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, size_t size,
  86                                 struct page **pages, size_t page_offset,
  87                                 int num_pages)
  88 {
  89         struct btrfs_key key;
  90         struct btrfs_path *path;
  91         struct extent_buffer *leaf;
  92         char *kaddr;
  93         unsigned long ptr;
  94         struct btrfs_file_extent_item *ei;
  95         struct page *page;
  96         u32 datasize;
  97         int err = 0;
  98         int ret;
  99         int i;
 100         ssize_t cur_size;
 101
 102         path = btrfs_alloc_path();
 103         if (!path)
 104                 return -ENOMEM;
 105
 106         btrfs_set_trans_block_group(trans, inode);
 107
 108         key.objectid = inode->i_ino;
 109         key.offset = offset;
 110         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 111         datasize = btrfs_file_extent_calc_inline_size(offset + size);
 112
 113         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 114         if (ret < 0) {
 115                 err = ret;
 116                 goto fail;
 117         }
 118         if (ret == 1) {
 119                 path->slots[0]--;
 120                 leaf = path->nodes[0];
 121                 ei = btrfs_item_ptr(leaf, path->slots[0],
 122                                     struct btrfs_file_extent_item);
 123
 124                 if (btrfs_file_extent_type(leaf, ei) !=
 125                     BTRFS_FILE_EXTENT_INLINE) {
 126                         goto insert;
 127                 }
 128                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 129                 ret = 0;
 130         }
 131         if (ret == 0) {
 132                 u32 found_size;
 133                 u64 found_start;
 134
 135                 leaf = path->nodes[0];
 136                 ei = btrfs_item_ptr(leaf, path->slots[0],
 137                                     struct btrfs_file_extent_item);
 138
 139                 if (btrfs_file_extent_type(leaf, ei) !=
 140                     BTRFS_FILE_EXTENT_INLINE) {
 141                         err = ret;
 142                         btrfs_print_leaf(root, leaf);
 143                         printk("found wasn't inline offset %Lu inode %lu\n",
 144                                offset, inode->i_ino);
 145                         goto fail;
 146                 }
 147                 found_start = key.offset;
 148                 found_size = btrfs_file_extent_inline_len(leaf,
 149                                           btrfs_item_nr(leaf, path->slots[0]));
 150
 151                 if (found_size < offset + size) {
 152                         btrfs_release_path(root, path);
 153                         ret = btrfs_search_slot(trans, root, &key, path,
 154                                                 offset + size - found_size -
 155                                                 found_start, 1);
 156                         BUG_ON(ret != 0);
 157                         ret = btrfs_extend_item(trans, root, path,
 158                                                 offset + size - found_size -
 159                                                 found_start);
 160                         if (ret) {
 161                                 err = ret;
 162                                 goto fail;
 163                         }
 164                         leaf = path->nodes[0];
 165                         ei = btrfs_item_ptr(leaf, path->slots[0],
 166                                             struct btrfs_file_extent_item);
 167                 }
 168         } else {
 169 insert:
 170                 btrfs_release_path(root, path);
 171                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 172                                               datasize);
 173                 if (ret) {
 174                         err = ret;
 175                         printk("got bad ret %d\n", ret);
 176                         goto fail;
 177                 }
 178                 leaf = path->nodes[0];
 179                 ei = btrfs_item_ptr(leaf, path->slots[0],
 180                                     struct btrfs_file_extent_item);
 181                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 182                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 183         }
 184         ptr = btrfs_file_extent_inline_start(ei) + offset;
 185
 186         cur_size = size;
 187         i = 0;
 188         while (size > 0) {
 189                 page = pages[i];
 190                 kaddr = kmap_atomic(page, KM_USER0);
 191                 cur_size = min(PAGE_CACHE_SIZE - page_offset, size);
 192                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 193                 kunmap_atomic(kaddr, KM_USER0);
 194                 page_offset = 0;
 195                 ptr += cur_size;
 196                 size -= cur_size;
 197                 if (i >= num_pages) {
 198                         printk("i %d num_pages %d\n", i, num_pages);
 199                 }
 200                 i++;
 201         }
 202         btrfs_mark_buffer_dirty(leaf);
 203 fail:
 204         btrfs_free_path(path);
 205         return err;
 206 }
 207
 208 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 209                                    struct btrfs_root *root,
 210                                    struct file *file,
 211                                    struct page **pages,
 212                                    size_t num_pages,
 213                                    loff_t pos,
 214                                    size_t write_bytes)
 215 {
 216         int err = 0;
 217         int i;
 218         struct inode *inode = file->f_path.dentry->d_inode;
 219         struct extent_map *em;
 220         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 221         u64 hint_byte;
 222         u64 num_bytes;
 223         u64 start_pos;
 224         u64 end_of_last_block;
 225         u64 end_pos = pos + write_bytes;
 226         u32 inline_size;
 227         loff_t isize = i_size_read(inode);
 228
 229         em = alloc_extent_map(GFP_NOFS);
 230         if (!em)
 231                 return -ENOMEM;
 232
 233         em->bdev = inode->i_sb->s_bdev;
 234
 235         start_pos = pos & ~((u64)root->sectorsize - 1);
 236         num_bytes = (write_bytes + pos - start_pos +
 237                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 238
 239         down_read(&BTRFS_I(inode)->root->snap_sem);
 240         end_of_last_block = start_pos + num_bytes - 1;
 241
 242         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 243         mutex_lock(&root->fs_info->fs_mutex);
 244         trans = btrfs_start_transaction(root, 1);
 245         if (!trans) {
 246                 err = -ENOMEM;
 247                 goto out_unlock;
 248         }
 249         btrfs_set_trans_block_group(trans, inode);
 250         inode->i_blocks += num_bytes >> 9;
 251         hint_byte = 0;
 252
 253         if ((end_of_last_block & 4095) == 0) {
 254                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 255         }
 256         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 257
 258         /* FIXME...EIEIO, ENOSPC and more */
 259
 260         /* insert any holes we need to create */
 261         if (inode->i_size < start_pos) {
 262                 u64 last_pos_in_file;
 263                 u64 hole_size;
 264                 u64 mask = root->sectorsize - 1;
 265                 last_pos_in_file = (isize + mask) & ~mask;
 266                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 267
 268                 if (last_pos_in_file < start_pos) {
 269                         err = btrfs_drop_extents(trans, root, inode,
 270                                                  last_pos_in_file,
 271                                                  last_pos_in_file + hole_size,
 272                                                  last_pos_in_file,
 273                                                  &hint_byte);
 274                         if (err)
 275                                 goto failed;
 276
 277                         err = btrfs_insert_file_extent(trans, root,
 278                                                        inode->i_ino,
 279                                                        last_pos_in_file,
 280                                                        0, 0, hole_size);
 281                 }
 282                 if (err)
 283                         goto failed;
 284         }
 285
 286         /*
 287          * either allocate an extent for the new bytes or setup the key
 288          * to show we are doing inline data in the extent
 289          */
 290         inline_size = end_pos;
 291         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 292             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 293                 u64 last_end;
 294
 295                 for (i = 0; i < num_pages; i++) {
 296                         struct page *p = pages[i];
 297                         SetPageUptodate(p);
 298                         set_page_dirty(p);
 299                 }
 300                 last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
 301                 last_end += PAGE_CACHE_SIZE - 1;
 302                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 303                                  GFP_NOFS);
 304         } else {
 305                 u64 aligned_end;
 306                 /* step one, delete the existing extents in this range */
 307                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 308                         ~((u64)root->sectorsize - 1);
 309                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 310                                          aligned_end, end_pos, &hint_byte);
 311                 if (err)
 312                         goto failed;
 313                 err = insert_inline_extent(trans, root, inode, start_pos,
 314                                            end_pos - start_pos, pages, 0,
 315                                            num_pages);
 316                 BUG_ON(err);
 317         }
 318         if (end_pos > isize) {
 319                 i_size_write(inode, end_pos);
 320                 btrfs_update_inode(trans, root, inode);
 321         }
 322 failed:
 323         err = btrfs_end_transaction(trans, root);
 324 out_unlock:
 325         mutex_unlock(&root->fs_info->fs_mutex);
 326         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 327         free_extent_map(em);
 328         up_read(&BTRFS_I(inode)->root->snap_sem);
 329         return err;
 330 }
 331
 332 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 333 {
 334         struct extent_map *em;
 335         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 336
 337         while(1) {
 338                 em = lookup_extent_mapping(em_tree, start, end);
 339                 if (!em)
 340                         break;
 341                 remove_extent_mapping(em_tree, em);
 342                 /* once for us */
 343                 free_extent_map(em);
 344                 /* once for the tree*/
 345                 free_extent_map(em);
 346         }
 347         return 0;
 348 }
 349
 350 /*
 351  * this is very complex, but the basic idea is to drop all extents
 352  * in the range start - end.  hint_block is filled in with a block number
 353  * that would be a good hint to the block allocator for this file.
 354  *
 355  * If an extent intersects the range but is not entirely inside the range
 356  * it is either truncated or split.  Anything entirely inside the range
 357  * is deleted from the tree.
 358  */
 359 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 360                        struct btrfs_root *root, struct inode *inode,
 361                        u64 start, u64 end, u64 inline_end, u64 *hint_byte)
 362 {
 363         int ret;
 364         struct btrfs_key key;
 365         struct extent_buffer *leaf;
 366         int slot;
 367         struct btrfs_file_extent_item *extent;
 368         u64 extent_end = 0;
 369         int keep;
 370         struct btrfs_file_extent_item old;
 371         struct btrfs_path *path;
 372         u64 search_start = start;
 373         int bookend;
 374         int found_type;
 375         int found_extent;
 376         int found_inline;
 377         int recow;
 378
 379         btrfs_drop_extent_cache(inode, start, end - 1);
 380
 381         path = btrfs_alloc_path();
 382         if (!path)
 383                 return -ENOMEM;
 384         while(1) {
 385                 recow = 0;
 386                 btrfs_release_path(root, path);
 387                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 388                                                search_start, -1);
 389                 if (ret < 0)
 390                         goto out;
 391                 if (ret > 0) {
 392                         if (path->slots[0] == 0) {
 393                                 ret = 0;
 394                                 goto out;
 395                         }
 396                         path->slots[0]--;
 397                 }
 398 next_slot:
 399                 keep = 0;
 400                 bookend = 0;
 401                 found_extent = 0;
 402                 found_inline = 0;
 403                 extent = NULL;
 404                 leaf = path->nodes[0];
 405                 slot = path->slots[0];
 406                 ret = 0;
 407                 btrfs_item_key_to_cpu(leaf, &key, slot);
 408                 if (key.offset >= end || key.objectid != inode->i_ino) {
 409                         goto out;
 410                 }
 411                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 412                         goto out;
 413                 }
 414                 if (recow) {
 415                         search_start = key.offset;
 416                         continue;
 417                 }
 418                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 419                         extent = btrfs_item_ptr(leaf, slot,
 420                                                 struct btrfs_file_extent_item);
 421                         found_type = btrfs_file_extent_type(leaf, extent);
 422                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 423                                 extent_end = key.offset +
 424                                      btrfs_file_extent_num_bytes(leaf, extent);
 425                                 found_extent = 1;
 426                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 427                                 struct btrfs_item *item;
 428                                 item = btrfs_item_nr(leaf, slot);
 429                                 found_inline = 1;
 430                                 extent_end = key.offset +
 431                                      btrfs_file_extent_inline_len(leaf, item);
 432                         }
 433                 } else {
 434                         extent_end = search_start;
 435                 }
 436
 437                 /* we found nothing we can drop */
 438                 if ((!found_extent && !found_inline) ||
 439                     search_start >= extent_end) {
 440                         int nextret;
 441                         u32 nritems;
 442                         nritems = btrfs_header_nritems(leaf);
 443                         if (slot >= nritems - 1) {
 444                                 nextret = btrfs_next_leaf(root, path);
 445                                 if (nextret)
 446                                         goto out;
 447                                 recow = 1;
 448                         } else {
 449                                 path->slots[0]++;
 450                         }
 451                         goto next_slot;
 452                 }
 453
 454                 /* FIXME, there's only one inline extent allowed right now */
 455                 if (found_inline) {
 456                         u64 mask = root->sectorsize - 1;
 457                         search_start = (extent_end + mask) & ~mask;
 458                 } else
 459                         search_start = extent_end;
 460
 461                 if (end < extent_end && end >= key.offset) {
 462                         if (found_extent) {
 463                                 u64 disk_bytenr =
 464                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 465                                 u64 disk_num_bytes =
 466                                     btrfs_file_extent_disk_num_bytes(leaf,
 467                                                                       extent);
 468                                 read_extent_buffer(leaf, &old,
 469                                                    (unsigned long)extent,
 470                                                    sizeof(old));
 471                                 if (disk_bytenr != 0) {
 472                                         ret = btrfs_inc_extent_ref(trans, root,
 473                                                  disk_bytenr, disk_num_bytes);
 474                                         BUG_ON(ret);
 475                                 }
 476                         }
 477                         if (!found_inline)
 478                                 bookend = 1;
 479                 }
 480                 /* truncate existing extent */
 481                 if (start > key.offset) {
 482                         u64 new_num;
 483                         u64 old_num;
 484                         keep = 1;
 485                         WARN_ON(start & (root->sectorsize - 1));
 486                         if (found_extent) {
 487                                 new_num = start - key.offset;
 488                                 old_num = btrfs_file_extent_num_bytes(leaf,
 489                                                                       extent);
 490                                 *hint_byte =
 491                                         btrfs_file_extent_disk_bytenr(leaf,
 492                                                                       extent);
 493                                 if (btrfs_file_extent_disk_bytenr(leaf,
 494                                                                   extent)) {
 495                                         inode->i_blocks -=
 496                                                 (old_num - new_num) >> 9;
 497                                 }
 498                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 499                                                                 new_num);
 500                                 btrfs_mark_buffer_dirty(leaf);
 501                         } else if (end > extent_end &&
 502                                    key.offset < inline_end &&
 503                                    inline_end < extent_end) {
 504                                 u32 new_size;
 505                                 new_size = btrfs_file_extent_calc_inline_size(
 506                                                    inline_end - key.offset);
 507                                 btrfs_truncate_item(trans, root, path,
 508                                                     new_size);
 509                         }
 510                 }
 511                 /* delete the entire extent */
 512                 if (!keep) {
 513                         u64 disk_bytenr = 0;
 514                         u64 disk_num_bytes = 0;
 515                         u64 extent_num_bytes = 0;
 516                         if (found_extent) {
 517                                 disk_bytenr =
 518                                       btrfs_file_extent_disk_bytenr(leaf,
 519                                                                      extent);
 520                                 disk_num_bytes =
 521                                       btrfs_file_extent_disk_num_bytes(leaf,
 522                                                                        extent);
 523                                 extent_num_bytes =
 524                                       btrfs_file_extent_num_bytes(leaf, extent);
 525                                 *hint_byte =
 526                                         btrfs_file_extent_disk_bytenr(leaf,
 527                                                                       extent);
 528                         }
 529                         ret = btrfs_del_item(trans, root, path);
 530                         /* TODO update progress marker and return */
 531                         BUG_ON(ret);
 532                         btrfs_release_path(root, path);
 533                         extent = NULL;
 534                         if (found_extent && disk_bytenr != 0) {
 535                                 inode->i_blocks -= extent_num_bytes >> 9;
 536                                 ret = btrfs_free_extent(trans, root,
 537                                                         disk_bytenr,
 538                                                         disk_num_bytes, 0);
 539                         }
 540
 541                         BUG_ON(ret);
 542                         if (!bookend && search_start >= end) {
 543                                 ret = 0;
 544                                 goto out;
 545                         }
 546                         if (!bookend)
 547                                 continue;
 548                 }
 549                 /* create bookend, splitting the extent in two */
 550                 if (bookend && found_extent) {
 551                         struct btrfs_key ins;
 552                         ins.objectid = inode->i_ino;
 553                         ins.offset = end;
 554                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 555                         btrfs_release_path(root, path);
 556                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 557                                                       sizeof(*extent));
 558
 559                         leaf = path->nodes[0];
 560                         if (ret) {
 561                                 btrfs_print_leaf(root, leaf);
 562                                 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);
 563                         }
 564                         BUG_ON(ret);
 565                         extent = btrfs_item_ptr(leaf, path->slots[0],
 566                                                 struct btrfs_file_extent_item);
 567                         write_extent_buffer(leaf, &old,
 568                                             (unsigned long)extent, sizeof(old));
 569
 570                         btrfs_set_file_extent_offset(leaf, extent,
 571                                     le64_to_cpu(old.offset) + end - key.offset);
 572                         WARN_ON(le64_to_cpu(old.num_bytes) <
 573                                 (extent_end - end));
 574                         btrfs_set_file_extent_num_bytes(leaf, extent,
 575                                                         extent_end - end);
 576                         btrfs_set_file_extent_type(leaf, extent,
 577                                                    BTRFS_FILE_EXTENT_REG);
 578
 579                         btrfs_mark_buffer_dirty(path->nodes[0]);
 580                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 581                                 inode->i_blocks +=
 582                                       btrfs_file_extent_num_bytes(leaf,
 583                                                                   extent) >> 9;
 584                         }
 585                         ret = 0;
 586                         goto out;
 587                 }
 588         }
 589 out:
 590         btrfs_free_path(path);
 591         return ret;
 592 }
 593
 594 /*
 595  * this gets pages into the page cache and locks them down
 596  */
 597 static int prepare_pages(struct btrfs_root *root,
 598                          struct file *file,
 599                          struct page **pages,
 600                          size_t num_pages,
 601                          loff_t pos,
 602                          unsigned long first_index,
 603                          unsigned long last_index,
 604                          size_t write_bytes)
 605 {
 606         int i;
 607         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 608         struct inode *inode = file->f_path.dentry->d_inode;
 609         int err = 0;
 610         u64 start_pos;
 611
 612         start_pos = pos & ~((u64)root->sectorsize - 1);
 613
 614         memset(pages, 0, num_pages * sizeof(struct page *));
 615
 616         for (i = 0; i < num_pages; i++) {
 617                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 618                 if (!pages[i]) {
 619                         err = -ENOMEM;
 620                         BUG_ON(1);
 621                 }
 622                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 623                 wait_on_page_writeback(pages[i]);
 624                 set_page_extent_mapped(pages[i]);
 625                 WARN_ON(!PageLocked(pages[i]));
 626         }
 627         return 0;
 628 }
 629
 630 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 631                                 size_t count, loff_t *ppos)
 632 {
 633         loff_t pos;
 634         size_t num_written = 0;
 635         int err = 0;
 636         int ret = 0;
 637         struct inode *inode = file->f_path.dentry->d_inode;
 638         struct btrfs_root *root = BTRFS_I(inode)->root;
 639         struct page **pages = NULL;
 640         int nrptrs;
 641         struct page *pinned[2];
 642         unsigned long first_index;
 643         unsigned long last_index;
 644
 645         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 646                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 647         pinned[0] = NULL;
 648         pinned[1] = NULL;
 649         if (file->f_flags & O_DIRECT)
 650                 return -EINVAL;
 651         pos = *ppos;
 652         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 653         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 654         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 655         if (err)
 656                 goto out;
 657         if (count == 0)
 658                 goto out;
 659         err = remove_suid(file->f_path.dentry);
 660         if (err)
 661                 goto out;
 662         file_update_time(file);
 663
 664         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 665
 666         mutex_lock(&inode->i_mutex);
 667         first_index = pos >> PAGE_CACHE_SHIFT;
 668         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 669
 670         /*
 671          * there are lots of better ways to do this, but this code
 672          * makes sure the first and last page in the file range are
 673          * up to date and ready for cow
 674          */
 675         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 676                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 677                 if (!PageUptodate(pinned[0])) {
 678                         ret = btrfs_readpage(NULL, pinned[0]);
 679                         BUG_ON(ret);
 680                         wait_on_page_locked(pinned[0]);
 681                 } else {
 682                         unlock_page(pinned[0]);
 683                 }
 684         }
 685         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 686                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 687                 if (!PageUptodate(pinned[1])) {
 688                         ret = btrfs_readpage(NULL, pinned[1]);
 689                         BUG_ON(ret);
 690                         wait_on_page_locked(pinned[1]);
 691                 } else {
 692                         unlock_page(pinned[1]);
 693                 }
 694         }
 695
 696         while(count > 0) {
 697                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 698                 size_t write_bytes = min(count, nrptrs *
 699                                         (size_t)PAGE_CACHE_SIZE -
 700                                          offset);
 701                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 702                                         PAGE_CACHE_SHIFT;
 703
 704                 WARN_ON(num_pages > nrptrs);
 705                 memset(pages, 0, sizeof(pages));
 706                 ret = prepare_pages(root, file, pages, num_pages,
 707                                     pos, first_index, last_index,
 708                                     write_bytes);
 709                 if (ret)
 710                         goto out;
 711
 712                 ret = btrfs_copy_from_user(pos, num_pages,
 713                                            write_bytes, pages, buf);
 714                 if (ret) {
 715                         btrfs_drop_pages(pages, num_pages);
 716                         goto out;
 717                 }
 718
 719                 ret = dirty_and_release_pages(NULL, root, file, pages,
 720                                               num_pages, pos, write_bytes);
 721                 btrfs_drop_pages(pages, num_pages);
 722                 if (ret)
 723                         goto out;
 724
 725                 buf += write_bytes;
 726                 count -= write_bytes;
 727                 pos += write_bytes;
 728                 num_written += write_bytes;
 729
 730                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 731                 btrfs_btree_balance_dirty(root, 1);
 732                 cond_resched();
 733         }
 734         mutex_unlock(&inode->i_mutex);
 735 out:
 736         kfree(pages);
 737         if (pinned[0])
 738                 page_cache_release(pinned[0]);
 739         if (pinned[1])
 740                 page_cache_release(pinned[1]);
 741         *ppos = pos;
 742         current->backing_dev_info = NULL;
 743         return num_written ? num_written : err;
 744 }
 745
 746 static int btrfs_sync_file(struct file *file,
 747                            struct dentry *dentry, int datasync)
 748 {
 749         struct inode *inode = dentry->d_inode;
 750         struct btrfs_root *root = BTRFS_I(inode)->root;
 751         int ret = 0;
 752         struct btrfs_trans_handle *trans;
 753
 754         /*
 755          * check the transaction that last modified this inode
 756          * and see if its already been committed
 757          */
 758         mutex_lock(&root->fs_info->fs_mutex);
 759         if (!BTRFS_I(inode)->last_trans)
 760                 goto out;
 761         mutex_lock(&root->fs_info->trans_mutex);
 762         if (BTRFS_I(inode)->last_trans <=
 763             root->fs_info->last_trans_committed) {
 764                 BTRFS_I(inode)->last_trans = 0;
 765                 mutex_unlock(&root->fs_info->trans_mutex);
 766                 goto out;
 767         }
 768         mutex_unlock(&root->fs_info->trans_mutex);
 769
 770         /*
 771          * ok we haven't committed the transaction yet, lets do a commit
 772          */
 773         trans = btrfs_start_transaction(root, 1);
 774         if (!trans) {
 775                 ret = -ENOMEM;
 776                 goto out;
 777         }
 778         ret = btrfs_commit_transaction(trans, root);
 779 out:
 780         mutex_unlock(&root->fs_info->fs_mutex);
 781         return ret > 0 ? EIO : ret;
 782 }
 783
 784 static struct vm_operations_struct btrfs_file_vm_ops = {
 785 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 786         .nopage         = filemap_nopage,
 787         .populate       = filemap_populate,
 788 #else
 789         .fault          = filemap_fault,
 790 #endif
 791         .page_mkwrite   = btrfs_page_mkwrite,
 792 };
 793
 794 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 795 {
 796         vma->vm_ops = &btrfs_file_vm_ops;
 797         file_accessed(filp);
 798         return 0;
 799 }
 800
 801 struct file_operations btrfs_file_operations = {
 802         .llseek         = generic_file_llseek,
 803         .read           = do_sync_read,
 804         .aio_read       = generic_file_aio_read,
 805         .write          = btrfs_file_write,
 806         .mmap           = btrfs_file_mmap,
 807         .open           = generic_file_open,
 808         .fsync          = btrfs_sync_file,
 809         .unlocked_ioctl = btrfs_ioctl,
 810 #ifdef CONFIG_COMPAT
 811         .compat_ioctl   = btrfs_ioctl,
 812 #endif
 813 };
 814