SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 "ordered-data.h"
  38 #include "ioctl.h"
  39 #include "print-tree.h"
  40
  41
  42 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  43                                 struct page **prepared_pages,
  44                                 const char __user * buf)
  45 {
  46         long page_fault = 0;
  47         int i;
  48         int offset = pos & (PAGE_CACHE_SIZE - 1);
  49
  50         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  51                 size_t count = min_t(size_t,
  52                                      PAGE_CACHE_SIZE - offset, write_bytes);
  53                 struct page *page = prepared_pages[i];
  54                 fault_in_pages_readable(buf, count);
  55
  56                 /* Copy data from userspace to the current page */
  57                 kmap(page);
  58                 page_fault = __copy_from_user(page_address(page) + offset,
  59                                               buf, count);
  60                 /* Flush processor's dcache for this page */
  61                 flush_dcache_page(page);
  62                 kunmap(page);
  63                 buf += count;
  64                 write_bytes -= count;
  65
  66                 if (page_fault)
  67                         break;
  68         }
  69         return page_fault ? -EFAULT : 0;
  70 }
  71
  72 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  73 {
  74         size_t i;
  75         for (i = 0; i < num_pages; i++) {
  76                 if (!pages[i])
  77                         break;
  78                 unlock_page(pages[i]);
  79                 mark_page_accessed(pages[i]);
  80                 page_cache_release(pages[i]);
  81         }
  82 }
  83
  84 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
  85                                 struct btrfs_root *root, struct inode *inode,
  86                                 u64 offset, size_t size,
  87                                 struct page **pages, size_t page_offset,
  88                                 int num_pages)
  89 {
  90         struct btrfs_key key;
  91         struct btrfs_path *path;
  92         struct extent_buffer *leaf;
  93         char *kaddr;
  94         unsigned long ptr;
  95         struct btrfs_file_extent_item *ei;
  96         struct page *page;
  97         u32 datasize;
  98         int err = 0;
  99         int ret;
 100         int i;
 101         ssize_t cur_size;
 102
 103         path = btrfs_alloc_path();
 104         if (!path)
 105                 return -ENOMEM;
 106
 107         btrfs_set_trans_block_group(trans, inode);
 108
 109         key.objectid = inode->i_ino;
 110         key.offset = offset;
 111         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 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                 struct btrfs_key found_key;
 120
 121                 if (path->slots[0] == 0)
 122                         goto insert;
 123
 124                 path->slots[0]--;
 125                 leaf = path->nodes[0];
 126                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 127
 128                 if (found_key.objectid != inode->i_ino)
 129                         goto insert;
 130
 131                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 132                         goto insert;
 133                 ei = btrfs_item_ptr(leaf, path->slots[0],
 134                                     struct btrfs_file_extent_item);
 135
 136                 if (btrfs_file_extent_type(leaf, ei) !=
 137                     BTRFS_FILE_EXTENT_INLINE) {
 138                         goto insert;
 139                 }
 140                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 141                 ret = 0;
 142         }
 143         if (ret == 0) {
 144                 u32 found_size;
 145                 u64 found_end;
 146
 147                 leaf = path->nodes[0];
 148                 ei = btrfs_item_ptr(leaf, path->slots[0],
 149                                     struct btrfs_file_extent_item);
 150
 151                 if (btrfs_file_extent_type(leaf, ei) !=
 152                     BTRFS_FILE_EXTENT_INLINE) {
 153                         err = ret;
 154                         btrfs_print_leaf(root, leaf);
 155                         printk("found wasn't inline offset %Lu inode %lu\n",
 156                                offset, inode->i_ino);
 157                         goto fail;
 158                 }
 159                 found_size = btrfs_file_extent_inline_len(leaf,
 160                                           btrfs_item_nr(leaf, path->slots[0]));
 161                 found_end = key.offset + found_size;
 162
 163                 if (found_end < offset + size) {
 164                         btrfs_release_path(root, path);
 165                         ret = btrfs_search_slot(trans, root, &key, path,
 166                                                 offset + size - found_end, 1);
 167                         BUG_ON(ret != 0);
 168
 169                         ret = btrfs_extend_item(trans, root, path,
 170                                                 offset + size - found_end);
 171                         if (ret) {
 172                                 err = ret;
 173                                 goto fail;
 174                         }
 175                         leaf = path->nodes[0];
 176                         ei = btrfs_item_ptr(leaf, path->slots[0],
 177                                             struct btrfs_file_extent_item);
 178                 }
 179                 if (found_end < offset) {
 180                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 181                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 182                 }
 183         } else {
 184 insert:
 185                 btrfs_release_path(root, path);
 186                 datasize = offset + size - key.offset;
 187                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 188                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 189                                               datasize);
 190                 if (ret) {
 191                         err = ret;
 192                         printk("got bad ret %d\n", ret);
 193                         goto fail;
 194                 }
 195                 leaf = path->nodes[0];
 196                 ei = btrfs_item_ptr(leaf, path->slots[0],
 197                                     struct btrfs_file_extent_item);
 198                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 199                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 200         }
 201         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 202
 203         cur_size = size;
 204         i = 0;
 205         while (size > 0) {
 206                 page = pages[i];
 207                 kaddr = kmap_atomic(page, KM_USER0);
 208                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 209                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 210                 kunmap_atomic(kaddr, KM_USER0);
 211                 page_offset = 0;
 212                 ptr += cur_size;
 213                 size -= cur_size;
 214                 if (i >= num_pages) {
 215                         printk("i %d num_pages %d\n", i, num_pages);
 216                 }
 217                 i++;
 218         }
 219         btrfs_mark_buffer_dirty(leaf);
 220 fail:
 221         btrfs_free_path(path);
 222         return err;
 223 }
 224
 225 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 226                                    struct btrfs_root *root,
 227                                    struct file *file,
 228                                    struct page **pages,
 229                                    size_t num_pages,
 230                                    loff_t pos,
 231                                    size_t write_bytes)
 232 {
 233         int err = 0;
 234         int i;
 235         struct inode *inode = fdentry(file)->d_inode;
 236         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_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
 245         start_pos = pos & ~((u64)root->sectorsize - 1);
 246         num_bytes = (write_bytes + pos - start_pos +
 247                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 248
 249         end_of_last_block = start_pos + num_bytes - 1;
 250
 251         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 252         mutex_lock(&root->fs_info->fs_mutex);
 253         trans = btrfs_start_transaction(root, 1);
 254         if (!trans) {
 255                 err = -ENOMEM;
 256                 goto out_unlock;
 257         }
 258         btrfs_set_trans_block_group(trans, inode);
 259         inode->i_blocks += num_bytes >> 9;
 260         hint_byte = 0;
 261
 262         if ((end_of_last_block & 4095) == 0) {
 263                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 264         }
 265         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 266
 267         /* FIXME...EIEIO, ENOSPC and more */
 268
 269         /* insert any holes we need to create */
 270         if (inode->i_size < start_pos) {
 271                 u64 last_pos_in_file;
 272                 u64 hole_size;
 273                 u64 mask = root->sectorsize - 1;
 274                 last_pos_in_file = (isize + mask) & ~mask;
 275                 hole_size = (end_pos - last_pos_in_file + mask) & ~mask;
 276
 277                 if (last_pos_in_file < start_pos) {
 278                         err = btrfs_drop_extents(trans, root, inode,
 279                                                  last_pos_in_file,
 280                                                  last_pos_in_file + hole_size,
 281                                                  last_pos_in_file,
 282                                                  &hint_byte);
 283                         if (err)
 284                                 goto failed;
 285
 286                         err = btrfs_insert_file_extent(trans, root,
 287                                                        inode->i_ino,
 288                                                        last_pos_in_file,
 289                                                        0, 0, hole_size);
 290                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 291                                         last_pos_in_file + hole_size -1);
 292                         btrfs_check_file(root, inode);
 293                 }
 294                 if (err)
 295                         goto failed;
 296         }
 297
 298         /*
 299          * either allocate an extent for the new bytes or setup the key
 300          * to show we are doing inline data in the extent
 301          */
 302         inline_size = end_pos;
 303         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 304             inline_size > root->fs_info->max_inline ||
 305             (inline_size & (root->sectorsize -1)) == 0 ||
 306             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 307                 u64 last_end;
 308                 u64 existing_delalloc = 0;
 309
 310                 for (i = 0; i < num_pages; i++) {
 311                         struct page *p = pages[i];
 312                         SetPageUptodate(p);
 313                         set_page_dirty(p);
 314                 }
 315                 last_end = (u64)(pages[num_pages -1]->index) <<
 316                                 PAGE_CACHE_SHIFT;
 317                 last_end += PAGE_CACHE_SIZE - 1;
 318                 if (start_pos < isize) {
 319                         u64 delalloc_start = start_pos;
 320                         existing_delalloc = count_range_bits(io_tree,
 321                                              &delalloc_start,
 322                                              end_of_last_block, (u64)-1,
 323                                              EXTENT_DELALLOC);
 324                 }
 325                 set_extent_delalloc(io_tree, start_pos, end_of_last_block,
 326                                  GFP_NOFS);
 327                 btrfs_add_ordered_inode(inode);
 328         } else {
 329                 u64 aligned_end;
 330                 /* step one, delete the existing extents in this range */
 331                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 332                         ~((u64)root->sectorsize - 1);
 333                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 334                                          aligned_end, aligned_end, &hint_byte);
 335                 if (err)
 336                         goto failed;
 337                 if (isize > inline_size)
 338                         inline_size = min_t(u64, isize, aligned_end);
 339                 inline_size -= start_pos;
 340                 err = insert_inline_extent(trans, root, inode, start_pos,
 341                                            inline_size, pages, 0, num_pages);
 342                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 343                 BUG_ON(err);
 344         }
 345         if (end_pos > isize) {
 346                 i_size_write(inode, end_pos);
 347                 btrfs_update_inode(trans, root, inode);
 348         }
 349 failed:
 350         err = btrfs_end_transaction(trans, root);
 351 out_unlock:
 352         mutex_unlock(&root->fs_info->fs_mutex);
 353         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 354         return err;
 355 }
 356
 357 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 358 {
 359         struct extent_map *em;
 360         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 361
 362         while(1) {
 363                 spin_lock(&em_tree->lock);
 364                 em = lookup_extent_mapping(em_tree, start, end);
 365                 if (!em) {
 366                         spin_unlock(&em_tree->lock);
 367                         break;
 368                 }
 369                 remove_extent_mapping(em_tree, em);
 370                 spin_unlock(&em_tree->lock);
 371
 372                 /* once for us */
 373                 free_extent_map(em);
 374                 /* once for the tree*/
 375                 free_extent_map(em);
 376         }
 377         return 0;
 378 }
 379
 380 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 381 {
 382         return 0;
 383 #if 0
 384         struct btrfs_path *path;
 385         struct btrfs_key found_key;
 386         struct extent_buffer *leaf;
 387         struct btrfs_file_extent_item *extent;
 388         u64 last_offset = 0;
 389         int nritems;
 390         int slot;
 391         int found_type;
 392         int ret;
 393         int err = 0;
 394         u64 extent_end = 0;
 395
 396         path = btrfs_alloc_path();
 397         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 398                                        last_offset, 0);
 399         while(1) {
 400                 nritems = btrfs_header_nritems(path->nodes[0]);
 401                 if (path->slots[0] >= nritems) {
 402                         ret = btrfs_next_leaf(root, path);
 403                         if (ret)
 404                                 goto out;
 405                         nritems = btrfs_header_nritems(path->nodes[0]);
 406                 }
 407                 slot = path->slots[0];
 408                 leaf = path->nodes[0];
 409                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 410                 if (found_key.objectid != inode->i_ino)
 411                         break;
 412                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 413                         goto out;
 414
 415                 if (found_key.offset != last_offset) {
 416                         WARN_ON(1);
 417                         btrfs_print_leaf(root, leaf);
 418                         printk("inode %lu found offset %Lu expected %Lu\n",
 419                                inode->i_ino, found_key.offset, last_offset);
 420                         err = 1;
 421                         goto out;
 422                 }
 423                 extent = btrfs_item_ptr(leaf, slot,
 424                                         struct btrfs_file_extent_item);
 425                 found_type = btrfs_file_extent_type(leaf, extent);
 426                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 427                         extent_end = found_key.offset +
 428                              btrfs_file_extent_num_bytes(leaf, extent);
 429                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 430                         struct btrfs_item *item;
 431                         item = btrfs_item_nr(leaf, slot);
 432                         extent_end = found_key.offset +
 433                              btrfs_file_extent_inline_len(leaf, item);
 434                         extent_end = (extent_end + root->sectorsize - 1) &
 435                                 ~((u64)root->sectorsize -1 );
 436                 }
 437                 last_offset = extent_end;
 438                 path->slots[0]++;
 439         }
 440         if (last_offset < inode->i_size) {
 441                 WARN_ON(1);
 442                 btrfs_print_leaf(root, leaf);
 443                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 444                        last_offset, inode->i_size);
 445                 err = 1;
 446
 447         }
 448 out:
 449         btrfs_free_path(path);
 450         return err;
 451 #endif
 452 }
 453
 454 /*
 455  * this is very complex, but the basic idea is to drop all extents
 456  * in the range start - end.  hint_block is filled in with a block number
 457  * that would be a good hint to the block allocator for this file.
 458  *
 459  * If an extent intersects the range but is not entirely inside the range
 460  * it is either truncated or split.  Anything entirely inside the range
 461  * is deleted from the tree.
 462  */
 463 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 464                        struct btrfs_root *root, struct inode *inode,
 465                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 466 {
 467         u64 extent_end = 0;
 468         u64 search_start = start;
 469         struct extent_buffer *leaf;
 470         struct btrfs_file_extent_item *extent;
 471         struct btrfs_path *path;
 472         struct btrfs_key key;
 473         struct btrfs_file_extent_item old;
 474         int keep;
 475         int slot;
 476         int bookend;
 477         int found_type;
 478         int found_extent;
 479         int found_inline;
 480         int recow;
 481         int ret;
 482
 483         btrfs_drop_extent_cache(inode, start, end - 1);
 484
 485         path = btrfs_alloc_path();
 486         if (!path)
 487                 return -ENOMEM;
 488         while(1) {
 489                 recow = 0;
 490                 btrfs_release_path(root, path);
 491                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 492                                                search_start, -1);
 493                 if (ret < 0)
 494                         goto out;
 495                 if (ret > 0) {
 496                         if (path->slots[0] == 0) {
 497                                 ret = 0;
 498                                 goto out;
 499                         }
 500                         path->slots[0]--;
 501                 }
 502 next_slot:
 503                 keep = 0;
 504                 bookend = 0;
 505                 found_extent = 0;
 506                 found_inline = 0;
 507                 extent = NULL;
 508                 leaf = path->nodes[0];
 509                 slot = path->slots[0];
 510                 ret = 0;
 511                 btrfs_item_key_to_cpu(leaf, &key, slot);
 512
 513                 if (key.offset >= end || key.objectid != inode->i_ino) {
 514                         goto out;
 515                 }
 516                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 517                         goto out;
 518                 }
 519                 if (recow) {
 520                         search_start = key.offset;
 521                         continue;
 522                 }
 523                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 524                         extent = btrfs_item_ptr(leaf, slot,
 525                                                 struct btrfs_file_extent_item);
 526                         found_type = btrfs_file_extent_type(leaf, extent);
 527                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 528                                 extent_end =
 529                                      btrfs_file_extent_disk_bytenr(leaf,
 530                                                                    extent);
 531                                 if (extent_end)
 532                                         *hint_byte = extent_end;
 533
 534                                 extent_end = key.offset +
 535                                      btrfs_file_extent_num_bytes(leaf, extent);
 536                                 found_extent = 1;
 537                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 538                                 struct btrfs_item *item;
 539                                 item = btrfs_item_nr(leaf, slot);
 540                                 found_inline = 1;
 541                                 extent_end = key.offset +
 542                                      btrfs_file_extent_inline_len(leaf, item);
 543                         }
 544                 } else {
 545                         extent_end = search_start;
 546                 }
 547
 548                 /* we found nothing we can drop */
 549                 if ((!found_extent && !found_inline) ||
 550                     search_start >= extent_end) {
 551                         int nextret;
 552                         u32 nritems;
 553                         nritems = btrfs_header_nritems(leaf);
 554                         if (slot >= nritems - 1) {
 555                                 nextret = btrfs_next_leaf(root, path);
 556                                 if (nextret)
 557                                         goto out;
 558                                 recow = 1;
 559                         } else {
 560                                 path->slots[0]++;
 561                         }
 562                         goto next_slot;
 563                 }
 564
 565                 if (found_inline) {
 566                         u64 mask = root->sectorsize - 1;
 567                         search_start = (extent_end + mask) & ~mask;
 568                 } else
 569                         search_start = extent_end;
 570                 if (end <= extent_end && start >= key.offset && found_inline) {
 571                         *hint_byte = EXTENT_MAP_INLINE;
 572                         continue;
 573                 }
 574                 if (end < extent_end && end >= key.offset) {
 575                         if (found_extent) {
 576                                 u64 disk_bytenr =
 577                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 578                                 u64 disk_num_bytes =
 579                                     btrfs_file_extent_disk_num_bytes(leaf,
 580                                                                       extent);
 581                                 read_extent_buffer(leaf, &old,
 582                                                    (unsigned long)extent,
 583                                                    sizeof(old));
 584                                 if (disk_bytenr != 0) {
 585                                         ret = btrfs_inc_extent_ref(trans, root,
 586                                                  disk_bytenr, disk_num_bytes,
 587                                                  root->root_key.objectid,
 588                                                  trans->transid,
 589                                                  key.objectid, end);
 590                                         BUG_ON(ret);
 591                                 }
 592                         }
 593                         bookend = 1;
 594                         if (found_inline && start <= key.offset &&
 595                             inline_limit < extent_end)
 596                                 keep = 1;
 597                 }
 598                 /* truncate existing extent */
 599                 if (start > key.offset) {
 600                         u64 new_num;
 601                         u64 old_num;
 602                         keep = 1;
 603                         WARN_ON(start & (root->sectorsize - 1));
 604                         if (found_extent) {
 605                                 new_num = start - key.offset;
 606                                 old_num = btrfs_file_extent_num_bytes(leaf,
 607                                                                       extent);
 608                                 *hint_byte =
 609                                         btrfs_file_extent_disk_bytenr(leaf,
 610                                                                       extent);
 611                                 if (btrfs_file_extent_disk_bytenr(leaf,
 612                                                                   extent)) {
 613                                         inode->i_blocks -=
 614                                                 (old_num - new_num) >> 9;
 615                                 }
 616                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 617                                                                 new_num);
 618                                 btrfs_mark_buffer_dirty(leaf);
 619                         } else if (key.offset < inline_limit &&
 620                                    (end > extent_end) &&
 621                                    (inline_limit < extent_end)) {
 622                                 u32 new_size;
 623                                 new_size = btrfs_file_extent_calc_inline_size(
 624                                                    inline_limit - key.offset);
 625                                 btrfs_truncate_item(trans, root, path,
 626                                                     new_size, 1);
 627                         }
 628                 }
 629                 /* delete the entire extent */
 630                 if (!keep) {
 631                         u64 disk_bytenr = 0;
 632                         u64 disk_num_bytes = 0;
 633                         u64 extent_num_bytes = 0;
 634                         u64 root_gen;
 635                         u64 root_owner;
 636
 637                         root_gen = btrfs_header_generation(leaf);
 638                         root_owner = btrfs_header_owner(leaf);
 639                         if (found_extent) {
 640                                 disk_bytenr =
 641                                       btrfs_file_extent_disk_bytenr(leaf,
 642                                                                      extent);
 643                                 disk_num_bytes =
 644                                       btrfs_file_extent_disk_num_bytes(leaf,
 645                                                                        extent);
 646                                 extent_num_bytes =
 647                                       btrfs_file_extent_num_bytes(leaf, extent);
 648                                 *hint_byte =
 649                                         btrfs_file_extent_disk_bytenr(leaf,
 650                                                                       extent);
 651                         }
 652                         ret = btrfs_del_item(trans, root, path);
 653                         /* TODO update progress marker and return */
 654                         BUG_ON(ret);
 655                         btrfs_release_path(root, path);
 656                         extent = NULL;
 657                         if (found_extent && disk_bytenr != 0) {
 658                                 inode->i_blocks -= extent_num_bytes >> 9;
 659                                 ret = btrfs_free_extent(trans, root,
 660                                                 disk_bytenr,
 661                                                 disk_num_bytes,
 662                                                 root_owner,
 663                                                 root_gen, inode->i_ino,
 664                                                 key.offset, 0);
 665                         }
 666
 667                         BUG_ON(ret);
 668                         if (!bookend && search_start >= end) {
 669                                 ret = 0;
 670                                 goto out;
 671                         }
 672                         if (!bookend)
 673                                 continue;
 674                 }
 675                 if (bookend && found_inline && start <= key.offset &&
 676                     inline_limit < extent_end && key.offset <= inline_limit) {
 677                         u32 new_size;
 678                         new_size = btrfs_file_extent_calc_inline_size(
 679                                                    extent_end - inline_limit);
 680                         btrfs_truncate_item(trans, root, path, new_size, 0);
 681                 }
 682                 /* create bookend, splitting the extent in two */
 683                 if (bookend && found_extent) {
 684                         struct btrfs_key ins;
 685                         ins.objectid = inode->i_ino;
 686                         ins.offset = end;
 687                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 688                         btrfs_release_path(root, path);
 689                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 690                                                       sizeof(*extent));
 691
 692                         leaf = path->nodes[0];
 693                         if (ret) {
 694                                 btrfs_print_leaf(root, leaf);
 695                                 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);
 696                         }
 697                         BUG_ON(ret);
 698                         extent = btrfs_item_ptr(leaf, path->slots[0],
 699                                                 struct btrfs_file_extent_item);
 700                         write_extent_buffer(leaf, &old,
 701                                             (unsigned long)extent, sizeof(old));
 702
 703                         btrfs_set_file_extent_offset(leaf, extent,
 704                                     le64_to_cpu(old.offset) + end - key.offset);
 705                         WARN_ON(le64_to_cpu(old.num_bytes) <
 706                                 (extent_end - end));
 707                         btrfs_set_file_extent_num_bytes(leaf, extent,
 708                                                         extent_end - end);
 709                         btrfs_set_file_extent_type(leaf, extent,
 710                                                    BTRFS_FILE_EXTENT_REG);
 711
 712                         btrfs_mark_buffer_dirty(path->nodes[0]);
 713                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 714                                 inode->i_blocks +=
 715                                       btrfs_file_extent_num_bytes(leaf,
 716                                                                   extent) >> 9;
 717                         }
 718                         ret = 0;
 719                         goto out;
 720                 }
 721         }
 722 out:
 723         btrfs_free_path(path);
 724         return ret;
 725 }
 726
 727 /*
 728  * this gets pages into the page cache and locks them down
 729  */
 730 static int prepare_pages(struct btrfs_root *root, struct file *file,
 731                          struct page **pages, size_t num_pages,
 732                          loff_t pos, unsigned long first_index,
 733                          unsigned long last_index, size_t write_bytes)
 734 {
 735         int i;
 736         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 737         struct inode *inode = fdentry(file)->d_inode;
 738         int err = 0;
 739         u64 start_pos;
 740
 741         start_pos = pos & ~((u64)root->sectorsize - 1);
 742
 743         memset(pages, 0, num_pages * sizeof(struct page *));
 744
 745         for (i = 0; i < num_pages; i++) {
 746                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 747                 if (!pages[i]) {
 748                         err = -ENOMEM;
 749                         BUG_ON(1);
 750                 }
 751 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 752                 ClearPageDirty(pages[i]);
 753 #else
 754                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 755 #endif
 756                 wait_on_page_writeback(pages[i]);
 757                 set_page_extent_mapped(pages[i]);
 758                 WARN_ON(!PageLocked(pages[i]));
 759         }
 760         return 0;
 761 }
 762
 763 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 764                                 size_t count, loff_t *ppos)
 765 {
 766         loff_t pos;
 767         loff_t start_pos;
 768         ssize_t num_written = 0;
 769         ssize_t err = 0;
 770         int ret = 0;
 771         struct inode *inode = fdentry(file)->d_inode;
 772         struct btrfs_root *root = BTRFS_I(inode)->root;
 773         struct page **pages = NULL;
 774         int nrptrs;
 775         struct page *pinned[2];
 776         unsigned long first_index;
 777         unsigned long last_index;
 778
 779         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 780                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 781         pinned[0] = NULL;
 782         pinned[1] = NULL;
 783         if (file->f_flags & O_DIRECT)
 784                 return -EINVAL;
 785
 786         pos = *ppos;
 787         start_pos = pos;
 788
 789         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 790         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 791         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 792         if (err)
 793                 goto out_nolock;
 794         if (count == 0)
 795                 goto out_nolock;
 796         err = remove_suid(fdentry(file));
 797         if (err)
 798                 goto out_nolock;
 799         file_update_time(file);
 800
 801         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 802
 803         mutex_lock(&inode->i_mutex);
 804         first_index = pos >> PAGE_CACHE_SHIFT;
 805         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 806
 807         /*
 808          * there are lots of better ways to do this, but this code
 809          * makes sure the first and last page in the file range are
 810          * up to date and ready for cow
 811          */
 812         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 813                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 814                 if (!PageUptodate(pinned[0])) {
 815                         ret = btrfs_readpage(NULL, pinned[0]);
 816                         BUG_ON(ret);
 817                         wait_on_page_locked(pinned[0]);
 818                 } else {
 819                         unlock_page(pinned[0]);
 820                 }
 821         }
 822         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 823                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 824                 if (!PageUptodate(pinned[1])) {
 825                         ret = btrfs_readpage(NULL, pinned[1]);
 826                         BUG_ON(ret);
 827                         wait_on_page_locked(pinned[1]);
 828                 } else {
 829                         unlock_page(pinned[1]);
 830                 }
 831         }
 832
 833         while(count > 0) {
 834                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 835                 size_t write_bytes = min(count, nrptrs *
 836                                         (size_t)PAGE_CACHE_SIZE -
 837                                          offset);
 838                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 839                                         PAGE_CACHE_SHIFT;
 840
 841                 WARN_ON(num_pages > nrptrs);
 842                 memset(pages, 0, sizeof(pages));
 843
 844                 mutex_lock(&root->fs_info->fs_mutex);
 845                 ret = btrfs_check_free_space(root, write_bytes, 0);
 846                 mutex_unlock(&root->fs_info->fs_mutex);
 847                 if (ret)
 848                         goto out;
 849
 850                 ret = prepare_pages(root, file, pages, num_pages,
 851                                     pos, first_index, last_index,
 852                                     write_bytes);
 853                 if (ret)
 854                         goto out;
 855
 856                 ret = btrfs_copy_from_user(pos, num_pages,
 857                                            write_bytes, pages, buf);
 858                 if (ret) {
 859                         btrfs_drop_pages(pages, num_pages);
 860                         goto out;
 861                 }
 862
 863                 ret = dirty_and_release_pages(NULL, root, file, pages,
 864                                               num_pages, pos, write_bytes);
 865                 btrfs_drop_pages(pages, num_pages);
 866                 if (ret)
 867                         goto out;
 868
 869                 buf += write_bytes;
 870                 count -= write_bytes;
 871                 pos += write_bytes;
 872                 num_written += write_bytes;
 873
 874                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 875                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 876                         btrfs_btree_balance_dirty(root, 1);
 877                 btrfs_throttle(root);
 878                 cond_resched();
 879         }
 880 out:
 881         mutex_unlock(&inode->i_mutex);
 882
 883 out_nolock:
 884         kfree(pages);
 885         if (pinned[0])
 886                 page_cache_release(pinned[0]);
 887         if (pinned[1])
 888                 page_cache_release(pinned[1]);
 889         *ppos = pos;
 890
 891         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 892                 err = sync_page_range(inode, inode->i_mapping,
 893                                       start_pos, num_written);
 894                 if (err < 0)
 895                         num_written = err;
 896         }
 897         current->backing_dev_info = NULL;
 898         return num_written ? num_written : err;
 899 }
 900
 901 static int btrfs_sync_file(struct file *file,
 902                            struct dentry *dentry, int datasync)
 903 {
 904         struct inode *inode = dentry->d_inode;
 905         struct btrfs_root *root = BTRFS_I(inode)->root;
 906         int ret = 0;
 907         struct btrfs_trans_handle *trans;
 908
 909         /*
 910          * check the transaction that last modified this inode
 911          * and see if its already been committed
 912          */
 913         mutex_lock(&root->fs_info->fs_mutex);
 914         if (!BTRFS_I(inode)->last_trans)
 915                 goto out;
 916         mutex_lock(&root->fs_info->trans_mutex);
 917         if (BTRFS_I(inode)->last_trans <=
 918             root->fs_info->last_trans_committed) {
 919                 BTRFS_I(inode)->last_trans = 0;
 920                 mutex_unlock(&root->fs_info->trans_mutex);
 921                 goto out;
 922         }
 923         mutex_unlock(&root->fs_info->trans_mutex);
 924
 925         /*
 926          * ok we haven't committed the transaction yet, lets do a commit
 927          */
 928         trans = btrfs_start_transaction(root, 1);
 929         if (!trans) {
 930                 ret = -ENOMEM;
 931                 goto out;
 932         }
 933         ret = btrfs_commit_transaction(trans, root);
 934 out:
 935         mutex_unlock(&root->fs_info->fs_mutex);
 936         return ret > 0 ? EIO : ret;
 937 }
 938
 939 static struct vm_operations_struct btrfs_file_vm_ops = {
 940 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 941         .nopage         = filemap_nopage,
 942         .populate       = filemap_populate,
 943 #else
 944         .fault          = filemap_fault,
 945 #endif
 946         .page_mkwrite   = btrfs_page_mkwrite,
 947 };
 948
 949 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 950 {
 951         vma->vm_ops = &btrfs_file_vm_ops;
 952         file_accessed(filp);
 953         return 0;
 954 }
 955
 956 struct file_operations btrfs_file_operations = {
 957         .llseek         = generic_file_llseek,
 958         .read           = do_sync_read,
 959         .aio_read       = generic_file_aio_read,
 960         .splice_read    = generic_file_splice_read,
 961 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 962         .sendfile       = generic_file_sendfile,
 963 #endif
 964         .write          = btrfs_file_write,
 965         .mmap           = btrfs_file_mmap,
 966         .open           = generic_file_open,
 967         .fsync          = btrfs_sync_file,
 968         .unlocked_ioctl = btrfs_ioctl,
 969 #ifdef CONFIG_COMPAT
 970         .compat_ioctl   = btrfs_ioctl,
 971 #endif
 972 };
 973