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         struct btrfs_key ins;
 147         u64 hint_block;
 148         u64 num_blocks;
 149         u64 start_pos;
 150         u64 end_of_last_block;
 151         u64 end_pos = pos + write_bytes;
 152         loff_t isize = i_size_read(inode);
 153
 154         em = alloc_extent_map(GFP_NOFS);
 155         if (!em)
 156                 return -ENOMEM;
 157
 158         em->bdev = inode->i_sb->s_bdev;
 159
 160         start_pos = pos & ~((u64)root->blocksize - 1);
 161         num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
 162                         inode->i_blkbits;
 163
 164         end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
 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         /* step one, delete the existing extents in this range */
 183         /* FIXME blocksize != pagesize */
 184         if (start_pos < inode->i_size) {
 185                 err = btrfs_drop_extents(trans, root, inode,
 186                          start_pos, (pos + write_bytes + root->blocksize -1) &
 187                          ~((u64)root->blocksize - 1), &hint_block);
 188                 if (err)
 189                         goto failed;
 190         }
 191
 192         /* insert any holes we need to create */
 193         if (inode->i_size < start_pos) {
 194                 u64 last_pos_in_file;
 195                 u64 hole_size;
 196                 u64 mask = root->blocksize - 1;
 197                 last_pos_in_file = (isize + mask) & ~mask;
 198                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 199                 hole_size >>= inode->i_blkbits;
 200                 if (last_pos_in_file < start_pos) {
 201                         err = btrfs_insert_file_extent(trans, root,
 202                                                        inode->i_ino,
 203                                                        last_pos_in_file,
 204                                                        0, 0, hole_size);
 205                 }
 206                 if (err)
 207                         goto failed;
 208         }
 209
 210         /*
 211          * either allocate an extent for the new bytes or setup the key
 212          * to show we are doing inline data in the extent
 213          */
 214         if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
 215             pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 216                 err = btrfs_alloc_extent(trans, root, inode->i_ino,
 217                                          num_blocks, 0, hint_block, (u64)-1,
 218                                          &ins, 1);
 219                 BUG_ON(err);
 220                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 221                                        start_pos, ins.objectid, ins.offset,
 222                                        ins.offset);
 223                 BUG_ON(err);
 224                 em->start = start_pos;
 225                 em->end = end_of_last_block;
 226                 em->block_start = ins.objectid << inode->i_blkbits;
 227                 em->block_end = em->block_start +
 228                         (ins.offset << inode->i_blkbits) - 1;
 229                 set_extent_dirty(em_tree, start_pos, end_of_last_block,
 230                                  GFP_NOFS);
 231                 err = add_extent_mapping(em_tree, em);
 232                 for (i = 0; i < num_pages; i++) {
 233                         struct page *p = pages[i];
 234                         SetPageUptodate(p);
 235                         __set_page_dirty_nobuffers(p);
 236                 }
 237         } else {
 238                 struct page *p = pages[0];
 239                 err = insert_inline_extent(trans, root, inode, start_pos,
 240                                            end_pos - start_pos, p, 0);
 241                 BUG_ON(err);
 242                 em->start = start_pos;
 243                 em->end = end_pos;
 244                 em->block_start = EXTENT_MAP_INLINE;
 245                 em->block_end = EXTENT_MAP_INLINE;
 246                 add_extent_mapping(em_tree, em);
 247         }
 248         if (end_pos > isize) {
 249                 i_size_write(inode, end_pos);
 250                 btrfs_update_inode(trans, root, inode);
 251         }
 252 failed:
 253         err = btrfs_end_transaction(trans, root);
 254 out_unlock:
 255         mutex_unlock(&root->fs_info->fs_mutex);
 256         free_extent_map(em);
 257         return err;
 258 }
 259
 260 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 261 {
 262         struct extent_map *em;
 263         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 264
 265         while(1) {
 266                 em = lookup_extent_mapping(em_tree, start, end);
 267                 if (!em)
 268                         break;
 269                 remove_extent_mapping(em_tree, em);
 270                 /* once for us */
 271                 free_extent_map(em);
 272                 /* once for the tree*/
 273                 free_extent_map(em);
 274         }
 275         return 0;
 276 }
 277
 278 /*
 279  * this is very complex, but the basic idea is to drop all extents
 280  * in the range start - end.  hint_block is filled in with a block number
 281  * that would be a good hint to the block allocator for this file.
 282  *
 283  * If an extent intersects the range but is not entirely inside the range
 284  * it is either truncated or split.  Anything entirely inside the range
 285  * is deleted from the tree.
 286  */
 287 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 288                        struct btrfs_root *root, struct inode *inode,
 289                        u64 start, u64 end, u64 *hint_block)
 290 {
 291         int ret;
 292         struct btrfs_key key;
 293         struct btrfs_leaf *leaf;
 294         int slot;
 295         struct btrfs_file_extent_item *extent;
 296         u64 extent_end = 0;
 297         int keep;
 298         struct btrfs_file_extent_item old;
 299         struct btrfs_path *path;
 300         u64 search_start = start;
 301         int bookend;
 302         int found_type;
 303         int found_extent;
 304         int found_inline;
 305         int recow;
 306
 307         btrfs_drop_extent_cache(inode, start, end - 1);
 308
 309         path = btrfs_alloc_path();
 310         if (!path)
 311                 return -ENOMEM;
 312         while(1) {
 313                 recow = 0;
 314                 btrfs_release_path(root, path);
 315                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 316                                                search_start, -1);
 317                 if (ret < 0)
 318                         goto out;
 319                 if (ret > 0) {
 320                         if (path->slots[0] == 0) {
 321                                 ret = 0;
 322                                 goto out;
 323                         }
 324                         path->slots[0]--;
 325                 }
 326 next_slot:
 327                 keep = 0;
 328                 bookend = 0;
 329                 found_extent = 0;
 330                 found_inline = 0;
 331                 extent = NULL;
 332                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 333                 slot = path->slots[0];
 334                 ret = 0;
 335                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
 336                 if (key.offset >= end || key.objectid != inode->i_ino) {
 337                         goto out;
 338                 }
 339                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 340                         goto out;
 341                 }
 342                 if (recow) {
 343                         search_start = key.offset;
 344                         continue;
 345                 }
 346                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 347                         extent = btrfs_item_ptr(leaf, slot,
 348                                                 struct btrfs_file_extent_item);
 349                         found_type = btrfs_file_extent_type(extent);
 350                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 351                                 extent_end = key.offset +
 352                                         (btrfs_file_extent_num_blocks(extent) <<
 353                                          inode->i_blkbits);
 354                                 found_extent = 1;
 355                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 356                                 found_inline = 1;
 357                                 extent_end = key.offset +
 358                                      btrfs_file_extent_inline_len(leaf->items +
 359                                                                   slot);
 360                         }
 361                 } else {
 362                         extent_end = search_start;
 363                 }
 364
 365                 /* we found nothing we can drop */
 366                 if ((!found_extent && !found_inline) ||
 367                     search_start >= extent_end) {
 368                         int nextret;
 369                         u32 nritems;
 370                         nritems = btrfs_header_nritems(
 371                                         btrfs_buffer_header(path->nodes[0]));
 372                         if (slot >= nritems - 1) {
 373                                 nextret = btrfs_next_leaf(root, path);
 374                                 if (nextret)
 375                                         goto out;
 376                                 recow = 1;
 377                         } else {
 378                                 path->slots[0]++;
 379                         }
 380                         goto next_slot;
 381                 }
 382
 383                 /* FIXME, there's only one inline extent allowed right now */
 384                 if (found_inline) {
 385                         u64 mask = root->blocksize - 1;
 386                         search_start = (extent_end + mask) & ~mask;
 387                 } else
 388                         search_start = extent_end;
 389
 390                 if (end < extent_end && end >= key.offset) {
 391                         if (found_extent) {
 392                                 u64 disk_blocknr =
 393                                         btrfs_file_extent_disk_blocknr(extent);
 394                                 u64 disk_num_blocks =
 395                                       btrfs_file_extent_disk_num_blocks(extent);
 396                                 memcpy(&old, extent, 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->blocksize - 1));
 412                         if (found_extent) {
 413                                 new_num = (start - key.offset) >>
 414                                         inode->i_blkbits;
 415                                 old_num = btrfs_file_extent_num_blocks(extent);
 416                                 *hint_block =
 417                                         btrfs_file_extent_disk_blocknr(extent);
 418                                 if (btrfs_file_extent_disk_blocknr(extent)) {
 419                                         inode->i_blocks -=
 420                                                 (old_num - new_num) << 3;
 421                                 }
 422                                 btrfs_set_file_extent_num_blocks(extent,
 423                                                                  new_num);
 424                                 btrfs_mark_buffer_dirty(path->nodes[0]);
 425                         } else {
 426                                 WARN_ON(1);
 427                         }
 428                 }
 429                 /* delete the entire extent */
 430                 if (!keep) {
 431                         u64 disk_blocknr = 0;
 432                         u64 disk_num_blocks = 0;
 433                         u64 extent_num_blocks = 0;
 434                         if (found_extent) {
 435                                 disk_blocknr =
 436                                       btrfs_file_extent_disk_blocknr(extent);
 437                                 disk_num_blocks =
 438                                       btrfs_file_extent_disk_num_blocks(extent);
 439                                 extent_num_blocks =
 440                                       btrfs_file_extent_num_blocks(extent);
 441                                 *hint_block =
 442                                         btrfs_file_extent_disk_blocknr(extent);
 443                         }
 444                         ret = btrfs_del_item(trans, root, path);
 445                         /* TODO update progress marker and return */
 446                         BUG_ON(ret);
 447                         btrfs_release_path(root, path);
 448                         extent = NULL;
 449                         if (found_extent && disk_blocknr != 0) {
 450                                 inode->i_blocks -= extent_num_blocks << 3;
 451                                 ret = btrfs_free_extent(trans, root,
 452                                                         disk_blocknr,
 453                                                         disk_num_blocks, 0);
 454                         }
 455
 456                         BUG_ON(ret);
 457                         if (!bookend && search_start >= end) {
 458                                 ret = 0;
 459                                 goto out;
 460                         }
 461                         if (!bookend)
 462                                 continue;
 463                 }
 464                 /* create bookend, splitting the extent in two */
 465                 if (bookend && found_extent) {
 466                         struct btrfs_key ins;
 467                         ins.objectid = inode->i_ino;
 468                         ins.offset = end;
 469                         ins.flags = 0;
 470                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 471                         btrfs_release_path(root, path);
 472                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 473                                                       sizeof(*extent));
 474
 475                         if (ret) {
 476                                 btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
 477                                 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);
 478                         }
 479                         BUG_ON(ret);
 480                         extent = btrfs_item_ptr(
 481                                     btrfs_buffer_leaf(path->nodes[0]),
 482                                     path->slots[0],
 483                                     struct btrfs_file_extent_item);
 484                         btrfs_set_file_extent_disk_blocknr(extent,
 485                                     btrfs_file_extent_disk_blocknr(&old));
 486                         btrfs_set_file_extent_disk_num_blocks(extent,
 487                                     btrfs_file_extent_disk_num_blocks(&old));
 488
 489                         btrfs_set_file_extent_offset(extent,
 490                                     btrfs_file_extent_offset(&old) +
 491                                     ((end - key.offset) >> inode->i_blkbits));
 492                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
 493                                 (extent_end - end) >> inode->i_blkbits);
 494                         btrfs_set_file_extent_num_blocks(extent,
 495                                     (extent_end - end) >> inode->i_blkbits);
 496
 497                         btrfs_set_file_extent_type(extent,
 498                                                    BTRFS_FILE_EXTENT_REG);
 499                         btrfs_set_file_extent_generation(extent,
 500                                     btrfs_file_extent_generation(&old));
 501                         btrfs_mark_buffer_dirty(path->nodes[0]);
 502                         if (btrfs_file_extent_disk_blocknr(&old) != 0) {
 503                                 inode->i_blocks +=
 504                                       btrfs_file_extent_num_blocks(extent) << 3;
 505                         }
 506                         ret = 0;
 507                         goto out;
 508                 }
 509         }
 510 out:
 511         btrfs_free_path(path);
 512         return ret;
 513 }
 514
 515 /*
 516  * this gets pages into the page cache and locks them down
 517  */
 518 static int prepare_pages(struct btrfs_root *root,
 519                          struct file *file,
 520                          struct page **pages,
 521                          size_t num_pages,
 522                          loff_t pos,
 523                          unsigned long first_index,
 524                          unsigned long last_index,
 525                          size_t write_bytes)
 526 {
 527         int i;
 528         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 529         struct inode *inode = file->f_path.dentry->d_inode;
 530         int err = 0;
 531         u64 num_blocks;
 532         u64 start_pos;
 533
 534         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
 535         num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
 536                         inode->i_blkbits;
 537
 538         memset(pages, 0, num_pages * sizeof(struct page *));
 539
 540         for (i = 0; i < num_pages; i++) {
 541                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 542                 if (!pages[i]) {
 543                         err = -ENOMEM;
 544                         BUG_ON(1);
 545                 }
 546                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 547                 wait_on_page_writeback(pages[i]);
 548                 if (!PagePrivate(pages[i])) {
 549                         SetPagePrivate(pages[i]);
 550                         set_page_private(pages[i], 1);
 551                         page_cache_get(pages[i]);
 552                 }
 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);
 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         .ioctl          = btrfs_ioctl,
 736         .fsync          = btrfs_sync_file,
 737 #ifdef CONFIG_COMPAT
 738         .compat_ioctl   = btrfs_compat_ioctl,
 739 #endif
 740 };
 741