto out;
2008                         }
2009                 }
2010                 if (!full_sync) {
2011                         ret = btrfs_wait_ordered_range(inode, start,
2012                                                        end - start + 1);
2013                         if (ret)
2014                                 goto out;
2015                 }
2016                 ret = btrfs_commit_transaction(trans, root);
2017         } else {
2018                 ret = btrfs_end_transaction(trans, root);
2019         }
2020 out:
2021         return ret > 0 ? -EIO : ret;
2022 }
2023
2024 static const struct vm_operations_struct btrfs_file_vm_ops = {
2025         .fault          = filemap_fault,
2026         .map_pages      = filemap_map_pages,
2027         .page_mkwrite   = btrfs_page_mkwrite,
2028         .remap_pages    = generic_file_remap_pages,
2029 };
2030
2031 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
2032 {
2033         struct address_space *mapping = filp->f_mapping;
2034
2035         if (!mapping->a_ops->readpage)
2036                 return -ENOEXEC;
2037
2038         file_accessed(filp);
2039         vma->vm_ops = &btrfs_file_vm_ops;
2040
2041         return 0;
2042 }
2043
2044 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2045                           int slot, u64 start, u64 end)
2046 {
2047         struct btrfs_file_extent_item *fi;
2048         struct btrfs_key key;
2049
2050         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2051                 return 0;
2052
2053         btrfs_item_key_to_cpu(leaf, &key, slot);
2054         if (key.objectid != btrfs_ino(inode) ||
2055             key.type != BTRFS_EXTENT_DATA_KEY)
2056                 return 0;
2057
2058         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2059
2060         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2061                 return 0;
2062
2063         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2064                 return 0;
2065
2066         if (key.offset == end)
2067                 return 1;
2068         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2069                 return 1;
2070         return 0;
2071 }
2072
2073 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2074                       struct btrfs_path *path, u64 offset, u64 end)
2075 {
2076         struct btrfs_root *root = BTRFS_I(inode)->root;
2077         struct extent_buffer *leaf;
2078         struct btrfs_file_extent_item *fi;
2079         struct extent_map *hole_em;
2080         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2081         struct btrfs_key key;
2082         int ret;
2083
2084         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2085                 goto out;
2086
2087         key.objectid = btrfs_ino(inode);
2088         key.type = BTRFS_EXTENT_DATA_KEY;
2089         key.offset = offset;
2090
2091         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2092         if (ret < 0)
2093                 return ret;
2094         BUG_ON(!ret);
2095
2096         leaf = path->nodes[0];
2097         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2098                 u64 num_bytes;
2099
2100                 path->slots[0]--;
2101                 fi = btrfs_item_ptr(leaf, path->slots[0],
2102                                     struct btrfs_file_extent_item);
2103                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2104                         end - offset;
2105                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2106                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2107                 btrfs_set_file_extent_offset(leaf, fi, 0);
2108                 btrfs_mark_buffer_dirty(leaf);
2109                 goto out;
2110         }
2111
2112         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2113                 u64 num_bytes;
2114
2115                 path->slots[0]++;
2116                 key.offset = offset;
2117                 btrfs_set_item_key_safe(root, path, &key);
2118                 fi = btrfs_item_ptr(leaf, path->slots[0],
2119                                     struct btrfs_file_extent_item);
2120                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2121                         offset;
2122                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2123                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2124                 btrfs_set_file_extent_offset(leaf, fi, 0);
2125                 btrfs_mark_buffer_dirty(leaf);
2126                 goto out;
2127         }
2128         btrfs_release_path(path);
2129
2130         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2131                                        0, 0, end - offset, 0, end - offset,
2132                                        0, 0, 0);
2133         if (ret)
2134                 return ret;
2135
2136 out:
2137         btrfs_release_path(path);
2138
2139         hole_em = alloc_extent_map();
2140         if (!hole_em) {
2141                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2142                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2143                         &BTRFS_I(inode)->runtime_flags);
2144         } else {
2145                 hole_em->start = offset;
2146                 hole_em->len = end - offset;
2147                 hole_em->ram_bytes = hole_em->len;
2148                 hole_em->orig_start = offset;
2149
2150                 hole_em->block_start = EXTENT_MAP_HOLE;
2151                 hole_em->block_len = 0;
2152                 hole_em->orig_block_len = 0;
2153                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2154                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2155                 hole_em->generation = trans->transid;
2156
2157                 do {
2158                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2159                         write_lock(&em_tree->lock);
2160                         ret = add_extent_mapping(em_tree, hole_em, 1);
2161                         write_unlock(&em_tree->lock);
2162                 } while (ret == -EEXIST);
2163                 free_extent_map(hole_em);
2164                 if (ret)
2165                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2166                                 &BTRFS_I(inode)->runtime_flags);
2167         }
2168
2169         return 0;
2170 }
2171
2172 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2173 {
2174         struct btrfs_root *root = BTRFS_I(inode)->root;
2175         struct extent_state *cached_state = NULL;
2176         struct btrfs_path *path;
2177         struct btrfs_block_rsv *rsv;
2178         struct btrfs_trans_handle *trans;
2179         u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2180         u64 lockend = round_down(offset + len,
2181                                  BTRFS_I(inode)->root->sectorsize) - 1;
2182         u64 cur_offset = lockstart;
2183         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2184         u64 drop_end;
2185         int ret = 0;
2186         int err = 0;
2187         int rsv_count;
2188         bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2189                           ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2190         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2191         u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2192
2193         ret = btrfs_wait_ordered_range(inode, offset, len);
2194         if (ret)
2195                 return ret;
2196
2197         mutex_lock(&inode->i_mutex);
2198         /*
2199          * We needn't truncate any page which is beyond the end of the file
2200          * because we are sure there is no data there.
2201          */
2202         /*
2203          * Only do this if we are in the same page and we aren't doing the
2204          * entire page.
2205          */
2206         if (same_page && len < PAGE_CACHE_SIZE) {
2207                 if (offset < ino_size)
2208                         ret = btrfs_truncate_page(inode, offset, len, 0);
2209                 mutex_unlock(&inode->i_mutex);
2210                 return ret;
2211         }
2212
2213         /* zero back part of the first page */
2214         if (offset < ino_size) {
2215                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2216                 if (ret) {
2217                         mutex_unlock(&inode->i_mutex);
2218                         return ret;
2219                 }
2220         }
2221
2222         /* zero the front end of the last page */
2223         if (offset + len < ino_size) {
2224                 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2225                 if (ret) {
2226                         mutex_unlock(&inode->i_mutex);
2227                         return ret;
2228                 }
2229         }
2230
2231         if (lockend < lockstart) {
2232                 mutex_unlock(&inode->i_mutex);
2233                 return 0;
2234         }
2235
2236         while (1) {
2237                 struct btrfs_ordered_extent *ordered;
2238
2239                 truncate_pagecache_range(inode, lockstart, lockend);
2240
2241                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2242                                  0, &cached_state);
2243                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2244
2245                 /*
2246                  * We need to make sure we have no ordered extents in this range
2247                  * and nobody raced in and read a page in this range, if we did
2248                  * we need to try again.
2249                  */
2250                 if ((!ordered ||
2251                     (ordered->file_offset + ordered->len <= lockstart ||
2252                      ordered->file_offset > lockend)) &&
2253                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2254                                      lockend, EXTENT_UPTODATE, 0,
2255                                      cached_state)) {
2256                         if (ordered)
2257                                 btrfs_put_ordered_extent(ordered);
2258                         break;
2259                 }
2260                 if (ordered)
2261                         btrfs_put_ordered_extent(ordered);
2262                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2263                                      lockend, &cached_state, GFP_NOFS);
2264                 ret = btrfs_wait_ordered_range(inode, lockstart,
2265                                                lockend - lockstart + 1);
2266                 if (ret) {
2267                         mutex_unlock(&inode->i_mutex);
2268                         return ret;
2269                 }
2270         }
2271
2272         path = btrfs_alloc_path();
2273         if (!path) {
2274                 ret = -ENOMEM;
2275                 goto out;
2276         }
2277
2278         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2279         if (!rsv) {
2280                 ret = -ENOMEM;
2281                 goto out_free;
2282         }
2283         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2284         rsv->failfast = 1;
2285
2286         /*
2287          * 1 - update the inode
2288          * 1 - removing the extents in the range
2289          * 1 - adding the hole extent if no_holes isn't set
2290          */
2291         rsv_count = no_holes ? 2 : 3;
2292         trans = btrfs_start_transaction(root, rsv_count);
2293         if (IS_ERR(trans)) {
2294                 err = PTR_ERR(trans);
2295                 goto out_free;
2296         }
2297
2298         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2299                                       min_size);
2300         BUG_ON(ret);
2301         trans->block_rsv = rsv;
2302
2303         while (cur_offset < lockend) {
2304                 ret = __btrfs_drop_extents(trans, root, inode, path,
2305                                            cur_offset, lockend + 1,
2306                                            &drop_end, 1, 0, 0, NULL);
2307                 if (ret != -ENOSPC)
2308                         break;
2309
2310                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2311
2312                 if (cur_offset < ino_size) {
2313                         ret = fill_holes(trans, inode, path, cur_offset,
2314                                          drop_end);
2315                         if (ret) {
2316                                 err = ret;
2317                                 break;
2318                         }
2319                 }
2320
2321                 cur_offset = drop_end;
2322
2323                 ret = btrfs_update_inode(trans, root, inode);
2324                 if (ret) {
2325                         err = ret;
2326                         break;
2327                 }
2328
2329                 btrfs_end_transaction(trans, root);
2330                 btrfs_btree_balance_dirty(root);
2331
2332                 trans = btrfs_start_transaction(root, rsv_count);
2333                 if (IS_ERR(trans)) {
2334                         ret = PTR_ERR(trans);
2335                         trans = NULL;
2336                         break;
2337                 }
2338
2339                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2340                                               rsv, min_size);
2341                 BUG_ON(ret);    /* shouldn't happen */
2342                 trans->block_rsv = rsv;
2343         }
2344
2345         if (ret) {
2346                 err = ret;
2347                 goto out_trans;
2348         }
2349
2350         trans->block_rsv = &root->fs_info->trans_block_rsv;
2351         if (cur_offset < ino_size) {
2352                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2353                 if (ret) {
2354                         err = ret;
2355                         goto out_trans;
2356                 }
2357         }
2358
2359 out_trans:
2360         if (!trans)
2361                 goto out_free;
2362
2363         inode_inc_iversion(inode);
2364         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2365
2366         trans->block_rsv = &root->fs_info->trans_block_rsv;
2367         ret = btrfs_update_inode(trans, root, inode);
2368         btrfs_end_transaction(trans, root);
2369         btrfs_btree_balance_dirty(root);
2370 out_free:
2371         btrfs_free_path(path);
2372         btrfs_free_block_rsv(root, rsv);
2373 out:
2374         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2375                              &cached_state, GFP_NOFS);
2376         mutex_unlock(&inode->i_mutex);
2377         if (ret && !err)
2378                 err = ret;
2379         return err;
2380 }
2381
2382 static long btrfs_fallocate(struct file *file, int mode,
2383                             loff_t offset, loff_t len)
2384 {
2385         struct inode *inode = file_inode(file);
2386         struct extent_state *cached_state = NULL;
2387         struct btrfs_root *root = BTRFS_I(inode)->root;
2388         u64 cur_offset;
2389         u64 last_byte;
2390         u64 alloc_start;
2391         u64 alloc_end;
2392         u64 alloc_hint = 0;
2393         u64 locked_end;
2394         struct extent_map *em;
2395         int blocksize = BTRFS_I(inode)->root->sectorsize;
2396         int ret;
2397
2398         alloc_start = round_down(offset, blocksize);
2399         alloc_end = round_up(offset + len, blocksize);
2400
2401         /* Make sure we aren't being give some crap mode */
2402         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2403                 return -EOPNOTSUPP;
2404
2405         if (mode & FALLOC_FL_PUNCH_HOLE)
2406                 return btrfs_punch_hole(inode, offset, len);
2407
2408         /*
2409          * Make sure we have enough space before we do the
2410          * allocation.
2411          */
2412         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2413         if (ret)
2414                 return ret;
2415         if (root->fs_info->quota_enabled) {
2416                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2417                 if (ret)
2418                         goto out_reserve_fail;
2419         }
2420
2421         mutex_lock(&inode->i_mutex);
2422         ret = inode_newsize_ok(inode, alloc_end);
2423         if (ret)
2424                 goto out;
2425
2426         if (alloc_start > inode->i_size) {
2427                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2428                                         alloc_start);
2429                 if (ret)
2430                         goto out;
2431         } else {
2432                 /*
2433                  * If we are fallocating from the end of the file onward we
2434                  * need to zero out the end of the page if i_size lands in the
2435                  * middle of a page.
2436                  */
2437                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2438                 if (ret)
2439                         goto out;
2440         }
2441
2442         /*
2443          * wait for ordered IO before we have any locks.  We'll loop again
2444          * below with the locks held.
2445          */
2446         ret = btrfs_wait_ordered_range(inode, alloc_start,
2447                                        alloc_end - alloc_start);
2448         if (ret)
2449                 goto out;
2450
2451         locked_end = alloc_end - 1;
2452         while (1) {
2453                 struct btrfs_ordered_extent *ordered;
2454
2455                 /* the extent lock is ordered inside the running
2456                  * transaction
2457                  */
2458                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2459                                  locked_end, 0, &cached_state);
2460                 ordered = btrfs_lookup_first_ordered_extent(inode,
2461                                                             alloc_end - 1);
2462                 if (ordered &&
2463                     ordered->file_offset + ordered->len > alloc_start &&
2464                     ordered->file_offset < alloc_end) {
2465                         btrfs_put_ordered_extent(ordered);
2466                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2467                                              alloc_start, locked_end,
2468                                              &cached_state, GFP_NOFS);
2469                         /*
2470                          * we can't wait on the range with the transaction
2471                          * running or with the extent lock held
2472                          */
2473                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2474                                                        alloc_end - alloc_start);
2475                         if (ret)
2476                                 goto out;
2477                 } else {
2478                         if (ordered)
2479                                 btrfs_put_ordered_extent(ordered);
2480                         break;
2481                 }
2482         }
2483
2484         cur_offset = alloc_start;
2485         while (1) {
2486                 u64 actual_end;
2487
2488                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2489                                       alloc_end - cur_offset, 0);
2490                 if (IS_ERR_OR_NULL(em)) {
2491                         if (!em)
2492                                 ret = -ENOMEM;
2493                         else
2494                                 ret = PTR_ERR(em);
2495                         break;
2496                 }
2497                 last_byte = min(extent_map_end(em), alloc_end);
2498                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2499                 last_byte = ALIGN(last_byte, blocksize);
2500
2501                 if (em->block_start == EXTENT_MAP_HOLE ||
2502                     (cur_offset >= inode->i_size &&
2503                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2504                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2505                                                         last_byte - cur_offset,
2506                                                         1 << inode->i_blkbits,
2507                                                         offset + len,
2508                                                         &alloc_hint);
2509
2510                         if (ret < 0) {
2511                                 free_extent_map(em);
2512                                 break;
2513                         }
2514                 } else if (actual_end > inode->i_size &&
2515                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2516                         /*
2517                          * We didn't need to allocate any more space, but we
2518                          * still extended the size of the file so we need to
2519                          * update i_size.
2520                          */
2521                         inode->i_ctime = CURRENT_TIME;
2522                         i_size_write(inode, actual_end);
2523                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2524                 }
2525                 free_extent_map(em);
2526
2527                 cur_offset = last_byte;
2528                 if (cur_offset >= alloc_end) {
2529                         ret = 0;
2530                         break;
2531                 }
2532         }
2533         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2534                              &cached_state, GFP_NOFS);
2535 out:
2536         mutex_unlock(&inode->i_mutex);
2537         if (root->fs_info->quota_enabled)
2538                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2539 out_reserve_fail:
2540         /* Let go of our reservation. */
2541         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2542         return ret;
2543 }
2544
2545 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2546 {
2547         struct btrfs_root *root = BTRFS_I(inode)->root;
2548         struct extent_map *em = NULL;
2549         struct extent_state *cached_state = NULL;
2550         u64 lockstart = *offset;
2551         u64 lockend = i_size_read(inode);
2552         u64 start = *offset;
2553         u64 len = i_size_read(inode);
2554         int ret = 0;
2555
2556         lockend = max_t(u64, root->sectorsize, lockend);
2557         if (lockend <= lockstart)
2558                 lockend = lockstart + root->sectorsize;
2559
2560         lockend--;
2561         len = lockend - lockstart + 1;
2562
2563         len = max_t(u64, len, root->sectorsize);
2564         if (inode->i_size == 0)
2565                 return -ENXIO;
2566
2567         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2568                          &cached_state);
2569
2570         while (start < inode->i_size) {
2571                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2572                 if (IS_ERR(em)) {
2573                         ret = PTR_ERR(em);
2574                         em = NULL;
2575                         break;
2576                 }
2577
2578                 if (whence == SEEK_HOLE &&
2579                     (em->block_start == EXTENT_MAP_HOLE ||
2580                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2581                         break;
2582                 else if (whence == SEEK_DATA &&
2583                            (em->block_start != EXTENT_MAP_HOLE &&
2584                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2585                         break;
2586
2587                 start = em->start + em->len;
2588                 free_extent_map(em);
2589                 em = NULL;
2590                 cond_resched();
2591         }
2592         free_extent_map(em);
2593         if (!ret) {
2594                 if (whence == SEEK_DATA && start >= inode->i_size)
2595                         ret = -ENXIO;
2596                 else
2597                         *offset = min_t(loff_t, start, inode->i_size);
2598         }
2599         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2600                              &cached_state, GFP_NOFS);
2601         return ret;
2602 }
2603
2604 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2605 {
2606         struct inode *inode = file->f_mapping->host;
2607         int ret;
2608
2609         mutex_lock(&inode->i_mutex);
2610         switch (whence) {
2611         case SEEK_END:
2612         case SEEK_CUR:
2613                 offset = generic_file_llseek(file, offset, whence);
2614                 goto out;
2615         case SEEK_DATA:
2616         case SEEK_HOLE:
2617                 if (offset >= i_size_read(inode)) {
2618                         mutex_unlock(&inode->i_mutex);
2619                         return -ENXIO;
2620                 }
2621
2622                 ret = find_desired_extent(inode, &offset, whence);
2623                 if (ret) {
2624                         mutex_unlock(&inode->i_mutex);
2625                         return ret;
2626                 }
2627         }
2628
2629         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2630 out:
2631         mutex_unlock(&inode->i_mutex);
2632         return offset;
2633 }
2634
2635 const struct file_operations btrfs_file_operations = {
2636         .llseek         = btrfs_file_llseek,
2637         .read           = do_sync_read,
2638         .write          = do_sync_write,
2639         .aio_read       = generic_file_aio_read,
2640         .splice_read    = generic_file_splice_read,
2641         .aio_write      = btrfs_file_aio_write,
2642         .mmap           = btrfs_file_mmap,
2643         .open           = generic_file_open,
2644         .release        = btrfs_release_file,
2645         .fsync          = btrfs_sync_file,
2646         .fallocate      = btrfs_fallocate,
2647         .unlocked_ioctl = btrfs_ioctl,
2648 #ifdef CONFIG_COMPAT
2649         .compat_ioctl   = btrfs_ioctl,
2650 #endif
2651 };
2652
2653 void btrfs_auto_defrag_exit(void)
2654 {
2655         if (btrfs_inode_defrag_cachep)
2656                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2657 }
2658
2659 int btrfs_auto_defrag_init(void)
2660 {
2661         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2662                                         sizeof(struct inode_defrag), 0,
2663                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2664                                         NULL);
2665         if (!btrfs_inode_defrag_cachep)
2666                 return -ENOMEM;
2667
2668         return 0;
2669 }