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/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/version.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
  40
  41 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  42                                 struct page **prepared_pages,
  43                                 const char __user * buf)
  44 {
  45         long page_fault = 0;
  46         int i;
  47         int offset = pos & (PAGE_CACHE_SIZE - 1);
  48
  49         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  50                 size_t count = min_t(size_t,
  51                                      PAGE_CACHE_SIZE - offset, write_bytes);
  52                 struct page *page = prepared_pages[i];
  53                 fault_in_pages_readable(buf, count);
  54
  55                 /* Copy data from userspace to the current page */
  56                 kmap(page);
  57                 page_fault = __copy_from_user(page_address(page) + offset,
  58                                               buf, count);
  59                 /* Flush processor's dcache for this page */
  60                 flush_dcache_page(page);
  61                 kunmap(page);
  62                 buf += count;
  63                 write_bytes -= count;
  64
  65                 if (page_fault)
  66                         break;
  67         }
  68         return page_fault ? -EFAULT : 0;
  69 }
  70
  71 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  72 {
  73         size_t i;
  74         for (i = 0; i < num_pages; i++) {
  75                 if (!pages[i])
  76                         break;
  77                 unlock_page(pages[i]);
  78                 mark_page_accessed(pages[i]);
  79                 page_cache_release(pages[i]);
  80         }
  81 }
  82
  83 static int insert_inline_extent(struct btrfs_trans_handle *trans,
  84                                 struct btrfs_root *root, struct inode *inode,
  85                                 u64 offset, ssize_t size,
  86                                 struct page *page, size_t page_offset)
  87 {
  88         struct btrfs_key key;
  89         struct btrfs_path *path;
  90         struct extent_buffer *leaf;
  91         char *kaddr;
  92         unsigned long ptr;
  93         struct btrfs_file_extent_item *ei;
  94         u32 datasize;
  95         int err = 0;
  96         int ret;
  97
  98         path = btrfs_alloc_path();
  99         if (!path)
 100                 return -ENOMEM;
 101
 102         btrfs_set_trans_block_group(trans, inode);
 103
 104         key.objectid = inode->i_ino;
 105         key.offset = offset;
 106         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 107         BUG_ON(size >= PAGE_CACHE_SIZE);
 108         datasize = btrfs_file_extent_calc_inline_size(size);
 109
 110         ret = btrfs_insert_empty_item(trans, root, path, &key,
 111                                       datasize);
 112         if (ret) {
 113                 err = ret;
 114                 goto fail;
 115         }
 116         leaf = path->nodes[0];
 117         ei = btrfs_item_ptr(leaf, path->slots[0],
 118                             struct btrfs_file_extent_item);
 119         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 120         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 121         ptr = btrfs_file_extent_inline_start(ei);
 122
 123         kaddr = kmap_atomic(page, KM_USER0);
 124         write_extent_buffer(leaf, kaddr + page_offset, ptr, size);
 125         kunmap_atomic(kaddr, KM_USER0);
 126         btrfs_mark_buffer_dirty(leaf);
 127 fail:
 128         btrfs_free_path(path);
 129         return err;
 130 }
 131
 132 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 133                                    struct btrfs_root *root,
 134                                    struct file *file,
 135                                    struct page **pages,
 136                                    size_t num_pages,
 137                                    loff_t pos,
 138                                    size_t write_bytes)
 139 {
 140         int err = 0;
 141         int i;
 142         struct inode *inode = file->f_path.dentry->d_inode;
 143         struct extent_map *em;
 144         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 145         u64 hint_block;
 146         u64 num_blocks;
 147         u64 start_pos;
 148         u64 end_of_last_block;
 149         u64 end_pos = pos + write_bytes;
 150         loff_t isize = i_size_read(inode);
 151
 152         em = alloc_extent_map(GFP_NOFS);
 153         if (!em)
 154                 return -ENOMEM;
 155
 156         em->bdev = inode->i_sb->s_bdev;
 157
 158         start_pos = pos & ~((u64)root->sectorsize - 1);
 159         num_blocks = (write_bytes + pos - start_pos + root->sectorsize - 1) >>
 160                         inode->i_blkbits;
 161
 162         down_read(&BTRFS_I(inode)->root->snap_sem);
 163         end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
 164         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 165         mutex_lock(&root->fs_info->fs_mutex);
 166         trans = btrfs_start_transaction(root, 1);
 167         if (!trans) {
 168                 err = -ENOMEM;
 169                 goto out_unlock;
 170         }
 171         btrfs_set_trans_block_group(trans, inode);
 172         inode->i_blocks += num_blocks << 3;
 173         hint_block = 0;
 174
 175         if ((end_of_last_block & 4095) == 0) {
 176                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 177         }
 178         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 179
 180         /* FIXME...EIEIO, ENOSPC and more */
 181
 182         /* insert any holes we need to create */
 183         if (inode->i_size < start_pos) {
 184                 u64 last_pos_in_file;
 185                 u64 hole_size;
 186                 u64 mask = root->sectorsize - 1;
 187                 last_pos_in_file = (isize + mask) & ~mask;
 188                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 189
 190                 if (last_pos_in_file < start_pos) {
 191                         err = btrfs_drop_extents(trans, root, inode,
 192                                                  last_pos_in_file,
 193                                                  last_pos_in_file + hole_size,
 194                                                  &hint_block);
 195                         if (err)
 196                                 goto failed;
 197
 198                         hole_size >>= inode->i_blkbits;
 199                         err = btrfs_insert_file_extent(trans, root,
 200                                                        inode->i_ino,
 201                                                        last_pos_in_file,
 202                                                        0, 0, hole_size);
 203                 }
 204                 if (err)
 205                         goto failed;
 206         }
 207
 208         /*
 209          * either allocate an extent for the new bytes or setup the key
 210          * to show we are doing inline data in the extent
 211          */
 212         if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
 213             pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 214                 u64 last_end;
 215                 for (i = 0; i < num_pages; i++) {
 216                         struct page *p = pages[i];
 217                         SetPageUptodate(p);
 218                         set_page_dirty(p);
 219                 }
 220                 last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
 221                 last_end += PAGE_CACHE_SIZE - 1;
 222                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 223                                  GFP_NOFS);
 224         } else {
 225                 struct page *p = pages[0];
 226                 /* step one, delete the existing extents in this range */
 227                 /* FIXME blocksize != pagesize */
 228                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 229                          (pos + write_bytes + root->sectorsize -1) &
 230                          ~((u64)root->sectorsize - 1), &hint_block);
 231                 if (err)
 232                         goto failed;
 233
 234                 err = insert_inline_extent(trans, root, inode, start_pos,
 235                                            end_pos - start_pos, p, 0);
 236                 BUG_ON(err);
 237                 em->start = start_pos;
 238                 em->end = end_pos - 1;
 239                 em->block_start = EXTENT_MAP_INLINE;
 240                 em->block_end = EXTENT_MAP_INLINE;
 241                 add_extent_mapping(em_tree, em);
 242         }
 243         if (end_pos > isize) {
 244                 i_size_write(inode, end_pos);
 245                 btrfs_update_inode(trans, root, inode);
 246         }
 247 failed:
 248         err = btrfs_end_transaction(trans, root);
 249 out_unlock:
 250         mutex_unlock(&root->fs_info->fs_mutex);
 251         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 252         free_extent_map(em);
 253         up_read(&BTRFS_I(inode)->root->snap_sem);
 254         return err;
 255 }
 256
 257 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 258 {
 259         struct extent_map *em;
 260         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 261
 262         while(1) {
 263                 em = lookup_extent_mapping(em_tree, start, end);
 264                 if (!em)
 265                         break;
 266                 remove_extent_mapping(em_tree, em);
 267                 /* once for us */
 268                 free_extent_map(em);
 269                 /* once for the tree*/
 270                 free_extent_map(em);
 271         }
 272         return 0;
 273 }
 274
 275 /*
 276  * this is very complex, but the basic idea is to drop all extents
 277  * in the range start - end.  hint_block is filled in with a block number
 278  * that would be a good hint to the block allocator for this file.
 279  *
 280  * If an extent intersects the range but is not entirely inside the range
 281  * it is either truncated or split.  Anything entirely inside the range
 282  * is deleted from the tree.
 283  */
 284 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 285                        struct btrfs_root *root, struct inode *inode,
 286                        u64 start, u64 end, u64 *hint_block)
 287 {
 288         int ret;
 289         struct btrfs_key key;
 290         struct extent_buffer *leaf;
 291         int slot;
 292         struct btrfs_file_extent_item *extent;
 293         u64 extent_end = 0;
 294         int keep;
 295         struct btrfs_file_extent_item old;
 296         struct btrfs_path *path;
 297         u64 search_start = start;
 298         int bookend;
 299         int found_type;
 300         int found_extent;
 301         int found_inline;
 302         int recow;
 303
 304         btrfs_drop_extent_cache(inode, start, end - 1);
 305
 306         path = btrfs_alloc_path();
 307         if (!path)
 308                 return -ENOMEM;
 309         while(1) {
 310                 recow = 0;
 311                 btrfs_release_path(root, path);
 312                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 313                                                search_start, -1);
 314                 if (ret < 0)
 315                         goto out;
 316                 if (ret > 0) {
 317                         if (path->slots[0] == 0) {
 318                                 ret = 0;
 319                                 goto out;
 320                         }
 321                         path->slots[0]--;
 322                 }
 323 next_slot:
 324                 keep = 0;
 325                 bookend = 0;
 326                 found_extent = 0;
 327                 found_inline = 0;
 328                 extent = NULL;
 329                 leaf = path->nodes[0];
 330                 slot = path->slots[0];
 331                 ret = 0;
 332                 btrfs_item_key_to_cpu(leaf, &key, slot);
 333                 if (key.offset >= end || key.objectid != inode->i_ino) {
 334                         goto out;
 335                 }
 336                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 337                         goto out;
 338                 }
 339                 if (recow) {
 340                         search_start = key.offset;
 341                         continue;
 342                 }
 343                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 344                         extent = btrfs_item_ptr(leaf, slot,
 345                                                 struct btrfs_file_extent_item);
 346                         found_type = btrfs_file_extent_type(leaf, extent);
 347                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 348                                 extent_end = key.offset +
 349                                  (btrfs_file_extent_num_blocks(leaf, extent) <<
 350                                          inode->i_blkbits);
 351                                 found_extent = 1;
 352                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 353                                 struct btrfs_item *item;
 354                                 item = btrfs_item_nr(leaf, slot);
 355                                 found_inline = 1;
 356                                 extent_end = key.offset +
 357                                      btrfs_file_extent_inline_len(leaf, item);
 358                         }
 359                 } else {
 360                         extent_end = search_start;
 361                 }
 362
 363                 /* we found nothing we can drop */
 364                 if ((!found_extent && !found_inline) ||
 365                     search_start >= extent_end) {
 366                         int nextret;
 367                         u32 nritems;
 368                         nritems = btrfs_header_nritems(leaf);
 369                         if (slot >= nritems - 1) {
 370                                 nextret = btrfs_next_leaf(root, path);
 371                                 if (nextret)
 372                                         goto out;
 373                                 recow = 1;
 374                         } else {
 375                                 path->slots[0]++;
 376                         }
 377                         goto next_slot;
 378                 }
 379
 380                 /* FIXME, there's only one inline extent allowed right now */
 381                 if (found_inline) {
 382                         u64 mask = root->sectorsize - 1;
 383                         search_start = (extent_end + mask) & ~mask;
 384                 } else
 385                         search_start = extent_end;
 386
 387                 if (end < extent_end && end >= key.offset) {
 388                         if (found_extent) {
 389                                 u64 disk_blocknr =
 390                                     btrfs_file_extent_disk_blocknr(leaf,extent);
 391                                 u64 disk_num_blocks =
 392                                     btrfs_file_extent_disk_num_blocks(leaf,
 393                                                                       extent);
 394                                 read_extent_buffer(leaf, &old,
 395                                                    (unsigned long)extent,
 396                                                    sizeof(old));
 397                                 if (disk_blocknr != 0) {
 398                                         ret = btrfs_inc_extent_ref(trans, root,
 399                                                  disk_blocknr, disk_num_blocks);
 400                                         BUG_ON(ret);
 401                                 }
 402                         }
 403                         WARN_ON(found_inline);
 404                         bookend = 1;
 405                 }
 406                 /* truncate existing extent */
 407                 if (start > key.offset) {
 408                         u64 new_num;
 409                         u64 old_num;
 410                         keep = 1;
 411                         WARN_ON(start & (root->sectorsize - 1));
 412                         if (found_extent) {
 413                                 new_num = (start - key.offset) >>
 414                                         inode->i_blkbits;
 415                                 old_num = btrfs_file_extent_num_blocks(leaf,
 416                                                                        extent);
 417                                 *hint_block =
 418                                         btrfs_file_extent_disk_blocknr(leaf,
 419                                                                        extent);
 420                                 if (btrfs_file_extent_disk_blocknr(leaf,
 421                                                                    extent)) {
 422                                         inode->i_blocks -=
 423                                                 (old_num - new_num) << 3;
 424                                 }
 425                                 btrfs_set_file_extent_num_blocks(leaf,
 426                                                                  extent,
 427                                                                  new_num);
 428                                 btrfs_mark_buffer_dirty(leaf);
 429                         } else {
 430                                 WARN_ON(1);
 431                         }
 432                 }
 433                 /* delete the entire extent */
 434                 if (!keep) {
 435                         u64 disk_blocknr = 0;
 436                         u64 disk_num_blocks = 0;
 437                         u64 extent_num_blocks = 0;
 438                         if (found_extent) {
 439                                 disk_blocknr =
 440                                       btrfs_file_extent_disk_blocknr(leaf,
 441                                                                      extent);
 442                                 disk_num_blocks =
 443                                       btrfs_file_extent_disk_num_blocks(leaf,
 444                                                                         extent);
 445                                 extent_num_blocks =
 446                                       btrfs_file_extent_num_blocks(leaf,
 447                                                                    extent);
 448                                 *hint_block =
 449                                         btrfs_file_extent_disk_blocknr(leaf,
 450                                                                        extent);
 451                         }
 452                         ret = btrfs_del_item(trans, root, path);
 453                         /* TODO update progress marker and return */
 454                         BUG_ON(ret);
 455                         btrfs_release_path(root, path);
 456                         extent = NULL;
 457                         if (found_extent && disk_blocknr != 0) {
 458                                 inode->i_blocks -= extent_num_blocks << 3;
 459                                 ret = btrfs_free_extent(trans, root,
 460                                                         disk_blocknr,
 461                                                         disk_num_blocks, 0);
 462                         }
 463
 464                         BUG_ON(ret);
 465                         if (!bookend && search_start >= end) {
 466                                 ret = 0;
 467                                 goto out;
 468                         }
 469                         if (!bookend)
 470                                 continue;
 471                 }
 472                 /* create bookend, splitting the extent in two */
 473                 if (bookend && found_extent) {
 474                         struct btrfs_key ins;
 475                         ins.objectid = inode->i_ino;
 476                         ins.offset = end;
 477                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 478                         btrfs_release_path(root, path);
 479                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 480                                                       sizeof(*extent));
 481
 482                         leaf = path->nodes[0];
 483                         if (ret) {
 484                                 btrfs_print_leaf(root, leaf);
 485                                 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.type, ins.offset, start, end, key.offset, extent_end, keep);
 486                         }
 487                         BUG_ON(ret);
 488                         extent = btrfs_item_ptr(leaf, path->slots[0],
 489                                                 struct btrfs_file_extent_item);
 490                         write_extent_buffer(leaf, &old,
 491                                             (unsigned long)extent, sizeof(old));
 492
 493                         btrfs_set_file_extent_offset(leaf, extent,
 494                                     le64_to_cpu(old.offset) +
 495                                     ((end - key.offset) >> inode->i_blkbits));
 496                         WARN_ON(le64_to_cpu(old.num_blocks) <
 497                                 (extent_end - end) >> inode->i_blkbits);
 498                         btrfs_set_file_extent_num_blocks(leaf, extent,
 499                                     (extent_end - end) >> inode->i_blkbits);
 500
 501                         btrfs_set_file_extent_type(leaf, extent,
 502                                                    BTRFS_FILE_EXTENT_REG);
 503                         btrfs_mark_buffer_dirty(path->nodes[0]);
 504                         if (le64_to_cpu(old.disk_blocknr) != 0) {
 505                                 inode->i_blocks +=
 506                                       btrfs_file_extent_num_blocks(leaf,
 507                                                                    extent) << 3;
 508                         }
 509                         ret = 0;
 510                         goto out;
 511                 }
 512         }
 513 out:
 514         btrfs_free_path(path);
 515         return ret;
 516 }
 517
 518 /*
 519  * this gets pages into the page cache and locks them down
 520  */
 521 static int prepare_pages(struct btrfs_root *root,
 522                          struct file *file,
 523                          struct page **pages,
 524                          size_t num_pages,
 525                          loff_t pos,
 526                          unsigned long first_index,
 527                          unsigned long last_index,
 528                          size_t write_bytes)
 529 {
 530         int i;
 531         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 532         struct inode *inode = file->f_path.dentry->d_inode;
 533         int err = 0;
 534         u64 num_blocks;
 535         u64 start_pos;
 536
 537         start_pos = pos & ~((u64)root->sectorsize - 1);
 538         num_blocks = (write_bytes + pos - start_pos + root->sectorsize - 1) >>
 539                         inode->i_blkbits;
 540
 541         memset(pages, 0, num_pages * sizeof(struct page *));
 542
 543         for (i = 0; i < num_pages; i++) {
 544                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 545                 if (!pages[i]) {
 546                         err = -ENOMEM;
 547                         BUG_ON(1);
 548                 }
 549                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 550                 wait_on_page_writeback(pages[i]);
 551                 set_page_extent_mapped(pages[i]);
 552                 WARN_ON(!PageLocked(pages[i]));
 553         }
 554         return 0;
 555 }
 556
 557 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 558                                 size_t count, loff_t *ppos)
 559 {
 560         loff_t pos;
 561         size_t num_written = 0;
 562         int err = 0;
 563         int ret = 0;
 564         struct inode *inode = file->f_path.dentry->d_inode;
 565         struct btrfs_root *root = BTRFS_I(inode)->root;
 566         struct page **pages = NULL;
 567         int nrptrs;
 568         struct page *pinned[2];
 569         unsigned long first_index;
 570         unsigned long last_index;
 571
 572         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 573                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 574         pinned[0] = NULL;
 575         pinned[1] = NULL;
 576         if (file->f_flags & O_DIRECT)
 577                 return -EINVAL;
 578         pos = *ppos;
 579         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 580         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 581         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 582         if (err)
 583                 goto out;
 584         if (count == 0)
 585                 goto out;
 586         err = remove_suid(file->f_path.dentry);
 587         if (err)
 588                 goto out;
 589         file_update_time(file);
 590
 591         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 592
 593         mutex_lock(&inode->i_mutex);
 594         first_index = pos >> PAGE_CACHE_SHIFT;
 595         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 596
 597         /*
 598          * there are lots of better ways to do this, but this code
 599          * makes sure the first and last page in the file range are
 600          * up to date and ready for cow
 601          */
 602         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 603                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 604                 if (!PageUptodate(pinned[0])) {
 605                         ret = btrfs_readpage(NULL, pinned[0]);
 606                         BUG_ON(ret);
 607                         wait_on_page_locked(pinned[0]);
 608                 } else {
 609                         unlock_page(pinned[0]);
 610                 }
 611         }
 612         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 613                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 614                 if (!PageUptodate(pinned[1])) {
 615                         ret = btrfs_readpage(NULL, pinned[1]);
 616                         BUG_ON(ret);
 617                         wait_on_page_locked(pinned[1]);
 618                 } else {
 619                         unlock_page(pinned[1]);
 620                 }
 621         }
 622
 623         while(count > 0) {
 624                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 625                 size_t write_bytes = min(count, nrptrs *
 626                                         (size_t)PAGE_CACHE_SIZE -
 627                                          offset);
 628                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 629                                         PAGE_CACHE_SHIFT;
 630
 631                 WARN_ON(num_pages > nrptrs);
 632                 memset(pages, 0, sizeof(pages));
 633                 ret = prepare_pages(root, file, pages, num_pages,
 634                                     pos, first_index, last_index,
 635                                     write_bytes);
 636                 if (ret)
 637                         goto out;
 638
 639                 ret = btrfs_copy_from_user(pos, num_pages,
 640                                            write_bytes, pages, buf);
 641                 if (ret) {
 642                         btrfs_drop_pages(pages, num_pages);
 643                         goto out;
 644                 }
 645
 646                 ret = dirty_and_release_pages(NULL, root, file, pages,
 647                                               num_pages, pos, write_bytes);
 648                 btrfs_drop_pages(pages, num_pages);
 649                 if (ret)
 650                         goto out;
 651
 652                 buf += write_bytes;
 653                 count -= write_bytes;
 654                 pos += write_bytes;
 655                 num_written += write_bytes;
 656
 657                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 658                 btrfs_btree_balance_dirty(root, 1);
 659                 cond_resched();
 660         }
 661         mutex_unlock(&inode->i_mutex);
 662 out:
 663         kfree(pages);
 664         if (pinned[0])
 665                 page_cache_release(pinned[0]);
 666         if (pinned[1])
 667                 page_cache_release(pinned[1]);
 668         *ppos = pos;
 669         current->backing_dev_info = NULL;
 670         return num_written ? num_written : err;
 671 }
 672
 673 static int btrfs_sync_file(struct file *file,
 674                            struct dentry *dentry, int datasync)
 675 {
 676         struct inode *inode = dentry->d_inode;
 677         struct btrfs_root *root = BTRFS_I(inode)->root;
 678         int ret = 0;
 679         struct btrfs_trans_handle *trans;
 680
 681         /*
 682          * check the transaction that last modified this inode
 683          * and see if its already been committed
 684          */
 685         mutex_lock(&root->fs_info->fs_mutex);
 686         if (!BTRFS_I(inode)->last_trans)
 687                 goto out;
 688         mutex_lock(&root->fs_info->trans_mutex);
 689         if (BTRFS_I(inode)->last_trans <=
 690             root->fs_info->last_trans_committed) {
 691                 BTRFS_I(inode)->last_trans = 0;
 692                 mutex_unlock(&root->fs_info->trans_mutex);
 693                 goto out;
 694         }
 695         mutex_unlock(&root->fs_info->trans_mutex);
 696
 697         /*
 698          * ok we haven't committed the transaction yet, lets do a commit
 699          */
 700         trans = btrfs_start_transaction(root, 1);
 701         if (!trans) {
 702                 ret = -ENOMEM;
 703                 goto out;
 704         }
 705         ret = btrfs_commit_transaction(trans, root);
 706 out:
 707         mutex_unlock(&root->fs_info->fs_mutex);
 708         return ret > 0 ? EIO : ret;
 709 }
 710
 711 static struct vm_operations_struct btrfs_file_vm_ops = {
 712 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 713         .nopage         = filemap_nopage,
 714         .populate       = filemap_populate,
 715 #else
 716         .fault          = filemap_fault,
 717 #endif
 718         .page_mkwrite   = btrfs_page_mkwrite,
 719 };
 720
 721 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 722 {
 723         vma->vm_ops = &btrfs_file_vm_ops;
 724         file_accessed(filp);
 725         return 0;
 726 }
 727
 728 struct file_operations btrfs_file_operations = {
 729         .llseek         = generic_file_llseek,
 730         .read           = do_sync_read,
 731         .aio_read       = generic_file_aio_read,
 732         .write          = btrfs_file_write,
 733         .mmap           = btrfs_file_mmap,
 734         .open           = generic_file_open,
 735         .fsync          = btrfs_sync_file,
 736         .unlocked_ioctl = btrfs_ioctl,
 737 #ifdef CONFIG_COMPAT
 738         .compat_ioctl   = btrfs_ioctl,
 739 #endif
 740 };
 741