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
 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 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         down_read(&BTRFS_I(inode)->root->snap_sem);
 255         end_of_last_block = start_pos + num_bytes - 1;
 256
 257         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 258         mutex_lock(&root->fs_info->fs_mutex);
 259         trans = btrfs_start_transaction(root, 1);
 260         if (!trans) {
 261                 err = -ENOMEM;
 262                 goto out_unlock;
 263         }
 264         btrfs_set_trans_block_group(trans, inode);
 265         inode->i_blocks += num_bytes >> 9;
 266         hint_byte = 0;
 267
 268         if ((end_of_last_block & 4095) == 0) {
 269                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 270         }
 271         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 272
 273         /* FIXME...EIEIO, ENOSPC and more */
 274
 275         /* insert any holes we need to create */
 276         if (inode->i_size < start_pos) {
 277                 u64 last_pos_in_file;
 278                 u64 hole_size;
 279                 u64 mask = root->sectorsize - 1;
 280                 last_pos_in_file = (isize + mask) & ~mask;
 281                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 282
 283                 if (last_pos_in_file < start_pos) {
 284                         err = btrfs_drop_extents(trans, root, inode,
 285                                                  last_pos_in_file,
 286                                                  last_pos_in_file + hole_size,
 287                                                  last_pos_in_file,
 288                                                  &hint_byte);
 289                         if (err)
 290                                 goto failed;
 291
 292                         err = btrfs_insert_file_extent(trans, root,
 293                                                        inode->i_ino,
 294                                                        last_pos_in_file,
 295                                                        0, 0, hole_size);
 296                 }
 297                 if (err)
 298                         goto failed;
 299         }
 300
 301         /*
 302          * either allocate an extent for the new bytes or setup the key
 303          * to show we are doing inline data in the extent
 304          */
 305         inline_size = end_pos;
 306         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 307             inline_size > 32768 ||
 308             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 309                 u64 last_end;
 310                 u64 existing_delalloc = 0;
 311
 312                 for (i = 0; i < num_pages; i++) {
 313                         struct page *p = pages[i];
 314                         SetPageUptodate(p);
 315                         set_page_dirty(p);
 316                 }
 317                 last_end = (u64)(pages[num_pages -1]->index) <<
 318                                 PAGE_CACHE_SHIFT;
 319                 last_end += PAGE_CACHE_SIZE - 1;
 320                 if (start_pos < isize) {
 321                         u64 delalloc_start = start_pos;
 322                         existing_delalloc = count_range_bits(em_tree,
 323                                              &delalloc_start,
 324                                              end_of_last_block, (u64)-1,
 325                                              EXTENT_DELALLOC);
 326                 }
 327                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 328                                  GFP_NOFS);
 329                 spin_lock(&root->fs_info->delalloc_lock);
 330                 root->fs_info->delalloc_bytes += (end_of_last_block + 1 -
 331                                           start_pos) - existing_delalloc;
 332                 spin_unlock(&root->fs_info->delalloc_lock);
 333         } else {
 334                 u64 aligned_end;
 335                 /* step one, delete the existing extents in this range */
 336                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 337                         ~((u64)root->sectorsize - 1);
 338                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 339                                          aligned_end, aligned_end, &hint_byte);
 340                 if (err)
 341                         goto failed;
 342                 if (isize > inline_size)
 343                         inline_size = min_t(u64, isize, aligned_end);
 344                 inline_size -= start_pos;
 345                 err = insert_inline_extent(trans, root, inode, start_pos,
 346                                            inline_size, pages, 0, num_pages);
 347                 BUG_ON(err);
 348         }
 349         if (end_pos > isize) {
 350                 i_size_write(inode, end_pos);
 351                 btrfs_update_inode(trans, root, inode);
 352         }
 353 failed:
 354         err = btrfs_end_transaction(trans, root);
 355 out_unlock:
 356         mutex_unlock(&root->fs_info->fs_mutex);
 357         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 358         free_extent_map(em);
 359         up_read(&BTRFS_I(inode)->root->snap_sem);
 360         return err;
 361 }
 362
 363 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 364 {
 365         struct extent_map *em;
 366         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 367
 368         while(1) {
 369                 em = lookup_extent_mapping(em_tree, start, end);
 370                 if (!em)
 371                         break;
 372                 remove_extent_mapping(em_tree, em);
 373                 /* once for us */
 374                 free_extent_map(em);
 375                 /* once for the tree*/
 376                 free_extent_map(em);
 377         }
 378         return 0;
 379 }
 380
 381 /*
 382  * this is very complex, but the basic idea is to drop all extents
 383  * in the range start - end.  hint_block is filled in with a block number
 384  * that would be a good hint to the block allocator for this file.
 385  *
 386  * If an extent intersects the range but is not entirely inside the range
 387  * it is either truncated or split.  Anything entirely inside the range
 388  * is deleted from the tree.
 389  */
 390 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 391                        struct btrfs_root *root, struct inode *inode,
 392                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 393 {
 394         u64 extent_end = 0;
 395         u64 search_start = start;
 396         struct extent_buffer *leaf;
 397         struct btrfs_file_extent_item *extent;
 398         struct btrfs_path *path;
 399         struct btrfs_key key;
 400         struct btrfs_file_extent_item old;
 401         int keep;
 402         int slot;
 403         int bookend;
 404         int found_type;
 405         int found_extent;
 406         int found_inline;
 407         int recow;
 408         int ret;
 409
 410         btrfs_drop_extent_cache(inode, start, end - 1);
 411
 412         path = btrfs_alloc_path();
 413         if (!path)
 414                 return -ENOMEM;
 415         while(1) {
 416                 recow = 0;
 417                 btrfs_release_path(root, path);
 418                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 419                                                search_start, -1);
 420                 if (ret < 0)
 421                         goto out;
 422                 if (ret > 0) {
 423                         if (path->slots[0] == 0) {
 424                                 ret = 0;
 425                                 goto out;
 426                         }
 427                         path->slots[0]--;
 428                 }
 429 next_slot:
 430                 keep = 0;
 431                 bookend = 0;
 432                 found_extent = 0;
 433                 found_inline = 0;
 434                 extent = NULL;
 435                 leaf = path->nodes[0];
 436                 slot = path->slots[0];
 437                 ret = 0;
 438                 btrfs_item_key_to_cpu(leaf, &key, slot);
 439                 if (key.offset >= end || key.objectid != inode->i_ino) {
 440                         goto out;
 441                 }
 442                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 443                         goto out;
 444                 }
 445                 if (recow) {
 446                         search_start = key.offset;
 447                         continue;
 448                 }
 449                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 450                         extent = btrfs_item_ptr(leaf, slot,
 451                                                 struct btrfs_file_extent_item);
 452                         found_type = btrfs_file_extent_type(leaf, extent);
 453                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 454                                 extent_end =
 455                                      btrfs_file_extent_disk_bytenr(leaf,
 456                                                                    extent);
 457                                 if (extent_end)
 458                                         *hint_byte = extent_end;
 459
 460                                 extent_end = key.offset +
 461                                      btrfs_file_extent_num_bytes(leaf, extent);
 462                                 found_extent = 1;
 463                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 464                                 struct btrfs_item *item;
 465                                 item = btrfs_item_nr(leaf, slot);
 466                                 found_inline = 1;
 467                                 extent_end = key.offset +
 468                                      btrfs_file_extent_inline_len(leaf, item);
 469                         }
 470                 } else {
 471                         extent_end = search_start;
 472                 }
 473
 474                 /* we found nothing we can drop */
 475                 if ((!found_extent && !found_inline) ||
 476                     search_start >= extent_end) {
 477                         int nextret;
 478                         u32 nritems;
 479                         nritems = btrfs_header_nritems(leaf);
 480                         if (slot >= nritems - 1) {
 481                                 nextret = btrfs_next_leaf(root, path);
 482                                 if (nextret)
 483                                         goto out;
 484                                 recow = 1;
 485                         } else {
 486                                 path->slots[0]++;
 487                         }
 488                         goto next_slot;
 489                 }
 490
 491                 if (found_inline) {
 492                         u64 mask = root->sectorsize - 1;
 493                         search_start = (extent_end + mask) & ~mask;
 494                 } else
 495                         search_start = extent_end;
 496                 if (end <= extent_end && start >= key.offset && found_inline) {
 497                         *hint_byte = EXTENT_MAP_INLINE;
 498                         continue;
 499                 }
 500                 if (end < extent_end && end >= key.offset) {
 501                         if (found_extent) {
 502                                 u64 disk_bytenr =
 503                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 504                                 u64 disk_num_bytes =
 505                                     btrfs_file_extent_disk_num_bytes(leaf,
 506                                                                       extent);
 507                                 read_extent_buffer(leaf, &old,
 508                                                    (unsigned long)extent,
 509                                                    sizeof(old));
 510                                 if (disk_bytenr != 0) {
 511                                         ret = btrfs_inc_extent_ref(trans, root,
 512                                                  disk_bytenr, disk_num_bytes,
 513                                                  root->root_key.objectid,
 514                                                  trans->transid,
 515                                                  key.objectid, end);
 516                                         BUG_ON(ret);
 517                                 }
 518                         }
 519                         bookend = 1;
 520                         if (found_inline && start <= key.offset &&
 521                             inline_limit < extent_end)
 522                                 keep = 1;
 523                 }
 524                 /* truncate existing extent */
 525                 if (start > key.offset) {
 526                         u64 new_num;
 527                         u64 old_num;
 528                         keep = 1;
 529                         WARN_ON(start & (root->sectorsize - 1));
 530                         if (found_extent) {
 531                                 new_num = start - key.offset;
 532                                 old_num = btrfs_file_extent_num_bytes(leaf,
 533                                                                       extent);
 534                                 *hint_byte =
 535                                         btrfs_file_extent_disk_bytenr(leaf,
 536                                                                       extent);
 537                                 if (btrfs_file_extent_disk_bytenr(leaf,
 538                                                                   extent)) {
 539                                         inode->i_blocks -=
 540                                                 (old_num - new_num) >> 9;
 541                                 }
 542                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 543                                                                 new_num);
 544                                 btrfs_mark_buffer_dirty(leaf);
 545                         } else if (key.offset < inline_limit &&
 546                                    (end > extent_end) &&
 547                                    (inline_limit < extent_end)) {
 548                                 u32 new_size;
 549                                 new_size = btrfs_file_extent_calc_inline_size(
 550                                                    inline_limit - key.offset);
 551                                 btrfs_truncate_item(trans, root, path,
 552                                                     new_size, 1);
 553                         }
 554                 }
 555                 /* delete the entire extent */
 556                 if (!keep) {
 557                         u64 disk_bytenr = 0;
 558                         u64 disk_num_bytes = 0;
 559                         u64 extent_num_bytes = 0;
 560                         u64 root_gen;
 561                         u64 root_owner;
 562
 563                         root_gen = btrfs_header_generation(leaf);
 564                         root_owner = btrfs_header_owner(leaf);
 565                         if (found_extent) {
 566                                 disk_bytenr =
 567                                       btrfs_file_extent_disk_bytenr(leaf,
 568                                                                      extent);
 569                                 disk_num_bytes =
 570                                       btrfs_file_extent_disk_num_bytes(leaf,
 571                                                                        extent);
 572                                 extent_num_bytes =
 573                                       btrfs_file_extent_num_bytes(leaf, extent);
 574                                 *hint_byte =
 575                                         btrfs_file_extent_disk_bytenr(leaf,
 576                                                                       extent);
 577                         }
 578                         ret = btrfs_del_item(trans, root, path);
 579                         /* TODO update progress marker and return */
 580                         BUG_ON(ret);
 581                         btrfs_release_path(root, path);
 582                         extent = NULL;
 583                         if (found_extent && disk_bytenr != 0) {
 584                                 inode->i_blocks -= extent_num_bytes >> 9;
 585                                 ret = btrfs_free_extent(trans, root,
 586                                                 disk_bytenr,
 587                                                 disk_num_bytes,
 588                                                 root_owner,
 589                                                 root_gen, inode->i_ino,
 590                                                 key.offset, 0);
 591                         }
 592
 593                         BUG_ON(ret);
 594                         if (!bookend && search_start >= end) {
 595                                 ret = 0;
 596                                 goto out;
 597                         }
 598                         if (!bookend)
 599                                 continue;
 600                 }
 601                 if (bookend && found_inline && start <= key.offset &&
 602                     inline_limit < extent_end && key.offset <= inline_limit) {
 603                         u32 new_size;
 604                         new_size = btrfs_file_extent_calc_inline_size(
 605                                                    extent_end - inline_limit);
 606                         btrfs_truncate_item(trans, root, path, new_size, 0);
 607                 }
 608                 /* create bookend, splitting the extent in two */
 609                 if (bookend && found_extent) {
 610                         struct btrfs_key ins;
 611                         ins.objectid = inode->i_ino;
 612                         ins.offset = end;
 613                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 614                         btrfs_release_path(root, path);
 615                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 616                                                       sizeof(*extent));
 617
 618                         leaf = path->nodes[0];
 619                         if (ret) {
 620                                 btrfs_print_leaf(root, leaf);
 621                                 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);
 622                         }
 623                         BUG_ON(ret);
 624                         extent = btrfs_item_ptr(leaf, path->slots[0],
 625                                                 struct btrfs_file_extent_item);
 626                         write_extent_buffer(leaf, &old,
 627                                             (unsigned long)extent, sizeof(old));
 628
 629                         btrfs_set_file_extent_offset(leaf, extent,
 630                                     le64_to_cpu(old.offset) + end - key.offset);
 631                         WARN_ON(le64_to_cpu(old.num_bytes) <
 632                                 (extent_end - end));
 633                         btrfs_set_file_extent_num_bytes(leaf, extent,
 634                                                         extent_end - end);
 635                         btrfs_set_file_extent_type(leaf, extent,
 636                                                    BTRFS_FILE_EXTENT_REG);
 637
 638                         btrfs_mark_buffer_dirty(path->nodes[0]);
 639                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 640                                 inode->i_blocks +=
 641                                       btrfs_file_extent_num_bytes(leaf,
 642                                                                   extent) >> 9;
 643                         }
 644                         ret = 0;
 645                         goto out;
 646                 }
 647         }
 648 out:
 649         btrfs_free_path(path);
 650         return ret;
 651 }
 652
 653 /*
 654  * this gets pages into the page cache and locks them down
 655  */
 656 static int prepare_pages(struct btrfs_root *root,
 657                          struct file *file,
 658                          struct page **pages,
 659                          size_t num_pages,
 660                          loff_t pos,
 661                          unsigned long first_index,
 662                          unsigned long last_index,
 663                          size_t write_bytes)
 664 {
 665         int i;
 666         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 667         struct inode *inode = fdentry(file)->d_inode;
 668         int err = 0;
 669         u64 start_pos;
 670
 671         start_pos = pos & ~((u64)root->sectorsize - 1);
 672
 673         memset(pages, 0, num_pages * sizeof(struct page *));
 674
 675         for (i = 0; i < num_pages; i++) {
 676                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 677                 if (!pages[i]) {
 678                         err = -ENOMEM;
 679                         BUG_ON(1);
 680                 }
 681 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 682                 ClearPageDirty(pages[i]);
 683 #else
 684                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 685 #endif
 686                 wait_on_page_writeback(pages[i]);
 687                 set_page_extent_mapped(pages[i]);
 688                 WARN_ON(!PageLocked(pages[i]));
 689         }
 690         return 0;
 691 }
 692
 693 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 694                                 size_t count, loff_t *ppos)
 695 {
 696         loff_t pos;
 697         loff_t start_pos;
 698         ssize_t num_written = 0;
 699         ssize_t err = 0;
 700         int ret = 0;
 701         struct inode *inode = fdentry(file)->d_inode;
 702         struct btrfs_root *root = BTRFS_I(inode)->root;
 703         struct page **pages = NULL;
 704         int nrptrs;
 705         struct page *pinned[2];
 706         unsigned long first_index;
 707         unsigned long last_index;
 708
 709         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 710                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 711         pinned[0] = NULL;
 712         pinned[1] = NULL;
 713         if (file->f_flags & O_DIRECT)
 714                 return -EINVAL;
 715
 716         pos = *ppos;
 717         start_pos = pos;
 718
 719         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 720         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 721         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 722         if (err)
 723                 goto out_nolock;
 724         if (count == 0)
 725                 goto out_nolock;
 726         err = remove_suid(fdentry(file));
 727         if (err)
 728                 goto out_nolock;
 729         file_update_time(file);
 730
 731         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 732
 733         mutex_lock(&inode->i_mutex);
 734         first_index = pos >> PAGE_CACHE_SHIFT;
 735         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 736
 737         /*
 738          * there are lots of better ways to do this, but this code
 739          * makes sure the first and last page in the file range are
 740          * up to date and ready for cow
 741          */
 742         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 743                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 744                 if (!PageUptodate(pinned[0])) {
 745                         ret = btrfs_readpage(NULL, pinned[0]);
 746                         BUG_ON(ret);
 747                         wait_on_page_locked(pinned[0]);
 748                 } else {
 749                         unlock_page(pinned[0]);
 750                 }
 751         }
 752         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 753                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 754                 if (!PageUptodate(pinned[1])) {
 755                         ret = btrfs_readpage(NULL, pinned[1]);
 756                         BUG_ON(ret);
 757                         wait_on_page_locked(pinned[1]);
 758                 } else {
 759                         unlock_page(pinned[1]);
 760                 }
 761         }
 762
 763         while(count > 0) {
 764                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 765                 size_t write_bytes = min(count, nrptrs *
 766                                         (size_t)PAGE_CACHE_SIZE -
 767                                          offset);
 768                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 769                                         PAGE_CACHE_SHIFT;
 770
 771                 WARN_ON(num_pages > nrptrs);
 772                 memset(pages, 0, sizeof(pages));
 773
 774                 mutex_lock(&root->fs_info->fs_mutex);
 775                 ret = btrfs_check_free_space(root, write_bytes, 0);
 776                 mutex_unlock(&root->fs_info->fs_mutex);
 777                 if (ret)
 778                         goto out;
 779
 780                 ret = prepare_pages(root, file, pages, num_pages,
 781                                     pos, first_index, last_index,
 782                                     write_bytes);
 783                 if (ret)
 784                         goto out;
 785
 786                 ret = btrfs_copy_from_user(pos, num_pages,
 787                                            write_bytes, pages, buf);
 788                 if (ret) {
 789                         btrfs_drop_pages(pages, num_pages);
 790                         goto out;
 791                 }
 792
 793                 ret = dirty_and_release_pages(NULL, root, file, pages,
 794                                               num_pages, pos, write_bytes);
 795                 btrfs_drop_pages(pages, num_pages);
 796                 if (ret)
 797                         goto out;
 798
 799                 buf += write_bytes;
 800                 count -= write_bytes;
 801                 pos += write_bytes;
 802                 num_written += write_bytes;
 803
 804                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 805                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 806                         btrfs_btree_balance_dirty(root, 1);
 807                 cond_resched();
 808         }
 809 out:
 810         mutex_unlock(&inode->i_mutex);
 811 out_nolock:
 812         kfree(pages);
 813         if (pinned[0])
 814                 page_cache_release(pinned[0]);
 815         if (pinned[1])
 816                 page_cache_release(pinned[1]);
 817         *ppos = pos;
 818
 819         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 820                 err = sync_page_range(inode, inode->i_mapping,
 821                                       start_pos, num_written);
 822                 if (err < 0)
 823                         num_written = err;
 824         }
 825         current->backing_dev_info = NULL;
 826         return num_written ? num_written : err;
 827 }
 828
 829 static int btrfs_sync_file(struct file *file,
 830                            struct dentry *dentry, int datasync)
 831 {
 832         struct inode *inode = dentry->d_inode;
 833         struct btrfs_root *root = BTRFS_I(inode)->root;
 834         int ret = 0;
 835         struct btrfs_trans_handle *trans;
 836
 837         /*
 838          * check the transaction that last modified this inode
 839          * and see if its already been committed
 840          */
 841         mutex_lock(&root->fs_info->fs_mutex);
 842         if (!BTRFS_I(inode)->last_trans)
 843                 goto out;
 844         mutex_lock(&root->fs_info->trans_mutex);
 845         if (BTRFS_I(inode)->last_trans <=
 846             root->fs_info->last_trans_committed) {
 847                 BTRFS_I(inode)->last_trans = 0;
 848                 mutex_unlock(&root->fs_info->trans_mutex);
 849                 goto out;
 850         }
 851         mutex_unlock(&root->fs_info->trans_mutex);
 852
 853         /*
 854          * ok we haven't committed the transaction yet, lets do a commit
 855          */
 856         trans = btrfs_start_transaction(root, 1);
 857         if (!trans) {
 858                 ret = -ENOMEM;
 859                 goto out;
 860         }
 861         ret = btrfs_commit_transaction(trans, root);
 862 out:
 863         mutex_unlock(&root->fs_info->fs_mutex);
 864         return ret > 0 ? EIO : ret;
 865 }
 866
 867 static struct vm_operations_struct btrfs_file_vm_ops = {
 868 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 869         .nopage         = filemap_nopage,
 870         .populate       = filemap_populate,
 871 #else
 872         .fault          = filemap_fault,
 873 #endif
 874         .page_mkwrite   = btrfs_page_mkwrite,
 875 };
 876
 877 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 878 {
 879         vma->vm_ops = &btrfs_file_vm_ops;
 880         file_accessed(filp);
 881         return 0;
 882 }
 883
 884 struct file_operations btrfs_file_operations = {
 885         .llseek         = generic_file_llseek,
 886         .read           = do_sync_read,
 887         .aio_read       = generic_file_aio_read,
 888         .splice_read    = generic_file_splice_read,
 889 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 890         .sendfile       = generic_file_sendfile,
 891 #endif
 892         .write          = btrfs_file_write,
 893         .mmap           = btrfs_file_mmap,
 894         .open           = generic_file_open,
 895         .fsync          = btrfs_sync_file,
 896         .unlocked_ioctl = btrfs_ioctl,
 897 #ifdef CONFIG_COMPAT
 898         .compat_ioctl   = btrfs_ioctl,
 899 #endif
 900 };
 901