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_USER1);
 124         write_extent_buffer(leaf, kaddr + page_offset, ptr, size);
 125         kunmap_atomic(kaddr, KM_USER1);
 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_byte;
 146         u64 num_bytes;
 147         u64 start_pos;
 148         u64 end_of_last_block;
 149         u64 end_pos = pos + write_bytes;
 150         u32 inline_size;
 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->sectorsize - 1);
 160         num_bytes = (write_bytes + pos - start_pos +
 161                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 162
 163         down_read(&BTRFS_I(inode)->root->snap_sem);
 164         end_of_last_block = start_pos + num_bytes - 1;
 165
 166         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 167         mutex_lock(&root->fs_info->fs_mutex);
 168         trans = btrfs_start_transaction(root, 1);
 169         if (!trans) {
 170                 err = -ENOMEM;
 171                 goto out_unlock;
 172         }
 173         btrfs_set_trans_block_group(trans, inode);
 174         inode->i_blocks += num_bytes >> 9;
 175         hint_byte = 0;
 176
 177         if ((end_of_last_block & 4095) == 0) {
 178                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 179         }
 180         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 181
 182         /* FIXME...EIEIO, ENOSPC and more */
 183
 184         /* insert any holes we need to create */
 185         if (inode->i_size < start_pos) {
 186                 u64 last_pos_in_file;
 187                 u64 hole_size;
 188                 u64 mask = root->sectorsize - 1;
 189                 last_pos_in_file = (isize + mask) & ~mask;
 190                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 191
 192                 if (last_pos_in_file < start_pos) {
 193                         err = btrfs_drop_extents(trans, root, inode,
 194                                                  last_pos_in_file,
 195                                                  last_pos_in_file + hole_size,
 196                                                  &hint_byte);
 197                         if (err)
 198                                 goto failed;
 199
 200                         err = btrfs_insert_file_extent(trans, root,
 201                                                        inode->i_ino,
 202                                                        last_pos_in_file,
 203                                                        0, 0, hole_size);
 204                 }
 205                 if (err)
 206                         goto failed;
 207         }
 208
 209         /*
 210          * either allocate an extent for the new bytes or setup the key
 211          * to show we are doing inline data in the extent
 212          */
 213         inline_size = end_pos - start_pos;
 214         if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
 215             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 216             inline_size >= PAGE_CACHE_SIZE) {
 217                 u64 last_end;
 218                 for (i = 0; i < num_pages; i++) {
 219                         struct page *p = pages[i];
 220                         SetPageUptodate(p);
 221                         set_page_dirty(p);
 222                 }
 223                 last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
 224                 last_end += PAGE_CACHE_SIZE - 1;
 225                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 226                                  GFP_NOFS);
 227         } else {
 228                 struct page *p = pages[0];
 229                 /* step one, delete the existing extents in this range */
 230                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 231                          (pos + write_bytes + root->sectorsize -1) &
 232                          ~((u64)root->sectorsize - 1), &hint_byte);
 233                 if (err)
 234                         goto failed;
 235
 236                 err = insert_inline_extent(trans, root, inode, start_pos,
 237                                            end_pos - start_pos, p, 0);
 238                 BUG_ON(err);
 239                 em->start = start_pos;
 240                 em->end = end_pos - 1;
 241                 em->block_start = EXTENT_MAP_INLINE;
 242                 em->block_end = EXTENT_MAP_INLINE;
 243                 add_extent_mapping(em_tree, em);
 244         }
 245         if (end_pos > isize) {
 246                 i_size_write(inode, end_pos);
 247                 btrfs_update_inode(trans, root, inode);
 248         }
 249 failed:
 250         err = btrfs_end_transaction(trans, root);
 251 out_unlock:
 252         mutex_unlock(&root->fs_info->fs_mutex);
 253         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 254         free_extent_map(em);
 255         up_read(&BTRFS_I(inode)->root->snap_sem);
 256         return err;
 257 }
 258
 259 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 260 {
 261         struct extent_map *em;
 262         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 263
 264         while(1) {
 265                 em = lookup_extent_mapping(em_tree, start, end);
 266                 if (!em)
 267                         break;
 268                 remove_extent_mapping(em_tree, em);
 269                 /* once for us */
 270                 free_extent_map(em);
 271                 /* once for the tree*/
 272                 free_extent_map(em);
 273         }
 274         return 0;
 275 }
 276
 277 /*
 278  * this is very complex, but the basic idea is to drop all extents
 279  * in the range start - end.  hint_block is filled in with a block number
 280  * that would be a good hint to the block allocator for this file.
 281  *
 282  * If an extent intersects the range but is not entirely inside the range
 283  * it is either truncated or split.  Anything entirely inside the range
 284  * is deleted from the tree.
 285  */
 286 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 287                        struct btrfs_root *root, struct inode *inode,
 288                        u64 start, u64 end, u64 *hint_byte)
 289 {
 290         int ret;
 291         struct btrfs_key key;
 292         struct extent_buffer *leaf;
 293         int slot;
 294         struct btrfs_file_extent_item *extent;
 295         u64 extent_end = 0;
 296         int keep;
 297         struct btrfs_file_extent_item old;
 298         struct btrfs_path *path;
 299         u64 search_start = start;
 300         int bookend;
 301         int found_type;
 302         int found_extent;
 303         int found_inline;
 304         int recow;
 305
 306         btrfs_drop_extent_cache(inode, start, end - 1);
 307
 308         path = btrfs_alloc_path();
 309         if (!path)
 310                 return -ENOMEM;
 311         while(1) {
 312                 recow = 0;
 313                 btrfs_release_path(root, path);
 314                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 315                                                search_start, -1);
 316                 if (ret < 0)
 317                         goto out;
 318                 if (ret > 0) {
 319                         if (path->slots[0] == 0) {
 320                                 ret = 0;
 321                                 goto out;
 322                         }
 323                         path->slots[0]--;
 324                 }
 325 next_slot:
 326                 keep = 0;
 327                 bookend = 0;
 328                 found_extent = 0;
 329                 found_inline = 0;
 330                 extent = NULL;
 331                 leaf = path->nodes[0];
 332                 slot = path->slots[0];
 333                 ret = 0;
 334                 btrfs_item_key_to_cpu(leaf, &key, slot);
 335                 if (key.offset >= end || key.objectid != inode->i_ino) {
 336                         goto out;
 337                 }
 338                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 339                         goto out;
 340                 }
 341                 if (recow) {
 342                         search_start = key.offset;
 343                         continue;
 344                 }
 345                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 346                         extent = btrfs_item_ptr(leaf, slot,
 347                                                 struct btrfs_file_extent_item);
 348                         found_type = btrfs_file_extent_type(leaf, extent);
 349                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 350                                 extent_end = key.offset +
 351                                      btrfs_file_extent_num_bytes(leaf, extent);
 352                                 found_extent = 1;
 353                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 354                                 struct btrfs_item *item;
 355                                 item = btrfs_item_nr(leaf, slot);
 356                                 found_inline = 1;
 357                                 extent_end = key.offset +
 358                                      btrfs_file_extent_inline_len(leaf, item);
 359                         }
 360                 } else {
 361                         extent_end = search_start;
 362                 }
 363
 364                 /* we found nothing we can drop */
 365                 if ((!found_extent && !found_inline) ||
 366                     search_start >= extent_end) {
 367                         int nextret;
 368                         u32 nritems;
 369                         nritems = btrfs_header_nritems(leaf);
 370                         if (slot >= nritems - 1) {
 371                                 nextret = btrfs_next_leaf(root, path);
 372                                 if (nextret)
 373                                         goto out;
 374                                 recow = 1;
 375                         } else {
 376                                 path->slots[0]++;
 377                         }
 378                         goto next_slot;
 379                 }
 380
 381                 /* FIXME, there's only one inline extent allowed right now */
 382                 if (found_inline) {
 383                         u64 mask = root->sectorsize - 1;
 384                         search_start = (extent_end + mask) & ~mask;
 385                 } else
 386                         search_start = extent_end;
 387
 388                 if (end < extent_end && end >= key.offset) {
 389                         if (found_extent) {
 390                                 u64 disk_bytenr =
 391                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 392                                 u64 disk_num_bytes =
 393                                     btrfs_file_extent_disk_num_bytes(leaf,
 394                                                                       extent);
 395                                 read_extent_buffer(leaf, &old,
 396                                                    (unsigned long)extent,
 397                                                    sizeof(old));
 398                                 if (disk_bytenr != 0) {
 399                                         ret = btrfs_inc_extent_ref(trans, root,
 400                                                  disk_bytenr, disk_num_bytes);
 401                                         BUG_ON(ret);
 402                                 }
 403                         }
 404                         WARN_ON(found_inline);
 405                         bookend = 1;
 406                 }
 407                 /* truncate existing extent */
 408                 if (start > key.offset) {
 409                         u64 new_num;
 410                         u64 old_num;
 411                         keep = 1;
 412                         WARN_ON(start & (root->sectorsize - 1));
 413                         if (found_extent) {
 414                                 new_num = start - key.offset;
 415                                 old_num = btrfs_file_extent_num_bytes(leaf,
 416                                                                       extent);
 417                                 *hint_byte =
 418                                         btrfs_file_extent_disk_bytenr(leaf,
 419                                                                       extent);
 420                                 if (btrfs_file_extent_disk_bytenr(leaf,
 421                                                                   extent)) {
 422                                         inode->i_blocks -=
 423                                                 (old_num - new_num) >> 9;
 424                                 }
 425                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 426                                                                 new_num);
 427                                 btrfs_mark_buffer_dirty(leaf);
 428                         } else {
 429                                 WARN_ON(1);
 430                         }
 431                 }
 432                 /* delete the entire extent */
 433                 if (!keep) {
 434                         u64 disk_bytenr = 0;
 435                         u64 disk_num_bytes = 0;
 436                         u64 extent_num_bytes = 0;
 437                         if (found_extent) {
 438                                 disk_bytenr =
 439                                       btrfs_file_extent_disk_bytenr(leaf,
 440                                                                      extent);
 441                                 disk_num_bytes =
 442                                       btrfs_file_extent_disk_num_bytes(leaf,
 443                                                                        extent);
 444                                 extent_num_bytes =
 445                                       btrfs_file_extent_num_bytes(leaf, extent);
 446                                 *hint_byte =
 447                                         btrfs_file_extent_disk_bytenr(leaf,
 448                                                                       extent);
 449                         }
 450                         ret = btrfs_del_item(trans, root, path);
 451                         /* TODO update progress marker and return */
 452                         BUG_ON(ret);
 453                         btrfs_release_path(root, path);
 454                         extent = NULL;
 455                         if (found_extent && disk_bytenr != 0) {
 456                                 inode->i_blocks -= extent_num_bytes >> 9;
 457                                 ret = btrfs_free_extent(trans, root,
 458                                                         disk_bytenr,
 459                                                         disk_num_bytes, 0);
 460                         }
 461
 462                         BUG_ON(ret);
 463                         if (!bookend && search_start >= end) {
 464                                 ret = 0;
 465                                 goto out;
 466                         }
 467                         if (!bookend)
 468                                 continue;
 469                 }
 470                 /* create bookend, splitting the extent in two */
 471                 if (bookend && found_extent) {
 472                         struct btrfs_key ins;
 473                         ins.objectid = inode->i_ino;
 474                         ins.offset = end;
 475                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 476                         btrfs_release_path(root, path);
 477                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 478                                                       sizeof(*extent));
 479
 480                         leaf = path->nodes[0];
 481                         if (ret) {
 482                                 btrfs_print_leaf(root, leaf);
 483                                 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);
 484                         }
 485                         BUG_ON(ret);
 486                         extent = btrfs_item_ptr(leaf, path->slots[0],
 487                                                 struct btrfs_file_extent_item);
 488                         write_extent_buffer(leaf, &old,
 489                                             (unsigned long)extent, sizeof(old));
 490
 491                         btrfs_set_file_extent_offset(leaf, extent,
 492                                     le64_to_cpu(old.offset) + end - key.offset);
 493                         WARN_ON(le64_to_cpu(old.num_bytes) <
 494                                 (extent_end - end));
 495                         btrfs_set_file_extent_num_bytes(leaf, extent,
 496                                                         extent_end - end);
 497                         btrfs_set_file_extent_type(leaf, extent,
 498                                                    BTRFS_FILE_EXTENT_REG);
 499
 500                         btrfs_mark_buffer_dirty(path->nodes[0]);
 501                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 502                                 inode->i_blocks +=
 503                                       btrfs_file_extent_num_bytes(leaf,
 504                                                                   extent) >> 9;
 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 start_pos;
 532
 533         start_pos = pos & ~((u64)root->sectorsize - 1);
 534
 535         memset(pages, 0, num_pages * sizeof(struct page *));
 536
 537         for (i = 0; i < num_pages; i++) {
 538                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 539                 if (!pages[i]) {
 540                         err = -ENOMEM;
 541                         BUG_ON(1);
 542                 }
 543                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 544                 wait_on_page_writeback(pages[i]);
 545                 set_page_extent_mapped(pages[i]);
 546                 WARN_ON(!PageLocked(pages[i]));
 547         }
 548         return 0;
 549 }
 550
 551 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 552                                 size_t count, loff_t *ppos)
 553 {
 554         loff_t pos;
 555         size_t num_written = 0;
 556         int err = 0;
 557         int ret = 0;
 558         struct inode *inode = file->f_path.dentry->d_inode;
 559         struct btrfs_root *root = BTRFS_I(inode)->root;
 560         struct page **pages = NULL;
 561         int nrptrs;
 562         struct page *pinned[2];
 563         unsigned long first_index;
 564         unsigned long last_index;
 565
 566         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 567                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 568         pinned[0] = NULL;
 569         pinned[1] = NULL;
 570         if (file->f_flags & O_DIRECT)
 571                 return -EINVAL;
 572         pos = *ppos;
 573         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 574         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 575         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 576         if (err)
 577                 goto out;
 578         if (count == 0)
 579                 goto out;
 580         err = remove_suid(file->f_path.dentry);
 581         if (err)
 582                 goto out;
 583         file_update_time(file);
 584
 585         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 586
 587         mutex_lock(&inode->i_mutex);
 588         first_index = pos >> PAGE_CACHE_SHIFT;
 589         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 590
 591         /*
 592          * there are lots of better ways to do this, but this code
 593          * makes sure the first and last page in the file range are
 594          * up to date and ready for cow
 595          */
 596         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 597                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 598                 if (!PageUptodate(pinned[0])) {
 599                         ret = btrfs_readpage(NULL, pinned[0]);
 600                         BUG_ON(ret);
 601                         wait_on_page_locked(pinned[0]);
 602                 } else {
 603                         unlock_page(pinned[0]);
 604                 }
 605         }
 606         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 607                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 608                 if (!PageUptodate(pinned[1])) {
 609                         ret = btrfs_readpage(NULL, pinned[1]);
 610                         BUG_ON(ret);
 611                         wait_on_page_locked(pinned[1]);
 612                 } else {
 613                         unlock_page(pinned[1]);
 614                 }
 615         }
 616
 617         while(count > 0) {
 618                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 619                 size_t write_bytes = min(count, nrptrs *
 620                                         (size_t)PAGE_CACHE_SIZE -
 621                                          offset);
 622                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 623                                         PAGE_CACHE_SHIFT;
 624
 625                 WARN_ON(num_pages > nrptrs);
 626                 memset(pages, 0, sizeof(pages));
 627                 ret = prepare_pages(root, file, pages, num_pages,
 628                                     pos, first_index, last_index,
 629                                     write_bytes);
 630                 if (ret)
 631                         goto out;
 632
 633                 ret = btrfs_copy_from_user(pos, num_pages,
 634                                            write_bytes, pages, buf);
 635                 if (ret) {
 636                         btrfs_drop_pages(pages, num_pages);
 637                         goto out;
 638                 }
 639
 640                 ret = dirty_and_release_pages(NULL, root, file, pages,
 641                                               num_pages, pos, write_bytes);
 642                 btrfs_drop_pages(pages, num_pages);
 643                 if (ret)
 644                         goto out;
 645
 646                 buf += write_bytes;
 647                 count -= write_bytes;
 648                 pos += write_bytes;
 649                 num_written += write_bytes;
 650
 651                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 652                 btrfs_btree_balance_dirty(root, 1);
 653                 cond_resched();
 654         }
 655         mutex_unlock(&inode->i_mutex);
 656 out:
 657         kfree(pages);
 658         if (pinned[0])
 659                 page_cache_release(pinned[0]);
 660         if (pinned[1])
 661                 page_cache_release(pinned[1]);
 662         *ppos = pos;
 663         current->backing_dev_info = NULL;
 664         return num_written ? num_written : err;
 665 }
 666
 667 static int btrfs_sync_file(struct file *file,
 668                            struct dentry *dentry, int datasync)
 669 {
 670         struct inode *inode = dentry->d_inode;
 671         struct btrfs_root *root = BTRFS_I(inode)->root;
 672         int ret = 0;
 673         struct btrfs_trans_handle *trans;
 674
 675         /*
 676          * check the transaction that last modified this inode
 677          * and see if its already been committed
 678          */
 679         mutex_lock(&root->fs_info->fs_mutex);
 680         if (!BTRFS_I(inode)->last_trans)
 681                 goto out;
 682         mutex_lock(&root->fs_info->trans_mutex);
 683         if (BTRFS_I(inode)->last_trans <=
 684             root->fs_info->last_trans_committed) {
 685                 BTRFS_I(inode)->last_trans = 0;
 686                 mutex_unlock(&root->fs_info->trans_mutex);
 687                 goto out;
 688         }
 689         mutex_unlock(&root->fs_info->trans_mutex);
 690
 691         /*
 692          * ok we haven't committed the transaction yet, lets do a commit
 693          */
 694         trans = btrfs_start_transaction(root, 1);
 695         if (!trans) {
 696                 ret = -ENOMEM;
 697                 goto out;
 698         }
 699         ret = btrfs_commit_transaction(trans, root);
 700 out:
 701         mutex_unlock(&root->fs_info->fs_mutex);
 702         return ret > 0 ? EIO : ret;
 703 }
 704
 705 static struct vm_operations_struct btrfs_file_vm_ops = {
 706 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 707         .nopage         = filemap_nopage,
 708         .populate       = filemap_populate,
 709 #else
 710         .fault          = filemap_fault,
 711 #endif
 712         .page_mkwrite   = btrfs_page_mkwrite,
 713 };
 714
 715 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 716 {
 717         vma->vm_ops = &btrfs_file_vm_ops;
 718         file_accessed(filp);
 719         return 0;
 720 }
 721
 722 struct file_operations btrfs_file_operations = {
 723         .llseek         = generic_file_llseek,
 724         .read           = do_sync_read,
 725         .aio_read       = generic_file_aio_read,
 726         .write          = btrfs_file_write,
 727         .mmap           = btrfs_file_mmap,
 728         .open           = generic_file_open,
 729         .fsync          = btrfs_sync_file,
 730         .unlocked_ioctl = btrfs_ioctl,
 731 #ifdef CONFIG_COMPAT
 732         .compat_ioctl   = btrfs_ioctl,
 733 #endif
 734 };
 735