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                         inode->i_blocks += (offset + size - found_end) >> 9;
 179                 }
 180                 if (found_end < offset) {
 181                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 182                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 183                 }
 184         } else {
 185 insert:
 186                 btrfs_release_path(root, path);
 187                 datasize = offset + size - key.offset;
 188                 inode->i_blocks += datasize >> 9;
 189                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 190                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 191                                               datasize);
 192                 if (ret) {
 193                         err = ret;
 194                         printk("got bad ret %d\n", ret);
 195                         goto fail;
 196                 }
 197                 leaf = path->nodes[0];
 198                 ei = btrfs_item_ptr(leaf, path->slots[0],
 199                                     struct btrfs_file_extent_item);
 200                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 201                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 202         }
 203         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 204
 205         cur_size = size;
 206         i = 0;
 207         while (size > 0) {
 208                 page = pages[i];
 209                 kaddr = kmap_atomic(page, KM_USER0);
 210                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 211                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 212                 kunmap_atomic(kaddr, KM_USER0);
 213                 page_offset = 0;
 214                 ptr += cur_size;
 215                 size -= cur_size;
 216                 if (i >= num_pages) {
 217                         printk("i %d num_pages %d\n", i, num_pages);
 218                 }
 219                 i++;
 220         }
 221         btrfs_mark_buffer_dirty(leaf);
 222 fail:
 223         btrfs_free_path(path);
 224         return err;
 225 }
 226
 227 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 228                                    struct btrfs_root *root,
 229                                    struct file *file,
 230                                    struct page **pages,
 231                                    size_t num_pages,
 232                                    loff_t pos,
 233                                    size_t write_bytes)
 234 {
 235         int err = 0;
 236         int i;
 237         struct inode *inode = fdentry(file)->d_inode;
 238         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 239         u64 hint_byte;
 240         u64 num_bytes;
 241         u64 start_pos;
 242         u64 end_of_last_block;
 243         u64 end_pos = pos + write_bytes;
 244         u64 inline_size;
 245         loff_t isize = i_size_read(inode);
 246
 247         start_pos = pos & ~((u64)root->sectorsize - 1);
 248         num_bytes = (write_bytes + pos - start_pos +
 249                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 250
 251         end_of_last_block = start_pos + num_bytes - 1;
 252
 253         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 254         mutex_lock(&root->fs_info->fs_mutex);
 255         trans = btrfs_start_transaction(root, 1);
 256         if (!trans) {
 257                 err = -ENOMEM;
 258                 goto out_unlock;
 259         }
 260         btrfs_set_trans_block_group(trans, inode);
 261         hint_byte = 0;
 262
 263         if ((end_of_last_block & 4095) == 0) {
 264                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 265         }
 266         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 267
 268         /* FIXME...EIEIO, ENOSPC and more */
 269         /* insert any holes we need to create */
 270         if (isize < end_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                 if (last_pos_in_file < end_pos) {
 277                         err = btrfs_drop_extents(trans, root, inode,
 278                                                  last_pos_in_file,
 279                                                  last_pos_in_file + hole_size,
 280                                                  last_pos_in_file,
 281                                                  &hint_byte);
 282                         if (err)
 283                                 goto failed;
 284
 285                         err = btrfs_insert_file_extent(trans, root,
 286                                                        inode->i_ino,
 287                                                        last_pos_in_file,
 288                                                        0, 0, hole_size);
 289                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 290                                         last_pos_in_file + hole_size -1);
 291                         btrfs_check_file(root, inode);
 292                 }
 293                 if (err)
 294                         goto failed;
 295         }
 296
 297         /*
 298          * either allocate an extent for the new bytes or setup the key
 299          * to show we are doing inline data in the extent
 300          */
 301         inline_size = end_pos;
 302         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 303             inline_size > root->fs_info->max_inline ||
 304             (inline_size & (root->sectorsize -1)) == 0 ||
 305             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 306                 u64 last_end;
 307                 u64 existing_delalloc = 0;
 308
 309                 for (i = 0; i < num_pages; i++) {
 310                         struct page *p = pages[i];
 311                         SetPageUptodate(p);
 312                         set_page_dirty(p);
 313                 }
 314                 last_end = (u64)(pages[num_pages -1]->index) <<
 315                                 PAGE_CACHE_SHIFT;
 316                 last_end += PAGE_CACHE_SIZE - 1;
 317                 if (start_pos < isize) {
 318                         u64 delalloc_start = start_pos;
 319                         existing_delalloc = count_range_bits(io_tree,
 320                                              &delalloc_start,
 321                                              end_of_last_block, (u64)-1,
 322                                              EXTENT_DELALLOC);
 323                 }
 324                 set_extent_delalloc(io_tree, start_pos, end_of_last_block,
 325                                  GFP_NOFS);
 326                 btrfs_add_ordered_inode(inode);
 327         } else {
 328                 u64 aligned_end;
 329                 /* step one, delete the existing extents in this range */
 330                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 331                         ~((u64)root->sectorsize - 1);
 332                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 333                                          aligned_end, aligned_end, &hint_byte);
 334                 if (err)
 335                         goto failed;
 336                 if (isize > inline_size)
 337                         inline_size = min_t(u64, isize, aligned_end);
 338                 inline_size -= start_pos;
 339                 err = insert_inline_extent(trans, root, inode, start_pos,
 340                                            inline_size, pages, 0, num_pages);
 341                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 342                 BUG_ON(err);
 343         }
 344         if (end_pos > isize) {
 345                 i_size_write(inode, end_pos);
 346                 btrfs_update_inode(trans, root, inode);
 347         }
 348 failed:
 349         err = btrfs_end_transaction(trans, root);
 350 out_unlock:
 351         mutex_unlock(&root->fs_info->fs_mutex);
 352         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 353         return err;
 354 }
 355
 356 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 357 {
 358         struct extent_map *em;
 359         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 360         u64 len = end - start + 1;
 361
 362         if (end == (u64)-1)
 363                 len = (u64)-1;
 364         while(1) {
 365                 spin_lock(&em_tree->lock);
 366                 em = lookup_extent_mapping(em_tree, start, len);
 367                 if (!em) {
 368                         spin_unlock(&em_tree->lock);
 369                         break;
 370                 }
 371                 remove_extent_mapping(em_tree, em);
 372                 spin_unlock(&em_tree->lock);
 373
 374                 /* once for us */
 375                 free_extent_map(em);
 376                 /* once for the tree*/
 377                 free_extent_map(em);
 378         }
 379         return 0;
 380 }
 381
 382 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 383 {
 384         return 0;
 385 #if 0
 386         struct btrfs_path *path;
 387         struct btrfs_key found_key;
 388         struct extent_buffer *leaf;
 389         struct btrfs_file_extent_item *extent;
 390         u64 last_offset = 0;
 391         int nritems;
 392         int slot;
 393         int found_type;
 394         int ret;
 395         int err = 0;
 396         u64 extent_end = 0;
 397
 398         path = btrfs_alloc_path();
 399         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 400                                        last_offset, 0);
 401         while(1) {
 402                 nritems = btrfs_header_nritems(path->nodes[0]);
 403                 if (path->slots[0] >= nritems) {
 404                         ret = btrfs_next_leaf(root, path);
 405                         if (ret)
 406                                 goto out;
 407                         nritems = btrfs_header_nritems(path->nodes[0]);
 408                 }
 409                 slot = path->slots[0];
 410                 leaf = path->nodes[0];
 411                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 412                 if (found_key.objectid != inode->i_ino)
 413                         break;
 414                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 415                         goto out;
 416
 417                 if (found_key.offset < last_offset) {
 418                         WARN_ON(1);
 419                         btrfs_print_leaf(root, leaf);
 420                         printk("inode %lu found offset %Lu expected %Lu\n",
 421                                inode->i_ino, found_key.offset, last_offset);
 422                         err = 1;
 423                         goto out;
 424                 }
 425                 extent = btrfs_item_ptr(leaf, slot,
 426                                         struct btrfs_file_extent_item);
 427                 found_type = btrfs_file_extent_type(leaf, extent);
 428                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 429                         extent_end = found_key.offset +
 430                              btrfs_file_extent_num_bytes(leaf, extent);
 431                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 432                         struct btrfs_item *item;
 433                         item = btrfs_item_nr(leaf, slot);
 434                         extent_end = found_key.offset +
 435                              btrfs_file_extent_inline_len(leaf, item);
 436                         extent_end = (extent_end + root->sectorsize - 1) &
 437                                 ~((u64)root->sectorsize -1 );
 438                 }
 439                 last_offset = extent_end;
 440                 path->slots[0]++;
 441         }
 442         if (0 && last_offset < inode->i_size) {
 443                 WARN_ON(1);
 444                 btrfs_print_leaf(root, leaf);
 445                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 446                        last_offset, inode->i_size);
 447                 err = 1;
 448
 449         }
 450 out:
 451         btrfs_free_path(path);
 452         return err;
 453 #endif
 454 }
 455
 456 /*
 457  * this is very complex, but the basic idea is to drop all extents
 458  * in the range start - end.  hint_block is filled in with a block number
 459  * that would be a good hint to the block allocator for this file.
 460  *
 461  * If an extent intersects the range but is not entirely inside the range
 462  * it is either truncated or split.  Anything entirely inside the range
 463  * is deleted from the tree.
 464  */
 465 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 466                        struct btrfs_root *root, struct inode *inode,
 467                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 468 {
 469         u64 extent_end = 0;
 470         u64 search_start = start;
 471         struct extent_buffer *leaf;
 472         struct btrfs_file_extent_item *extent;
 473         struct btrfs_path *path;
 474         struct btrfs_key key;
 475         struct btrfs_file_extent_item old;
 476         int keep;
 477         int slot;
 478         int bookend;
 479         int found_type;
 480         int found_extent;
 481         int found_inline;
 482         int recow;
 483         int ret;
 484
 485         btrfs_drop_extent_cache(inode, start, end - 1);
 486
 487         path = btrfs_alloc_path();
 488         if (!path)
 489                 return -ENOMEM;
 490         while(1) {
 491                 recow = 0;
 492                 btrfs_release_path(root, path);
 493                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 494                                                search_start, -1);
 495                 if (ret < 0)
 496                         goto out;
 497                 if (ret > 0) {
 498                         if (path->slots[0] == 0) {
 499                                 ret = 0;
 500                                 goto out;
 501                         }
 502                         path->slots[0]--;
 503                 }
 504 next_slot:
 505                 keep = 0;
 506                 bookend = 0;
 507                 found_extent = 0;
 508                 found_inline = 0;
 509                 extent = NULL;
 510                 leaf = path->nodes[0];
 511                 slot = path->slots[0];
 512                 ret = 0;
 513                 btrfs_item_key_to_cpu(leaf, &key, slot);
 514                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 515                     key.offset >= end) {
 516                         goto out;
 517                 }
 518                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 519                     key.objectid != inode->i_ino) {
 520                         goto out;
 521                 }
 522                 if (recow) {
 523                         search_start = key.offset;
 524                         continue;
 525                 }
 526                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 527                         extent = btrfs_item_ptr(leaf, slot,
 528                                                 struct btrfs_file_extent_item);
 529                         found_type = btrfs_file_extent_type(leaf, extent);
 530                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 531                                 extent_end =
 532                                      btrfs_file_extent_disk_bytenr(leaf,
 533                                                                    extent);
 534                                 if (extent_end)
 535                                         *hint_byte = extent_end;
 536
 537                                 extent_end = key.offset +
 538                                      btrfs_file_extent_num_bytes(leaf, extent);
 539                                 found_extent = 1;
 540                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 541                                 struct btrfs_item *item;
 542                                 item = btrfs_item_nr(leaf, slot);
 543                                 found_inline = 1;
 544                                 extent_end = key.offset +
 545                                      btrfs_file_extent_inline_len(leaf, item);
 546                         }
 547                 } else {
 548                         extent_end = search_start;
 549                 }
 550
 551                 /* we found nothing we can drop */
 552                 if ((!found_extent && !found_inline) ||
 553                     search_start >= extent_end) {
 554                         int nextret;
 555                         u32 nritems;
 556                         nritems = btrfs_header_nritems(leaf);
 557                         if (slot >= nritems - 1) {
 558                                 nextret = btrfs_next_leaf(root, path);
 559                                 if (nextret)
 560                                         goto out;
 561                                 recow = 1;
 562                         } else {
 563                                 path->slots[0]++;
 564                         }
 565                         goto next_slot;
 566                 }
 567
 568                 if (found_inline) {
 569                         u64 mask = root->sectorsize - 1;
 570                         search_start = (extent_end + mask) & ~mask;
 571                 } else
 572                         search_start = extent_end;
 573                 if (end <= extent_end && start >= key.offset && found_inline) {
 574                         *hint_byte = EXTENT_MAP_INLINE;
 575                         continue;
 576                 }
 577                 if (end < extent_end && end >= key.offset) {
 578                         if (found_extent) {
 579                                 u64 disk_bytenr =
 580                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 581                                 u64 disk_num_bytes =
 582                                     btrfs_file_extent_disk_num_bytes(leaf,
 583                                                                       extent);
 584                                 read_extent_buffer(leaf, &old,
 585                                                    (unsigned long)extent,
 586                                                    sizeof(old));
 587                                 if (disk_bytenr != 0) {
 588                                         ret = btrfs_inc_extent_ref(trans, root,
 589                                                  disk_bytenr, disk_num_bytes,
 590                                                  root->root_key.objectid,
 591                                                  trans->transid,
 592                                                  key.objectid, end);
 593                                         BUG_ON(ret);
 594                                 }
 595                         }
 596                         bookend = 1;
 597                         if (found_inline && start <= key.offset)
 598                                 keep = 1;
 599                 }
 600                 /* truncate existing extent */
 601                 if (start > key.offset) {
 602                         u64 new_num;
 603                         u64 old_num;
 604                         keep = 1;
 605                         WARN_ON(start & (root->sectorsize - 1));
 606                         if (found_extent) {
 607                                 new_num = start - key.offset;
 608                                 old_num = btrfs_file_extent_num_bytes(leaf,
 609                                                                       extent);
 610                                 *hint_byte =
 611                                         btrfs_file_extent_disk_bytenr(leaf,
 612                                                                       extent);
 613                                 if (btrfs_file_extent_disk_bytenr(leaf,
 614                                                                   extent)) {
 615                                         dec_i_blocks(inode, old_num - new_num);
 616                                 }
 617                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 618                                                                 new_num);
 619                                 btrfs_mark_buffer_dirty(leaf);
 620                         } else if (key.offset < inline_limit &&
 621                                    (end > extent_end) &&
 622                                    (inline_limit < extent_end)) {
 623                                 u32 new_size;
 624                                 new_size = btrfs_file_extent_calc_inline_size(
 625                                                    inline_limit - key.offset);
 626                                 dec_i_blocks(inode, (extent_end - key.offset) -
 627                                         (inline_limit - key.offset));
 628                                 btrfs_truncate_item(trans, root, path,
 629                                                     new_size, 1);
 630                         }
 631                 }
 632                 /* delete the entire extent */
 633                 if (!keep) {
 634                         u64 disk_bytenr = 0;
 635                         u64 disk_num_bytes = 0;
 636                         u64 extent_num_bytes = 0;
 637                         u64 root_gen;
 638                         u64 root_owner;
 639
 640                         root_gen = btrfs_header_generation(leaf);
 641                         root_owner = btrfs_header_owner(leaf);
 642                         if (found_extent) {
 643                                 disk_bytenr =
 644                                       btrfs_file_extent_disk_bytenr(leaf,
 645                                                                      extent);
 646                                 disk_num_bytes =
 647                                       btrfs_file_extent_disk_num_bytes(leaf,
 648                                                                        extent);
 649                                 extent_num_bytes =
 650                                       btrfs_file_extent_num_bytes(leaf, extent);
 651                                 *hint_byte =
 652                                         btrfs_file_extent_disk_bytenr(leaf,
 653                                                                       extent);
 654                         }
 655                         ret = btrfs_del_item(trans, root, path);
 656                         /* TODO update progress marker and return */
 657                         BUG_ON(ret);
 658                         btrfs_release_path(root, path);
 659                         extent = NULL;
 660                         if (found_extent && disk_bytenr != 0) {
 661                                 dec_i_blocks(inode, extent_num_bytes);
 662                                 ret = btrfs_free_extent(trans, root,
 663                                                 disk_bytenr,
 664                                                 disk_num_bytes,
 665                                                 root_owner,
 666                                                 root_gen, inode->i_ino,
 667                                                 key.offset, 0);
 668                         }
 669
 670                         BUG_ON(ret);
 671                         if (!bookend && search_start >= end) {
 672                                 ret = 0;
 673                                 goto out;
 674                         }
 675                         if (!bookend)
 676                                 continue;
 677                 }
 678                 if (bookend && found_inline && start <= key.offset) {
 679                         u32 new_size;
 680                         new_size = btrfs_file_extent_calc_inline_size(
 681                                                    extent_end - end);
 682                         dec_i_blocks(inode, (extent_end - key.offset) -
 683                                         (extent_end - end));
 684                         btrfs_truncate_item(trans, root, path, new_size, 0);
 685                 }
 686                 /* create bookend, splitting the extent in two */
 687                 if (bookend && found_extent) {
 688                         struct btrfs_key ins;
 689                         ins.objectid = inode->i_ino;
 690                         ins.offset = end;
 691                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 692                         btrfs_release_path(root, path);
 693                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 694                                                       sizeof(*extent));
 695
 696                         leaf = path->nodes[0];
 697                         if (ret) {
 698                                 btrfs_print_leaf(root, leaf);
 699                                 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);
 700                         }
 701                         BUG_ON(ret);
 702                         extent = btrfs_item_ptr(leaf, path->slots[0],
 703                                                 struct btrfs_file_extent_item);
 704                         write_extent_buffer(leaf, &old,
 705                                             (unsigned long)extent, sizeof(old));
 706
 707                         btrfs_set_file_extent_offset(leaf, extent,
 708                                     le64_to_cpu(old.offset) + end - key.offset);
 709                         WARN_ON(le64_to_cpu(old.num_bytes) <
 710                                 (extent_end - end));
 711                         btrfs_set_file_extent_num_bytes(leaf, extent,
 712                                                         extent_end - end);
 713                         btrfs_set_file_extent_type(leaf, extent,
 714                                                    BTRFS_FILE_EXTENT_REG);
 715
 716                         btrfs_mark_buffer_dirty(path->nodes[0]);
 717                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 718                                 inode->i_blocks +=
 719                                       btrfs_file_extent_num_bytes(leaf,
 720                                                                   extent) >> 9;
 721                         }
 722                         ret = 0;
 723                         goto out;
 724                 }
 725         }
 726 out:
 727         btrfs_free_path(path);
 728         btrfs_check_file(root, inode);
 729         return ret;
 730 }
 731
 732 /*
 733  * this gets pages into the page cache and locks them down
 734  */
 735 static int prepare_pages(struct btrfs_root *root, struct file *file,
 736                          struct page **pages, size_t num_pages,
 737                          loff_t pos, unsigned long first_index,
 738                          unsigned long last_index, size_t write_bytes)
 739 {
 740         int i;
 741         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 742         struct inode *inode = fdentry(file)->d_inode;
 743         int err = 0;
 744         u64 start_pos;
 745
 746         start_pos = pos & ~((u64)root->sectorsize - 1);
 747
 748         memset(pages, 0, num_pages * sizeof(struct page *));
 749
 750         for (i = 0; i < num_pages; i++) {
 751                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 752                 if (!pages[i]) {
 753                         err = -ENOMEM;
 754                         BUG_ON(1);
 755                 }
 756 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 757                 ClearPageDirty(pages[i]);
 758 #else
 759                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 760 #endif
 761                 wait_on_page_writeback(pages[i]);
 762                 set_page_extent_mapped(pages[i]);
 763                 WARN_ON(!PageLocked(pages[i]));
 764         }
 765         if (start_pos < inode->i_size) {
 766                 u64 last_pos;
 767                 last_pos = (index + num_pages) << PAGE_CACHE_SHIFT;
 768                 lock_extent(&BTRFS_I(inode)->io_tree,
 769                             start_pos, last_pos - 1, GFP_NOFS);
 770                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 771                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 772                                   GFP_NOFS);
 773                 unlock_extent(&BTRFS_I(inode)->io_tree,
 774                               start_pos, last_pos - 1, GFP_NOFS);
 775         }
 776         return 0;
 777 }
 778
 779 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 780                                 size_t count, loff_t *ppos)
 781 {
 782         loff_t pos;
 783         loff_t start_pos;
 784         ssize_t num_written = 0;
 785         ssize_t err = 0;
 786         int ret = 0;
 787         struct inode *inode = fdentry(file)->d_inode;
 788         struct btrfs_root *root = BTRFS_I(inode)->root;
 789         struct page **pages = NULL;
 790         int nrptrs;
 791         struct page *pinned[2];
 792         unsigned long first_index;
 793         unsigned long last_index;
 794
 795         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 796                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 797         pinned[0] = NULL;
 798         pinned[1] = NULL;
 799         if (file->f_flags & O_DIRECT)
 800                 return -EINVAL;
 801
 802         pos = *ppos;
 803         start_pos = pos;
 804
 805         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 806         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 807         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 808         if (err)
 809                 goto out_nolock;
 810         if (count == 0)
 811                 goto out_nolock;
 812         err = remove_suid(fdentry(file));
 813         if (err)
 814                 goto out_nolock;
 815         file_update_time(file);
 816
 817         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 818
 819         mutex_lock(&inode->i_mutex);
 820         first_index = pos >> PAGE_CACHE_SHIFT;
 821         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 822
 823         /*
 824          * there are lots of better ways to do this, but this code
 825          * makes sure the first and last page in the file range are
 826          * up to date and ready for cow
 827          */
 828         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 829                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 830                 if (!PageUptodate(pinned[0])) {
 831                         ret = btrfs_readpage(NULL, pinned[0]);
 832                         BUG_ON(ret);
 833                         wait_on_page_locked(pinned[0]);
 834                 } else {
 835                         unlock_page(pinned[0]);
 836                 }
 837         }
 838         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 839                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 840                 if (!PageUptodate(pinned[1])) {
 841                         ret = btrfs_readpage(NULL, pinned[1]);
 842                         BUG_ON(ret);
 843                         wait_on_page_locked(pinned[1]);
 844                 } else {
 845                         unlock_page(pinned[1]);
 846                 }
 847         }
 848
 849         while(count > 0) {
 850                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 851                 size_t write_bytes = min(count, nrptrs *
 852                                         (size_t)PAGE_CACHE_SIZE -
 853                                          offset);
 854                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 855                                         PAGE_CACHE_SHIFT;
 856
 857                 WARN_ON(num_pages > nrptrs);
 858                 memset(pages, 0, sizeof(pages));
 859
 860                 mutex_lock(&root->fs_info->fs_mutex);
 861                 ret = btrfs_check_free_space(root, write_bytes, 0);
 862                 mutex_unlock(&root->fs_info->fs_mutex);
 863                 if (ret)
 864                         goto out;
 865
 866                 ret = prepare_pages(root, file, pages, num_pages,
 867                                     pos, first_index, last_index,
 868                                     write_bytes);
 869                 if (ret)
 870                         goto out;
 871
 872                 ret = btrfs_copy_from_user(pos, num_pages,
 873                                            write_bytes, pages, buf);
 874                 if (ret) {
 875                         btrfs_drop_pages(pages, num_pages);
 876                         goto out;
 877                 }
 878
 879                 ret = dirty_and_release_pages(NULL, root, file, pages,
 880                                               num_pages, pos, write_bytes);
 881                 btrfs_drop_pages(pages, num_pages);
 882                 if (ret)
 883                         goto out;
 884
 885                 buf += write_bytes;
 886                 count -= write_bytes;
 887                 pos += write_bytes;
 888                 num_written += write_bytes;
 889
 890                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 891                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 892                         btrfs_btree_balance_dirty(root, 1);
 893                 btrfs_throttle(root);
 894                 cond_resched();
 895         }
 896 out:
 897         mutex_unlock(&inode->i_mutex);
 898
 899 out_nolock:
 900         kfree(pages);
 901         if (pinned[0])
 902                 page_cache_release(pinned[0]);
 903         if (pinned[1])
 904                 page_cache_release(pinned[1]);
 905         *ppos = pos;
 906
 907         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 908                 err = sync_page_range(inode, inode->i_mapping,
 909                                       start_pos, num_written);
 910                 if (err < 0)
 911                         num_written = err;
 912         }
 913         current->backing_dev_info = NULL;
 914         return num_written ? num_written : err;
 915 }
 916
 917 static int btrfs_sync_file(struct file *file,
 918                            struct dentry *dentry, int datasync)
 919 {
 920         struct inode *inode = dentry->d_inode;
 921         struct btrfs_root *root = BTRFS_I(inode)->root;
 922         int ret = 0;
 923         struct btrfs_trans_handle *trans;
 924
 925         /*
 926          * check the transaction that last modified this inode
 927          * and see if its already been committed
 928          */
 929         mutex_lock(&root->fs_info->fs_mutex);
 930         if (!BTRFS_I(inode)->last_trans)
 931                 goto out;
 932         mutex_lock(&root->fs_info->trans_mutex);
 933         if (BTRFS_I(inode)->last_trans <=
 934             root->fs_info->last_trans_committed) {
 935                 BTRFS_I(inode)->last_trans = 0;
 936                 mutex_unlock(&root->fs_info->trans_mutex);
 937                 goto out;
 938         }
 939         mutex_unlock(&root->fs_info->trans_mutex);
 940
 941         /*
 942          * ok we haven't committed the transaction yet, lets do a commit
 943          */
 944         trans = btrfs_start_transaction(root, 1);
 945         if (!trans) {
 946                 ret = -ENOMEM;
 947                 goto out;
 948         }
 949         ret = btrfs_commit_transaction(trans, root);
 950 out:
 951         mutex_unlock(&root->fs_info->fs_mutex);
 952         return ret > 0 ? EIO : ret;
 953 }
 954
 955 static struct vm_operations_struct btrfs_file_vm_ops = {
 956 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 957         .nopage         = filemap_nopage,
 958         .populate       = filemap_populate,
 959 #else
 960         .fault          = filemap_fault,
 961 #endif
 962         .page_mkwrite   = btrfs_page_mkwrite,
 963 };
 964
 965 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 966 {
 967         vma->vm_ops = &btrfs_file_vm_ops;
 968         file_accessed(filp);
 969         return 0;
 970 }
 971
 972 struct file_operations btrfs_file_operations = {
 973         .llseek         = generic_file_llseek,
 974         .read           = do_sync_read,
 975         .aio_read       = generic_file_aio_read,
 976         .splice_read    = generic_file_splice_read,
 977 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 978         .sendfile       = generic_file_sendfile,
 979 #endif
 980         .write          = btrfs_file_write,
 981         .mmap           = btrfs_file_mmap,
 982         .open           = generic_file_open,
 983         .fsync          = btrfs_sync_file,
 984         .unlocked_ioctl = btrfs_ioctl,
 985 #ifdef CONFIG_COMPAT
 986         .compat_ioctl   = btrfs_ioctl,
 987 #endif
 988 };
 989