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         /* insert any holes we need to create */
 269         if (isize < end_pos) {
 270                 u64 last_pos_in_file;
 271                 u64 hole_size;
 272                 u64 mask = root->sectorsize - 1;
 273                 last_pos_in_file = (isize + mask) & ~mask;
 274                 hole_size = (end_pos - last_pos_in_file + mask) & ~mask;
 275                 if (last_pos_in_file < end_pos) {
 276                         err = btrfs_drop_extents(trans, root, inode,
 277                                                  last_pos_in_file,
 278                                                  last_pos_in_file + hole_size,
 279                                                  last_pos_in_file,
 280                                                  &hint_byte);
 281                         if (err)
 282                                 goto failed;
 283
 284                         err = btrfs_insert_file_extent(trans, root,
 285                                                        inode->i_ino,
 286                                                        last_pos_in_file,
 287                                                        0, 0, hole_size);
 288                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 289                                         last_pos_in_file + hole_size -1);
 290                         btrfs_check_file(root, inode);
 291                 }
 292                 if (err)
 293                         goto failed;
 294         }
 295
 296         /*
 297          * either allocate an extent for the new bytes or setup the key
 298          * to show we are doing inline data in the extent
 299          */
 300         inline_size = end_pos;
 301         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 302             inline_size > root->fs_info->max_inline ||
 303             (inline_size & (root->sectorsize -1)) == 0 ||
 304             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 305                 u64 last_end;
 306                 u64 existing_delalloc = 0;
 307
 308                 for (i = 0; i < num_pages; i++) {
 309                         struct page *p = pages[i];
 310                         SetPageUptodate(p);
 311                         set_page_dirty(p);
 312                 }
 313                 last_end = (u64)(pages[num_pages -1]->index) <<
 314                                 PAGE_CACHE_SHIFT;
 315                 last_end += PAGE_CACHE_SIZE - 1;
 316                 if (start_pos < isize) {
 317                         u64 delalloc_start = start_pos;
 318                         existing_delalloc = count_range_bits(io_tree,
 319                                              &delalloc_start,
 320                                              end_of_last_block, (u64)-1,
 321                                              EXTENT_DELALLOC);
 322                 }
 323                 set_extent_delalloc(io_tree, start_pos, end_of_last_block,
 324                                  GFP_NOFS);
 325                 btrfs_add_ordered_inode(inode);
 326         } else {
 327                 u64 aligned_end;
 328                 /* step one, delete the existing extents in this range */
 329                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 330                         ~((u64)root->sectorsize - 1);
 331                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 332                                          aligned_end, aligned_end, &hint_byte);
 333                 if (err)
 334                         goto failed;
 335                 if (isize > inline_size)
 336                         inline_size = min_t(u64, isize, aligned_end);
 337                 inline_size -= start_pos;
 338                 err = insert_inline_extent(trans, root, inode, start_pos,
 339                                            inline_size, pages, 0, num_pages);
 340                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 341                 BUG_ON(err);
 342         }
 343         if (end_pos > isize) {
 344                 i_size_write(inode, end_pos);
 345                 btrfs_update_inode(trans, root, inode);
 346         }
 347 failed:
 348         err = btrfs_end_transaction(trans, root);
 349 out_unlock:
 350         mutex_unlock(&root->fs_info->fs_mutex);
 351         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 352         return err;
 353 }
 354
 355 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 356 {
 357         struct extent_map *em;
 358         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 359
 360         while(1) {
 361                 spin_lock(&em_tree->lock);
 362                 em = lookup_extent_mapping(em_tree, start, end);
 363                 if (!em) {
 364                         spin_unlock(&em_tree->lock);
 365                         break;
 366                 }
 367                 remove_extent_mapping(em_tree, em);
 368                 spin_unlock(&em_tree->lock);
 369
 370                 /* once for us */
 371                 free_extent_map(em);
 372                 /* once for the tree*/
 373                 free_extent_map(em);
 374         }
 375         return 0;
 376 }
 377
 378 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 379 {
 380         return 0;
 381 #if 0
 382         struct btrfs_path *path;
 383         struct btrfs_key found_key;
 384         struct extent_buffer *leaf;
 385         struct btrfs_file_extent_item *extent;
 386         u64 last_offset = 0;
 387         int nritems;
 388         int slot;
 389         int found_type;
 390         int ret;
 391         int err = 0;
 392         u64 extent_end = 0;
 393
 394         path = btrfs_alloc_path();
 395         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 396                                        last_offset, 0);
 397         while(1) {
 398                 nritems = btrfs_header_nritems(path->nodes[0]);
 399                 if (path->slots[0] >= nritems) {
 400                         ret = btrfs_next_leaf(root, path);
 401                         if (ret)
 402                                 goto out;
 403                         nritems = btrfs_header_nritems(path->nodes[0]);
 404                 }
 405                 slot = path->slots[0];
 406                 leaf = path->nodes[0];
 407                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 408                 if (found_key.objectid != inode->i_ino)
 409                         break;
 410                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 411                         goto out;
 412
 413                 if (found_key.offset != last_offset) {
 414                         WARN_ON(1);
 415                         btrfs_print_leaf(root, leaf);
 416                         printk("inode %lu found offset %Lu expected %Lu\n",
 417                                inode->i_ino, found_key.offset, last_offset);
 418                         err = 1;
 419                         goto out;
 420                 }
 421                 extent = btrfs_item_ptr(leaf, slot,
 422                                         struct btrfs_file_extent_item);
 423                 found_type = btrfs_file_extent_type(leaf, extent);
 424                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 425                         extent_end = found_key.offset +
 426                              btrfs_file_extent_num_bytes(leaf, extent);
 427                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 428                         struct btrfs_item *item;
 429                         item = btrfs_item_nr(leaf, slot);
 430                         extent_end = found_key.offset +
 431                              btrfs_file_extent_inline_len(leaf, item);
 432                         extent_end = (extent_end + root->sectorsize - 1) &
 433                                 ~((u64)root->sectorsize -1 );
 434                 }
 435                 last_offset = extent_end;
 436                 path->slots[0]++;
 437         }
 438         if (last_offset < inode->i_size) {
 439                 WARN_ON(1);
 440                 btrfs_print_leaf(root, leaf);
 441                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 442                        last_offset, inode->i_size);
 443                 err = 1;
 444
 445         }
 446 out:
 447         btrfs_free_path(path);
 448         return err;
 449 #endif
 450 }
 451
 452 /*
 453  * this is very complex, but the basic idea is to drop all extents
 454  * in the range start - end.  hint_block is filled in with a block number
 455  * that would be a good hint to the block allocator for this file.
 456  *
 457  * If an extent intersects the range but is not entirely inside the range
 458  * it is either truncated or split.  Anything entirely inside the range
 459  * is deleted from the tree.
 460  */
 461 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 462                        struct btrfs_root *root, struct inode *inode,
 463                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 464 {
 465         u64 extent_end = 0;
 466         u64 search_start = start;
 467         struct extent_buffer *leaf;
 468         struct btrfs_file_extent_item *extent;
 469         struct btrfs_path *path;
 470         struct btrfs_key key;
 471         struct btrfs_file_extent_item old;
 472         int keep;
 473         int slot;
 474         int bookend;
 475         int found_type;
 476         int found_extent;
 477         int found_inline;
 478         int recow;
 479         int ret;
 480
 481         btrfs_drop_extent_cache(inode, start, end - 1);
 482
 483         path = btrfs_alloc_path();
 484         if (!path)
 485                 return -ENOMEM;
 486         while(1) {
 487                 recow = 0;
 488                 btrfs_release_path(root, path);
 489                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 490                                                search_start, -1);
 491                 if (ret < 0)
 492                         goto out;
 493                 if (ret > 0) {
 494                         if (path->slots[0] == 0) {
 495                                 ret = 0;
 496                                 goto out;
 497                         }
 498                         path->slots[0]--;
 499                 }
 500 next_slot:
 501                 keep = 0;
 502                 bookend = 0;
 503                 found_extent = 0;
 504                 found_inline = 0;
 505                 extent = NULL;
 506                 leaf = path->nodes[0];
 507                 slot = path->slots[0];
 508                 ret = 0;
 509                 btrfs_item_key_to_cpu(leaf, &key, slot);
 510
 511                 if (key.offset >= end || key.objectid != inode->i_ino) {
 512                         goto out;
 513                 }
 514                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 515                         goto out;
 516                 }
 517                 if (recow) {
 518                         search_start = key.offset;
 519                         continue;
 520                 }
 521                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 522                         extent = btrfs_item_ptr(leaf, slot,
 523                                                 struct btrfs_file_extent_item);
 524                         found_type = btrfs_file_extent_type(leaf, extent);
 525                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 526                                 extent_end =
 527                                      btrfs_file_extent_disk_bytenr(leaf,
 528                                                                    extent);
 529                                 if (extent_end)
 530                                         *hint_byte = extent_end;
 531
 532                                 extent_end = key.offset +
 533                                      btrfs_file_extent_num_bytes(leaf, extent);
 534                                 found_extent = 1;
 535                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 536                                 struct btrfs_item *item;
 537                                 item = btrfs_item_nr(leaf, slot);
 538                                 found_inline = 1;
 539                                 extent_end = key.offset +
 540                                      btrfs_file_extent_inline_len(leaf, item);
 541                         }
 542                 } else {
 543                         extent_end = search_start;
 544                 }
 545
 546                 /* we found nothing we can drop */
 547                 if ((!found_extent && !found_inline) ||
 548                     search_start >= extent_end) {
 549                         int nextret;
 550                         u32 nritems;
 551                         nritems = btrfs_header_nritems(leaf);
 552                         if (slot >= nritems - 1) {
 553                                 nextret = btrfs_next_leaf(root, path);
 554                                 if (nextret)
 555                                         goto out;
 556                                 recow = 1;
 557                         } else {
 558                                 path->slots[0]++;
 559                         }
 560                         goto next_slot;
 561                 }
 562
 563                 if (found_inline) {
 564                         u64 mask = root->sectorsize - 1;
 565                         search_start = (extent_end + mask) & ~mask;
 566                 } else
 567                         search_start = extent_end;
 568                 if (end <= extent_end && start >= key.offset && found_inline) {
 569                         *hint_byte = EXTENT_MAP_INLINE;
 570                         continue;
 571                 }
 572                 if (end < extent_end && end >= key.offset) {
 573                         if (found_extent) {
 574                                 u64 disk_bytenr =
 575                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 576                                 u64 disk_num_bytes =
 577                                     btrfs_file_extent_disk_num_bytes(leaf,
 578                                                                       extent);
 579                                 read_extent_buffer(leaf, &old,
 580                                                    (unsigned long)extent,
 581                                                    sizeof(old));
 582                                 if (disk_bytenr != 0) {
 583                                         ret = btrfs_inc_extent_ref(trans, root,
 584                                                  disk_bytenr, disk_num_bytes,
 585                                                  root->root_key.objectid,
 586                                                  trans->transid,
 587                                                  key.objectid, end);
 588                                         BUG_ON(ret);
 589                                 }
 590                         }
 591                         bookend = 1;
 592                         if (found_inline && start <= key.offset)
 593                                 keep = 1;
 594                 }
 595                 /* truncate existing extent */
 596                 if (start > key.offset) {
 597                         u64 new_num;
 598                         u64 old_num;
 599                         keep = 1;
 600                         WARN_ON(start & (root->sectorsize - 1));
 601                         if (found_extent) {
 602                                 new_num = start - key.offset;
 603                                 old_num = btrfs_file_extent_num_bytes(leaf,
 604                                                                       extent);
 605                                 *hint_byte =
 606                                         btrfs_file_extent_disk_bytenr(leaf,
 607                                                                       extent);
 608                                 if (btrfs_file_extent_disk_bytenr(leaf,
 609                                                                   extent)) {
 610                                         inode->i_blocks -=
 611                                                 (old_num - new_num) >> 9;
 612                                 }
 613                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 614                                                                 new_num);
 615                                 btrfs_mark_buffer_dirty(leaf);
 616                         } else if (key.offset < inline_limit &&
 617                                    (end > extent_end) &&
 618                                    (inline_limit < extent_end)) {
 619                                 u32 new_size;
 620                                 new_size = btrfs_file_extent_calc_inline_size(
 621                                                    inline_limit - key.offset);
 622                                 btrfs_truncate_item(trans, root, path,
 623                                                     new_size, 1);
 624                         }
 625                 }
 626                 /* delete the entire extent */
 627                 if (!keep) {
 628                         u64 disk_bytenr = 0;
 629                         u64 disk_num_bytes = 0;
 630                         u64 extent_num_bytes = 0;
 631                         u64 root_gen;
 632                         u64 root_owner;
 633
 634                         root_gen = btrfs_header_generation(leaf);
 635                         root_owner = btrfs_header_owner(leaf);
 636                         if (found_extent) {
 637                                 disk_bytenr =
 638                                       btrfs_file_extent_disk_bytenr(leaf,
 639                                                                      extent);
 640                                 disk_num_bytes =
 641                                       btrfs_file_extent_disk_num_bytes(leaf,
 642                                                                        extent);
 643                                 extent_num_bytes =
 644                                       btrfs_file_extent_num_bytes(leaf, extent);
 645                                 *hint_byte =
 646                                         btrfs_file_extent_disk_bytenr(leaf,
 647                                                                       extent);
 648                         }
 649                         ret = btrfs_del_item(trans, root, path);
 650                         /* TODO update progress marker and return */
 651                         BUG_ON(ret);
 652                         btrfs_release_path(root, path);
 653                         extent = NULL;
 654                         if (found_extent && disk_bytenr != 0) {
 655                                 inode->i_blocks -= extent_num_bytes >> 9;
 656                                 ret = btrfs_free_extent(trans, root,
 657                                                 disk_bytenr,
 658                                                 disk_num_bytes,
 659                                                 root_owner,
 660                                                 root_gen, inode->i_ino,
 661                                                 key.offset, 0);
 662                         }
 663
 664                         BUG_ON(ret);
 665                         if (!bookend && search_start >= end) {
 666                                 ret = 0;
 667                                 goto out;
 668                         }
 669                         if (!bookend)
 670                                 continue;
 671                 }
 672                 if (bookend && found_inline && start <= key.offset) {
 673                         u32 new_size;
 674                         new_size = btrfs_file_extent_calc_inline_size(
 675                                                    extent_end - end);
 676                         btrfs_truncate_item(trans, root, path, new_size, 0);
 677                 }
 678                 /* create bookend, splitting the extent in two */
 679                 if (bookend && found_extent) {
 680                         struct btrfs_key ins;
 681                         ins.objectid = inode->i_ino;
 682                         ins.offset = end;
 683                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 684                         btrfs_release_path(root, path);
 685                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 686                                                       sizeof(*extent));
 687
 688                         leaf = path->nodes[0];
 689                         if (ret) {
 690                                 btrfs_print_leaf(root, leaf);
 691                                 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);
 692                         }
 693                         BUG_ON(ret);
 694                         extent = btrfs_item_ptr(leaf, path->slots[0],
 695                                                 struct btrfs_file_extent_item);
 696                         write_extent_buffer(leaf, &old,
 697                                             (unsigned long)extent, sizeof(old));
 698
 699                         btrfs_set_file_extent_offset(leaf, extent,
 700                                     le64_to_cpu(old.offset) + end - key.offset);
 701                         WARN_ON(le64_to_cpu(old.num_bytes) <
 702                                 (extent_end - end));
 703                         btrfs_set_file_extent_num_bytes(leaf, extent,
 704                                                         extent_end - end);
 705                         btrfs_set_file_extent_type(leaf, extent,
 706                                                    BTRFS_FILE_EXTENT_REG);
 707
 708                         btrfs_mark_buffer_dirty(path->nodes[0]);
 709                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 710                                 inode->i_blocks +=
 711                                       btrfs_file_extent_num_bytes(leaf,
 712                                                                   extent) >> 9;
 713                         }
 714                         ret = 0;
 715                         goto out;
 716                 }
 717         }
 718 out:
 719         btrfs_free_path(path);
 720         return ret;
 721 }
 722
 723 /*
 724  * this gets pages into the page cache and locks them down
 725  */
 726 static int prepare_pages(struct btrfs_root *root, struct file *file,
 727                          struct page **pages, size_t num_pages,
 728                          loff_t pos, unsigned long first_index,
 729                          unsigned long last_index, size_t write_bytes)
 730 {
 731         int i;
 732         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 733         struct inode *inode = fdentry(file)->d_inode;
 734         int err = 0;
 735         u64 start_pos;
 736
 737         start_pos = pos & ~((u64)root->sectorsize - 1);
 738
 739         memset(pages, 0, num_pages * sizeof(struct page *));
 740
 741         for (i = 0; i < num_pages; i++) {
 742                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 743                 if (!pages[i]) {
 744                         err = -ENOMEM;
 745                         BUG_ON(1);
 746                 }
 747 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 748                 ClearPageDirty(pages[i]);
 749 #else
 750                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 751 #endif
 752                 wait_on_page_writeback(pages[i]);
 753                 set_page_extent_mapped(pages[i]);
 754                 WARN_ON(!PageLocked(pages[i]));
 755         }
 756         return 0;
 757 }
 758
 759 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 760                                 size_t count, loff_t *ppos)
 761 {
 762         loff_t pos;
 763         loff_t start_pos;
 764         ssize_t num_written = 0;
 765         ssize_t err = 0;
 766         int ret = 0;
 767         struct inode *inode = fdentry(file)->d_inode;
 768         struct btrfs_root *root = BTRFS_I(inode)->root;
 769         struct page **pages = NULL;
 770         int nrptrs;
 771         struct page *pinned[2];
 772         unsigned long first_index;
 773         unsigned long last_index;
 774
 775         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 776                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 777         pinned[0] = NULL;
 778         pinned[1] = NULL;
 779         if (file->f_flags & O_DIRECT)
 780                 return -EINVAL;
 781
 782         pos = *ppos;
 783         start_pos = pos;
 784
 785         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 786         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 787         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 788         if (err)
 789                 goto out_nolock;
 790         if (count == 0)
 791                 goto out_nolock;
 792         err = remove_suid(fdentry(file));
 793         if (err)
 794                 goto out_nolock;
 795         file_update_time(file);
 796
 797         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 798
 799         mutex_lock(&inode->i_mutex);
 800         first_index = pos >> PAGE_CACHE_SHIFT;
 801         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 802
 803         /*
 804          * there are lots of better ways to do this, but this code
 805          * makes sure the first and last page in the file range are
 806          * up to date and ready for cow
 807          */
 808         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 809                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 810                 if (!PageUptodate(pinned[0])) {
 811                         ret = btrfs_readpage(NULL, pinned[0]);
 812                         BUG_ON(ret);
 813                         wait_on_page_locked(pinned[0]);
 814                 } else {
 815                         unlock_page(pinned[0]);
 816                 }
 817         }
 818         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 819                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 820                 if (!PageUptodate(pinned[1])) {
 821                         ret = btrfs_readpage(NULL, pinned[1]);
 822                         BUG_ON(ret);
 823                         wait_on_page_locked(pinned[1]);
 824                 } else {
 825                         unlock_page(pinned[1]);
 826                 }
 827         }
 828
 829         while(count > 0) {
 830                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 831                 size_t write_bytes = min(count, nrptrs *
 832                                         (size_t)PAGE_CACHE_SIZE -
 833                                          offset);
 834                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 835                                         PAGE_CACHE_SHIFT;
 836
 837                 WARN_ON(num_pages > nrptrs);
 838                 memset(pages, 0, sizeof(pages));
 839
 840                 mutex_lock(&root->fs_info->fs_mutex);
 841                 ret = btrfs_check_free_space(root, write_bytes, 0);
 842                 mutex_unlock(&root->fs_info->fs_mutex);
 843                 if (ret)
 844                         goto out;
 845
 846                 ret = prepare_pages(root, file, pages, num_pages,
 847                                     pos, first_index, last_index,
 848                                     write_bytes);
 849                 if (ret)
 850                         goto out;
 851
 852                 ret = btrfs_copy_from_user(pos, num_pages,
 853                                            write_bytes, pages, buf);
 854                 if (ret) {
 855                         btrfs_drop_pages(pages, num_pages);
 856                         goto out;
 857                 }
 858
 859                 ret = dirty_and_release_pages(NULL, root, file, pages,
 860                                               num_pages, pos, write_bytes);
 861                 btrfs_drop_pages(pages, num_pages);
 862                 if (ret)
 863                         goto out;
 864
 865                 buf += write_bytes;
 866                 count -= write_bytes;
 867                 pos += write_bytes;
 868                 num_written += write_bytes;
 869
 870                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 871                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 872                         btrfs_btree_balance_dirty(root, 1);
 873                 btrfs_throttle(root);
 874                 cond_resched();
 875         }
 876 out:
 877         mutex_unlock(&inode->i_mutex);
 878
 879 out_nolock:
 880         kfree(pages);
 881         if (pinned[0])
 882                 page_cache_release(pinned[0]);
 883         if (pinned[1])
 884                 page_cache_release(pinned[1]);
 885         *ppos = pos;
 886
 887         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 888                 err = sync_page_range(inode, inode->i_mapping,
 889                                       start_pos, num_written);
 890                 if (err < 0)
 891                         num_written = err;
 892         }
 893         current->backing_dev_info = NULL;
 894         return num_written ? num_written : err;
 895 }
 896
 897 static int btrfs_sync_file(struct file *file,
 898                            struct dentry *dentry, int datasync)
 899 {
 900         struct inode *inode = dentry->d_inode;
 901         struct btrfs_root *root = BTRFS_I(inode)->root;
 902         int ret = 0;
 903         struct btrfs_trans_handle *trans;
 904
 905         /*
 906          * check the transaction that last modified this inode
 907          * and see if its already been committed
 908          */
 909         mutex_lock(&root->fs_info->fs_mutex);
 910         if (!BTRFS_I(inode)->last_trans)
 911                 goto out;
 912         mutex_lock(&root->fs_info->trans_mutex);
 913         if (BTRFS_I(inode)->last_trans <=
 914             root->fs_info->last_trans_committed) {
 915                 BTRFS_I(inode)->last_trans = 0;
 916                 mutex_unlock(&root->fs_info->trans_mutex);
 917                 goto out;
 918         }
 919         mutex_unlock(&root->fs_info->trans_mutex);
 920
 921         /*
 922          * ok we haven't committed the transaction yet, lets do a commit
 923          */
 924         trans = btrfs_start_transaction(root, 1);
 925         if (!trans) {
 926                 ret = -ENOMEM;
 927                 goto out;
 928         }
 929         ret = btrfs_commit_transaction(trans, root);
 930 out:
 931         mutex_unlock(&root->fs_info->fs_mutex);
 932         return ret > 0 ? EIO : ret;
 933 }
 934
 935 static struct vm_operations_struct btrfs_file_vm_ops = {
 936 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 937         .nopage         = filemap_nopage,
 938         .populate       = filemap_populate,
 939 #else
 940         .fault          = filemap_fault,
 941 #endif
 942         .page_mkwrite   = btrfs_page_mkwrite,
 943 };
 944
 945 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 946 {
 947         vma->vm_ops = &btrfs_file_vm_ops;
 948         file_accessed(filp);
 949         return 0;
 950 }
 951
 952 struct file_operations btrfs_file_operations = {
 953         .llseek         = generic_file_llseek,
 954         .read           = do_sync_read,
 955         .aio_read       = generic_file_aio_read,
 956         .splice_read    = generic_file_splice_read,
 957 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 958         .sendfile       = generic_file_sendfile,
 959 #endif
 960         .write          = btrfs_file_write,
 961         .mmap           = btrfs_file_mmap,
 962         .open           = generic_file_open,
 963         .fsync          = btrfs_sync_file,
 964         .unlocked_ioctl = btrfs_ioctl,
 965 #ifdef CONFIG_COMPAT
 966         .compat_ioctl   = btrfs_ioctl,
 967 #endif
 968 };
 969