Pileus Git - ~andy/linux/blob - 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/falloc.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/slab.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39 #include "tree-log.h"
  40 #include "locking.h"
  41 #include "compat.h"
  42 #include "volumes.h"
  43
  44 /*
  45  * when auto defrag is enabled we
  46  * queue up these defrag structs to remember which
  47  * inodes need defragging passes
  48  */
  49 struct inode_defrag {
  50         struct rb_node rb_node;
  51         /* objectid */
  52         u64 ino;
  53         /*
  54          * transid where the defrag was added, we search for
  55          * extents newer than this
  56          */
  57         u64 transid;
  58
  59         /* root objectid */
  60         u64 root;
  61
  62         /* last offset we were able to defrag */
  63         u64 last_offset;
  64
  65         /* if we've wrapped around back to zero once already */
  66         int cycled;
  67 };
  68
  69 static int __compare_inode_defrag(struct inode_defrag *defrag1,
  70                                   struct inode_defrag *defrag2)
  71 {
  72         if (defrag1->root > defrag2->root)
  73                 return 1;
  74         else if (defrag1->root < defrag2->root)
  75                 return -1;
  76         else if (defrag1->ino > defrag2->ino)
  77                 return 1;
  78         else if (defrag1->ino < defrag2->ino)
  79                 return -1;
  80         else
  81                 return 0;
  82 }
  83
  84 /* pop a record for an inode into the defrag tree.  The lock
  85  * must be held already
  86  *
  87  * If you're inserting a record for an older transid than an
  88  * existing record, the transid already in the tree is lowered
  89  *
  90  * If an existing record is found the defrag item you
  91  * pass in is freed
  92  */
  93 static void __btrfs_add_inode_defrag(struct inode *inode,
  94                                     struct inode_defrag *defrag)
  95 {
  96         struct btrfs_root *root = BTRFS_I(inode)->root;
  97         struct inode_defrag *entry;
  98         struct rb_node **p;
  99         struct rb_node *parent = NULL;
 100         int ret;
 101
 102         p = &root->fs_info->defrag_inodes.rb_node;
 103         while (*p) {
 104                 parent = *p;
 105                 entry = rb_entry(parent, struct inode_defrag, rb_node);
 106
 107                 ret = __compare_inode_defrag(defrag, entry);
 108                 if (ret < 0)
 109                         p = &parent->rb_left;
 110                 else if (ret > 0)
 111                         p = &parent->rb_right;
 112                 else {
 113                         /* if we're reinserting an entry for
 114                          * an old defrag run, make sure to
 115                          * lower the transid of our existing record
 116                          */
 117                         if (defrag->transid < entry->transid)
 118                                 entry->transid = defrag->transid;
 119                         if (defrag->last_offset > entry->last_offset)
 120                                 entry->last_offset = defrag->last_offset;
 121                         goto exists;
 122                 }
 123         }
 124         set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
 125         rb_link_node(&defrag->rb_node, parent, p);
 126         rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
 127         return;
 128
 129 exists:
 130         kfree(defrag);
 131         return;
 132
 133 }
 134
 135 /*
 136  * insert a defrag record for this inode if auto defrag is
 137  * enabled
 138  */
 139 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
 140                            struct inode *inode)
 141 {
 142         struct btrfs_root *root = BTRFS_I(inode)->root;
 143         struct inode_defrag *defrag;
 144         u64 transid;
 145
 146         if (!btrfs_test_opt(root, AUTO_DEFRAG))
 147                 return 0;
 148
 149         if (btrfs_fs_closing(root->fs_info))
 150                 return 0;
 151
 152         if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
 153                 return 0;
 154
 155         if (trans)
 156                 transid = trans->transid;
 157         else
 158                 transid = BTRFS_I(inode)->root->last_trans;
 159
 160         defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
 161         if (!defrag)
 162                 return -ENOMEM;
 163
 164         defrag->ino = btrfs_ino(inode);
 165         defrag->transid = transid;
 166         defrag->root = root->root_key.objectid;
 167
 168         spin_lock(&root->fs_info->defrag_inodes_lock);
 169         if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
 170                 __btrfs_add_inode_defrag(inode, defrag);
 171         else
 172                 kfree(defrag);
 173         spin_unlock(&root->fs_info->defrag_inodes_lock);
 174         return 0;
 175 }
 176
 177 /*
 178  * must be called with the defrag_inodes lock held
 179  */
 180 struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info,
 181                                              u64 root, u64 ino,
 182                                              struct rb_node **next)
 183 {
 184         struct inode_defrag *entry = NULL;
 185         struct inode_defrag tmp;
 186         struct rb_node *p;
 187         struct rb_node *parent = NULL;
 188         int ret;
 189
 190         tmp.ino = ino;
 191         tmp.root = root;
 192
 193         p = info->defrag_inodes.rb_node;
 194         while (p) {
 195                 parent = p;
 196                 entry = rb_entry(parent, struct inode_defrag, rb_node);
 197
 198                 ret = __compare_inode_defrag(&tmp, entry);
 199                 if (ret < 0)
 200                         p = parent->rb_left;
 201                 else if (ret > 0)
 202                         p = parent->rb_right;
 203                 else
 204                         return entry;
 205         }
 206
 207         if (next) {
 208                 while (parent && __compare_inode_defrag(&tmp, entry) > 0) {
 209                         parent = rb_next(parent);
 210                         entry = rb_entry(parent, struct inode_defrag, rb_node);
 211                 }
 212                 *next = parent;
 213         }
 214         return NULL;
 215 }
 216
 217 /*
 218  * run through the list of inodes in the FS that need
 219  * defragging
 220  */
 221 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
 222 {
 223         struct inode_defrag *defrag;
 224         struct btrfs_root *inode_root;
 225         struct inode *inode;
 226         struct rb_node *n;
 227         struct btrfs_key key;
 228         struct btrfs_ioctl_defrag_range_args range;
 229         u64 first_ino = 0;
 230         u64 root_objectid = 0;
 231         int num_defrag;
 232         int defrag_batch = 1024;
 233
 234         memset(&range, 0, sizeof(range));
 235         range.len = (u64)-1;
 236
 237         atomic_inc(&fs_info->defrag_running);
 238         spin_lock(&fs_info->defrag_inodes_lock);
 239         while(1) {
 240                 n = NULL;
 241
 242                 /* find an inode to defrag */
 243                 defrag = btrfs_find_defrag_inode(fs_info, root_objectid,
 244                                                  first_ino, &n);
 245                 if (!defrag) {
 246                         if (n) {
 247                                 defrag = rb_entry(n, struct inode_defrag,
 248                                                   rb_node);
 249                         } else if (root_objectid || first_ino) {
 250                                 root_objectid = 0;
 251                                 first_ino = 0;
 252                                 continue;
 253                         } else {
 254                                 break;
 255                         }
 256                 }
 257
 258                 /* remove it from the rbtree */
 259                 first_ino = defrag->ino + 1;
 260                 root_objectid = defrag->root;
 261                 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
 262
 263                 if (btrfs_fs_closing(fs_info))
 264                         goto next_free;
 265
 266                 spin_unlock(&fs_info->defrag_inodes_lock);
 267
 268                 /* get the inode */
 269                 key.objectid = defrag->root;
 270                 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 271                 key.offset = (u64)-1;
 272                 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
 273                 if (IS_ERR(inode_root))
 274                         goto next;
 275
 276                 key.objectid = defrag->ino;
 277                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 278                 key.offset = 0;
 279
 280                 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
 281                 if (IS_ERR(inode))
 282                         goto next;
 283
 284                 /* do a chunk of defrag */
 285                 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
 286                 range.start = defrag->last_offset;
 287                 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
 288                                                defrag_batch);
 289                 /*
 290                  * if we filled the whole defrag batch, there
 291                  * must be more work to do.  Queue this defrag
 292                  * again
 293                  */
 294                 if (num_defrag == defrag_batch) {
 295                         defrag->last_offset = range.start;
 296                         __btrfs_add_inode_defrag(inode, defrag);
 297                         /*
 298                          * we don't want to kfree defrag, we added it back to
 299                          * the rbtree
 300                          */
 301                         defrag = NULL;
 302                 } else if (defrag->last_offset && !defrag->cycled) {
 303                         /*
 304                          * we didn't fill our defrag batch, but
 305                          * we didn't start at zero.  Make sure we loop
 306                          * around to the start of the file.
 307                          */
 308                         defrag->last_offset = 0;
 309                         defrag->cycled = 1;
 310                         __btrfs_add_inode_defrag(inode, defrag);
 311                         defrag = NULL;
 312                 }
 313
 314                 iput(inode);
 315 next:
 316                 spin_lock(&fs_info->defrag_inodes_lock);
 317 next_free:
 318                 kfree(defrag);
 319         }
 320         spin_unlock(&fs_info->defrag_inodes_lock);
 321
 322         atomic_dec(&fs_info->defrag_running);
 323
 324         /*
 325          * during unmount, we use the transaction_wait queue to
 326          * wait for the defragger to stop
 327          */
 328         wake_up(&fs_info->transaction_wait);
 329         return 0;
 330 }
 331
 332 /* simple helper to fault in pages and copy.  This should go away
 333  * and be replaced with calls into generic code.
 334  */
 335 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
 336                                          size_t write_bytes,
 337                                          struct page **prepared_pages,
 338                                          struct iov_iter *i)
 339 {
 340         size_t copied = 0;
 341         size_t total_copied = 0;
 342         int pg = 0;
 343         int offset = pos & (PAGE_CACHE_SIZE - 1);
 344
 345         while (write_bytes > 0) {
 346                 size_t count = min_t(size_t,
 347                                      PAGE_CACHE_SIZE - offset, write_bytes);
 348                 struct page *page = prepared_pages[pg];
 349                 /*
 350                  * Copy data from userspace to the current page
 351                  *
 352                  * Disable pagefault to avoid recursive lock since
 353                  * the pages are already locked
 354                  */
 355                 pagefault_disable();
 356                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
 357                 pagefault_enable();
 358
 359                 /* Flush processor's dcache for this page */
 360                 flush_dcache_page(page);
 361
 362                 /*
 363                  * if we get a partial write, we can end up with
 364                  * partially up to date pages.  These add
 365                  * a lot of complexity, so make sure they don't
 366                  * happen by forcing this copy to be retried.
 367                  *
 368                  * The rest of the btrfs_file_write code will fall
 369                  * back to page at a time copies after we return 0.
 370                  */
 371                 if (!PageUptodate(page) && copied < count)
 372                         copied = 0;
 373
 374                 iov_iter_advance(i, copied);
 375                 write_bytes -= copied;
 376                 total_copied += copied;
 377
 378                 /* Return to btrfs_file_aio_write to fault page */
 379                 if (unlikely(copied == 0))
 380                         break;
 381
 382                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
 383                         offset += copied;
 384                 } else {
 385                         pg++;
 386                         offset = 0;
 387                 }
 388         }
 389         return total_copied;
 390 }
 391
 392 /*
 393  * unlocks pages after btrfs_file_write is done with them
 394  */
 395 void btrfs_drop_pages(struct page **pages, size_t num_pages)
 396 {
 397         size_t i;
 398         for (i = 0; i < num_pages; i++) {
 399                 /* page checked is some magic around finding pages that
 400                  * have been modified without going through btrfs_set_page_dirty
 401                  * clear it here
 402                  */
 403                 ClearPageChecked(pages[i]);
 404                 unlock_page(pages[i]);
 405                 mark_page_accessed(pages[i]);
 406                 page_cache_release(pages[i]);
 407         }
 408 }
 409
 410 /*
 411  * after copy_from_user, pages need to be dirtied and we need to make
 412  * sure holes are created between the current EOF and the start of
 413  * any next extents (if required).
 414  *
 415  * this also makes the decision about creating an inline extent vs
 416  * doing real data extents, marking pages dirty and delalloc as required.
 417  */
 418 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
 419                       struct page **pages, size_t num_pages,
 420                       loff_t pos, size_t write_bytes,
 421                       struct extent_state **cached)
 422 {
 423         int err = 0;
 424         int i;
 425         u64 num_bytes;
 426         u64 start_pos;
 427         u64 end_of_last_block;
 428         u64 end_pos = pos + write_bytes;
 429         loff_t isize = i_size_read(inode);
 430
 431         start_pos = pos & ~((u64)root->sectorsize - 1);
 432         num_bytes = (write_bytes + pos - start_pos +
 433                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 434
 435         end_of_last_block = start_pos + num_bytes - 1;
 436         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 437                                         cached);
 438         if (err)
 439                 return err;
 440
 441         for (i = 0; i < num_pages; i++) {
 442                 struct page *p = pages[i];
 443                 SetPageUptodate(p);
 444                 ClearPageChecked(p);
 445                 set_page_dirty(p);
 446         }
 447
 448         /*
 449          * we've only changed i_size in ram, and we haven't updated
 450          * the disk i_size.  There is no need to log the inode
 451          * at this time.
 452          */
 453         if (end_pos > isize)
 454                 i_size_write(inode, end_pos);
 455         return 0;
 456 }
 457
 458 /*
 459  * this drops all the extents in the cache that intersect the range
 460  * [start, end].  Existing extents are split as required.
 461  */
 462 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 463                              int skip_pinned)
 464 {
 465         struct extent_map *em;
 466         struct extent_map *split = NULL;
 467         struct extent_map *split2 = NULL;
 468         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 469         u64 len = end - start + 1;
 470         u64 gen;
 471         int ret;
 472         int testend = 1;
 473         unsigned long flags;
 474         int compressed = 0;
 475
 476         WARN_ON(end < start);
 477         if (end == (u64)-1) {
 478                 len = (u64)-1;
 479                 testend = 0;
 480         }
 481         while (1) {
 482                 int no_splits = 0;
 483
 484                 if (!split)
 485                         split = alloc_extent_map();
 486                 if (!split2)
 487                         split2 = alloc_extent_map();
 488                 if (!split || !split2)
 489                         no_splits = 1;
 490
 491                 write_lock(&em_tree->lock);
 492                 em = lookup_extent_mapping(em_tree, start, len);
 493                 if (!em) {
 494                         write_unlock(&em_tree->lock);
 495                         break;
 496                 }
 497                 flags = em->flags;
 498                 gen = em->generation;
 499                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 500                         if (testend && em->start + em->len >= start + len) {
 501                                 free_extent_map(em);
 502                                 write_unlock(&em_tree->lock);
 503                                 break;
 504                         }
 505                         start = em->start + em->len;
 506                         if (testend)
 507                                 len = start + len - (em->start + em->len);
 508                         free_extent_map(em);
 509                         write_unlock(&em_tree->lock);
 510                         continue;
 511                 }
 512                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 513                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 514                 remove_extent_mapping(em_tree, em);
 515                 if (no_splits)
 516                         goto next;
 517
 518                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 519                     em->start < start) {
 520                         split->start = em->start;
 521                         split->len = start - em->start;
 522                         split->orig_start = em->orig_start;
 523                         split->block_start = em->block_start;
 524
 525                         if (compressed)
 526                                 split->block_len = em->block_len;
 527                         else
 528                                 split->block_len = split->len;
 529                         split->generation = gen;
 530                         split->bdev = em->bdev;
 531                         split->flags = flags;
 532                         split->compress_type = em->compress_type;
 533                         ret = add_extent_mapping(em_tree, split);
 534                         BUG_ON(ret); /* Logic error */
 535                         list_move(&split->list, &em_tree->modified_extents);
 536                         free_extent_map(split);
 537                         split = split2;
 538                         split2 = NULL;
 539                 }
 540                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 541                     testend && em->start + em->len > start + len) {
 542                         u64 diff = start + len - em->start;
 543
 544                         split->start = start + len;
 545                         split->len = em->start + em->len - (start + len);
 546                         split->bdev = em->bdev;
 547                         split->flags = flags;
 548                         split->compress_type = em->compress_type;
 549                         split->generation = gen;
 550
 551                         if (compressed) {
 552                                 split->block_len = em->block_len;
 553                                 split->block_start = em->block_start;
 554                                 split->orig_start = em->orig_start;
 555                         } else {
 556                                 split->block_len = split->len;
 557                                 split->block_start = em->block_start + diff;
 558                                 split->orig_start = split->start;
 559                         }
 560
 561                         ret = add_extent_mapping(em_tree, split);
 562                         BUG_ON(ret); /* Logic error */
 563                         list_move(&split->list, &em_tree->modified_extents);
 564                         free_extent_map(split);
 565                         split = NULL;
 566                 }
 567 next:
 568                 write_unlock(&em_tree->lock);
 569
 570                 /* once for us */
 571                 free_extent_map(em);
 572                 /* once for the tree*/
 573                 free_extent_map(em);
 574         }
 575         if (split)
 576                 free_extent_map(split);
 577         if (split2)
 578                 free_extent_map(split2);
 579 }
 580
 581 /*
 582  * this is very complex, but the basic idea is to drop all extents
 583  * in the range start - end.  hint_block is filled in with a block number
 584  * that would be a good hint to the block allocator for this file.
 585  *
 586  * If an extent intersects the range but is not entirely inside the range
 587  * it is either truncated or split.  Anything entirely inside the range
 588  * is deleted from the tree.
 589  */
 590 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 591                          struct btrfs_root *root, struct inode *inode,
 592                          struct btrfs_path *path, u64 start, u64 end,
 593                          u64 *drop_end, int drop_cache)
 594 {
 595         struct extent_buffer *leaf;
 596         struct btrfs_file_extent_item *fi;
 597         struct btrfs_key key;
 598         struct btrfs_key new_key;
 599         u64 ino = btrfs_ino(inode);
 600         u64 search_start = start;
 601         u64 disk_bytenr = 0;
 602         u64 num_bytes = 0;
 603         u64 extent_offset = 0;
 604         u64 extent_end = 0;
 605         int del_nr = 0;
 606         int del_slot = 0;
 607         int extent_type;
 608         int recow;
 609         int ret;
 610         int modify_tree = -1;
 611         int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
 612         int found = 0;
 613
 614         if (drop_cache)
 615                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 616
 617         if (start >= BTRFS_I(inode)->disk_i_size)
 618                 modify_tree = 0;
 619
 620         while (1) {
 621                 recow = 0;
 622                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
 623                                                search_start, modify_tree);
 624                 if (ret < 0)
 625                         break;
 626                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 627                         leaf = path->nodes[0];
 628                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 629                         if (key.objectid == ino &&
 630                             key.type == BTRFS_EXTENT_DATA_KEY)
 631                                 path->slots[0]--;
 632                 }
 633                 ret = 0;
 634 next_slot:
 635                 leaf = path->nodes[0];
 636                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 637                         BUG_ON(del_nr > 0);
 638                         ret = btrfs_next_leaf(root, path);
 639                         if (ret < 0)
 640                                 break;
 641                         if (ret > 0) {
 642                                 ret = 0;
 643                                 break;
 644                         }
 645                         leaf = path->nodes[0];
 646                         recow = 1;
 647                 }
 648
 649                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 650                 if (key.objectid > ino ||
 651                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 652                         break;
 653
 654                 fi = btrfs_item_ptr(leaf, path->slots[0],
 655                                     struct btrfs_file_extent_item);
 656                 extent_type = btrfs_file_extent_type(leaf, fi);
 657
 658                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 659                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 660                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 661                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 662                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 663                         extent_end = key.offset +
 664                                 btrfs_file_extent_num_bytes(leaf, fi);
 665                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 666                         extent_end = key.offset +
 667                                 btrfs_file_extent_inline_len(leaf, fi);
 668                 } else {
 669                         WARN_ON(1);
 670                         extent_end = search_start;
 671                 }
 672
 673                 if (extent_end <= search_start) {
 674                         path->slots[0]++;
 675                         goto next_slot;
 676                 }
 677
 678                 found = 1;
 679                 search_start = max(key.offset, start);
 680                 if (recow || !modify_tree) {
 681                         modify_tree = -1;
 682                         btrfs_release_path(path);
 683                         continue;
 684                 }
 685
 686                 /*
 687                  *     | - range to drop - |
 688                  *  | -------- extent -------- |
 689                  */
 690                 if (start > key.offset && end < extent_end) {
 691                         BUG_ON(del_nr > 0);
 692                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 693
 694                         memcpy(&new_key, &key, sizeof(new_key));
 695                         new_key.offset = start;
 696                         ret = btrfs_duplicate_item(trans, root, path,
 697                                                    &new_key);
 698                         if (ret == -EAGAIN) {
 699                                 btrfs_release_path(path);
 700                                 continue;
 701                         }
 702                         if (ret < 0)
 703                                 break;
 704
 705                         leaf = path->nodes[0];
 706                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 707                                             struct btrfs_file_extent_item);
 708                         btrfs_set_file_extent_num_bytes(leaf, fi,
 709                                                         start - key.offset);
 710
 711                         fi = btrfs_item_ptr(leaf, path->slots[0],
 712                                             struct btrfs_file_extent_item);
 713
 714                         extent_offset += start - key.offset;
 715                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 716                         btrfs_set_file_extent_num_bytes(leaf, fi,
 717                                                         extent_end - start);
 718                         btrfs_mark_buffer_dirty(leaf);
 719
 720                         if (update_refs && disk_bytenr > 0) {
 721                                 ret = btrfs_inc_extent_ref(trans, root,
 722                                                 disk_bytenr, num_bytes, 0,
 723                                                 root->root_key.objectid,
 724                                                 new_key.objectid,
 725                                                 start - extent_offset, 0);
 726                                 BUG_ON(ret); /* -ENOMEM */
 727                         }
 728                         key.offset = start;
 729                 }
 730                 /*
 731                  *  | ---- range to drop ----- |
 732                  *      | -------- extent -------- |
 733                  */
 734                 if (start <= key.offset && end < extent_end) {
 735                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 736
 737                         memcpy(&new_key, &key, sizeof(new_key));
 738                         new_key.offset = end;
 739                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 740
 741                         extent_offset += end - key.offset;
 742                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 743                         btrfs_set_file_extent_num_bytes(leaf, fi,
 744                                                         extent_end - end);
 745                         btrfs_mark_buffer_dirty(leaf);
 746                         if (update_refs && disk_bytenr > 0)
 747                                 inode_sub_bytes(inode, end - key.offset);
 748                         break;
 749                 }
 750
 751                 search_start = extent_end;
 752                 /*
 753                  *       | ---- range to drop ----- |
 754                  *  | -------- extent -------- |
 755                  */
 756                 if (start > key.offset && end >= extent_end) {
 757                         BUG_ON(del_nr > 0);
 758                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 759
 760                         btrfs_set_file_extent_num_bytes(leaf, fi,
 761                                                         start - key.offset);
 762                         btrfs_mark_buffer_dirty(leaf);
 763                         if (update_refs && disk_bytenr > 0)
 764                                 inode_sub_bytes(inode, extent_end - start);
 765                         if (end == extent_end)
 766                                 break;
 767
 768                         path->slots[0]++;
 769                         goto next_slot;
 770                 }
 771
 772                 /*
 773                  *  | ---- range to drop ----- |
 774                  *    | ------ extent ------ |
 775                  */
 776                 if (start <= key.offset && end >= extent_end) {
 777                         if (del_nr == 0) {
 778                                 del_slot = path->slots[0];
 779                                 del_nr = 1;
 780                         } else {
 781                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 782                                 del_nr++;
 783                         }
 784
 785                         if (update_refs &&
 786                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
 787                                 inode_sub_bytes(inode,
 788                                                 extent_end - key.offset);
 789                                 extent_end = ALIGN(extent_end,
 790                                                    root->sectorsize);
 791                         } else if (update_refs && disk_bytenr > 0) {
 792                                 ret = btrfs_free_extent(trans, root,
 793                                                 disk_bytenr, num_bytes, 0,
 794                                                 root->root_key.objectid,
 795                                                 key.objectid, key.offset -
 796                                                 extent_offset, 0);
 797                                 BUG_ON(ret); /* -ENOMEM */
 798                                 inode_sub_bytes(inode,
 799                                                 extent_end - key.offset);
 800                         }
 801
 802                         if (end == extent_end)
 803                                 break;
 804
 805                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 806                                 path->slots[0]++;
 807                                 goto next_slot;
 808                         }
 809
 810                         ret = btrfs_del_items(trans, root, path, del_slot,
 811                                               del_nr);
 812                         if (ret) {
 813                                 btrfs_abort_transaction(trans, root, ret);
 814                                 break;
 815                         }
 816
 817                         del_nr = 0;
 818                         del_slot = 0;
 819
 820                         btrfs_release_path(path);
 821                         continue;
 822                 }
 823
 824                 BUG_ON(1);
 825         }
 826
 827         if (!ret && del_nr > 0) {
 828                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 829                 if (ret)
 830                         btrfs_abort_transaction(trans, root, ret);
 831         }
 832
 833         if (drop_end)
 834                 *drop_end = found ? min(end, extent_end) : end;
 835         btrfs_release_path(path);
 836         return ret;
 837 }
 838
 839 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 840                        struct btrfs_root *root, struct inode *inode, u64 start,
 841                        u64 end, int drop_cache)
 842 {
 843         struct btrfs_path *path;
 844         int ret;
 845
 846         path = btrfs_alloc_path();
 847         if (!path)
 848                 return -ENOMEM;
 849         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
 850                                    drop_cache);
 851         btrfs_free_path(path);
 852         return ret;
 853 }
 854
 855 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 856                             u64 objectid, u64 bytenr, u64 orig_offset,
 857                             u64 *start, u64 *end)
 858 {
 859         struct btrfs_file_extent_item *fi;
 860         struct btrfs_key key;
 861         u64 extent_end;
 862
 863         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 864                 return 0;
 865
 866         btrfs_item_key_to_cpu(leaf, &key, slot);
 867         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 868                 return 0;
 869
 870         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 871         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 872             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 873             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 874             btrfs_file_extent_compression(leaf, fi) ||
 875             btrfs_file_extent_encryption(leaf, fi) ||
 876             btrfs_file_extent_other_encoding(leaf, fi))
 877                 return 0;
 878
 879         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 880         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 881                 return 0;
 882
 883         *start = key.offset;
 884         *end = extent_end;
 885         return 1;
 886 }
 887
 888 /*
 889  * Mark extent in the range start - end as written.
 890  *
 891  * This changes extent type from 'pre-allocated' to 'regular'. If only
 892  * part of extent is marked as written, the extent will be split into
 893  * two or three.
 894  */
 895 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 896                               struct inode *inode, u64 start, u64 end)
 897 {
 898         struct btrfs_root *root = BTRFS_I(inode)->root;
 899         struct extent_buffer *leaf;
 900         struct btrfs_path *path;
 901         struct btrfs_file_extent_item *fi;
 902         struct btrfs_key key;
 903         struct btrfs_key new_key;
 904         u64 bytenr;
 905         u64 num_bytes;
 906         u64 extent_end;
 907         u64 orig_offset;
 908         u64 other_start;
 909         u64 other_end;
 910         u64 split;
 911         int del_nr = 0;
 912         int del_slot = 0;
 913         int recow;
 914         int ret;
 915         u64 ino = btrfs_ino(inode);
 916
 917         path = btrfs_alloc_path();
 918         if (!path)
 919                 return -ENOMEM;
 920 again:
 921         recow = 0;
 922         split = start;
 923         key.objectid = ino;
 924         key.type = BTRFS_EXTENT_DATA_KEY;
 925         key.offset = split;
 926
 927         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 928         if (ret < 0)
 929                 goto out;
 930         if (ret > 0 && path->slots[0] > 0)
 931                 path->slots[0]--;
 932
 933         leaf = path->nodes[0];
 934         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 935         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
 936         fi = btrfs_item_ptr(leaf, path->slots[0],
 937                             struct btrfs_file_extent_item);
 938         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 939                BTRFS_FILE_EXTENT_PREALLOC);
 940         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 941         BUG_ON(key.offset > start || extent_end < end);
 942
 943         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 944         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 945         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 946         memcpy(&new_key, &key, sizeof(new_key));
 947
 948         if (start == key.offset && end < extent_end) {
 949                 other_start = 0;
 950                 other_end = start;
 951                 if (extent_mergeable(leaf, path->slots[0] - 1,
 952                                      ino, bytenr, orig_offset,
 953                                      &other_start, &other_end)) {
 954                         new_key.offset = end;
 955                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 956                         fi = btrfs_item_ptr(leaf, path->slots[0],
 957                                             struct btrfs_file_extent_item);
 958                         btrfs_set_file_extent_generation(leaf, fi,
 959                                                          trans->transid);
 960                         btrfs_set_file_extent_num_bytes(leaf, fi,
 961                                                         extent_end - end);
 962                         btrfs_set_file_extent_offset(leaf, fi,
 963                                                      end - orig_offset);
 964                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 965                                             struct btrfs_file_extent_item);
 966                         btrfs_set_file_extent_generation(leaf, fi,
 967                                                          trans->transid);
 968                         btrfs_set_file_extent_num_bytes(leaf, fi,
 969                                                         end - other_start);
 970                         btrfs_mark_buffer_dirty(leaf);
 971                         goto out;
 972                 }
 973         }
 974
 975         if (start > key.offset && end == extent_end) {
 976                 other_start = end;
 977                 other_end = 0;
 978                 if (extent_mergeable(leaf, path->slots[0] + 1,
 979                                      ino, bytenr, orig_offset,
 980                                      &other_start, &other_end)) {
 981                         fi = btrfs_item_ptr(leaf, path->slots[0],
 982                                             struct btrfs_file_extent_item);
 983                         btrfs_set_file_extent_num_bytes(leaf, fi,
 984                                                         start - key.offset);
 985                         btrfs_set_file_extent_generation(leaf, fi,
 986                                                          trans->transid);
 987                         path->slots[0]++;
 988                         new_key.offset = start;
 989                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 990
 991                         fi = btrfs_item_ptr(leaf, path->slots[0],
 992                                             struct btrfs_file_extent_item);
 993                         btrfs_set_file_extent_generation(leaf, fi,
 994                                                          trans->transid);
 995                         btrfs_set_file_extent_num_bytes(leaf, fi,
 996                                                         other_end - start);
 997                         btrfs_set_file_extent_offset(leaf, fi,
 998                                                      start - orig_offset);
 999                         btrfs_mark_buffer_dirty(leaf);
1000                         goto out;
1001                 }
1002         }
1003
1004         while (start > key.offset || end < extent_end) {
1005                 if (key.offset == start)
1006                         split = end;
1007
1008                 new_key.offset = split;
1009                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1010                 if (ret == -EAGAIN) {
1011                         btrfs_release_path(path);
1012                         goto again;
1013                 }
1014                 if (ret < 0) {
1015                         btrfs_abort_transaction(trans, root, ret);
1016                         goto out;
1017                 }
1018
1019                 leaf = path->nodes[0];
1020                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1021                                     struct btrfs_file_extent_item);
1022                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1023                 btrfs_set_file_extent_num_bytes(leaf, fi,
1024                                                 split - key.offset);
1025
1026                 fi = btrfs_item_ptr(leaf, path->slots[0],
1027                                     struct btrfs_file_extent_item);
1028
1029                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1030                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1031                 btrfs_set_file_extent_num_bytes(leaf, fi,
1032                                                 extent_end - split);
1033                 btrfs_mark_buffer_dirty(leaf);
1034
1035                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1036                                            root->root_key.objectid,
1037                                            ino, orig_offset, 0);
1038                 BUG_ON(ret); /* -ENOMEM */
1039
1040                 if (split == start) {
1041                         key.offset = start;
1042                 } else {
1043                         BUG_ON(start != key.offset);
1044                         path->slots[0]--;
1045                         extent_end = end;
1046                 }
1047                 recow = 1;
1048         }
1049
1050         other_start = end;
1051         other_end = 0;
1052         if (extent_mergeable(leaf, path->slots[0] + 1,
1053                              ino, bytenr, orig_offset,
1054                              &other_start, &other_end)) {
1055                 if (recow) {
1056                         btrfs_release_path(path);
1057                         goto again;
1058                 }
1059                 extent_end = other_end;
1060                 del_slot = path->slots[0] + 1;
1061                 del_nr++;
1062                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1063                                         0, root->root_key.objectid,
1064                                         ino, orig_offset, 0);
1065                 BUG_ON(ret); /* -ENOMEM */
1066         }
1067         other_start = 0;
1068         other_end = start;
1069         if (extent_mergeable(leaf, path->slots[0] - 1,
1070                              ino, bytenr, orig_offset,
1071                              &other_start, &other_end)) {
1072                 if (recow) {
1073                         btrfs_release_path(path);
1074                         goto again;
1075                 }
1076                 key.offset = other_start;
1077                 del_slot = path->slots[0];
1078                 del_nr++;
1079                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1080                                         0, root->root_key.objectid,
1081                                         ino, orig_offset, 0);
1082                 BUG_ON(ret); /* -ENOMEM */
1083         }
1084         if (del_nr == 0) {
1085                 fi = btrfs_item_ptr(leaf, path->slots[0],
1086                            struct btrfs_file_extent_item);
1087                 btrfs_set_file_extent_type(leaf, fi,
1088                                            BTRFS_FILE_EXTENT_REG);
1089                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1090                 btrfs_mark_buffer_dirty(leaf);
1091         } else {
1092                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1093                            struct btrfs_file_extent_item);
1094                 btrfs_set_file_extent_type(leaf, fi,
1095                                            BTRFS_FILE_EXTENT_REG);
1096                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1097                 btrfs_set_file_extent_num_bytes(leaf, fi,
1098                                                 extent_end - key.offset);
1099                 btrfs_mark_buffer_dirty(leaf);
1100
1101                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1102                 if (ret < 0) {
1103                         btrfs_abort_transaction(trans, root, ret);
1104                         goto out;
1105                 }
1106         }
1107 out:
1108         btrfs_free_path(path);
1109         return 0;
1110 }
1111
1112 /*
1113  * on error we return an unlocked page and the error value
1114  * on success we return a locked page and 0
1115  */
1116 static int prepare_uptodate_page(struct page *page, u64 pos,
1117                                  bool force_uptodate)
1118 {
1119         int ret = 0;
1120
1121         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1122             !PageUptodate(page)) {
1123                 ret = btrfs_readpage(NULL, page);
1124                 if (ret)
1125                         return ret;
1126                 lock_page(page);
1127                 if (!PageUptodate(page)) {
1128                         unlock_page(page);
1129                         return -EIO;
1130                 }
1131         }
1132         return 0;
1133 }
1134
1135 /*
1136  * this gets pages into the page cache and locks them down, it also properly
1137  * waits for data=ordered extents to finish before allowing the pages to be
1138  * modified.
1139  */
1140 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1141                          struct page **pages, size_t num_pages,
1142                          loff_t pos, unsigned long first_index,
1143                          size_t write_bytes, bool force_uptodate)
1144 {
1145         struct extent_state *cached_state = NULL;
1146         int i;
1147         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1148         struct inode *inode = fdentry(file)->d_inode;
1149         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1150         int err = 0;
1151         int faili = 0;
1152         u64 start_pos;
1153         u64 last_pos;
1154
1155         start_pos = pos & ~((u64)root->sectorsize - 1);
1156         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1157
1158 again:
1159         for (i = 0; i < num_pages; i++) {
1160                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1161                                                mask | __GFP_WRITE);
1162                 if (!pages[i]) {
1163                         faili = i - 1;
1164                         err = -ENOMEM;
1165                         goto fail;
1166                 }
1167
1168                 if (i == 0)
1169                         err = prepare_uptodate_page(pages[i], pos,
1170                                                     force_uptodate);
1171                 if (i == num_pages - 1)
1172                         err = prepare_uptodate_page(pages[i],
1173                                                     pos + write_bytes, false);
1174                 if (err) {
1175                         page_cache_release(pages[i]);
1176                         faili = i - 1;
1177                         goto fail;
1178                 }
1179                 wait_on_page_writeback(pages[i]);
1180         }
1181         err = 0;
1182         if (start_pos < inode->i_size) {
1183                 struct btrfs_ordered_extent *ordered;
1184                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1185                                  start_pos, last_pos - 1, 0, &cached_state);
1186                 ordered = btrfs_lookup_first_ordered_extent(inode,
1187                                                             last_pos - 1);
1188                 if (ordered &&
1189                     ordered->file_offset + ordered->len > start_pos &&
1190                     ordered->file_offset < last_pos) {
1191                         btrfs_put_ordered_extent(ordered);
1192                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1193                                              start_pos, last_pos - 1,
1194                                              &cached_state, GFP_NOFS);
1195                         for (i = 0; i < num_pages; i++) {
1196                                 unlock_page(pages[i]);
1197                                 page_cache_release(pages[i]);
1198                         }
1199                         btrfs_wait_ordered_range(inode, start_pos,
1200                                                  last_pos - start_pos);
1201                         goto again;
1202                 }
1203                 if (ordered)
1204                         btrfs_put_ordered_extent(ordered);
1205
1206                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1207                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1208                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1209                                   0, 0, &cached_state, GFP_NOFS);
1210                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1211                                      start_pos, last_pos - 1, &cached_state,
1212                                      GFP_NOFS);
1213         }
1214         for (i = 0; i < num_pages; i++) {
1215                 if (clear_page_dirty_for_io(pages[i]))
1216                         account_page_redirty(pages[i]);
1217                 set_page_extent_mapped(pages[i]);
1218                 WARN_ON(!PageLocked(pages[i]));
1219         }
1220         return 0;
1221 fail:
1222         while (faili >= 0) {
1223                 unlock_page(pages[faili]);
1224                 page_cache_release(pages[faili]);
1225                 faili--;
1226         }
1227         return err;
1228
1229 }
1230
1231 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1232                                                struct iov_iter *i,
1233                                                loff_t pos)
1234 {
1235         struct inode *inode = fdentry(file)->d_inode;
1236         struct btrfs_root *root = BTRFS_I(inode)->root;
1237         struct page **pages = NULL;
1238         unsigned long first_index;
1239         size_t num_written = 0;
1240         int nrptrs;
1241         int ret = 0;
1242         bool force_page_uptodate = false;
1243
1244         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1245                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1246                      (sizeof(struct page *)));
1247         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1248         nrptrs = max(nrptrs, 8);
1249         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1250         if (!pages)
1251                 return -ENOMEM;
1252
1253         first_index = pos >> PAGE_CACHE_SHIFT;
1254
1255         while (iov_iter_count(i) > 0) {
1256                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1257                 size_t write_bytes = min(iov_iter_count(i),
1258                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1259                                          offset);
1260                 size_t num_pages = (write_bytes + offset +
1261                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1262                 size_t dirty_pages;
1263                 size_t copied;
1264
1265                 WARN_ON(num_pages > nrptrs);
1266
1267                 /*
1268                  * Fault pages before locking them in prepare_pages
1269                  * to avoid recursive lock
1270                  */
1271                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1272                         ret = -EFAULT;
1273                         break;
1274                 }
1275
1276                 ret = btrfs_delalloc_reserve_space(inode,
1277                                         num_pages << PAGE_CACHE_SHIFT);
1278                 if (ret)
1279                         break;
1280
1281                 /*
1282                  * This is going to setup the pages array with the number of
1283                  * pages we want, so we don't really need to worry about the
1284                  * contents of pages from loop to loop
1285                  */
1286                 ret = prepare_pages(root, file, pages, num_pages,
1287                                     pos, first_index, write_bytes,
1288                                     force_page_uptodate);
1289                 if (ret) {
1290                         btrfs_delalloc_release_space(inode,
1291                                         num_pages << PAGE_CACHE_SHIFT);
1292                         break;
1293                 }
1294
1295                 copied = btrfs_copy_from_user(pos, num_pages,
1296                                            write_bytes, pages, i);
1297
1298                 /*
1299                  * if we have trouble faulting in the pages, fall
1300                  * back to one page at a time
1301                  */
1302                 if (copied < write_bytes)
1303                         nrptrs = 1;
1304
1305                 if (copied == 0) {
1306                         force_page_uptodate = true;
1307                         dirty_pages = 0;
1308                 } else {
1309                         force_page_uptodate = false;
1310                         dirty_pages = (copied + offset +
1311                                        PAGE_CACHE_SIZE - 1) >>
1312                                        PAGE_CACHE_SHIFT;
1313                 }
1314
1315                 /*
1316                  * If we had a short copy we need to release the excess delaloc
1317                  * bytes we reserved.  We need to increment outstanding_extents
1318                  * because btrfs_delalloc_release_space will decrement it, but
1319                  * we still have an outstanding extent for the chunk we actually
1320                  * managed to copy.
1321                  */
1322                 if (num_pages > dirty_pages) {
1323                         if (copied > 0) {
1324                                 spin_lock(&BTRFS_I(inode)->lock);
1325                                 BTRFS_I(inode)->outstanding_extents++;
1326                                 spin_unlock(&BTRFS_I(inode)->lock);
1327                         }
1328                         btrfs_delalloc_release_space(inode,
1329                                         (num_pages - dirty_pages) <<
1330                                         PAGE_CACHE_SHIFT);
1331                 }
1332
1333                 if (copied > 0) {
1334                         ret = btrfs_dirty_pages(root, inode, pages,
1335                                                 dirty_pages, pos, copied,
1336                                                 NULL);
1337                         if (ret) {
1338                                 btrfs_delalloc_release_space(inode,
1339                                         dirty_pages << PAGE_CACHE_SHIFT);
1340                                 btrfs_drop_pages(pages, num_pages);
1341                                 break;
1342                         }
1343                 }
1344
1345                 btrfs_drop_pages(pages, num_pages);
1346
1347                 cond_resched();
1348
1349                 balance_dirty_pages_ratelimited(inode->i_mapping);
1350                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1351                         btrfs_btree_balance_dirty(root, 1);
1352
1353                 pos += copied;
1354                 num_written += copied;
1355         }
1356
1357         kfree(pages);
1358
1359         return num_written ? num_written : ret;
1360 }
1361
1362 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1363                                     const struct iovec *iov,
1364                                     unsigned long nr_segs, loff_t pos,
1365                                     loff_t *ppos, size_t count, size_t ocount)
1366 {
1367         struct file *file = iocb->ki_filp;
1368         struct iov_iter i;
1369         ssize_t written;
1370         ssize_t written_buffered;
1371         loff_t endbyte;
1372         int err;
1373
1374         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1375                                             count, ocount);
1376
1377         if (written < 0 || written == count)
1378                 return written;
1379
1380         pos += written;
1381         count -= written;
1382         iov_iter_init(&i, iov, nr_segs, count, written);
1383         written_buffered = __btrfs_buffered_write(file, &i, pos);
1384         if (written_buffered < 0) {
1385                 err = written_buffered;
1386                 goto out;
1387         }
1388         endbyte = pos + written_buffered - 1;
1389         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1390         if (err)
1391                 goto out;
1392         written += written_buffered;
1393         *ppos = pos + written_buffered;
1394         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1395                                  endbyte >> PAGE_CACHE_SHIFT);
1396 out:
1397         return written ? written : err;
1398 }
1399
1400 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1401                                     const struct iovec *iov,
1402                                     unsigned long nr_segs, loff_t pos)
1403 {
1404         struct file *file = iocb->ki_filp;
1405         struct inode *inode = fdentry(file)->d_inode;
1406         struct btrfs_root *root = BTRFS_I(inode)->root;
1407         loff_t *ppos = &iocb->ki_pos;
1408         u64 start_pos;
1409         ssize_t num_written = 0;
1410         ssize_t err = 0;
1411         size_t count, ocount;
1412
1413         sb_start_write(inode->i_sb);
1414
1415         mutex_lock(&inode->i_mutex);
1416
1417         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1418         if (err) {
1419                 mutex_unlock(&inode->i_mutex);
1420                 goto out;
1421         }
1422         count = ocount;
1423
1424         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1425         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1426         if (err) {
1427                 mutex_unlock(&inode->i_mutex);
1428                 goto out;
1429         }
1430
1431         if (count == 0) {
1432                 mutex_unlock(&inode->i_mutex);
1433                 goto out;
1434         }
1435
1436         err = file_remove_suid(file);
1437         if (err) {
1438                 mutex_unlock(&inode->i_mutex);
1439                 goto out;
1440         }
1441
1442         /*
1443          * If BTRFS flips readonly due to some impossible error
1444          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1445          * although we have opened a file as writable, we have
1446          * to stop this write operation to ensure FS consistency.
1447          */
1448         if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1449                 mutex_unlock(&inode->i_mutex);
1450                 err = -EROFS;
1451                 goto out;
1452         }
1453
1454         err = file_update_time(file);
1455         if (err) {
1456                 mutex_unlock(&inode->i_mutex);
1457                 goto out;
1458         }
1459
1460         start_pos = round_down(pos, root->sectorsize);
1461         if (start_pos > i_size_read(inode)) {
1462                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1463                 if (err) {
1464                         mutex_unlock(&inode->i_mutex);
1465                         goto out;
1466                 }
1467         }
1468
1469         if (unlikely(file->f_flags & O_DIRECT)) {
1470                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1471                                                    pos, ppos, count, ocount);
1472         } else {
1473                 struct iov_iter i;
1474
1475                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1476
1477                 num_written = __btrfs_buffered_write(file, &i, pos);
1478                 if (num_written > 0)
1479                         *ppos = pos + num_written;
1480         }
1481
1482         mutex_unlock(&inode->i_mutex);
1483
1484         /*
1485          * we want to make sure fsync finds this change
1486          * but we haven't joined a transaction running right now.
1487          *
1488          * Later on, someone is sure to update the inode and get the
1489          * real transid recorded.
1490          *
1491          * We set last_trans now to the fs_info generation + 1,
1492          * this will either be one more than the running transaction
1493          * or the generation used for the next transaction if there isn't
1494          * one running right now.
1495          */
1496         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1497         if (num_written > 0 || num_written == -EIOCBQUEUED) {
1498                 err = generic_write_sync(file, pos, num_written);
1499                 if (err < 0 && num_written > 0)
1500                         num_written = err;
1501         }
1502 out:
1503         sb_end_write(inode->i_sb);
1504         current->backing_dev_info = NULL;
1505         return num_written ? num_written : err;
1506 }
1507
1508 int btrfs_release_file(struct inode *inode, struct file *filp)
1509 {
1510         /*
1511          * ordered_data_close is set by settattr when we are about to truncate
1512          * a file from a non-zero size to a zero size.  This tries to
1513          * flush down new bytes that may have been written if the
1514          * application were using truncate to replace a file in place.
1515          */
1516         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1517                                &BTRFS_I(inode)->runtime_flags)) {
1518                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1519                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1520                         filemap_flush(inode->i_mapping);
1521         }
1522         if (filp->private_data)
1523                 btrfs_ioctl_trans_end(filp);
1524         return 0;
1525 }
1526
1527 /*
1528  * fsync call for both files and directories.  This logs the inode into
1529  * the tree log instead of forcing full commits whenever possible.
1530  *
1531  * It needs to call filemap_fdatawait so that all ordered extent updates are
1532  * in the metadata btree are up to date for copying to the log.
1533  *
1534  * It drops the inode mutex before doing the tree log commit.  This is an
1535  * important optimization for directories because holding the mutex prevents
1536  * new operations on the dir while we write to disk.
1537  */
1538 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1539 {
1540         struct dentry *dentry = file->f_path.dentry;
1541         struct inode *inode = dentry->d_inode;
1542         struct btrfs_root *root = BTRFS_I(inode)->root;
1543         int ret = 0;
1544         struct btrfs_trans_handle *trans;
1545
1546         trace_btrfs_sync_file(file, datasync);
1547
1548         /*
1549          * We write the dirty pages in the range and wait until they complete
1550          * out of the ->i_mutex. If so, we can flush the dirty pages by
1551          * multi-task, and make the performance up.
1552          */
1553         ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1554         if (ret)
1555                 return ret;
1556
1557         mutex_lock(&inode->i_mutex);
1558
1559         /*
1560          * We flush the dirty pages again to avoid some dirty pages in the
1561          * range being left.
1562          */
1563         atomic_inc(&root->log_batch);
1564         btrfs_wait_ordered_range(inode, start, end);
1565         atomic_inc(&root->log_batch);
1566
1567         /*
1568          * check the transaction that last modified this inode
1569          * and see if its already been committed
1570          */
1571         if (!BTRFS_I(inode)->last_trans) {
1572                 mutex_unlock(&inode->i_mutex);
1573                 goto out;
1574         }
1575
1576         /*
1577          * if the last transaction that changed this file was before
1578          * the current transaction, we can bail out now without any
1579          * syncing
1580          */
1581         smp_mb();
1582         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1583             BTRFS_I(inode)->last_trans <=
1584             root->fs_info->last_trans_committed) {
1585                 BTRFS_I(inode)->last_trans = 0;
1586
1587                 /*
1588                  * We'v had everything committed since the last time we were
1589                  * modified so clear this flag in case it was set for whatever
1590                  * reason, it's no longer relevant.
1591                  */
1592                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1593                           &BTRFS_I(inode)->runtime_flags);
1594                 mutex_unlock(&inode->i_mutex);
1595                 goto out;
1596         }
1597
1598         /*
1599          * ok we haven't committed the transaction yet, lets do a commit
1600          */
1601         if (file->private_data)
1602                 btrfs_ioctl_trans_end(file);
1603
1604         trans = btrfs_start_transaction(root, 0);
1605         if (IS_ERR(trans)) {
1606                 ret = PTR_ERR(trans);
1607                 mutex_unlock(&inode->i_mutex);
1608                 goto out;
1609         }
1610
1611         ret = btrfs_log_dentry_safe(trans, root, dentry);
1612         if (ret < 0) {
1613                 mutex_unlock(&inode->i_mutex);
1614                 goto out;
1615         }
1616
1617         /* we've logged all the items and now have a consistent
1618          * version of the file in the log.  It is possible that
1619          * someone will come in and modify the file, but that's
1620          * fine because the log is consistent on disk, and we
1621          * have references to all of the file's extents
1622          *
1623          * It is possible that someone will come in and log the
1624          * file again, but that will end up using the synchronization
1625          * inside btrfs_sync_log to keep things safe.
1626          */
1627         mutex_unlock(&inode->i_mutex);
1628
1629         if (ret != BTRFS_NO_LOG_SYNC) {
1630                 if (ret > 0) {
1631                         ret = btrfs_commit_transaction(trans, root);
1632                 } else {
1633                         ret = btrfs_sync_log(trans, root);
1634                         if (ret == 0)
1635                                 ret = btrfs_end_transaction(trans, root);
1636                         else
1637                                 ret = btrfs_commit_transaction(trans, root);
1638                 }
1639         } else {
1640                 ret = btrfs_end_transaction(trans, root);
1641         }
1642 out:
1643         return ret > 0 ? -EIO : ret;
1644 }
1645
1646 static const struct vm_operations_struct btrfs_file_vm_ops = {
1647         .fault          = filemap_fault,
1648         .page_mkwrite   = btrfs_page_mkwrite,
1649         .remap_pages    = generic_file_remap_pages,
1650 };
1651
1652 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1653 {
1654         struct address_space *mapping = filp->f_mapping;
1655
1656         if (!mapping->a_ops->readpage)
1657                 return -ENOEXEC;
1658
1659         file_accessed(filp);
1660         vma->vm_ops = &btrfs_file_vm_ops;
1661
1662         return 0;
1663 }
1664
1665 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1666                           int slot, u64 start, u64 end)
1667 {
1668         struct btrfs_file_extent_item *fi;
1669         struct btrfs_key key;
1670
1671         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1672                 return 0;
1673
1674         btrfs_item_key_to_cpu(leaf, &key, slot);
1675         if (key.objectid != btrfs_ino(inode) ||
1676             key.type != BTRFS_EXTENT_DATA_KEY)
1677                 return 0;
1678
1679         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1680
1681         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1682                 return 0;
1683
1684         if (btrfs_file_extent_disk_bytenr(leaf, fi))
1685                 return 0;
1686
1687         if (key.offset == end)
1688                 return 1;
1689         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1690                 return 1;
1691         return 0;
1692 }
1693
1694 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1695                       struct btrfs_path *path, u64 offset, u64 end)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct extent_buffer *leaf;
1699         struct btrfs_file_extent_item *fi;
1700         struct extent_map *hole_em;
1701         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1702         struct btrfs_key key;
1703         int ret;
1704
1705         key.objectid = btrfs_ino(inode);
1706         key.type = BTRFS_EXTENT_DATA_KEY;
1707         key.offset = offset;
1708
1709
1710         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1711         if (ret < 0)
1712                 return ret;
1713         BUG_ON(!ret);
1714
1715         leaf = path->nodes[0];
1716         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
1717                 u64 num_bytes;
1718
1719                 path->slots[0]--;
1720                 fi = btrfs_item_ptr(leaf, path->slots[0],
1721                                     struct btrfs_file_extent_item);
1722                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
1723                         end - offset;
1724                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1725                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1726                 btrfs_set_file_extent_offset(leaf, fi, 0);
1727                 btrfs_mark_buffer_dirty(leaf);
1728                 goto out;
1729         }
1730
1731         if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
1732                 u64 num_bytes;
1733
1734                 path->slots[0]++;
1735                 key.offset = offset;
1736                 btrfs_set_item_key_safe(trans, root, path, &key);
1737                 fi = btrfs_item_ptr(leaf, path->slots[0],
1738                                     struct btrfs_file_extent_item);
1739                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
1740                         offset;
1741                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1742                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1743                 btrfs_set_file_extent_offset(leaf, fi, 0);
1744                 btrfs_mark_buffer_dirty(leaf);
1745                 goto out;
1746         }
1747         btrfs_release_path(path);
1748
1749         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
1750                                        0, 0, end - offset, 0, end - offset,
1751                                        0, 0, 0);
1752         if (ret)
1753                 return ret;
1754
1755 out:
1756         btrfs_release_path(path);
1757
1758         hole_em = alloc_extent_map();
1759         if (!hole_em) {
1760                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
1761                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1762                         &BTRFS_I(inode)->runtime_flags);
1763         } else {
1764                 hole_em->start = offset;
1765                 hole_em->len = end - offset;
1766                 hole_em->orig_start = offset;
1767
1768                 hole_em->block_start = EXTENT_MAP_HOLE;
1769                 hole_em->block_len = 0;
1770                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
1771                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
1772                 hole_em->generation = trans->transid;
1773
1774                 do {
1775                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
1776                         write_lock(&em_tree->lock);
1777                         ret = add_extent_mapping(em_tree, hole_em);
1778                         if (!ret)
1779                                 list_move(&hole_em->list,
1780                                           &em_tree->modified_extents);
1781                         write_unlock(&em_tree->lock);
1782                 } while (ret == -EEXIST);
1783                 free_extent_map(hole_em);
1784                 if (ret)
1785                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1786                                 &BTRFS_I(inode)->runtime_flags);
1787         }
1788
1789         return 0;
1790 }
1791
1792 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
1793 {
1794         struct btrfs_root *root = BTRFS_I(inode)->root;
1795         struct extent_state *cached_state = NULL;
1796         struct btrfs_path *path;
1797         struct btrfs_block_rsv *rsv;
1798         struct btrfs_trans_handle *trans;
1799         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1800         u64 lockstart = (offset + mask) & ~mask;
1801         u64 lockend = ((offset + len) & ~mask) - 1;
1802         u64 cur_offset = lockstart;
1803         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
1804         u64 drop_end;
1805         unsigned long nr;
1806         int ret = 0;
1807         int err = 0;
1808         bool same_page = (offset >> PAGE_CACHE_SHIFT) ==
1809                 ((offset + len) >> PAGE_CACHE_SHIFT);
1810
1811         btrfs_wait_ordered_range(inode, offset, len);
1812
1813         mutex_lock(&inode->i_mutex);
1814         if (offset >= inode->i_size) {
1815                 mutex_unlock(&inode->i_mutex);
1816                 return 0;
1817         }
1818
1819         /*
1820          * Only do this if we are in the same page and we aren't doing the
1821          * entire page.
1822          */
1823         if (same_page && len < PAGE_CACHE_SIZE) {
1824                 ret = btrfs_truncate_page(inode, offset, len, 0);
1825                 mutex_unlock(&inode->i_mutex);
1826                 return ret;
1827         }
1828
1829         /* zero back part of the first page */
1830         ret = btrfs_truncate_page(inode, offset, 0, 0);
1831         if (ret) {
1832                 mutex_unlock(&inode->i_mutex);
1833                 return ret;
1834         }
1835
1836         /* zero the front end of the last page */
1837         ret = btrfs_truncate_page(inode, offset + len, 0, 1);
1838         if (ret) {
1839                 mutex_unlock(&inode->i_mutex);
1840                 return ret;
1841         }
1842
1843         if (lockend < lockstart) {
1844                 mutex_unlock(&inode->i_mutex);
1845                 return 0;
1846         }
1847
1848         while (1) {
1849                 struct btrfs_ordered_extent *ordered;
1850
1851                 truncate_pagecache_range(inode, lockstart, lockend);
1852
1853                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1854                                  0, &cached_state);
1855                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
1856
1857                 /*
1858                  * We need to make sure we have no ordered extents in this range
1859                  * and nobody raced in and read a page in this range, if we did
1860                  * we need to try again.
1861                  */
1862                 if ((!ordered ||
1863                     (ordered->file_offset + ordered->len < lockstart ||
1864                      ordered->file_offset > lockend)) &&
1865                      !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
1866                                      lockend, EXTENT_UPTODATE, 0,
1867                                      cached_state)) {
1868                         if (ordered)
1869                                 btrfs_put_ordered_extent(ordered);
1870                         break;
1871                 }
1872                 if (ordered)
1873                         btrfs_put_ordered_extent(ordered);
1874                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
1875                                      lockend, &cached_state, GFP_NOFS);
1876                 btrfs_wait_ordered_range(inode, lockstart,
1877                                          lockend - lockstart + 1);
1878         }
1879
1880         path = btrfs_alloc_path();
1881         if (!path) {
1882                 ret = -ENOMEM;
1883                 goto out;
1884         }
1885
1886         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
1887         if (!rsv) {
1888                 ret = -ENOMEM;
1889                 goto out_free;
1890         }
1891         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
1892         rsv->failfast = 1;
1893
1894         /*
1895          * 1 - update the inode
1896          * 1 - removing the extents in the range
1897          * 1 - adding the hole extent
1898          */
1899         trans = btrfs_start_transaction(root, 3);
1900         if (IS_ERR(trans)) {
1901                 err = PTR_ERR(trans);
1902                 goto out_free;
1903         }
1904
1905         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
1906                                       min_size);
1907         BUG_ON(ret);
1908         trans->block_rsv = rsv;
1909
1910         while (cur_offset < lockend) {
1911                 ret = __btrfs_drop_extents(trans, root, inode, path,
1912                                            cur_offset, lockend + 1,
1913                                            &drop_end, 1);
1914                 if (ret != -ENOSPC)
1915                         break;
1916
1917                 trans->block_rsv = &root->fs_info->trans_block_rsv;
1918
1919                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
1920                 if (ret) {
1921                         err = ret;
1922                         break;
1923                 }
1924
1925                 cur_offset = drop_end;
1926
1927                 ret = btrfs_update_inode(trans, root, inode);
1928                 if (ret) {
1929                         err = ret;
1930                         break;
1931                 }
1932
1933                 nr = trans->blocks_used;
1934                 btrfs_end_transaction(trans, root);
1935                 btrfs_btree_balance_dirty(root, nr);
1936
1937                 trans = btrfs_start_transaction(root, 3);
1938                 if (IS_ERR(trans)) {
1939                         ret = PTR_ERR(trans);
1940                         trans = NULL;
1941                         break;
1942                 }
1943
1944                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
1945                                               rsv, min_size);
1946                 BUG_ON(ret);    /* shouldn't happen */
1947                 trans->block_rsv = rsv;
1948         }
1949
1950         if (ret) {
1951                 err = ret;
1952                 goto out_trans;
1953         }
1954
1955         trans->block_rsv = &root->fs_info->trans_block_rsv;
1956         ret = fill_holes(trans, inode, path, cur_offset, drop_end);
1957         if (ret) {
1958                 err = ret;
1959                 goto out_trans;
1960         }
1961
1962 out_trans:
1963         if (!trans)
1964                 goto out_free;
1965
1966         trans->block_rsv = &root->fs_info->trans_block_rsv;
1967         ret = btrfs_update_inode(trans, root, inode);
1968         nr = trans->blocks_used;
1969         btrfs_end_transaction(trans, root);
1970         btrfs_btree_balance_dirty(root, nr);
1971 out_free:
1972         btrfs_free_path(path);
1973         btrfs_free_block_rsv(root, rsv);
1974 out:
1975         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1976                              &cached_state, GFP_NOFS);
1977         mutex_unlock(&inode->i_mutex);
1978         if (ret && !err)
1979                 err = ret;
1980         return err;
1981 }
1982
1983 static long btrfs_fallocate(struct file *file, int mode,
1984                             loff_t offset, loff_t len)
1985 {
1986         struct inode *inode = file->f_path.dentry->d_inode;
1987         struct extent_state *cached_state = NULL;
1988         u64 cur_offset;
1989         u64 last_byte;
1990         u64 alloc_start;
1991         u64 alloc_end;
1992         u64 alloc_hint = 0;
1993         u64 locked_end;
1994         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1995         struct extent_map *em;
1996         int ret;
1997
1998         alloc_start = offset & ~mask;
1999         alloc_end =  (offset + len + mask) & ~mask;
2000
2001         /* Make sure we aren't being give some crap mode */
2002         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2003                 return -EOPNOTSUPP;
2004
2005         if (mode & FALLOC_FL_PUNCH_HOLE)
2006                 return btrfs_punch_hole(inode, offset, len);
2007
2008         /*
2009          * Make sure we have enough space before we do the
2010          * allocation.
2011          */
2012         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start + 1);
2013         if (ret)
2014                 return ret;
2015
2016         /*
2017          * wait for ordered IO before we have any locks.  We'll loop again
2018          * below with the locks held.
2019          */
2020         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
2021
2022         mutex_lock(&inode->i_mutex);
2023         ret = inode_newsize_ok(inode, alloc_end);
2024         if (ret)
2025                 goto out;
2026
2027         if (alloc_start > inode->i_size) {
2028                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2029                                         alloc_start);
2030                 if (ret)
2031                         goto out;
2032         }
2033
2034         locked_end = alloc_end - 1;
2035         while (1) {
2036                 struct btrfs_ordered_extent *ordered;
2037
2038                 /* the extent lock is ordered inside the running
2039                  * transaction
2040                  */
2041                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2042                                  locked_end, 0, &cached_state);
2043                 ordered = btrfs_lookup_first_ordered_extent(inode,
2044                                                             alloc_end - 1);
2045                 if (ordered &&
2046                     ordered->file_offset + ordered->len > alloc_start &&
2047                     ordered->file_offset < alloc_end) {
2048                         btrfs_put_ordered_extent(ordered);
2049                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2050                                              alloc_start, locked_end,
2051                                              &cached_state, GFP_NOFS);
2052                         /*
2053                          * we can't wait on the range with the transaction
2054                          * running or with the extent lock held
2055                          */
2056                         btrfs_wait_ordered_range(inode, alloc_start,
2057                                                  alloc_end - alloc_start);
2058                 } else {
2059                         if (ordered)
2060                                 btrfs_put_ordered_extent(ordered);
2061                         break;
2062                 }
2063         }
2064
2065         cur_offset = alloc_start;
2066         while (1) {
2067                 u64 actual_end;
2068
2069                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2070                                       alloc_end - cur_offset, 0);
2071                 if (IS_ERR_OR_NULL(em)) {
2072                         if (!em)
2073                                 ret = -ENOMEM;
2074                         else
2075                                 ret = PTR_ERR(em);
2076                         break;
2077                 }
2078                 last_byte = min(extent_map_end(em), alloc_end);
2079                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2080                 last_byte = (last_byte + mask) & ~mask;
2081
2082                 if (em->block_start == EXTENT_MAP_HOLE ||
2083                     (cur_offset >= inode->i_size &&
2084                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2085                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2086                                                         last_byte - cur_offset,
2087                                                         1 << inode->i_blkbits,
2088                                                         offset + len,
2089                                                         &alloc_hint);
2090
2091                         if (ret < 0) {
2092                                 free_extent_map(em);
2093                                 break;
2094                         }
2095                 } else if (actual_end > inode->i_size &&
2096                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2097                         /*
2098                          * We didn't need to allocate any more space, but we
2099                          * still extended the size of the file so we need to
2100                          * update i_size.
2101                          */
2102                         inode->i_ctime = CURRENT_TIME;
2103                         i_size_write(inode, actual_end);
2104                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2105                 }
2106                 free_extent_map(em);
2107
2108                 cur_offset = last_byte;
2109                 if (cur_offset >= alloc_end) {
2110                         ret = 0;
2111                         break;
2112                 }
2113         }
2114         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2115                              &cached_state, GFP_NOFS);
2116 out:
2117         mutex_unlock(&inode->i_mutex);
2118         /* Let go of our reservation. */
2119         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start + 1);
2120         return ret;
2121 }
2122
2123 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2124 {
2125         struct btrfs_root *root = BTRFS_I(inode)->root;
2126         struct extent_map *em;
2127         struct extent_state *cached_state = NULL;
2128         u64 lockstart = *offset;
2129         u64 lockend = i_size_read(inode);
2130         u64 start = *offset;
2131         u64 orig_start = *offset;
2132         u64 len = i_size_read(inode);
2133         u64 last_end = 0;
2134         int ret = 0;
2135
2136         lockend = max_t(u64, root->sectorsize, lockend);
2137         if (lockend <= lockstart)
2138                 lockend = lockstart + root->sectorsize;
2139
2140         len = lockend - lockstart + 1;
2141
2142         len = max_t(u64, len, root->sectorsize);
2143         if (inode->i_size == 0)
2144                 return -ENXIO;
2145
2146         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2147                          &cached_state);
2148
2149         /*
2150          * Delalloc is such a pain.  If we have a hole and we have pending
2151          * delalloc for a portion of the hole we will get back a hole that
2152          * exists for the entire range since it hasn't been actually written
2153          * yet.  So to take care of this case we need to look for an extent just
2154          * before the position we want in case there is outstanding delalloc
2155          * going on here.
2156          */
2157         if (whence == SEEK_HOLE && start != 0) {
2158                 if (start <= root->sectorsize)
2159                         em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
2160                                                      root->sectorsize, 0);
2161                 else
2162                         em = btrfs_get_extent_fiemap(inode, NULL, 0,
2163                                                      start - root->sectorsize,
2164                                                      root->sectorsize, 0);
2165                 if (IS_ERR(em)) {
2166                         ret = PTR_ERR(em);
2167                         goto out;
2168                 }
2169                 last_end = em->start + em->len;
2170                 if (em->block_start == EXTENT_MAP_DELALLOC)
2171                         last_end = min_t(u64, last_end, inode->i_size);
2172                 free_extent_map(em);
2173         }
2174
2175         while (1) {
2176                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2177                 if (IS_ERR(em)) {
2178                         ret = PTR_ERR(em);
2179                         break;
2180                 }
2181
2182                 if (em->block_start == EXTENT_MAP_HOLE) {
2183                         if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2184                                 if (last_end <= orig_start) {
2185                                         free_extent_map(em);
2186                                         ret = -ENXIO;
2187                                         break;
2188                                 }
2189                         }
2190
2191                         if (whence == SEEK_HOLE) {
2192                                 *offset = start;
2193                                 free_extent_map(em);
2194                                 break;
2195                         }
2196                 } else {
2197                         if (whence == SEEK_DATA) {
2198                                 if (em->block_start == EXTENT_MAP_DELALLOC) {
2199                                         if (start >= inode->i_size) {
2200                                                 free_extent_map(em);
2201                                                 ret = -ENXIO;
2202                                                 break;
2203                                         }
2204                                 }
2205
2206                                 *offset = start;
2207                                 free_extent_map(em);
2208                                 break;
2209                         }
2210                 }
2211
2212                 start = em->start + em->len;
2213                 last_end = em->start + em->len;
2214
2215                 if (em->block_start == EXTENT_MAP_DELALLOC)
2216                         last_end = min_t(u64, last_end, inode->i_size);
2217
2218                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2219                         free_extent_map(em);
2220                         ret = -ENXIO;
2221                         break;
2222                 }
2223                 free_extent_map(em);
2224                 cond_resched();
2225         }
2226         if (!ret)
2227                 *offset = min(*offset, inode->i_size);
2228 out:
2229         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2230                              &cached_state, GFP_NOFS);
2231         return ret;
2232 }
2233
2234 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2235 {
2236         struct inode *inode = file->f_mapping->host;
2237         int ret;
2238
2239         mutex_lock(&inode->i_mutex);
2240         switch (whence) {
2241         case SEEK_END:
2242         case SEEK_CUR:
2243                 offset = generic_file_llseek(file, offset, whence);
2244                 goto out;
2245         case SEEK_DATA:
2246         case SEEK_HOLE:
2247                 if (offset >= i_size_read(inode)) {
2248                         mutex_unlock(&inode->i_mutex);
2249                         return -ENXIO;
2250                 }
2251
2252                 ret = find_desired_extent(inode, &offset, whence);
2253                 if (ret) {
2254                         mutex_unlock(&inode->i_mutex);
2255                         return ret;
2256                 }
2257         }
2258
2259         if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
2260                 offset = -EINVAL;
2261                 goto out;
2262         }
2263         if (offset > inode->i_sb->s_maxbytes) {
2264                 offset = -EINVAL;
2265                 goto out;
2266         }
2267
2268         /* Special lock needed here? */
2269         if (offset != file->f_pos) {
2270                 file->f_pos = offset;
2271                 file->f_version = 0;
2272         }
2273 out:
2274         mutex_unlock(&inode->i_mutex);
2275         return offset;
2276 }
2277
2278 const struct file_operations btrfs_file_operations = {
2279         .llseek         = btrfs_file_llseek,
2280         .read           = do_sync_read,
2281         .write          = do_sync_write,
2282         .aio_read       = generic_file_aio_read,
2283         .splice_read    = generic_file_splice_read,
2284         .aio_write      = btrfs_file_aio_write,
2285         .mmap           = btrfs_file_mmap,
2286         .open           = generic_file_open,
2287         .release        = btrfs_release_file,
2288         .fsync          = btrfs_sync_file,
2289         .fallocate      = btrfs_fallocate,
2290         .unlocked_ioctl = btrfs_ioctl,
2291 #ifdef CONFIG_COMPAT
2292         .compat_ioctl   = btrfs_ioctl,
2293 #endif
2294 };