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
  43
  44 /* simple helper to fault in pages and copy.  This should go away
  45  * and be replaced with calls into generic code.
  46  */
  47 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  48                                          size_t write_bytes,
  49                                          struct page **prepared_pages,
  50                                          struct iov_iter *i)
  51 {
  52         size_t copied = 0;
  53         size_t total_copied = 0;
  54         int pg = 0;
  55         int offset = pos & (PAGE_CACHE_SIZE - 1);
  56
  57         while (write_bytes > 0) {
  58                 size_t count = min_t(size_t,
  59                                      PAGE_CACHE_SIZE - offset, write_bytes);
  60                 struct page *page = prepared_pages[pg];
  61                 /*
  62                  * Copy data from userspace to the current page
  63                  *
  64                  * Disable pagefault to avoid recursive lock since
  65                  * the pages are already locked
  66                  */
  67                 pagefault_disable();
  68                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
  69                 pagefault_enable();
  70
  71                 /* Flush processor's dcache for this page */
  72                 flush_dcache_page(page);
  73
  74                 /*
  75                  * if we get a partial write, we can end up with
  76                  * partially up to date pages.  These add
  77                  * a lot of complexity, so make sure they don't
  78                  * happen by forcing this copy to be retried.
  79                  *
  80                  * The rest of the btrfs_file_write code will fall
  81                  * back to page at a time copies after we return 0.
  82                  */
  83                 if (!PageUptodate(page) && copied < count)
  84                         copied = 0;
  85
  86                 iov_iter_advance(i, copied);
  87                 write_bytes -= copied;
  88                 total_copied += copied;
  89
  90                 /* Return to btrfs_file_aio_write to fault page */
  91                 if (unlikely(copied == 0))
  92                         break;
  93
  94                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
  95                         offset += copied;
  96                 } else {
  97                         pg++;
  98                         offset = 0;
  99                 }
 100         }
 101         return total_copied;
 102 }
 103
 104 /*
 105  * unlocks pages after btrfs_file_write is done with them
 106  */
 107 void btrfs_drop_pages(struct page **pages, size_t num_pages)
 108 {
 109         size_t i;
 110         for (i = 0; i < num_pages; i++) {
 111                 /* page checked is some magic around finding pages that
 112                  * have been modified without going through btrfs_set_page_dirty
 113                  * clear it here
 114                  */
 115                 ClearPageChecked(pages[i]);
 116                 unlock_page(pages[i]);
 117                 mark_page_accessed(pages[i]);
 118                 page_cache_release(pages[i]);
 119         }
 120 }
 121
 122 /*
 123  * after copy_from_user, pages need to be dirtied and we need to make
 124  * sure holes are created between the current EOF and the start of
 125  * any next extents (if required).
 126  *
 127  * this also makes the decision about creating an inline extent vs
 128  * doing real data extents, marking pages dirty and delalloc as required.
 129  */
 130 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
 131                       struct page **pages, size_t num_pages,
 132                       loff_t pos, size_t write_bytes,
 133                       struct extent_state **cached)
 134 {
 135         int err = 0;
 136         int i;
 137         u64 num_bytes;
 138         u64 start_pos;
 139         u64 end_of_last_block;
 140         u64 end_pos = pos + write_bytes;
 141         loff_t isize = i_size_read(inode);
 142
 143         start_pos = pos & ~((u64)root->sectorsize - 1);
 144         num_bytes = (write_bytes + pos - start_pos +
 145                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 146
 147         end_of_last_block = start_pos + num_bytes - 1;
 148         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 149                                         cached);
 150         if (err)
 151                 return err;
 152
 153         for (i = 0; i < num_pages; i++) {
 154                 struct page *p = pages[i];
 155                 SetPageUptodate(p);
 156                 ClearPageChecked(p);
 157                 set_page_dirty(p);
 158         }
 159
 160         /*
 161          * we've only changed i_size in ram, and we haven't updated
 162          * the disk i_size.  There is no need to log the inode
 163          * at this time.
 164          */
 165         if (end_pos > isize)
 166                 i_size_write(inode, end_pos);
 167         return 0;
 168 }
 169
 170 /*
 171  * this drops all the extents in the cache that intersect the range
 172  * [start, end].  Existing extents are split as required.
 173  */
 174 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 175                             int skip_pinned)
 176 {
 177         struct extent_map *em;
 178         struct extent_map *split = NULL;
 179         struct extent_map *split2 = NULL;
 180         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 181         u64 len = end - start + 1;
 182         int ret;
 183         int testend = 1;
 184         unsigned long flags;
 185         int compressed = 0;
 186
 187         WARN_ON(end < start);
 188         if (end == (u64)-1) {
 189                 len = (u64)-1;
 190                 testend = 0;
 191         }
 192         while (1) {
 193                 if (!split)
 194                         split = alloc_extent_map(GFP_NOFS);
 195                 if (!split2)
 196                         split2 = alloc_extent_map(GFP_NOFS);
 197                 BUG_ON(!split || !split2);
 198
 199                 write_lock(&em_tree->lock);
 200                 em = lookup_extent_mapping(em_tree, start, len);
 201                 if (!em) {
 202                         write_unlock(&em_tree->lock);
 203                         break;
 204                 }
 205                 flags = em->flags;
 206                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 207                         if (testend && em->start + em->len >= start + len) {
 208                                 free_extent_map(em);
 209                                 write_unlock(&em_tree->lock);
 210                                 break;
 211                         }
 212                         start = em->start + em->len;
 213                         if (testend)
 214                                 len = start + len - (em->start + em->len);
 215                         free_extent_map(em);
 216                         write_unlock(&em_tree->lock);
 217                         continue;
 218                 }
 219                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 220                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 221                 remove_extent_mapping(em_tree, em);
 222
 223                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 224                     em->start < start) {
 225                         split->start = em->start;
 226                         split->len = start - em->start;
 227                         split->orig_start = em->orig_start;
 228                         split->block_start = em->block_start;
 229
 230                         if (compressed)
 231                                 split->block_len = em->block_len;
 232                         else
 233                                 split->block_len = split->len;
 234
 235                         split->bdev = em->bdev;
 236                         split->flags = flags;
 237                         split->compress_type = em->compress_type;
 238                         ret = add_extent_mapping(em_tree, split);
 239                         BUG_ON(ret);
 240                         free_extent_map(split);
 241                         split = split2;
 242                         split2 = NULL;
 243                 }
 244                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 245                     testend && em->start + em->len > start + len) {
 246                         u64 diff = start + len - em->start;
 247
 248                         split->start = start + len;
 249                         split->len = em->start + em->len - (start + len);
 250                         split->bdev = em->bdev;
 251                         split->flags = flags;
 252                         split->compress_type = em->compress_type;
 253
 254                         if (compressed) {
 255                                 split->block_len = em->block_len;
 256                                 split->block_start = em->block_start;
 257                                 split->orig_start = em->orig_start;
 258                         } else {
 259                                 split->block_len = split->len;
 260                                 split->block_start = em->block_start + diff;
 261                                 split->orig_start = split->start;
 262                         }
 263
 264                         ret = add_extent_mapping(em_tree, split);
 265                         BUG_ON(ret);
 266                         free_extent_map(split);
 267                         split = NULL;
 268                 }
 269                 write_unlock(&em_tree->lock);
 270
 271                 /* once for us */
 272                 free_extent_map(em);
 273                 /* once for the tree*/
 274                 free_extent_map(em);
 275         }
 276         if (split)
 277                 free_extent_map(split);
 278         if (split2)
 279                 free_extent_map(split2);
 280         return 0;
 281 }
 282
 283 /*
 284  * this is very complex, but the basic idea is to drop all extents
 285  * in the range start - end.  hint_block is filled in with a block number
 286  * that would be a good hint to the block allocator for this file.
 287  *
 288  * If an extent intersects the range but is not entirely inside the range
 289  * it is either truncated or split.  Anything entirely inside the range
 290  * is deleted from the tree.
 291  */
 292 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
 293                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
 294 {
 295         struct btrfs_root *root = BTRFS_I(inode)->root;
 296         struct extent_buffer *leaf;
 297         struct btrfs_file_extent_item *fi;
 298         struct btrfs_path *path;
 299         struct btrfs_key key;
 300         struct btrfs_key new_key;
 301         u64 search_start = start;
 302         u64 disk_bytenr = 0;
 303         u64 num_bytes = 0;
 304         u64 extent_offset = 0;
 305         u64 extent_end = 0;
 306         int del_nr = 0;
 307         int del_slot = 0;
 308         int extent_type;
 309         int recow;
 310         int ret;
 311
 312         if (drop_cache)
 313                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 314
 315         path = btrfs_alloc_path();
 316         if (!path)
 317                 return -ENOMEM;
 318
 319         while (1) {
 320                 recow = 0;
 321                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 322                                                search_start, -1);
 323                 if (ret < 0)
 324                         break;
 325                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 326                         leaf = path->nodes[0];
 327                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 328                         if (key.objectid == inode->i_ino &&
 329                             key.type == BTRFS_EXTENT_DATA_KEY)
 330                                 path->slots[0]--;
 331                 }
 332                 ret = 0;
 333 next_slot:
 334                 leaf = path->nodes[0];
 335                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 336                         BUG_ON(del_nr > 0);
 337                         ret = btrfs_next_leaf(root, path);
 338                         if (ret < 0)
 339                                 break;
 340                         if (ret > 0) {
 341                                 ret = 0;
 342                                 break;
 343                         }
 344                         leaf = path->nodes[0];
 345                         recow = 1;
 346                 }
 347
 348                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 349                 if (key.objectid > inode->i_ino ||
 350                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 351                         break;
 352
 353                 fi = btrfs_item_ptr(leaf, path->slots[0],
 354                                     struct btrfs_file_extent_item);
 355                 extent_type = btrfs_file_extent_type(leaf, fi);
 356
 357                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 358                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 359                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 360                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 361                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 362                         extent_end = key.offset +
 363                                 btrfs_file_extent_num_bytes(leaf, fi);
 364                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 365                         extent_end = key.offset +
 366                                 btrfs_file_extent_inline_len(leaf, fi);
 367                 } else {
 368                         WARN_ON(1);
 369                         extent_end = search_start;
 370                 }
 371
 372                 if (extent_end <= search_start) {
 373                         path->slots[0]++;
 374                         goto next_slot;
 375                 }
 376
 377                 search_start = max(key.offset, start);
 378                 if (recow) {
 379                         btrfs_release_path(root, path);
 380                         continue;
 381                 }
 382
 383                 /*
 384                  *     | - range to drop - |
 385                  *  | -------- extent -------- |
 386                  */
 387                 if (start > key.offset && end < extent_end) {
 388                         BUG_ON(del_nr > 0);
 389                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 390
 391                         memcpy(&new_key, &key, sizeof(new_key));
 392                         new_key.offset = start;
 393                         ret = btrfs_duplicate_item(trans, root, path,
 394                                                    &new_key);
 395                         if (ret == -EAGAIN) {
 396                                 btrfs_release_path(root, path);
 397                                 continue;
 398                         }
 399                         if (ret < 0)
 400                                 break;
 401
 402                         leaf = path->nodes[0];
 403                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 404                                             struct btrfs_file_extent_item);
 405                         btrfs_set_file_extent_num_bytes(leaf, fi,
 406                                                         start - key.offset);
 407
 408                         fi = btrfs_item_ptr(leaf, path->slots[0],
 409                                             struct btrfs_file_extent_item);
 410
 411                         extent_offset += start - key.offset;
 412                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 413                         btrfs_set_file_extent_num_bytes(leaf, fi,
 414                                                         extent_end - start);
 415                         btrfs_mark_buffer_dirty(leaf);
 416
 417                         if (disk_bytenr > 0) {
 418                                 ret = btrfs_inc_extent_ref(trans, root,
 419                                                 disk_bytenr, num_bytes, 0,
 420                                                 root->root_key.objectid,
 421                                                 new_key.objectid,
 422                                                 start - extent_offset);
 423                                 BUG_ON(ret);
 424                                 *hint_byte = disk_bytenr;
 425                         }
 426                         key.offset = start;
 427                 }
 428                 /*
 429                  *  | ---- range to drop ----- |
 430                  *      | -------- extent -------- |
 431                  */
 432                 if (start <= key.offset && end < extent_end) {
 433                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 434
 435                         memcpy(&new_key, &key, sizeof(new_key));
 436                         new_key.offset = end;
 437                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 438
 439                         extent_offset += end - key.offset;
 440                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 441                         btrfs_set_file_extent_num_bytes(leaf, fi,
 442                                                         extent_end - end);
 443                         btrfs_mark_buffer_dirty(leaf);
 444                         if (disk_bytenr > 0) {
 445                                 inode_sub_bytes(inode, end - key.offset);
 446                                 *hint_byte = disk_bytenr;
 447                         }
 448                         break;
 449                 }
 450
 451                 search_start = extent_end;
 452                 /*
 453                  *       | ---- range to drop ----- |
 454                  *  | -------- extent -------- |
 455                  */
 456                 if (start > key.offset && end >= extent_end) {
 457                         BUG_ON(del_nr > 0);
 458                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 459
 460                         btrfs_set_file_extent_num_bytes(leaf, fi,
 461                                                         start - key.offset);
 462                         btrfs_mark_buffer_dirty(leaf);
 463                         if (disk_bytenr > 0) {
 464                                 inode_sub_bytes(inode, extent_end - start);
 465                                 *hint_byte = disk_bytenr;
 466                         }
 467                         if (end == extent_end)
 468                                 break;
 469
 470                         path->slots[0]++;
 471                         goto next_slot;
 472                 }
 473
 474                 /*
 475                  *  | ---- range to drop ----- |
 476                  *    | ------ extent ------ |
 477                  */
 478                 if (start <= key.offset && end >= extent_end) {
 479                         if (del_nr == 0) {
 480                                 del_slot = path->slots[0];
 481                                 del_nr = 1;
 482                         } else {
 483                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 484                                 del_nr++;
 485                         }
 486
 487                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 488                                 inode_sub_bytes(inode,
 489                                                 extent_end - key.offset);
 490                                 extent_end = ALIGN(extent_end,
 491                                                    root->sectorsize);
 492                         } else if (disk_bytenr > 0) {
 493                                 ret = btrfs_free_extent(trans, root,
 494                                                 disk_bytenr, num_bytes, 0,
 495                                                 root->root_key.objectid,
 496                                                 key.objectid, key.offset -
 497                                                 extent_offset);
 498                                 BUG_ON(ret);
 499                                 inode_sub_bytes(inode,
 500                                                 extent_end - key.offset);
 501                                 *hint_byte = disk_bytenr;
 502                         }
 503
 504                         if (end == extent_end)
 505                                 break;
 506
 507                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 508                                 path->slots[0]++;
 509                                 goto next_slot;
 510                         }
 511
 512                         ret = btrfs_del_items(trans, root, path, del_slot,
 513                                               del_nr);
 514                         BUG_ON(ret);
 515
 516                         del_nr = 0;
 517                         del_slot = 0;
 518
 519                         btrfs_release_path(root, path);
 520                         continue;
 521                 }
 522
 523                 BUG_ON(1);
 524         }
 525
 526         if (del_nr > 0) {
 527                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 528                 BUG_ON(ret);
 529         }
 530
 531         btrfs_free_path(path);
 532         return ret;
 533 }
 534
 535 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 536                             u64 objectid, u64 bytenr, u64 orig_offset,
 537                             u64 *start, u64 *end)
 538 {
 539         struct btrfs_file_extent_item *fi;
 540         struct btrfs_key key;
 541         u64 extent_end;
 542
 543         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 544                 return 0;
 545
 546         btrfs_item_key_to_cpu(leaf, &key, slot);
 547         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 548                 return 0;
 549
 550         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 551         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 552             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 553             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 554             btrfs_file_extent_compression(leaf, fi) ||
 555             btrfs_file_extent_encryption(leaf, fi) ||
 556             btrfs_file_extent_other_encoding(leaf, fi))
 557                 return 0;
 558
 559         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 560         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 561                 return 0;
 562
 563         *start = key.offset;
 564         *end = extent_end;
 565         return 1;
 566 }
 567
 568 /*
 569  * Mark extent in the range start - end as written.
 570  *
 571  * This changes extent type from 'pre-allocated' to 'regular'. If only
 572  * part of extent is marked as written, the extent will be split into
 573  * two or three.
 574  */
 575 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 576                               struct inode *inode, u64 start, u64 end)
 577 {
 578         struct btrfs_root *root = BTRFS_I(inode)->root;
 579         struct extent_buffer *leaf;
 580         struct btrfs_path *path;
 581         struct btrfs_file_extent_item *fi;
 582         struct btrfs_key key;
 583         struct btrfs_key new_key;
 584         u64 bytenr;
 585         u64 num_bytes;
 586         u64 extent_end;
 587         u64 orig_offset;
 588         u64 other_start;
 589         u64 other_end;
 590         u64 split;
 591         int del_nr = 0;
 592         int del_slot = 0;
 593         int recow;
 594         int ret;
 595
 596         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 597
 598         path = btrfs_alloc_path();
 599         BUG_ON(!path);
 600 again:
 601         recow = 0;
 602         split = start;
 603         key.objectid = inode->i_ino;
 604         key.type = BTRFS_EXTENT_DATA_KEY;
 605         key.offset = split;
 606
 607         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 608         if (ret < 0)
 609                 goto out;
 610         if (ret > 0 && path->slots[0] > 0)
 611                 path->slots[0]--;
 612
 613         leaf = path->nodes[0];
 614         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 615         BUG_ON(key.objectid != inode->i_ino ||
 616                key.type != BTRFS_EXTENT_DATA_KEY);
 617         fi = btrfs_item_ptr(leaf, path->slots[0],
 618                             struct btrfs_file_extent_item);
 619         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 620                BTRFS_FILE_EXTENT_PREALLOC);
 621         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 622         BUG_ON(key.offset > start || extent_end < end);
 623
 624         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 625         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 626         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 627         memcpy(&new_key, &key, sizeof(new_key));
 628
 629         if (start == key.offset && end < extent_end) {
 630                 other_start = 0;
 631                 other_end = start;
 632                 if (extent_mergeable(leaf, path->slots[0] - 1,
 633                                      inode->i_ino, bytenr, orig_offset,
 634                                      &other_start, &other_end)) {
 635                         new_key.offset = end;
 636                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 637                         fi = btrfs_item_ptr(leaf, path->slots[0],
 638                                             struct btrfs_file_extent_item);
 639                         btrfs_set_file_extent_num_bytes(leaf, fi,
 640                                                         extent_end - end);
 641                         btrfs_set_file_extent_offset(leaf, fi,
 642                                                      end - orig_offset);
 643                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 644                                             struct btrfs_file_extent_item);
 645                         btrfs_set_file_extent_num_bytes(leaf, fi,
 646                                                         end - other_start);
 647                         btrfs_mark_buffer_dirty(leaf);
 648                         goto out;
 649                 }
 650         }
 651
 652         if (start > key.offset && end == extent_end) {
 653                 other_start = end;
 654                 other_end = 0;
 655                 if (extent_mergeable(leaf, path->slots[0] + 1,
 656                                      inode->i_ino, bytenr, orig_offset,
 657                                      &other_start, &other_end)) {
 658                         fi = btrfs_item_ptr(leaf, path->slots[0],
 659                                             struct btrfs_file_extent_item);
 660                         btrfs_set_file_extent_num_bytes(leaf, fi,
 661                                                         start - key.offset);
 662                         path->slots[0]++;
 663                         new_key.offset = start;
 664                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 665
 666                         fi = btrfs_item_ptr(leaf, path->slots[0],
 667                                             struct btrfs_file_extent_item);
 668                         btrfs_set_file_extent_num_bytes(leaf, fi,
 669                                                         other_end - start);
 670                         btrfs_set_file_extent_offset(leaf, fi,
 671                                                      start - orig_offset);
 672                         btrfs_mark_buffer_dirty(leaf);
 673                         goto out;
 674                 }
 675         }
 676
 677         while (start > key.offset || end < extent_end) {
 678                 if (key.offset == start)
 679                         split = end;
 680
 681                 new_key.offset = split;
 682                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
 683                 if (ret == -EAGAIN) {
 684                         btrfs_release_path(root, path);
 685                         goto again;
 686                 }
 687                 BUG_ON(ret < 0);
 688
 689                 leaf = path->nodes[0];
 690                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 691                                     struct btrfs_file_extent_item);
 692                 btrfs_set_file_extent_num_bytes(leaf, fi,
 693                                                 split - key.offset);
 694
 695                 fi = btrfs_item_ptr(leaf, path->slots[0],
 696                                     struct btrfs_file_extent_item);
 697
 698                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 699                 btrfs_set_file_extent_num_bytes(leaf, fi,
 700                                                 extent_end - split);
 701                 btrfs_mark_buffer_dirty(leaf);
 702
 703                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 704                                            root->root_key.objectid,
 705                                            inode->i_ino, orig_offset);
 706                 BUG_ON(ret);
 707
 708                 if (split == start) {
 709                         key.offset = start;
 710                 } else {
 711                         BUG_ON(start != key.offset);
 712                         path->slots[0]--;
 713                         extent_end = end;
 714                 }
 715                 recow = 1;
 716         }
 717
 718         other_start = end;
 719         other_end = 0;
 720         if (extent_mergeable(leaf, path->slots[0] + 1,
 721                              inode->i_ino, bytenr, orig_offset,
 722                              &other_start, &other_end)) {
 723                 if (recow) {
 724                         btrfs_release_path(root, path);
 725                         goto again;
 726                 }
 727                 extent_end = other_end;
 728                 del_slot = path->slots[0] + 1;
 729                 del_nr++;
 730                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 731                                         0, root->root_key.objectid,
 732                                         inode->i_ino, orig_offset);
 733                 BUG_ON(ret);
 734         }
 735         other_start = 0;
 736         other_end = start;
 737         if (extent_mergeable(leaf, path->slots[0] - 1,
 738                              inode->i_ino, bytenr, orig_offset,
 739                              &other_start, &other_end)) {
 740                 if (recow) {
 741                         btrfs_release_path(root, path);
 742                         goto again;
 743                 }
 744                 key.offset = other_start;
 745                 del_slot = path->slots[0];
 746                 del_nr++;
 747                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 748                                         0, root->root_key.objectid,
 749                                         inode->i_ino, orig_offset);
 750                 BUG_ON(ret);
 751         }
 752         if (del_nr == 0) {
 753                 fi = btrfs_item_ptr(leaf, path->slots[0],
 754                            struct btrfs_file_extent_item);
 755                 btrfs_set_file_extent_type(leaf, fi,
 756                                            BTRFS_FILE_EXTENT_REG);
 757                 btrfs_mark_buffer_dirty(leaf);
 758         } else {
 759                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 760                            struct btrfs_file_extent_item);
 761                 btrfs_set_file_extent_type(leaf, fi,
 762                                            BTRFS_FILE_EXTENT_REG);
 763                 btrfs_set_file_extent_num_bytes(leaf, fi,
 764                                                 extent_end - key.offset);
 765                 btrfs_mark_buffer_dirty(leaf);
 766
 767                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 768                 BUG_ON(ret);
 769         }
 770 out:
 771         btrfs_free_path(path);
 772         return 0;
 773 }
 774
 775 /*
 776  * on error we return an unlocked page and the error value
 777  * on success we return a locked page and 0
 778  */
 779 static int prepare_uptodate_page(struct page *page, u64 pos)
 780 {
 781         int ret = 0;
 782
 783         if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
 784                 ret = btrfs_readpage(NULL, page);
 785                 if (ret)
 786                         return ret;
 787                 lock_page(page);
 788                 if (!PageUptodate(page)) {
 789                         unlock_page(page);
 790                         return -EIO;
 791                 }
 792         }
 793         return 0;
 794 }
 795
 796 /*
 797  * this gets pages into the page cache and locks them down, it also properly
 798  * waits for data=ordered extents to finish before allowing the pages to be
 799  * modified.
 800  */
 801 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 802                          struct page **pages, size_t num_pages,
 803                          loff_t pos, unsigned long first_index,
 804                          unsigned long last_index, size_t write_bytes)
 805 {
 806         struct extent_state *cached_state = NULL;
 807         int i;
 808         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 809         struct inode *inode = fdentry(file)->d_inode;
 810         int err = 0;
 811         int faili = 0;
 812         u64 start_pos;
 813         u64 last_pos;
 814
 815         start_pos = pos & ~((u64)root->sectorsize - 1);
 816         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 817
 818         if (start_pos > inode->i_size) {
 819                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
 820                 if (err)
 821                         return err;
 822         }
 823
 824 again:
 825         for (i = 0; i < num_pages; i++) {
 826                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 827                 if (!pages[i]) {
 828                         faili = i - 1;
 829                         err = -ENOMEM;
 830                         goto fail;
 831                 }
 832
 833                 if (i == 0)
 834                         err = prepare_uptodate_page(pages[i], pos);
 835                 if (i == num_pages - 1)
 836                         err = prepare_uptodate_page(pages[i],
 837                                                     pos + write_bytes);
 838                 if (err) {
 839                         page_cache_release(pages[i]);
 840                         faili = i - 1;
 841                         goto fail;
 842                 }
 843                 wait_on_page_writeback(pages[i]);
 844         }
 845         err = 0;
 846         if (start_pos < inode->i_size) {
 847                 struct btrfs_ordered_extent *ordered;
 848                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
 849                                  start_pos, last_pos - 1, 0, &cached_state,
 850                                  GFP_NOFS);
 851                 ordered = btrfs_lookup_first_ordered_extent(inode,
 852                                                             last_pos - 1);
 853                 if (ordered &&
 854                     ordered->file_offset + ordered->len > start_pos &&
 855                     ordered->file_offset < last_pos) {
 856                         btrfs_put_ordered_extent(ordered);
 857                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 858                                              start_pos, last_pos - 1,
 859                                              &cached_state, GFP_NOFS);
 860                         for (i = 0; i < num_pages; i++) {
 861                                 unlock_page(pages[i]);
 862                                 page_cache_release(pages[i]);
 863                         }
 864                         btrfs_wait_ordered_range(inode, start_pos,
 865                                                  last_pos - start_pos);
 866                         goto again;
 867                 }
 868                 if (ordered)
 869                         btrfs_put_ordered_extent(ordered);
 870
 871                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
 872                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 873                                   EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
 874                                   GFP_NOFS);
 875                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 876                                      start_pos, last_pos - 1, &cached_state,
 877                                      GFP_NOFS);
 878         }
 879         for (i = 0; i < num_pages; i++) {
 880                 clear_page_dirty_for_io(pages[i]);
 881                 set_page_extent_mapped(pages[i]);
 882                 WARN_ON(!PageLocked(pages[i]));
 883         }
 884         return 0;
 885 fail:
 886         while (faili >= 0) {
 887                 unlock_page(pages[faili]);
 888                 page_cache_release(pages[faili]);
 889                 faili--;
 890         }
 891         return err;
 892
 893 }
 894
 895 static noinline ssize_t __btrfs_buffered_write(struct file *file,
 896                                                struct iov_iter *i,
 897                                                loff_t pos)
 898 {
 899         struct inode *inode = fdentry(file)->d_inode;
 900         struct btrfs_root *root = BTRFS_I(inode)->root;
 901         struct page **pages = NULL;
 902         unsigned long first_index;
 903         unsigned long last_index;
 904         size_t num_written = 0;
 905         int nrptrs;
 906         int ret = 0;
 907
 908         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
 909                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
 910                      (sizeof(struct page *)));
 911         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 912         if (!pages)
 913                 return -ENOMEM;
 914
 915         first_index = pos >> PAGE_CACHE_SHIFT;
 916         last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT;
 917
 918         while (iov_iter_count(i) > 0) {
 919                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 920                 size_t write_bytes = min(iov_iter_count(i),
 921                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
 922                                          offset);
 923                 size_t num_pages = (write_bytes + offset +
 924                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 925                 size_t dirty_pages;
 926                 size_t copied;
 927
 928                 WARN_ON(num_pages > nrptrs);
 929
 930                 /*
 931                  * Fault pages before locking them in prepare_pages
 932                  * to avoid recursive lock
 933                  */
 934                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
 935                         ret = -EFAULT;
 936                         break;
 937                 }
 938
 939                 ret = btrfs_delalloc_reserve_space(inode,
 940                                         num_pages << PAGE_CACHE_SHIFT);
 941                 if (ret)
 942                         break;
 943
 944                 /*
 945                  * This is going to setup the pages array with the number of
 946                  * pages we want, so we don't really need to worry about the
 947                  * contents of pages from loop to loop
 948                  */
 949                 ret = prepare_pages(root, file, pages, num_pages,
 950                                     pos, first_index, last_index,
 951                                     write_bytes);
 952                 if (ret) {
 953                         btrfs_delalloc_release_space(inode,
 954                                         num_pages << PAGE_CACHE_SHIFT);
 955                         break;
 956                 }
 957
 958                 copied = btrfs_copy_from_user(pos, num_pages,
 959                                            write_bytes, pages, i);
 960
 961                 /*
 962                  * if we have trouble faulting in the pages, fall
 963                  * back to one page at a time
 964                  */
 965                 if (copied < write_bytes)
 966                         nrptrs = 1;
 967
 968                 if (copied == 0)
 969                         dirty_pages = 0;
 970                 else
 971                         dirty_pages = (copied + offset +
 972                                        PAGE_CACHE_SIZE - 1) >>
 973                                        PAGE_CACHE_SHIFT;
 974
 975                 /*
 976                  * If we had a short copy we need to release the excess delaloc
 977                  * bytes we reserved.  We need to increment outstanding_extents
 978                  * because btrfs_delalloc_release_space will decrement it, but
 979                  * we still have an outstanding extent for the chunk we actually
 980                  * managed to copy.
 981                  */
 982                 if (num_pages > dirty_pages) {
 983                         if (copied > 0)
 984                                 atomic_inc(
 985                                         &BTRFS_I(inode)->outstanding_extents);
 986                         btrfs_delalloc_release_space(inode,
 987                                         (num_pages - dirty_pages) <<
 988                                         PAGE_CACHE_SHIFT);
 989                 }
 990
 991                 if (copied > 0) {
 992                         ret = btrfs_dirty_pages(root, inode, pages,
 993                                                 dirty_pages, pos, copied,
 994                                                 NULL);
 995                         if (ret) {
 996                                 btrfs_delalloc_release_space(inode,
 997                                         dirty_pages << PAGE_CACHE_SHIFT);
 998                                 btrfs_drop_pages(pages, num_pages);
 999                                 break;
1000                         }
1001                 }
1002
1003                 btrfs_drop_pages(pages, num_pages);
1004
1005                 cond_resched();
1006
1007                 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1008                                                    dirty_pages);
1009                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1010                         btrfs_btree_balance_dirty(root, 1);
1011                 btrfs_throttle(root);
1012
1013                 pos += copied;
1014                 num_written += copied;
1015         }
1016
1017         kfree(pages);
1018
1019         return num_written ? num_written : ret;
1020 }
1021
1022 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1023                                     const struct iovec *iov,
1024                                     unsigned long nr_segs, loff_t pos,
1025                                     loff_t *ppos, size_t count, size_t ocount)
1026 {
1027         struct file *file = iocb->ki_filp;
1028         struct inode *inode = fdentry(file)->d_inode;
1029         struct iov_iter i;
1030         ssize_t written;
1031         ssize_t written_buffered;
1032         loff_t endbyte;
1033         int err;
1034
1035         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1036                                             count, ocount);
1037
1038         /*
1039          * the generic O_DIRECT will update in-memory i_size after the
1040          * DIOs are done.  But our endio handlers that update the on
1041          * disk i_size never update past the in memory i_size.  So we
1042          * need one more update here to catch any additions to the
1043          * file
1044          */
1045         if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1046                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1047                 mark_inode_dirty(inode);
1048         }
1049
1050         if (written < 0 || written == count)
1051                 return written;
1052
1053         pos += written;
1054         count -= written;
1055         iov_iter_init(&i, iov, nr_segs, count, written);
1056         written_buffered = __btrfs_buffered_write(file, &i, pos);
1057         if (written_buffered < 0) {
1058                 err = written_buffered;
1059                 goto out;
1060         }
1061         endbyte = pos + written_buffered - 1;
1062         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1063         if (err)
1064                 goto out;
1065         written += written_buffered;
1066         *ppos = pos + written_buffered;
1067         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1068                                  endbyte >> PAGE_CACHE_SHIFT);
1069 out:
1070         return written ? written : err;
1071 }
1072
1073 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1074                                     const struct iovec *iov,
1075                                     unsigned long nr_segs, loff_t pos)
1076 {
1077         struct file *file = iocb->ki_filp;
1078         struct inode *inode = fdentry(file)->d_inode;
1079         struct btrfs_root *root = BTRFS_I(inode)->root;
1080         loff_t *ppos = &iocb->ki_pos;
1081         ssize_t num_written = 0;
1082         ssize_t err = 0;
1083         size_t count, ocount;
1084
1085         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1086
1087         mutex_lock(&inode->i_mutex);
1088
1089         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1090         if (err) {
1091                 mutex_unlock(&inode->i_mutex);
1092                 goto out;
1093         }
1094         count = ocount;
1095
1096         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1097         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1098         if (err) {
1099                 mutex_unlock(&inode->i_mutex);
1100                 goto out;
1101         }
1102
1103         if (count == 0) {
1104                 mutex_unlock(&inode->i_mutex);
1105                 goto out;
1106         }
1107
1108         err = file_remove_suid(file);
1109         if (err) {
1110                 mutex_unlock(&inode->i_mutex);
1111                 goto out;
1112         }
1113
1114         /*
1115          * If BTRFS flips readonly due to some impossible error
1116          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1117          * although we have opened a file as writable, we have
1118          * to stop this write operation to ensure FS consistency.
1119          */
1120         if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1121                 mutex_unlock(&inode->i_mutex);
1122                 err = -EROFS;
1123                 goto out;
1124         }
1125
1126         file_update_time(file);
1127         BTRFS_I(inode)->sequence++;
1128
1129         if (unlikely(file->f_flags & O_DIRECT)) {
1130                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1131                                                    pos, ppos, count, ocount);
1132         } else {
1133                 struct iov_iter i;
1134
1135                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1136
1137                 num_written = __btrfs_buffered_write(file, &i, pos);
1138                 if (num_written > 0)
1139                         *ppos = pos + num_written;
1140         }
1141
1142         mutex_unlock(&inode->i_mutex);
1143
1144         /*
1145          * we want to make sure fsync finds this change
1146          * but we haven't joined a transaction running right now.
1147          *
1148          * Later on, someone is sure to update the inode and get the
1149          * real transid recorded.
1150          *
1151          * We set last_trans now to the fs_info generation + 1,
1152          * this will either be one more than the running transaction
1153          * or the generation used for the next transaction if there isn't
1154          * one running right now.
1155          */
1156         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1157         if (num_written > 0 || num_written == -EIOCBQUEUED) {
1158                 err = generic_write_sync(file, pos, num_written);
1159                 if (err < 0 && num_written > 0)
1160                         num_written = err;
1161         }
1162 out:
1163         current->backing_dev_info = NULL;
1164         return num_written ? num_written : err;
1165 }
1166
1167 int btrfs_release_file(struct inode *inode, struct file *filp)
1168 {
1169         /*
1170          * ordered_data_close is set by settattr when we are about to truncate
1171          * a file from a non-zero size to a zero size.  This tries to
1172          * flush down new bytes that may have been written if the
1173          * application were using truncate to replace a file in place.
1174          */
1175         if (BTRFS_I(inode)->ordered_data_close) {
1176                 BTRFS_I(inode)->ordered_data_close = 0;
1177                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1178                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1179                         filemap_flush(inode->i_mapping);
1180         }
1181         if (filp->private_data)
1182                 btrfs_ioctl_trans_end(filp);
1183         return 0;
1184 }
1185
1186 /*
1187  * fsync call for both files and directories.  This logs the inode into
1188  * the tree log instead of forcing full commits whenever possible.
1189  *
1190  * It needs to call filemap_fdatawait so that all ordered extent updates are
1191  * in the metadata btree are up to date for copying to the log.
1192  *
1193  * It drops the inode mutex before doing the tree log commit.  This is an
1194  * important optimization for directories because holding the mutex prevents
1195  * new operations on the dir while we write to disk.
1196  */
1197 int btrfs_sync_file(struct file *file, int datasync)
1198 {
1199         struct dentry *dentry = file->f_path.dentry;
1200         struct inode *inode = dentry->d_inode;
1201         struct btrfs_root *root = BTRFS_I(inode)->root;
1202         int ret = 0;
1203         struct btrfs_trans_handle *trans;
1204
1205         trace_btrfs_sync_file(file, datasync);
1206
1207         /* we wait first, since the writeback may change the inode */
1208         root->log_batch++;
1209         /* the VFS called filemap_fdatawrite for us */
1210         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1211         root->log_batch++;
1212
1213         /*
1214          * check the transaction that last modified this inode
1215          * and see if its already been committed
1216          */
1217         if (!BTRFS_I(inode)->last_trans)
1218                 goto out;
1219
1220         /*
1221          * if the last transaction that changed this file was before
1222          * the current transaction, we can bail out now without any
1223          * syncing
1224          */
1225         mutex_lock(&root->fs_info->trans_mutex);
1226         if (BTRFS_I(inode)->last_trans <=
1227             root->fs_info->last_trans_committed) {
1228                 BTRFS_I(inode)->last_trans = 0;
1229                 mutex_unlock(&root->fs_info->trans_mutex);
1230                 goto out;
1231         }
1232         mutex_unlock(&root->fs_info->trans_mutex);
1233
1234         /*
1235          * ok we haven't committed the transaction yet, lets do a commit
1236          */
1237         if (file->private_data)
1238                 btrfs_ioctl_trans_end(file);
1239
1240         trans = btrfs_start_transaction(root, 0);
1241         if (IS_ERR(trans)) {
1242                 ret = PTR_ERR(trans);
1243                 goto out;
1244         }
1245
1246         ret = btrfs_log_dentry_safe(trans, root, dentry);
1247         if (ret < 0)
1248                 goto out;
1249
1250         /* we've logged all the items and now have a consistent
1251          * version of the file in the log.  It is possible that
1252          * someone will come in and modify the file, but that's
1253          * fine because the log is consistent on disk, and we
1254          * have references to all of the file's extents
1255          *
1256          * It is possible that someone will come in and log the
1257          * file again, but that will end up using the synchronization
1258          * inside btrfs_sync_log to keep things safe.
1259          */
1260         mutex_unlock(&dentry->d_inode->i_mutex);
1261
1262         if (ret != BTRFS_NO_LOG_SYNC) {
1263                 if (ret > 0) {
1264                         ret = btrfs_commit_transaction(trans, root);
1265                 } else {
1266                         ret = btrfs_sync_log(trans, root);
1267                         if (ret == 0)
1268                                 ret = btrfs_end_transaction(trans, root);
1269                         else
1270                                 ret = btrfs_commit_transaction(trans, root);
1271                 }
1272         } else {
1273                 ret = btrfs_end_transaction(trans, root);
1274         }
1275         mutex_lock(&dentry->d_inode->i_mutex);
1276 out:
1277         return ret > 0 ? -EIO : ret;
1278 }
1279
1280 static const struct vm_operations_struct btrfs_file_vm_ops = {
1281         .fault          = filemap_fault,
1282         .page_mkwrite   = btrfs_page_mkwrite,
1283 };
1284
1285 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1286 {
1287         struct address_space *mapping = filp->f_mapping;
1288
1289         if (!mapping->a_ops->readpage)
1290                 return -ENOEXEC;
1291
1292         file_accessed(filp);
1293         vma->vm_ops = &btrfs_file_vm_ops;
1294         vma->vm_flags |= VM_CAN_NONLINEAR;
1295
1296         return 0;
1297 }
1298
1299 static long btrfs_fallocate(struct file *file, int mode,
1300                             loff_t offset, loff_t len)
1301 {
1302         struct inode *inode = file->f_path.dentry->d_inode;
1303         struct extent_state *cached_state = NULL;
1304         u64 cur_offset;
1305         u64 last_byte;
1306         u64 alloc_start;
1307         u64 alloc_end;
1308         u64 alloc_hint = 0;
1309         u64 locked_end;
1310         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1311         struct extent_map *em;
1312         int ret;
1313
1314         alloc_start = offset & ~mask;
1315         alloc_end =  (offset + len + mask) & ~mask;
1316
1317         /* We only support the FALLOC_FL_KEEP_SIZE mode */
1318         if (mode & ~FALLOC_FL_KEEP_SIZE)
1319                 return -EOPNOTSUPP;
1320
1321         /*
1322          * wait for ordered IO before we have any locks.  We'll loop again
1323          * below with the locks held.
1324          */
1325         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1326
1327         mutex_lock(&inode->i_mutex);
1328         ret = inode_newsize_ok(inode, alloc_end);
1329         if (ret)
1330                 goto out;
1331
1332         if (alloc_start > inode->i_size) {
1333                 ret = btrfs_cont_expand(inode, i_size_read(inode),
1334                                         alloc_start);
1335                 if (ret)
1336                         goto out;
1337         }
1338
1339         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1340         if (ret)
1341                 goto out;
1342
1343         locked_end = alloc_end - 1;
1344         while (1) {
1345                 struct btrfs_ordered_extent *ordered;
1346
1347                 /* the extent lock is ordered inside the running
1348                  * transaction
1349                  */
1350                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1351                                  locked_end, 0, &cached_state, GFP_NOFS);
1352                 ordered = btrfs_lookup_first_ordered_extent(inode,
1353                                                             alloc_end - 1);
1354                 if (ordered &&
1355                     ordered->file_offset + ordered->len > alloc_start &&
1356                     ordered->file_offset < alloc_end) {
1357                         btrfs_put_ordered_extent(ordered);
1358                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1359                                              alloc_start, locked_end,
1360                                              &cached_state, GFP_NOFS);
1361                         /*
1362                          * we can't wait on the range with the transaction
1363                          * running or with the extent lock held
1364                          */
1365                         btrfs_wait_ordered_range(inode, alloc_start,
1366                                                  alloc_end - alloc_start);
1367                 } else {
1368                         if (ordered)
1369                                 btrfs_put_ordered_extent(ordered);
1370                         break;
1371                 }
1372         }
1373
1374         cur_offset = alloc_start;
1375         while (1) {
1376                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1377                                       alloc_end - cur_offset, 0);
1378                 BUG_ON(IS_ERR(em) || !em);
1379                 last_byte = min(extent_map_end(em), alloc_end);
1380                 last_byte = (last_byte + mask) & ~mask;
1381                 if (em->block_start == EXTENT_MAP_HOLE ||
1382                     (cur_offset >= inode->i_size &&
1383                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1384                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1385                                                         last_byte - cur_offset,
1386                                                         1 << inode->i_blkbits,
1387                                                         offset + len,
1388                                                         &alloc_hint);
1389                         if (ret < 0) {
1390                                 free_extent_map(em);
1391                                 break;
1392                         }
1393                 }
1394                 free_extent_map(em);
1395
1396                 cur_offset = last_byte;
1397                 if (cur_offset >= alloc_end) {
1398                         ret = 0;
1399                         break;
1400                 }
1401         }
1402         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1403                              &cached_state, GFP_NOFS);
1404
1405         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1406 out:
1407         mutex_unlock(&inode->i_mutex);
1408         return ret;
1409 }
1410
1411 const struct file_operations btrfs_file_operations = {
1412         .llseek         = generic_file_llseek,
1413         .read           = do_sync_read,
1414         .write          = do_sync_write,
1415         .aio_read       = generic_file_aio_read,
1416         .splice_read    = generic_file_splice_read,
1417         .aio_write      = btrfs_file_aio_write,
1418         .mmap           = btrfs_file_mmap,
1419         .open           = generic_file_open,
1420         .release        = btrfs_release_file,
1421         .fsync          = btrfs_sync_file,
1422         .fallocate      = btrfs_fallocate,
1423         .unlocked_ioctl = btrfs_ioctl,
1424 #ifdef CONFIG_COMPAT
1425         .compat_ioctl   = btrfs_ioctl,
1426 #endif
1427 };