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 noinline 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
 112         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 113         if (ret < 0) {
 114                 err = ret;
 115                 goto fail;
 116         }
 117         if (ret == 1) {
 118                 struct btrfs_key found_key;
 119
 120                 if (path->slots[0] == 0)
 121                         goto insert;
 122
 123                 path->slots[0]--;
 124                 leaf = path->nodes[0];
 125                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 126
 127                 if (found_key.objectid != inode->i_ino)
 128                         goto insert;
 129
 130                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 131                         goto insert;
 132                 ei = btrfs_item_ptr(leaf, path->slots[0],
 133                                     struct btrfs_file_extent_item);
 134
 135                 if (btrfs_file_extent_type(leaf, ei) !=
 136                     BTRFS_FILE_EXTENT_INLINE) {
 137                         goto insert;
 138                 }
 139                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 140                 ret = 0;
 141         }
 142         if (ret == 0) {
 143                 u32 found_size;
 144                 u64 found_end;
 145
 146                 leaf = path->nodes[0];
 147                 ei = btrfs_item_ptr(leaf, path->slots[0],
 148                                     struct btrfs_file_extent_item);
 149
 150                 if (btrfs_file_extent_type(leaf, ei) !=
 151                     BTRFS_FILE_EXTENT_INLINE) {
 152                         err = ret;
 153                         btrfs_print_leaf(root, leaf);
 154                         printk("found wasn't inline offset %Lu inode %lu\n",
 155                                offset, inode->i_ino);
 156                         goto fail;
 157                 }
 158                 found_size = btrfs_file_extent_inline_len(leaf,
 159                                           btrfs_item_nr(leaf, path->slots[0]));
 160                 found_end = key.offset + found_size;
 161
 162                 if (found_end < offset + size) {
 163                         btrfs_release_path(root, path);
 164                         ret = btrfs_search_slot(trans, root, &key, path,
 165                                                 offset + size - found_end, 1);
 166                         BUG_ON(ret != 0);
 167
 168                         ret = btrfs_extend_item(trans, root, path,
 169                                                 offset + size - found_end);
 170                         if (ret) {
 171                                 err = ret;
 172                                 goto fail;
 173                         }
 174                         leaf = path->nodes[0];
 175                         ei = btrfs_item_ptr(leaf, path->slots[0],
 176                                             struct btrfs_file_extent_item);
 177                 }
 178                 if (found_end < offset) {
 179                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 180                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 181                 }
 182         } else {
 183 insert:
 184                 btrfs_release_path(root, path);
 185                 datasize = offset + size - key.offset;
 186                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 187                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 188                                               datasize);
 189                 if (ret) {
 190                         err = ret;
 191                         printk("got bad ret %d\n", ret);
 192                         goto fail;
 193                 }
 194                 leaf = path->nodes[0];
 195                 ei = btrfs_item_ptr(leaf, path->slots[0],
 196                                     struct btrfs_file_extent_item);
 197                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 198                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 199         }
 200         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 201
 202         cur_size = size;
 203         i = 0;
 204         while (size > 0) {
 205                 page = pages[i];
 206                 kaddr = kmap_atomic(page, KM_USER0);
 207                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 208                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 209                 kunmap_atomic(kaddr, KM_USER0);
 210                 page_offset = 0;
 211                 ptr += cur_size;
 212                 size -= cur_size;
 213                 if (i >= num_pages) {
 214                         printk("i %d num_pages %d\n", i, num_pages);
 215                 }
 216                 i++;
 217         }
 218         btrfs_mark_buffer_dirty(leaf);
 219 fail:
 220         btrfs_free_path(path);
 221         return err;
 222 }
 223
 224 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 225                                    struct btrfs_root *root,
 226                                    struct file *file,
 227                                    struct page **pages,
 228                                    size_t num_pages,
 229                                    loff_t pos,
 230                                    size_t write_bytes)
 231 {
 232         int err = 0;
 233         int i;
 234         struct inode *inode = fdentry(file)->d_inode;
 235         struct extent_map *em;
 236         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 237         u64 hint_byte;
 238         u64 num_bytes;
 239         u64 start_pos;
 240         u64 end_of_last_block;
 241         u64 end_pos = pos + write_bytes;
 242         u64 inline_size;
 243         loff_t isize = i_size_read(inode);
 244         em = alloc_extent_map(GFP_NOFS);
 245         if (!em)
 246                 return -ENOMEM;
 247
 248         em->bdev = inode->i_sb->s_bdev;
 249
 250         start_pos = pos & ~((u64)root->sectorsize - 1);
 251         num_bytes = (write_bytes + pos - start_pos +
 252                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 253
 254         end_of_last_block = start_pos + num_bytes - 1;
 255
 256         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 257         mutex_lock(&root->fs_info->fs_mutex);
 258         trans = btrfs_start_transaction(root, 1);
 259         if (!trans) {
 260                 err = -ENOMEM;
 261                 goto out_unlock;
 262         }
 263         btrfs_set_trans_block_group(trans, inode);
 264         inode->i_blocks += num_bytes >> 9;
 265         hint_byte = 0;
 266
 267         if ((end_of_last_block & 4095) == 0) {
 268                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 269         }
 270         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 271
 272         /* FIXME...EIEIO, ENOSPC and more */
 273
 274         /* insert any holes we need to create */
 275         if (inode->i_size < start_pos) {
 276                 u64 last_pos_in_file;
 277                 u64 hole_size;
 278                 u64 mask = root->sectorsize - 1;
 279                 last_pos_in_file = (isize + mask) & ~mask;
 280                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 281
 282                 if (last_pos_in_file < start_pos) {
 283                         err = btrfs_drop_extents(trans, root, inode,
 284                                                  last_pos_in_file,
 285                                                  last_pos_in_file + hole_size,
 286                                                  last_pos_in_file,
 287                                                  &hint_byte);
 288                         if (err)
 289                                 goto failed;
 290
 291                         err = btrfs_insert_file_extent(trans, root,
 292                                                        inode->i_ino,
 293                                                        last_pos_in_file,
 294                                                        0, 0, hole_size);
 295                 }
 296                 if (err)
 297                         goto failed;
 298         }
 299
 300         /*
 301          * either allocate an extent for the new bytes or setup the key
 302          * to show we are doing inline data in the extent
 303          */
 304         inline_size = end_pos;
 305         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 306             inline_size > 8192 ||
 307             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 308                 u64 last_end;
 309                 u64 existing_delalloc = 0;
 310
 311                 for (i = 0; i < num_pages; i++) {
 312                         struct page *p = pages[i];
 313                         SetPageUptodate(p);
 314                         set_page_dirty(p);
 315                 }
 316                 last_end = (u64)(pages[num_pages -1]->index) <<
 317                                 PAGE_CACHE_SHIFT;
 318                 last_end += PAGE_CACHE_SIZE - 1;
 319                 if (start_pos < isize) {
 320                         u64 delalloc_start = start_pos;
 321                         existing_delalloc = count_range_bits(em_tree,
 322                                              &delalloc_start,
 323                                              end_of_last_block, (u64)-1,
 324                                              EXTENT_DELALLOC);
 325                 }
 326                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 327                                  GFP_NOFS);
 328                 spin_lock(&root->fs_info->delalloc_lock);
 329                 root->fs_info->delalloc_bytes += (end_of_last_block + 1 -
 330                                           start_pos) - existing_delalloc;
 331                 spin_unlock(&root->fs_info->delalloc_lock);
 332         } else {
 333                 u64 aligned_end;
 334                 /* step one, delete the existing extents in this range */
 335                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 336                         ~((u64)root->sectorsize - 1);
 337                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 338                                          aligned_end, aligned_end, &hint_byte);
 339                 if (err)
 340                         goto failed;
 341                 if (isize > inline_size)
 342                         inline_size = min_t(u64, isize, aligned_end);
 343                 inline_size -= start_pos;
 344                 err = insert_inline_extent(trans, root, inode, start_pos,
 345                                            inline_size, pages, 0, num_pages);
 346                 BUG_ON(err);
 347         }
 348         if (end_pos > isize) {
 349                 i_size_write(inode, end_pos);
 350                 btrfs_update_inode(trans, root, inode);
 351         }
 352 failed:
 353         err = btrfs_end_transaction(trans, root);
 354 out_unlock:
 355         mutex_unlock(&root->fs_info->fs_mutex);
 356         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 357         free_extent_map(em);
 358         return err;
 359 }
 360
 361 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 362 {
 363         struct extent_map *em;
 364         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 365
 366         while(1) {
 367                 em = lookup_extent_mapping(em_tree, start, end);
 368                 if (!em)
 369                         break;
 370                 remove_extent_mapping(em_tree, em);
 371                 /* once for us */
 372                 free_extent_map(em);
 373                 /* once for the tree*/
 374                 free_extent_map(em);
 375         }
 376         return 0;
 377 }
 378
 379 /*
 380  * this is very complex, but the basic idea is to drop all extents
 381  * in the range start - end.  hint_block is filled in with a block number
 382  * that would be a good hint to the block allocator for this file.
 383  *
 384  * If an extent intersects the range but is not entirely inside the range
 385  * it is either truncated or split.  Anything entirely inside the range
 386  * is deleted from the tree.
 387  */
 388 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 389                        struct btrfs_root *root, struct inode *inode,
 390                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 391 {
 392         u64 extent_end = 0;
 393         u64 search_start = start;
 394         struct extent_buffer *leaf;
 395         struct btrfs_file_extent_item *extent;
 396         struct btrfs_path *path;
 397         struct btrfs_key key;
 398         struct btrfs_file_extent_item old;
 399         int keep;
 400         int slot;
 401         int bookend;
 402         int found_type;
 403         int found_extent;
 404         int found_inline;
 405         int recow;
 406         int ret;
 407
 408         btrfs_drop_extent_cache(inode, start, end - 1);
 409
 410         path = btrfs_alloc_path();
 411         if (!path)
 412                 return -ENOMEM;
 413         while(1) {
 414                 recow = 0;
 415                 btrfs_release_path(root, path);
 416                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 417                                                search_start, -1);
 418                 if (ret < 0)
 419                         goto out;
 420                 if (ret > 0) {
 421                         if (path->slots[0] == 0) {
 422                                 ret = 0;
 423                                 goto out;
 424                         }
 425                         path->slots[0]--;
 426                 }
 427 next_slot:
 428                 keep = 0;
 429                 bookend = 0;
 430                 found_extent = 0;
 431                 found_inline = 0;
 432                 extent = NULL;
 433                 leaf = path->nodes[0];
 434                 slot = path->slots[0];
 435                 ret = 0;
 436                 btrfs_item_key_to_cpu(leaf, &key, slot);
 437                 if (key.offset >= end || key.objectid != inode->i_ino) {
 438                         goto out;
 439                 }
 440                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 441                         goto out;
 442                 }
 443                 if (recow) {
 444                         search_start = key.offset;
 445                         continue;
 446                 }
 447                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 448                         extent = btrfs_item_ptr(leaf, slot,
 449                                                 struct btrfs_file_extent_item);
 450                         found_type = btrfs_file_extent_type(leaf, extent);
 451                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 452                                 extent_end =
 453                                      btrfs_file_extent_disk_bytenr(leaf,
 454                                                                    extent);
 455                                 if (extent_end)
 456                                         *hint_byte = extent_end;
 457
 458                                 extent_end = key.offset +
 459                                      btrfs_file_extent_num_bytes(leaf, extent);
 460                                 found_extent = 1;
 461                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 462                                 struct btrfs_item *item;
 463                                 item = btrfs_item_nr(leaf, slot);
 464                                 found_inline = 1;
 465                                 extent_end = key.offset +
 466                                      btrfs_file_extent_inline_len(leaf, item);
 467                         }
 468                 } else {
 469                         extent_end = search_start;
 470                 }
 471
 472                 /* we found nothing we can drop */
 473                 if ((!found_extent && !found_inline) ||
 474                     search_start >= extent_end) {
 475                         int nextret;
 476                         u32 nritems;
 477                         nritems = btrfs_header_nritems(leaf);
 478                         if (slot >= nritems - 1) {
 479                                 nextret = btrfs_next_leaf(root, path);
 480                                 if (nextret)
 481                                         goto out;
 482                                 recow = 1;
 483                         } else {
 484                                 path->slots[0]++;
 485                         }
 486                         goto next_slot;
 487                 }
 488
 489                 if (found_inline) {
 490                         u64 mask = root->sectorsize - 1;
 491                         search_start = (extent_end + mask) & ~mask;
 492                 } else
 493                         search_start = extent_end;
 494                 if (end <= extent_end && start >= key.offset && found_inline) {
 495                         *hint_byte = EXTENT_MAP_INLINE;
 496                         continue;
 497                 }
 498                 if (end < extent_end && end >= key.offset) {
 499                         if (found_extent) {
 500                                 u64 disk_bytenr =
 501                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 502                                 u64 disk_num_bytes =
 503                                     btrfs_file_extent_disk_num_bytes(leaf,
 504                                                                       extent);
 505                                 read_extent_buffer(leaf, &old,
 506                                                    (unsigned long)extent,
 507                                                    sizeof(old));
 508                                 if (disk_bytenr != 0) {
 509                                         ret = btrfs_inc_extent_ref(trans, root,
 510                                                  disk_bytenr, disk_num_bytes,
 511                                                  root->root_key.objectid,
 512                                                  trans->transid,
 513                                                  key.objectid, end);
 514                                         BUG_ON(ret);
 515                                 }
 516                         }
 517                         bookend = 1;
 518                         if (found_inline && start <= key.offset &&
 519                             inline_limit < extent_end)
 520                                 keep = 1;
 521                 }
 522                 /* truncate existing extent */
 523                 if (start > key.offset) {
 524                         u64 new_num;
 525                         u64 old_num;
 526                         keep = 1;
 527                         WARN_ON(start & (root->sectorsize - 1));
 528                         if (found_extent) {
 529                                 new_num = start - key.offset;
 530                                 old_num = btrfs_file_extent_num_bytes(leaf,
 531                                                                       extent);
 532                                 *hint_byte =
 533                                         btrfs_file_extent_disk_bytenr(leaf,
 534                                                                       extent);
 535                                 if (btrfs_file_extent_disk_bytenr(leaf,
 536                                                                   extent)) {
 537                                         inode->i_blocks -=
 538                                                 (old_num - new_num) >> 9;
 539                                 }
 540                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 541                                                                 new_num);
 542                                 btrfs_mark_buffer_dirty(leaf);
 543                         } else if (key.offset < inline_limit &&
 544                                    (end > extent_end) &&
 545                                    (inline_limit < extent_end)) {
 546                                 u32 new_size;
 547                                 new_size = btrfs_file_extent_calc_inline_size(
 548                                                    inline_limit - key.offset);
 549                                 btrfs_truncate_item(trans, root, path,
 550                                                     new_size, 1);
 551                         }
 552                 }
 553                 /* delete the entire extent */
 554                 if (!keep) {
 555                         u64 disk_bytenr = 0;
 556                         u64 disk_num_bytes = 0;
 557                         u64 extent_num_bytes = 0;
 558                         u64 root_gen;
 559                         u64 root_owner;
 560
 561                         root_gen = btrfs_header_generation(leaf);
 562                         root_owner = btrfs_header_owner(leaf);
 563                         if (found_extent) {
 564                                 disk_bytenr =
 565                                       btrfs_file_extent_disk_bytenr(leaf,
 566                                                                      extent);
 567                                 disk_num_bytes =
 568                                       btrfs_file_extent_disk_num_bytes(leaf,
 569                                                                        extent);
 570                                 extent_num_bytes =
 571                                       btrfs_file_extent_num_bytes(leaf, extent);
 572                                 *hint_byte =
 573                                         btrfs_file_extent_disk_bytenr(leaf,
 574                                                                       extent);
 575                         }
 576                         ret = btrfs_del_item(trans, root, path);
 577                         /* TODO update progress marker and return */
 578                         BUG_ON(ret);
 579                         btrfs_release_path(root, path);
 580                         extent = NULL;
 581                         if (found_extent && disk_bytenr != 0) {
 582                                 inode->i_blocks -= extent_num_bytes >> 9;
 583                                 ret = btrfs_free_extent(trans, root,
 584                                                 disk_bytenr,
 585                                                 disk_num_bytes,
 586                                                 root_owner,
 587                                                 root_gen, inode->i_ino,
 588                                                 key.offset, 0);
 589                         }
 590
 591                         BUG_ON(ret);
 592                         if (!bookend && search_start >= end) {
 593                                 ret = 0;
 594                                 goto out;
 595                         }
 596                         if (!bookend)
 597                                 continue;
 598                 }
 599                 if (bookend && found_inline && start <= key.offset &&
 600                     inline_limit < extent_end && key.offset <= inline_limit) {
 601                         u32 new_size;
 602                         new_size = btrfs_file_extent_calc_inline_size(
 603                                                    extent_end - inline_limit);
 604                         btrfs_truncate_item(trans, root, path, new_size, 0);
 605                 }
 606                 /* create bookend, splitting the extent in two */
 607                 if (bookend && found_extent) {
 608                         struct btrfs_key ins;
 609                         ins.objectid = inode->i_ino;
 610                         ins.offset = end;
 611                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 612                         btrfs_release_path(root, path);
 613                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 614                                                       sizeof(*extent));
 615
 616                         leaf = path->nodes[0];
 617                         if (ret) {
 618                                 btrfs_print_leaf(root, leaf);
 619                                 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);
 620                         }
 621                         BUG_ON(ret);
 622                         extent = btrfs_item_ptr(leaf, path->slots[0],
 623                                                 struct btrfs_file_extent_item);
 624                         write_extent_buffer(leaf, &old,
 625                                             (unsigned long)extent, sizeof(old));
 626
 627                         btrfs_set_file_extent_offset(leaf, extent,
 628                                     le64_to_cpu(old.offset) + end - key.offset);
 629                         WARN_ON(le64_to_cpu(old.num_bytes) <
 630                                 (extent_end - end));
 631                         btrfs_set_file_extent_num_bytes(leaf, extent,
 632                                                         extent_end - end);
 633                         btrfs_set_file_extent_type(leaf, extent,
 634                                                    BTRFS_FILE_EXTENT_REG);
 635
 636                         btrfs_mark_buffer_dirty(path->nodes[0]);
 637                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 638                                 inode->i_blocks +=
 639                                       btrfs_file_extent_num_bytes(leaf,
 640                                                                   extent) >> 9;
 641                         }
 642                         ret = 0;
 643                         goto out;
 644                 }
 645         }
 646 out:
 647         btrfs_free_path(path);
 648         return ret;
 649 }
 650
 651 /*
 652  * this gets pages into the page cache and locks them down
 653  */
 654 static int prepare_pages(struct btrfs_root *root, struct file *file,
 655                          struct page **pages, size_t num_pages,
 656                          loff_t pos, unsigned long first_index,
 657                          unsigned long last_index, size_t write_bytes)
 658 {
 659         int i;
 660         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 661         struct inode *inode = fdentry(file)->d_inode;
 662         int err = 0;
 663         u64 start_pos;
 664
 665         start_pos = pos & ~((u64)root->sectorsize - 1);
 666
 667         memset(pages, 0, num_pages * sizeof(struct page *));
 668
 669         for (i = 0; i < num_pages; i++) {
 670                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 671                 if (!pages[i]) {
 672                         err = -ENOMEM;
 673                         BUG_ON(1);
 674                 }
 675 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 676                 ClearPageDirty(pages[i]);
 677 #else
 678                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 679 #endif
 680                 wait_on_page_writeback(pages[i]);
 681                 set_page_extent_mapped(pages[i]);
 682                 WARN_ON(!PageLocked(pages[i]));
 683         }
 684         return 0;
 685 }
 686
 687 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 688                                 size_t count, loff_t *ppos)
 689 {
 690         loff_t pos;
 691         loff_t start_pos;
 692         ssize_t num_written = 0;
 693         ssize_t err = 0;
 694         int ret = 0;
 695         struct inode *inode = fdentry(file)->d_inode;
 696         struct btrfs_root *root = BTRFS_I(inode)->root;
 697         struct page **pages = NULL;
 698         int nrptrs;
 699         struct page *pinned[2];
 700         unsigned long first_index;
 701         unsigned long last_index;
 702
 703         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 704                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 705         pinned[0] = NULL;
 706         pinned[1] = NULL;
 707         if (file->f_flags & O_DIRECT)
 708                 return -EINVAL;
 709
 710         pos = *ppos;
 711         start_pos = pos;
 712
 713         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 714         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 715         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 716         if (err)
 717                 goto out_nolock;
 718         if (count == 0)
 719                 goto out_nolock;
 720         err = remove_suid(fdentry(file));
 721         if (err)
 722                 goto out_nolock;
 723         file_update_time(file);
 724
 725         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 726
 727         down_read(&BTRFS_I(inode)->root->snap_sem);
 728
 729         mutex_lock(&inode->i_mutex);
 730         first_index = pos >> PAGE_CACHE_SHIFT;
 731         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 732
 733         /*
 734          * there are lots of better ways to do this, but this code
 735          * makes sure the first and last page in the file range are
 736          * up to date and ready for cow
 737          */
 738         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 739                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 740                 if (!PageUptodate(pinned[0])) {
 741                         ret = btrfs_readpage(NULL, pinned[0]);
 742                         BUG_ON(ret);
 743                         wait_on_page_locked(pinned[0]);
 744                 } else {
 745                         unlock_page(pinned[0]);
 746                 }
 747         }
 748         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 749                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 750                 if (!PageUptodate(pinned[1])) {
 751                         ret = btrfs_readpage(NULL, pinned[1]);
 752                         BUG_ON(ret);
 753                         wait_on_page_locked(pinned[1]);
 754                 } else {
 755                         unlock_page(pinned[1]);
 756                 }
 757         }
 758
 759         while(count > 0) {
 760                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 761                 size_t write_bytes = min(count, nrptrs *
 762                                         (size_t)PAGE_CACHE_SIZE -
 763                                          offset);
 764                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 765                                         PAGE_CACHE_SHIFT;
 766
 767                 WARN_ON(num_pages > nrptrs);
 768                 memset(pages, 0, sizeof(pages));
 769
 770                 mutex_lock(&root->fs_info->fs_mutex);
 771                 ret = btrfs_check_free_space(root, write_bytes, 0);
 772                 mutex_unlock(&root->fs_info->fs_mutex);
 773                 if (ret)
 774                         goto out;
 775
 776                 ret = prepare_pages(root, file, pages, num_pages,
 777                                     pos, first_index, last_index,
 778                                     write_bytes);
 779                 if (ret)
 780                         goto out;
 781
 782                 ret = btrfs_copy_from_user(pos, num_pages,
 783                                            write_bytes, pages, buf);
 784                 if (ret) {
 785                         btrfs_drop_pages(pages, num_pages);
 786                         goto out;
 787                 }
 788
 789                 ret = dirty_and_release_pages(NULL, root, file, pages,
 790                                               num_pages, pos, write_bytes);
 791                 btrfs_drop_pages(pages, num_pages);
 792                 if (ret)
 793                         goto out;
 794
 795                 buf += write_bytes;
 796                 count -= write_bytes;
 797                 pos += write_bytes;
 798                 num_written += write_bytes;
 799
 800                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 801                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 802                         btrfs_btree_balance_dirty(root, 1);
 803                 cond_resched();
 804         }
 805 out:
 806         mutex_unlock(&inode->i_mutex);
 807         up_read(&BTRFS_I(inode)->root->snap_sem);
 808
 809 out_nolock:
 810         kfree(pages);
 811         if (pinned[0])
 812                 page_cache_release(pinned[0]);
 813         if (pinned[1])
 814                 page_cache_release(pinned[1]);
 815         *ppos = pos;
 816
 817         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 818                 err = sync_page_range(inode, inode->i_mapping,
 819                                       start_pos, num_written);
 820                 if (err < 0)
 821                         num_written = err;
 822         }
 823         current->backing_dev_info = NULL;
 824         return num_written ? num_written : err;
 825 }
 826
 827 static int btrfs_sync_file(struct file *file,
 828                            struct dentry *dentry, int datasync)
 829 {
 830         struct inode *inode = dentry->d_inode;
 831         struct btrfs_root *root = BTRFS_I(inode)->root;
 832         int ret = 0;
 833         struct btrfs_trans_handle *trans;
 834
 835         /*
 836          * check the transaction that last modified this inode
 837          * and see if its already been committed
 838          */
 839         mutex_lock(&root->fs_info->fs_mutex);
 840         if (!BTRFS_I(inode)->last_trans)
 841                 goto out;
 842         mutex_lock(&root->fs_info->trans_mutex);
 843         if (BTRFS_I(inode)->last_trans <=
 844             root->fs_info->last_trans_committed) {
 845                 BTRFS_I(inode)->last_trans = 0;
 846                 mutex_unlock(&root->fs_info->trans_mutex);
 847                 goto out;
 848         }
 849         mutex_unlock(&root->fs_info->trans_mutex);
 850
 851         /*
 852          * ok we haven't committed the transaction yet, lets do a commit
 853          */
 854         trans = btrfs_start_transaction(root, 1);
 855         if (!trans) {
 856                 ret = -ENOMEM;
 857                 goto out;
 858         }
 859         ret = btrfs_commit_transaction(trans, root);
 860 out:
 861         mutex_unlock(&root->fs_info->fs_mutex);
 862         return ret > 0 ? EIO : ret;
 863 }
 864
 865 static struct vm_operations_struct btrfs_file_vm_ops = {
 866 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 867         .nopage         = filemap_nopage,
 868         .populate       = filemap_populate,
 869 #else
 870         .fault          = filemap_fault,
 871 #endif
 872         .page_mkwrite   = btrfs_page_mkwrite,
 873 };
 874
 875 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 876 {
 877         vma->vm_ops = &btrfs_file_vm_ops;
 878         file_accessed(filp);
 879         return 0;
 880 }
 881
 882 struct file_operations btrfs_file_operations = {
 883         .llseek         = generic_file_llseek,
 884         .read           = do_sync_read,
 885         .aio_read       = generic_file_aio_read,
 886         .splice_read    = generic_file_splice_read,
 887 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 888         .sendfile       = generic_file_sendfile,
 889 #endif
 890         .write          = btrfs_file_write,
 891         .mmap           = btrfs_file_mmap,
 892         .open           = generic_file_open,
 893         .fsync          = btrfs_sync_file,
 894         .unlocked_ioctl = btrfs_ioctl,
 895 #ifdef CONFIG_COMPAT
 896         .compat_ioctl   = btrfs_ioctl,
 897 #endif
 898 };
 899