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