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 static noinline 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 static noinline int dirty_and_release_pages(struct btrfs_root *root,
 131                                             struct file *file,
 132                                             struct page **pages,
 133                                             size_t num_pages,
 134                                             loff_t pos,
 135                                             size_t write_bytes)
 136 {
 137         int err = 0;
 138         int i;
 139         struct inode *inode = fdentry(file)->d_inode;
 140         u64 num_bytes;
 141         u64 start_pos;
 142         u64 end_of_last_block;
 143         u64 end_pos = pos + write_bytes;
 144         loff_t isize = i_size_read(inode);
 145
 146         start_pos = pos & ~((u64)root->sectorsize - 1);
 147         num_bytes = (write_bytes + pos - start_pos +
 148                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 149
 150         end_of_last_block = start_pos + num_bytes - 1;
 151         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 152                                         NULL);
 153         if (err)
 154                 return err;
 155
 156         for (i = 0; i < num_pages; i++) {
 157                 struct page *p = pages[i];
 158                 SetPageUptodate(p);
 159                 ClearPageChecked(p);
 160                 set_page_dirty(p);
 161         }
 162
 163         /*
 164          * we've only changed i_size in ram, and we haven't updated
 165          * the disk i_size.  There is no need to log the inode
 166          * at this time.
 167          */
 168         if (end_pos > isize)
 169                 i_size_write(inode, end_pos);
 170         return 0;
 171 }
 172
 173 /*
 174  * this drops all the extents in the cache that intersect the range
 175  * [start, end].  Existing extents are split as required.
 176  */
 177 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 178                             int skip_pinned)
 179 {
 180         struct extent_map *em;
 181         struct extent_map *split = NULL;
 182         struct extent_map *split2 = NULL;
 183         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 184         u64 len = end - start + 1;
 185         int ret;
 186         int testend = 1;
 187         unsigned long flags;
 188         int compressed = 0;
 189
 190         WARN_ON(end < start);
 191         if (end == (u64)-1) {
 192                 len = (u64)-1;
 193                 testend = 0;
 194         }
 195         while (1) {
 196                 if (!split)
 197                         split = alloc_extent_map(GFP_NOFS);
 198                 if (!split2)
 199                         split2 = alloc_extent_map(GFP_NOFS);
 200                 BUG_ON(!split || !split2);
 201
 202                 write_lock(&em_tree->lock);
 203                 em = lookup_extent_mapping(em_tree, start, len);
 204                 if (!em) {
 205                         write_unlock(&em_tree->lock);
 206                         break;
 207                 }
 208                 flags = em->flags;
 209                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 210                         if (testend && em->start + em->len >= start + len) {
 211                                 free_extent_map(em);
 212                                 write_unlock(&em_tree->lock);
 213                                 break;
 214                         }
 215                         start = em->start + em->len;
 216                         if (testend)
 217                                 len = start + len - (em->start + em->len);
 218                         free_extent_map(em);
 219                         write_unlock(&em_tree->lock);
 220                         continue;
 221                 }
 222                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 223                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 224                 remove_extent_mapping(em_tree, em);
 225
 226                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 227                     em->start < start) {
 228                         split->start = em->start;
 229                         split->len = start - em->start;
 230                         split->orig_start = em->orig_start;
 231                         split->block_start = em->block_start;
 232
 233                         if (compressed)
 234                                 split->block_len = em->block_len;
 235                         else
 236                                 split->block_len = split->len;
 237
 238                         split->bdev = em->bdev;
 239                         split->flags = flags;
 240                         split->compress_type = em->compress_type;
 241                         ret = add_extent_mapping(em_tree, split);
 242                         BUG_ON(ret);
 243                         free_extent_map(split);
 244                         split = split2;
 245                         split2 = NULL;
 246                 }
 247                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 248                     testend && em->start + em->len > start + len) {
 249                         u64 diff = start + len - em->start;
 250
 251                         split->start = start + len;
 252                         split->len = em->start + em->len - (start + len);
 253                         split->bdev = em->bdev;
 254                         split->flags = flags;
 255                         split->compress_type = em->compress_type;
 256
 257                         if (compressed) {
 258                                 split->block_len = em->block_len;
 259                                 split->block_start = em->block_start;
 260                                 split->orig_start = em->orig_start;
 261                         } else {
 262                                 split->block_len = split->len;
 263                                 split->block_start = em->block_start + diff;
 264                                 split->orig_start = split->start;
 265                         }
 266
 267                         ret = add_extent_mapping(em_tree, split);
 268                         BUG_ON(ret);
 269                         free_extent_map(split);
 270                         split = NULL;
 271                 }
 272                 write_unlock(&em_tree->lock);
 273
 274                 /* once for us */
 275                 free_extent_map(em);
 276                 /* once for the tree*/
 277                 free_extent_map(em);
 278         }
 279         if (split)
 280                 free_extent_map(split);
 281         if (split2)
 282                 free_extent_map(split2);
 283         return 0;
 284 }
 285
 286 /*
 287  * this is very complex, but the basic idea is to drop all extents
 288  * in the range start - end.  hint_block is filled in with a block number
 289  * that would be a good hint to the block allocator for this file.
 290  *
 291  * If an extent intersects the range but is not entirely inside the range
 292  * it is either truncated or split.  Anything entirely inside the range
 293  * is deleted from the tree.
 294  */
 295 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
 296                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
 297 {
 298         struct btrfs_root *root = BTRFS_I(inode)->root;
 299         struct extent_buffer *leaf;
 300         struct btrfs_file_extent_item *fi;
 301         struct btrfs_path *path;
 302         struct btrfs_key key;
 303         struct btrfs_key new_key;
 304         u64 search_start = start;
 305         u64 disk_bytenr = 0;
 306         u64 num_bytes = 0;
 307         u64 extent_offset = 0;
 308         u64 extent_end = 0;
 309         int del_nr = 0;
 310         int del_slot = 0;
 311         int extent_type;
 312         int recow;
 313         int ret;
 314
 315         if (drop_cache)
 316                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 317
 318         path = btrfs_alloc_path();
 319         if (!path)
 320                 return -ENOMEM;
 321
 322         while (1) {
 323                 recow = 0;
 324                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 325                                                search_start, -1);
 326                 if (ret < 0)
 327                         break;
 328                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 329                         leaf = path->nodes[0];
 330                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 331                         if (key.objectid == inode->i_ino &&
 332                             key.type == BTRFS_EXTENT_DATA_KEY)
 333                                 path->slots[0]--;
 334                 }
 335                 ret = 0;
 336 next_slot:
 337                 leaf = path->nodes[0];
 338                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 339                         BUG_ON(del_nr > 0);
 340                         ret = btrfs_next_leaf(root, path);
 341                         if (ret < 0)
 342                                 break;
 343                         if (ret > 0) {
 344                                 ret = 0;
 345                                 break;
 346                         }
 347                         leaf = path->nodes[0];
 348                         recow = 1;
 349                 }
 350
 351                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 352                 if (key.objectid > inode->i_ino ||
 353                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 354                         break;
 355
 356                 fi = btrfs_item_ptr(leaf, path->slots[0],
 357                                     struct btrfs_file_extent_item);
 358                 extent_type = btrfs_file_extent_type(leaf, fi);
 359
 360                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 361                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 362                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 363                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 364                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 365                         extent_end = key.offset +
 366                                 btrfs_file_extent_num_bytes(leaf, fi);
 367                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 368                         extent_end = key.offset +
 369                                 btrfs_file_extent_inline_len(leaf, fi);
 370                 } else {
 371                         WARN_ON(1);
 372                         extent_end = search_start;
 373                 }
 374
 375                 if (extent_end <= search_start) {
 376                         path->slots[0]++;
 377                         goto next_slot;
 378                 }
 379
 380                 search_start = max(key.offset, start);
 381                 if (recow) {
 382                         btrfs_release_path(root, path);
 383                         continue;
 384                 }
 385
 386                 /*
 387                  *     | - range to drop - |
 388                  *  | -------- extent -------- |
 389                  */
 390                 if (start > key.offset && end < extent_end) {
 391                         BUG_ON(del_nr > 0);
 392                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 393
 394                         memcpy(&new_key, &key, sizeof(new_key));
 395                         new_key.offset = start;
 396                         ret = btrfs_duplicate_item(trans, root, path,
 397                                                    &new_key);
 398                         if (ret == -EAGAIN) {
 399                                 btrfs_release_path(root, path);
 400                                 continue;
 401                         }
 402                         if (ret < 0)
 403                                 break;
 404
 405                         leaf = path->nodes[0];
 406                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 407                                             struct btrfs_file_extent_item);
 408                         btrfs_set_file_extent_num_bytes(leaf, fi,
 409                                                         start - key.offset);
 410
 411                         fi = btrfs_item_ptr(leaf, path->slots[0],
 412                                             struct btrfs_file_extent_item);
 413
 414                         extent_offset += start - key.offset;
 415                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 416                         btrfs_set_file_extent_num_bytes(leaf, fi,
 417                                                         extent_end - start);
 418                         btrfs_mark_buffer_dirty(leaf);
 419
 420                         if (disk_bytenr > 0) {
 421                                 ret = btrfs_inc_extent_ref(trans, root,
 422                                                 disk_bytenr, num_bytes, 0,
 423                                                 root->root_key.objectid,
 424                                                 new_key.objectid,
 425                                                 start - extent_offset);
 426                                 BUG_ON(ret);
 427                                 *hint_byte = disk_bytenr;
 428                         }
 429                         key.offset = start;
 430                 }
 431                 /*
 432                  *  | ---- range to drop ----- |
 433                  *      | -------- extent -------- |
 434                  */
 435                 if (start <= key.offset && end < extent_end) {
 436                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 437
 438                         memcpy(&new_key, &key, sizeof(new_key));
 439                         new_key.offset = end;
 440                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 441
 442                         extent_offset += end - key.offset;
 443                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 444                         btrfs_set_file_extent_num_bytes(leaf, fi,
 445                                                         extent_end - end);
 446                         btrfs_mark_buffer_dirty(leaf);
 447                         if (disk_bytenr > 0) {
 448                                 inode_sub_bytes(inode, end - key.offset);
 449                                 *hint_byte = disk_bytenr;
 450                         }
 451                         break;
 452                 }
 453
 454                 search_start = extent_end;
 455                 /*
 456                  *       | ---- range to drop ----- |
 457                  *  | -------- extent -------- |
 458                  */
 459                 if (start > key.offset && end >= extent_end) {
 460                         BUG_ON(del_nr > 0);
 461                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 462
 463                         btrfs_set_file_extent_num_bytes(leaf, fi,
 464                                                         start - key.offset);
 465                         btrfs_mark_buffer_dirty(leaf);
 466                         if (disk_bytenr > 0) {
 467                                 inode_sub_bytes(inode, extent_end - start);
 468                                 *hint_byte = disk_bytenr;
 469                         }
 470                         if (end == extent_end)
 471                                 break;
 472
 473                         path->slots[0]++;
 474                         goto next_slot;
 475                 }
 476
 477                 /*
 478                  *  | ---- range to drop ----- |
 479                  *    | ------ extent ------ |
 480                  */
 481                 if (start <= key.offset && end >= extent_end) {
 482                         if (del_nr == 0) {
 483                                 del_slot = path->slots[0];
 484                                 del_nr = 1;
 485                         } else {
 486                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 487                                 del_nr++;
 488                         }
 489
 490                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 491                                 inode_sub_bytes(inode,
 492                                                 extent_end - key.offset);
 493                                 extent_end = ALIGN(extent_end,
 494                                                    root->sectorsize);
 495                         } else if (disk_bytenr > 0) {
 496                                 ret = btrfs_free_extent(trans, root,
 497                                                 disk_bytenr, num_bytes, 0,
 498                                                 root->root_key.objectid,
 499                                                 key.objectid, key.offset -
 500                                                 extent_offset);
 501                                 BUG_ON(ret);
 502                                 inode_sub_bytes(inode,
 503                                                 extent_end - key.offset);
 504                                 *hint_byte = disk_bytenr;
 505                         }
 506
 507                         if (end == extent_end)
 508                                 break;
 509
 510                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 511                                 path->slots[0]++;
 512                                 goto next_slot;
 513                         }
 514
 515                         ret = btrfs_del_items(trans, root, path, del_slot,
 516                                               del_nr);
 517                         BUG_ON(ret);
 518
 519                         del_nr = 0;
 520                         del_slot = 0;
 521
 522                         btrfs_release_path(root, path);
 523                         continue;
 524                 }
 525
 526                 BUG_ON(1);
 527         }
 528
 529         if (del_nr > 0) {
 530                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 531                 BUG_ON(ret);
 532         }
 533
 534         btrfs_free_path(path);
 535         return ret;
 536 }
 537
 538 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 539                             u64 objectid, u64 bytenr, u64 orig_offset,
 540                             u64 *start, u64 *end)
 541 {
 542         struct btrfs_file_extent_item *fi;
 543         struct btrfs_key key;
 544         u64 extent_end;
 545
 546         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 547                 return 0;
 548
 549         btrfs_item_key_to_cpu(leaf, &key, slot);
 550         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 551                 return 0;
 552
 553         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 554         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 555             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 556             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 557             btrfs_file_extent_compression(leaf, fi) ||
 558             btrfs_file_extent_encryption(leaf, fi) ||
 559             btrfs_file_extent_other_encoding(leaf, fi))
 560                 return 0;
 561
 562         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 563         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 564                 return 0;
 565
 566         *start = key.offset;
 567         *end = extent_end;
 568         return 1;
 569 }
 570
 571 /*
 572  * Mark extent in the range start - end as written.
 573  *
 574  * This changes extent type from 'pre-allocated' to 'regular'. If only
 575  * part of extent is marked as written, the extent will be split into
 576  * two or three.
 577  */
 578 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 579                               struct inode *inode, u64 start, u64 end)
 580 {
 581         struct btrfs_root *root = BTRFS_I(inode)->root;
 582         struct extent_buffer *leaf;
 583         struct btrfs_path *path;
 584         struct btrfs_file_extent_item *fi;
 585         struct btrfs_key key;
 586         struct btrfs_key new_key;
 587         u64 bytenr;
 588         u64 num_bytes;
 589         u64 extent_end;
 590         u64 orig_offset;
 591         u64 other_start;
 592         u64 other_end;
 593         u64 split;
 594         int del_nr = 0;
 595         int del_slot = 0;
 596         int recow;
 597         int ret;
 598
 599         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 600
 601         path = btrfs_alloc_path();
 602         BUG_ON(!path);
 603 again:
 604         recow = 0;
 605         split = start;
 606         key.objectid = inode->i_ino;
 607         key.type = BTRFS_EXTENT_DATA_KEY;
 608         key.offset = split;
 609
 610         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 611         if (ret < 0)
 612                 goto out;
 613         if (ret > 0 && path->slots[0] > 0)
 614                 path->slots[0]--;
 615
 616         leaf = path->nodes[0];
 617         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 618         BUG_ON(key.objectid != inode->i_ino ||
 619                key.type != BTRFS_EXTENT_DATA_KEY);
 620         fi = btrfs_item_ptr(leaf, path->slots[0],
 621                             struct btrfs_file_extent_item);
 622         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 623                BTRFS_FILE_EXTENT_PREALLOC);
 624         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 625         BUG_ON(key.offset > start || extent_end < end);
 626
 627         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 628         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 629         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 630         memcpy(&new_key, &key, sizeof(new_key));
 631
 632         if (start == key.offset && end < extent_end) {
 633                 other_start = 0;
 634                 other_end = start;
 635                 if (extent_mergeable(leaf, path->slots[0] - 1,
 636                                      inode->i_ino, bytenr, orig_offset,
 637                                      &other_start, &other_end)) {
 638                         new_key.offset = end;
 639                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 640                         fi = btrfs_item_ptr(leaf, path->slots[0],
 641                                             struct btrfs_file_extent_item);
 642                         btrfs_set_file_extent_num_bytes(leaf, fi,
 643                                                         extent_end - end);
 644                         btrfs_set_file_extent_offset(leaf, fi,
 645                                                      end - orig_offset);
 646                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 647                                             struct btrfs_file_extent_item);
 648                         btrfs_set_file_extent_num_bytes(leaf, fi,
 649                                                         end - other_start);
 650                         btrfs_mark_buffer_dirty(leaf);
 651                         goto out;
 652                 }
 653         }
 654
 655         if (start > key.offset && end == extent_end) {
 656                 other_start = end;
 657                 other_end = 0;
 658                 if (extent_mergeable(leaf, path->slots[0] + 1,
 659                                      inode->i_ino, bytenr, orig_offset,
 660                                      &other_start, &other_end)) {
 661                         fi = btrfs_item_ptr(leaf, path->slots[0],
 662                                             struct btrfs_file_extent_item);
 663                         btrfs_set_file_extent_num_bytes(leaf, fi,
 664                                                         start - key.offset);
 665                         path->slots[0]++;
 666                         new_key.offset = start;
 667                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 668
 669                         fi = btrfs_item_ptr(leaf, path->slots[0],
 670                                             struct btrfs_file_extent_item);
 671                         btrfs_set_file_extent_num_bytes(leaf, fi,
 672                                                         other_end - start);
 673                         btrfs_set_file_extent_offset(leaf, fi,
 674                                                      start - orig_offset);
 675                         btrfs_mark_buffer_dirty(leaf);
 676                         goto out;
 677                 }
 678         }
 679
 680         while (start > key.offset || end < extent_end) {
 681                 if (key.offset == start)
 682                         split = end;
 683
 684                 new_key.offset = split;
 685                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
 686                 if (ret == -EAGAIN) {
 687                         btrfs_release_path(root, path);
 688                         goto again;
 689                 }
 690                 BUG_ON(ret < 0);
 691
 692                 leaf = path->nodes[0];
 693                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 694                                     struct btrfs_file_extent_item);
 695                 btrfs_set_file_extent_num_bytes(leaf, fi,
 696                                                 split - key.offset);
 697
 698                 fi = btrfs_item_ptr(leaf, path->slots[0],
 699                                     struct btrfs_file_extent_item);
 700
 701                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 702                 btrfs_set_file_extent_num_bytes(leaf, fi,
 703                                                 extent_end - split);
 704                 btrfs_mark_buffer_dirty(leaf);
 705
 706                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 707                                            root->root_key.objectid,
 708                                            inode->i_ino, orig_offset);
 709                 BUG_ON(ret);
 710
 711                 if (split == start) {
 712                         key.offset = start;
 713                 } else {
 714                         BUG_ON(start != key.offset);
 715                         path->slots[0]--;
 716                         extent_end = end;
 717                 }
 718                 recow = 1;
 719         }
 720
 721         other_start = end;
 722         other_end = 0;
 723         if (extent_mergeable(leaf, path->slots[0] + 1,
 724                              inode->i_ino, bytenr, orig_offset,
 725                              &other_start, &other_end)) {
 726                 if (recow) {
 727                         btrfs_release_path(root, path);
 728                         goto again;
 729                 }
 730                 extent_end = other_end;
 731                 del_slot = path->slots[0] + 1;
 732                 del_nr++;
 733                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 734                                         0, root->root_key.objectid,
 735                                         inode->i_ino, orig_offset);
 736                 BUG_ON(ret);
 737         }
 738         other_start = 0;
 739         other_end = start;
 740         if (extent_mergeable(leaf, path->slots[0] - 1,
 741                              inode->i_ino, bytenr, orig_offset,
 742                              &other_start, &other_end)) {
 743                 if (recow) {
 744                         btrfs_release_path(root, path);
 745                         goto again;
 746                 }
 747                 key.offset = other_start;
 748                 del_slot = path->slots[0];
 749                 del_nr++;
 750                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 751                                         0, root->root_key.objectid,
 752                                         inode->i_ino, orig_offset);
 753                 BUG_ON(ret);
 754         }
 755         if (del_nr == 0) {
 756                 fi = btrfs_item_ptr(leaf, path->slots[0],
 757                            struct btrfs_file_extent_item);
 758                 btrfs_set_file_extent_type(leaf, fi,
 759                                            BTRFS_FILE_EXTENT_REG);
 760                 btrfs_mark_buffer_dirty(leaf);
 761         } else {
 762                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 763                            struct btrfs_file_extent_item);
 764                 btrfs_set_file_extent_type(leaf, fi,
 765                                            BTRFS_FILE_EXTENT_REG);
 766                 btrfs_set_file_extent_num_bytes(leaf, fi,
 767                                                 extent_end - key.offset);
 768                 btrfs_mark_buffer_dirty(leaf);
 769
 770                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 771                 BUG_ON(ret);
 772         }
 773 out:
 774         btrfs_free_path(path);
 775         return 0;
 776 }
 777
 778 /*
 779  * on error we return an unlocked page and the error value
 780  * on success we return a locked page and 0
 781  */
 782 static int prepare_uptodate_page(struct page *page, u64 pos)
 783 {
 784         int ret = 0;
 785
 786         if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
 787                 ret = btrfs_readpage(NULL, page);
 788                 if (ret)
 789                         return ret;
 790                 lock_page(page);
 791                 if (!PageUptodate(page)) {
 792                         unlock_page(page);
 793                         return -EIO;
 794                 }
 795         }
 796         return 0;
 797 }
 798
 799 /*
 800  * this gets pages into the page cache and locks them down, it also properly
 801  * waits for data=ordered extents to finish before allowing the pages to be
 802  * modified.
 803  */
 804 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 805                          struct page **pages, size_t num_pages,
 806                          loff_t pos, unsigned long first_index,
 807                          unsigned long last_index, size_t write_bytes)
 808 {
 809         struct extent_state *cached_state = NULL;
 810         int i;
 811         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 812         struct inode *inode = fdentry(file)->d_inode;
 813         int err = 0;
 814         int faili = 0;
 815         u64 start_pos;
 816         u64 last_pos;
 817
 818         start_pos = pos & ~((u64)root->sectorsize - 1);
 819         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 820
 821         if (start_pos > inode->i_size) {
 822                 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
 823                 if (err)
 824                         return err;
 825         }
 826
 827 again:
 828         for (i = 0; i < num_pages; i++) {
 829                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 830                 if (!pages[i]) {
 831                         faili = i - 1;
 832                         err = -ENOMEM;
 833                         goto fail;
 834                 }
 835
 836                 if (i == 0)
 837                         err = prepare_uptodate_page(pages[i], pos);
 838                 if (i == num_pages - 1)
 839                         err = prepare_uptodate_page(pages[i],
 840                                                     pos + write_bytes);
 841                 if (err) {
 842                         page_cache_release(pages[i]);
 843                         faili = i - 1;
 844                         goto fail;
 845                 }
 846                 wait_on_page_writeback(pages[i]);
 847         }
 848         err = 0;
 849         if (start_pos < inode->i_size) {
 850                 struct btrfs_ordered_extent *ordered;
 851                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
 852                                  start_pos, last_pos - 1, 0, &cached_state,
 853                                  GFP_NOFS);
 854                 ordered = btrfs_lookup_first_ordered_extent(inode,
 855                                                             last_pos - 1);
 856                 if (ordered &&
 857                     ordered->file_offset + ordered->len > start_pos &&
 858                     ordered->file_offset < last_pos) {
 859                         btrfs_put_ordered_extent(ordered);
 860                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 861                                              start_pos, last_pos - 1,
 862                                              &cached_state, GFP_NOFS);
 863                         for (i = 0; i < num_pages; i++) {
 864                                 unlock_page(pages[i]);
 865                                 page_cache_release(pages[i]);
 866                         }
 867                         btrfs_wait_ordered_range(inode, start_pos,
 868                                                  last_pos - start_pos);
 869                         goto again;
 870                 }
 871                 if (ordered)
 872                         btrfs_put_ordered_extent(ordered);
 873
 874                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
 875                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 876                                   EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
 877                                   GFP_NOFS);
 878                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 879                                      start_pos, last_pos - 1, &cached_state,
 880                                      GFP_NOFS);
 881         }
 882         for (i = 0; i < num_pages; i++) {
 883                 clear_page_dirty_for_io(pages[i]);
 884                 set_page_extent_mapped(pages[i]);
 885                 WARN_ON(!PageLocked(pages[i]));
 886         }
 887         return 0;
 888 fail:
 889         while (faili >= 0) {
 890                 unlock_page(pages[faili]);
 891                 page_cache_release(pages[faili]);
 892                 faili--;
 893         }
 894         return err;
 895
 896 }
 897
 898 static noinline ssize_t __btrfs_buffered_write(struct file *file,
 899                                                struct iov_iter *i,
 900                                                loff_t pos)
 901 {
 902         struct inode *inode = fdentry(file)->d_inode;
 903         struct btrfs_root *root = BTRFS_I(inode)->root;
 904         struct page **pages = NULL;
 905         unsigned long first_index;
 906         unsigned long last_index;
 907         size_t num_written = 0;
 908         int nrptrs;
 909         int ret = 0;
 910
 911         nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
 912                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
 913                      (sizeof(struct page *)));
 914         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 915         if (!pages)
 916                 return -ENOMEM;
 917
 918         first_index = pos >> PAGE_CACHE_SHIFT;
 919         last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT;
 920
 921         while (iov_iter_count(i) > 0) {
 922                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 923                 size_t write_bytes = min(iov_iter_count(i),
 924                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
 925                                          offset);
 926                 size_t num_pages = (write_bytes + offset +
 927                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 928                 size_t dirty_pages;
 929                 size_t copied;
 930
 931                 WARN_ON(num_pages > nrptrs);
 932
 933                 /*
 934                  * Fault pages before locking them in prepare_pages
 935                  * to avoid recursive lock
 936                  */
 937                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
 938                         ret = -EFAULT;
 939                         break;
 940                 }
 941
 942                 ret = btrfs_delalloc_reserve_space(inode,
 943                                         num_pages << PAGE_CACHE_SHIFT);
 944                 if (ret)
 945                         break;
 946
 947                 /*
 948                  * This is going to setup the pages array with the number of
 949                  * pages we want, so we don't really need to worry about the
 950                  * contents of pages from loop to loop
 951                  */
 952                 ret = prepare_pages(root, file, pages, num_pages,
 953                                     pos, first_index, last_index,
 954                                     write_bytes);
 955                 if (ret) {
 956                         btrfs_delalloc_release_space(inode,
 957                                         num_pages << PAGE_CACHE_SHIFT);
 958                         break;
 959                 }
 960
 961                 copied = btrfs_copy_from_user(pos, num_pages,
 962                                            write_bytes, pages, i);
 963
 964                 /*
 965                  * if we have trouble faulting in the pages, fall
 966                  * back to one page at a time
 967                  */
 968                 if (copied < write_bytes)
 969                         nrptrs = 1;
 970
 971                 if (copied == 0)
 972                         dirty_pages = 0;
 973                 else
 974                         dirty_pages = (copied + offset +
 975                                        PAGE_CACHE_SIZE - 1) >>
 976                                        PAGE_CACHE_SHIFT;
 977
 978                 /*
 979                  * If we had a short copy we need to release the excess delaloc
 980                  * bytes we reserved.  We need to increment outstanding_extents
 981                  * because btrfs_delalloc_release_space will decrement it, but
 982                  * we still have an outstanding extent for the chunk we actually
 983                  * managed to copy.
 984                  */
 985                 if (num_pages > dirty_pages) {
 986                         if (copied > 0)
 987                                 atomic_inc(
 988                                         &BTRFS_I(inode)->outstanding_extents);
 989                         btrfs_delalloc_release_space(inode,
 990                                         (num_pages - dirty_pages) <<
 991                                         PAGE_CACHE_SHIFT);
 992                 }
 993
 994                 if (copied > 0) {
 995                         ret = dirty_and_release_pages(root, file, pages,
 996                                                       dirty_pages, pos,
 997                                                       copied);
 998                         if (ret) {
 999                                 btrfs_delalloc_release_space(inode,
1000                                         dirty_pages << PAGE_CACHE_SHIFT);
1001                                 btrfs_drop_pages(pages, num_pages);
1002                                 break;
1003                         }
1004                 }
1005
1006                 btrfs_drop_pages(pages, num_pages);
1007
1008                 cond_resched();
1009
1010                 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1011                                                    dirty_pages);
1012                 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1013                         btrfs_btree_balance_dirty(root, 1);
1014                 btrfs_throttle(root);
1015
1016                 pos += copied;
1017                 num_written += copied;
1018         }
1019
1020         kfree(pages);
1021
1022         return num_written ? num_written : ret;
1023 }
1024
1025 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1026                                     const struct iovec *iov,
1027                                     unsigned long nr_segs, loff_t pos,
1028                                     loff_t *ppos, size_t count, size_t ocount)
1029 {
1030         struct file *file = iocb->ki_filp;
1031         struct inode *inode = fdentry(file)->d_inode;
1032         struct iov_iter i;
1033         ssize_t written;
1034         ssize_t written_buffered;
1035         loff_t endbyte;
1036         int err;
1037
1038         written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1039                                             count, ocount);
1040
1041         /*
1042          * the generic O_DIRECT will update in-memory i_size after the
1043          * DIOs are done.  But our endio handlers that update the on
1044          * disk i_size never update past the in memory i_size.  So we
1045          * need one more update here to catch any additions to the
1046          * file
1047          */
1048         if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1049                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1050                 mark_inode_dirty(inode);
1051         }
1052
1053         if (written < 0 || written == count)
1054                 return written;
1055
1056         pos += written;
1057         count -= written;
1058         iov_iter_init(&i, iov, nr_segs, count, written);
1059         written_buffered = __btrfs_buffered_write(file, &i, pos);
1060         if (written_buffered < 0) {
1061                 err = written_buffered;
1062                 goto out;
1063         }
1064         endbyte = pos + written_buffered - 1;
1065         err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1066         if (err)
1067                 goto out;
1068         written += written_buffered;
1069         *ppos = pos + written_buffered;
1070         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1071                                  endbyte >> PAGE_CACHE_SHIFT);
1072 out:
1073         return written ? written : err;
1074 }
1075
1076 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1077                                     const struct iovec *iov,
1078                                     unsigned long nr_segs, loff_t pos)
1079 {
1080         struct file *file = iocb->ki_filp;
1081         struct inode *inode = fdentry(file)->d_inode;
1082         struct btrfs_root *root = BTRFS_I(inode)->root;
1083         loff_t *ppos = &iocb->ki_pos;
1084         ssize_t num_written = 0;
1085         ssize_t err = 0;
1086         size_t count, ocount;
1087
1088         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1089
1090         mutex_lock(&inode->i_mutex);
1091
1092         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1093         if (err) {
1094                 mutex_unlock(&inode->i_mutex);
1095                 goto out;
1096         }
1097         count = ocount;
1098
1099         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1100         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1101         if (err) {
1102                 mutex_unlock(&inode->i_mutex);
1103                 goto out;
1104         }
1105
1106         if (count == 0) {
1107                 mutex_unlock(&inode->i_mutex);
1108                 goto out;
1109         }
1110
1111         err = file_remove_suid(file);
1112         if (err) {
1113                 mutex_unlock(&inode->i_mutex);
1114                 goto out;
1115         }
1116
1117         /*
1118          * If BTRFS flips readonly due to some impossible error
1119          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1120          * although we have opened a file as writable, we have
1121          * to stop this write operation to ensure FS consistency.
1122          */
1123         if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1124                 mutex_unlock(&inode->i_mutex);
1125                 err = -EROFS;
1126                 goto out;
1127         }
1128
1129         file_update_time(file);
1130         BTRFS_I(inode)->sequence++;
1131
1132         if (unlikely(file->f_flags & O_DIRECT)) {
1133                 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1134                                                    pos, ppos, count, ocount);
1135         } else {
1136                 struct iov_iter i;
1137
1138                 iov_iter_init(&i, iov, nr_segs, count, num_written);
1139
1140                 num_written = __btrfs_buffered_write(file, &i, pos);
1141                 if (num_written > 0)
1142                         *ppos = pos + num_written;
1143         }
1144
1145         mutex_unlock(&inode->i_mutex);
1146
1147         /*
1148          * we want to make sure fsync finds this change
1149          * but we haven't joined a transaction running right now.
1150          *
1151          * Later on, someone is sure to update the inode and get the
1152          * real transid recorded.
1153          *
1154          * We set last_trans now to the fs_info generation + 1,
1155          * this will either be one more than the running transaction
1156          * or the generation used for the next transaction if there isn't
1157          * one running right now.
1158          */
1159         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1160         if (num_written > 0 || num_written == -EIOCBQUEUED) {
1161                 err = generic_write_sync(file, pos, num_written);
1162                 if (err < 0 && num_written > 0)
1163                         num_written = err;
1164         }
1165 out:
1166         current->backing_dev_info = NULL;
1167         return num_written ? num_written : err;
1168 }
1169
1170 int btrfs_release_file(struct inode *inode, struct file *filp)
1171 {
1172         /*
1173          * ordered_data_close is set by settattr when we are about to truncate
1174          * a file from a non-zero size to a zero size.  This tries to
1175          * flush down new bytes that may have been written if the
1176          * application were using truncate to replace a file in place.
1177          */
1178         if (BTRFS_I(inode)->ordered_data_close) {
1179                 BTRFS_I(inode)->ordered_data_close = 0;
1180                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1181                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1182                         filemap_flush(inode->i_mapping);
1183         }
1184         if (filp->private_data)
1185                 btrfs_ioctl_trans_end(filp);
1186         return 0;
1187 }
1188
1189 /*
1190  * fsync call for both files and directories.  This logs the inode into
1191  * the tree log instead of forcing full commits whenever possible.
1192  *
1193  * It needs to call filemap_fdatawait so that all ordered extent updates are
1194  * in the metadata btree are up to date for copying to the log.
1195  *
1196  * It drops the inode mutex before doing the tree log commit.  This is an
1197  * important optimization for directories because holding the mutex prevents
1198  * new operations on the dir while we write to disk.
1199  */
1200 int btrfs_sync_file(struct file *file, int datasync)
1201 {
1202         struct dentry *dentry = file->f_path.dentry;
1203         struct inode *inode = dentry->d_inode;
1204         struct btrfs_root *root = BTRFS_I(inode)->root;
1205         int ret = 0;
1206         struct btrfs_trans_handle *trans;
1207
1208         trace_btrfs_sync_file(file, datasync);
1209
1210         /* we wait first, since the writeback may change the inode */
1211         root->log_batch++;
1212         /* the VFS called filemap_fdatawrite for us */
1213         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1214         root->log_batch++;
1215
1216         /*
1217          * check the transaction that last modified this inode
1218          * and see if its already been committed
1219          */
1220         if (!BTRFS_I(inode)->last_trans)
1221                 goto out;
1222
1223         /*
1224          * if the last transaction that changed this file was before
1225          * the current transaction, we can bail out now without any
1226          * syncing
1227          */
1228         mutex_lock(&root->fs_info->trans_mutex);
1229         if (BTRFS_I(inode)->last_trans <=
1230             root->fs_info->last_trans_committed) {
1231                 BTRFS_I(inode)->last_trans = 0;
1232                 mutex_unlock(&root->fs_info->trans_mutex);
1233                 goto out;
1234         }
1235         mutex_unlock(&root->fs_info->trans_mutex);
1236
1237         /*
1238          * ok we haven't committed the transaction yet, lets do a commit
1239          */
1240         if (file->private_data)
1241                 btrfs_ioctl_trans_end(file);
1242
1243         trans = btrfs_start_transaction(root, 0);
1244         if (IS_ERR(trans)) {
1245                 ret = PTR_ERR(trans);
1246                 goto out;
1247         }
1248
1249         ret = btrfs_log_dentry_safe(trans, root, dentry);
1250         if (ret < 0)
1251                 goto out;
1252
1253         /* we've logged all the items and now have a consistent
1254          * version of the file in the log.  It is possible that
1255          * someone will come in and modify the file, but that's
1256          * fine because the log is consistent on disk, and we
1257          * have references to all of the file's extents
1258          *
1259          * It is possible that someone will come in and log the
1260          * file again, but that will end up using the synchronization
1261          * inside btrfs_sync_log to keep things safe.
1262          */
1263         mutex_unlock(&dentry->d_inode->i_mutex);
1264
1265         if (ret != BTRFS_NO_LOG_SYNC) {
1266                 if (ret > 0) {
1267                         ret = btrfs_commit_transaction(trans, root);
1268                 } else {
1269                         ret = btrfs_sync_log(trans, root);
1270                         if (ret == 0)
1271                                 ret = btrfs_end_transaction(trans, root);
1272                         else
1273                                 ret = btrfs_commit_transaction(trans, root);
1274                 }
1275         } else {
1276                 ret = btrfs_end_transaction(trans, root);
1277         }
1278         mutex_lock(&dentry->d_inode->i_mutex);
1279 out:
1280         return ret > 0 ? -EIO : ret;
1281 }
1282
1283 static const struct vm_operations_struct btrfs_file_vm_ops = {
1284         .fault          = filemap_fault,
1285         .page_mkwrite   = btrfs_page_mkwrite,
1286 };
1287
1288 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1289 {
1290         struct address_space *mapping = filp->f_mapping;
1291
1292         if (!mapping->a_ops->readpage)
1293                 return -ENOEXEC;
1294
1295         file_accessed(filp);
1296         vma->vm_ops = &btrfs_file_vm_ops;
1297         vma->vm_flags |= VM_CAN_NONLINEAR;
1298
1299         return 0;
1300 }
1301
1302 static long btrfs_fallocate(struct file *file, int mode,
1303                             loff_t offset, loff_t len)
1304 {
1305         struct inode *inode = file->f_path.dentry->d_inode;
1306         struct extent_state *cached_state = NULL;
1307         u64 cur_offset;
1308         u64 last_byte;
1309         u64 alloc_start;
1310         u64 alloc_end;
1311         u64 alloc_hint = 0;
1312         u64 locked_end;
1313         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1314         struct extent_map *em;
1315         int ret;
1316
1317         alloc_start = offset & ~mask;
1318         alloc_end =  (offset + len + mask) & ~mask;
1319
1320         /* We only support the FALLOC_FL_KEEP_SIZE mode */
1321         if (mode & ~FALLOC_FL_KEEP_SIZE)
1322                 return -EOPNOTSUPP;
1323
1324         /*
1325          * wait for ordered IO before we have any locks.  We'll loop again
1326          * below with the locks held.
1327          */
1328         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1329
1330         mutex_lock(&inode->i_mutex);
1331         ret = inode_newsize_ok(inode, alloc_end);
1332         if (ret)
1333                 goto out;
1334
1335         if (alloc_start > inode->i_size) {
1336                 ret = btrfs_cont_expand(inode, i_size_read(inode),
1337                                         alloc_start);
1338                 if (ret)
1339                         goto out;
1340         }
1341
1342         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1343         if (ret)
1344                 goto out;
1345
1346         locked_end = alloc_end - 1;
1347         while (1) {
1348                 struct btrfs_ordered_extent *ordered;
1349
1350                 /* the extent lock is ordered inside the running
1351                  * transaction
1352                  */
1353                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1354                                  locked_end, 0, &cached_state, GFP_NOFS);
1355                 ordered = btrfs_lookup_first_ordered_extent(inode,
1356                                                             alloc_end - 1);
1357                 if (ordered &&
1358                     ordered->file_offset + ordered->len > alloc_start &&
1359                     ordered->file_offset < alloc_end) {
1360                         btrfs_put_ordered_extent(ordered);
1361                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1362                                              alloc_start, locked_end,
1363                                              &cached_state, GFP_NOFS);
1364                         /*
1365                          * we can't wait on the range with the transaction
1366                          * running or with the extent lock held
1367                          */
1368                         btrfs_wait_ordered_range(inode, alloc_start,
1369                                                  alloc_end - alloc_start);
1370                 } else {
1371                         if (ordered)
1372                                 btrfs_put_ordered_extent(ordered);
1373                         break;
1374                 }
1375         }
1376
1377         cur_offset = alloc_start;
1378         while (1) {
1379                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1380                                       alloc_end - cur_offset, 0);
1381                 BUG_ON(IS_ERR(em) || !em);
1382                 last_byte = min(extent_map_end(em), alloc_end);
1383                 last_byte = (last_byte + mask) & ~mask;
1384                 if (em->block_start == EXTENT_MAP_HOLE ||
1385                     (cur_offset >= inode->i_size &&
1386                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1387                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1388                                                         last_byte - cur_offset,
1389                                                         1 << inode->i_blkbits,
1390                                                         offset + len,
1391                                                         &alloc_hint);
1392                         if (ret < 0) {
1393                                 free_extent_map(em);
1394                                 break;
1395                         }
1396                 }
1397                 free_extent_map(em);
1398
1399                 cur_offset = last_byte;
1400                 if (cur_offset >= alloc_end) {
1401                         ret = 0;
1402                         break;
1403                 }
1404         }
1405         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1406                              &cached_state, GFP_NOFS);
1407
1408         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1409 out:
1410         mutex_unlock(&inode->i_mutex);
1411         return ret;
1412 }
1413
1414 const struct file_operations btrfs_file_operations = {
1415         .llseek         = generic_file_llseek,
1416         .read           = do_sync_read,
1417         .write          = do_sync_write,
1418         .aio_read       = generic_file_aio_read,
1419         .splice_read    = generic_file_splice_read,
1420         .aio_write      = btrfs_file_aio_write,
1421         .mmap           = btrfs_file_mmap,
1422         .open           = generic_file_open,
1423         .release        = btrfs_release_file,
1424         .fsync          = btrfs_sync_file,
1425         .fallocate      = btrfs_fallocate,
1426         .unlocked_ioctl = btrfs_ioctl,
1427 #ifdef CONFIG_COMPAT
1428         .compat_ioctl   = btrfs_ioctl,
1429 #endif
1430 };