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 "ioctl.h"
  38 #include "print-tree.h"
  39 #include "compat.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         trans = btrfs_start_transaction(root, 1);
 255         if (!trans) {
 256                 err = -ENOMEM;
 257                 goto out_unlock;
 258         }
 259         btrfs_set_trans_block_group(trans, inode);
 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 < start_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 = (start_pos - last_pos_in_file + mask) & ~mask;
 275                 if (hole_size > 0) {
 276                         btrfs_wait_ordered_range(inode, last_pos_in_file,
 277                                                  last_pos_in_file + hole_size);
 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, 0);
 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                 /* check for reserved extents on each page, we don't want
 308                  * to reset the delalloc bit on things that already have
 309                  * extents reserved.
 310                  */
 311                 set_extent_delalloc(io_tree, start_pos,
 312                                     end_of_last_block, GFP_NOFS);
 313                 for (i = 0; i < num_pages; i++) {
 314                         struct page *p = pages[i];
 315                         SetPageUptodate(p);
 316                         set_page_dirty(p);
 317                 }
 318         } else {
 319                 u64 aligned_end;
 320                 /* step one, delete the existing extents in this range */
 321                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 322                         ~((u64)root->sectorsize - 1);
 323                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 324                                          aligned_end, aligned_end, &hint_byte);
 325                 if (err)
 326                         goto failed;
 327                 if (isize > inline_size)
 328                         inline_size = min_t(u64, isize, aligned_end);
 329                 inline_size -= start_pos;
 330                 err = insert_inline_extent(trans, root, inode, start_pos,
 331                                            inline_size, pages, 0, num_pages);
 332                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 333                 BUG_ON(err);
 334         }
 335         if (end_pos > isize) {
 336                 i_size_write(inode, end_pos);
 337                 btrfs_update_inode(trans, root, inode);
 338         }
 339 failed:
 340         err = btrfs_end_transaction(trans, root);
 341 out_unlock:
 342         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 343         return err;
 344 }
 345
 346 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 347 {
 348         struct extent_map *em;
 349         struct extent_map *split = NULL;
 350         struct extent_map *split2 = NULL;
 351         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 352         struct extent_map *tmp;
 353         u64 len = end - start + 1;
 354         u64 next_start;
 355         int ret;
 356         int testend = 1;
 357
 358         WARN_ON(end < start);
 359         if (end == (u64)-1) {
 360                 len = (u64)-1;
 361                 testend = 0;
 362         }
 363         while(1) {
 364                 if (!split)
 365                         split = alloc_extent_map(GFP_NOFS);
 366                 if (!split2)
 367                         split2 = alloc_extent_map(GFP_NOFS);
 368
 369                 spin_lock(&em_tree->lock);
 370                 em = lookup_extent_mapping(em_tree, start, len);
 371                 if (!em) {
 372                         spin_unlock(&em_tree->lock);
 373                         break;
 374                 }
 375                 tmp = rb_entry(&em->rb_node, struct extent_map, rb_node);
 376                 next_start = tmp->start;
 377                 remove_extent_mapping(em_tree, em);
 378
 379                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 380                     em->start < start) {
 381                         split->start = em->start;
 382                         split->len = start - em->start;
 383                         split->block_start = em->block_start;
 384                         split->bdev = em->bdev;
 385                         split->flags = em->flags;
 386                         ret = add_extent_mapping(em_tree, split);
 387                         BUG_ON(ret);
 388                         free_extent_map(split);
 389                         split = split2;
 390                         split2 = NULL;
 391                 }
 392                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 393                     testend && em->start + em->len > start + len) {
 394                         u64 diff = start + len - em->start;
 395
 396                         split->start = start + len;
 397                         split->len = em->start + em->len - (start + len);
 398                         split->bdev = em->bdev;
 399                         split->flags = em->flags;
 400
 401                         split->block_start = em->block_start + diff;
 402
 403                         ret = add_extent_mapping(em_tree, split);
 404                         BUG_ON(ret);
 405                         free_extent_map(split);
 406                         split = NULL;
 407                 }
 408                 spin_unlock(&em_tree->lock);
 409
 410                 /* once for us */
 411                 free_extent_map(em);
 412                 /* once for the tree*/
 413                 free_extent_map(em);
 414         }
 415         if (split)
 416                 free_extent_map(split);
 417         if (split2)
 418                 free_extent_map(split2);
 419         return 0;
 420 }
 421
 422 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 423 {
 424         return 0;
 425 #if 0
 426         struct btrfs_path *path;
 427         struct btrfs_key found_key;
 428         struct extent_buffer *leaf;
 429         struct btrfs_file_extent_item *extent;
 430         u64 last_offset = 0;
 431         int nritems;
 432         int slot;
 433         int found_type;
 434         int ret;
 435         int err = 0;
 436         u64 extent_end = 0;
 437
 438         path = btrfs_alloc_path();
 439         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 440                                        last_offset, 0);
 441         while(1) {
 442                 nritems = btrfs_header_nritems(path->nodes[0]);
 443                 if (path->slots[0] >= nritems) {
 444                         ret = btrfs_next_leaf(root, path);
 445                         if (ret)
 446                                 goto out;
 447                         nritems = btrfs_header_nritems(path->nodes[0]);
 448                 }
 449                 slot = path->slots[0];
 450                 leaf = path->nodes[0];
 451                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 452                 if (found_key.objectid != inode->i_ino)
 453                         break;
 454                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 455                         goto out;
 456
 457                 if (found_key.offset < last_offset) {
 458                         WARN_ON(1);
 459                         btrfs_print_leaf(root, leaf);
 460                         printk("inode %lu found offset %Lu expected %Lu\n",
 461                                inode->i_ino, found_key.offset, last_offset);
 462                         err = 1;
 463                         goto out;
 464                 }
 465                 extent = btrfs_item_ptr(leaf, slot,
 466                                         struct btrfs_file_extent_item);
 467                 found_type = btrfs_file_extent_type(leaf, extent);
 468                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 469                         extent_end = found_key.offset +
 470                              btrfs_file_extent_num_bytes(leaf, extent);
 471                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 472                         struct btrfs_item *item;
 473                         item = btrfs_item_nr(leaf, slot);
 474                         extent_end = found_key.offset +
 475                              btrfs_file_extent_inline_len(leaf, item);
 476                         extent_end = (extent_end + root->sectorsize - 1) &
 477                                 ~((u64)root->sectorsize -1 );
 478                 }
 479                 last_offset = extent_end;
 480                 path->slots[0]++;
 481         }
 482         if (0 && last_offset < inode->i_size) {
 483                 WARN_ON(1);
 484                 btrfs_print_leaf(root, leaf);
 485                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 486                        last_offset, inode->i_size);
 487                 err = 1;
 488
 489         }
 490 out:
 491         btrfs_free_path(path);
 492         return err;
 493 #endif
 494 }
 495
 496 /*
 497  * this is very complex, but the basic idea is to drop all extents
 498  * in the range start - end.  hint_block is filled in with a block number
 499  * that would be a good hint to the block allocator for this file.
 500  *
 501  * If an extent intersects the range but is not entirely inside the range
 502  * it is either truncated or split.  Anything entirely inside the range
 503  * is deleted from the tree.
 504  */
 505 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 506                        struct btrfs_root *root, struct inode *inode,
 507                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 508 {
 509         u64 extent_end = 0;
 510         u64 search_start = start;
 511         struct extent_buffer *leaf;
 512         struct btrfs_file_extent_item *extent;
 513         struct btrfs_path *path;
 514         struct btrfs_key key;
 515         struct btrfs_file_extent_item old;
 516         int keep;
 517         int slot;
 518         int bookend;
 519         int found_type;
 520         int found_extent;
 521         int found_inline;
 522         int recow;
 523         int ret;
 524
 525         btrfs_drop_extent_cache(inode, start, end - 1);
 526
 527         path = btrfs_alloc_path();
 528         if (!path)
 529                 return -ENOMEM;
 530         while(1) {
 531                 recow = 0;
 532                 btrfs_release_path(root, path);
 533                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 534                                                search_start, -1);
 535                 if (ret < 0)
 536                         goto out;
 537                 if (ret > 0) {
 538                         if (path->slots[0] == 0) {
 539                                 ret = 0;
 540                                 goto out;
 541                         }
 542                         path->slots[0]--;
 543                 }
 544 next_slot:
 545                 keep = 0;
 546                 bookend = 0;
 547                 found_extent = 0;
 548                 found_inline = 0;
 549                 extent = NULL;
 550                 leaf = path->nodes[0];
 551                 slot = path->slots[0];
 552                 ret = 0;
 553                 btrfs_item_key_to_cpu(leaf, &key, slot);
 554                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 555                     key.offset >= end) {
 556                         goto out;
 557                 }
 558                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 559                     key.objectid != inode->i_ino) {
 560                         goto out;
 561                 }
 562                 if (recow) {
 563                         search_start = key.offset;
 564                         continue;
 565                 }
 566                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 567                         extent = btrfs_item_ptr(leaf, slot,
 568                                                 struct btrfs_file_extent_item);
 569                         found_type = btrfs_file_extent_type(leaf, extent);
 570                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 571                                 extent_end =
 572                                      btrfs_file_extent_disk_bytenr(leaf,
 573                                                                    extent);
 574                                 if (extent_end)
 575                                         *hint_byte = extent_end;
 576
 577                                 extent_end = key.offset +
 578                                      btrfs_file_extent_num_bytes(leaf, extent);
 579                                 found_extent = 1;
 580                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 581                                 struct btrfs_item *item;
 582                                 item = btrfs_item_nr(leaf, slot);
 583                                 found_inline = 1;
 584                                 extent_end = key.offset +
 585                                      btrfs_file_extent_inline_len(leaf, item);
 586                         }
 587                 } else {
 588                         extent_end = search_start;
 589                 }
 590
 591                 /* we found nothing we can drop */
 592                 if ((!found_extent && !found_inline) ||
 593                     search_start >= extent_end) {
 594                         int nextret;
 595                         u32 nritems;
 596                         nritems = btrfs_header_nritems(leaf);
 597                         if (slot >= nritems - 1) {
 598                                 nextret = btrfs_next_leaf(root, path);
 599                                 if (nextret)
 600                                         goto out;
 601                                 recow = 1;
 602                         } else {
 603                                 path->slots[0]++;
 604                         }
 605                         goto next_slot;
 606                 }
 607
 608                 if (found_inline) {
 609                         u64 mask = root->sectorsize - 1;
 610                         search_start = (extent_end + mask) & ~mask;
 611                 } else
 612                         search_start = extent_end;
 613                 if (end <= extent_end && start >= key.offset && found_inline) {
 614                         *hint_byte = EXTENT_MAP_INLINE;
 615                         continue;
 616                 }
 617                 if (end < extent_end && end >= key.offset) {
 618                         if (found_extent) {
 619                                 u64 disk_bytenr =
 620                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 621                                 u64 disk_num_bytes =
 622                                     btrfs_file_extent_disk_num_bytes(leaf,
 623                                                                       extent);
 624                                 read_extent_buffer(leaf, &old,
 625                                                    (unsigned long)extent,
 626                                                    sizeof(old));
 627                                 if (disk_bytenr != 0) {
 628                                         ret = btrfs_inc_extent_ref(trans, root,
 629                                                  disk_bytenr, disk_num_bytes,
 630                                                  root->root_key.objectid,
 631                                                  trans->transid,
 632                                                  key.objectid, end);
 633                                         BUG_ON(ret);
 634                                 }
 635                         }
 636                         bookend = 1;
 637                         if (found_inline && start <= key.offset)
 638                                 keep = 1;
 639                 }
 640                 /* truncate existing extent */
 641                 if (start > key.offset) {
 642                         u64 new_num;
 643                         u64 old_num;
 644                         keep = 1;
 645                         WARN_ON(start & (root->sectorsize - 1));
 646                         if (found_extent) {
 647                                 new_num = start - key.offset;
 648                                 old_num = btrfs_file_extent_num_bytes(leaf,
 649                                                                       extent);
 650                                 *hint_byte =
 651                                         btrfs_file_extent_disk_bytenr(leaf,
 652                                                                       extent);
 653                                 if (btrfs_file_extent_disk_bytenr(leaf,
 654                                                                   extent)) {
 655                                         dec_i_blocks(inode, old_num - new_num);
 656                                 }
 657                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 658                                                                 new_num);
 659                                 btrfs_mark_buffer_dirty(leaf);
 660                         } else if (key.offset < inline_limit &&
 661                                    (end > extent_end) &&
 662                                    (inline_limit < extent_end)) {
 663                                 u32 new_size;
 664                                 new_size = btrfs_file_extent_calc_inline_size(
 665                                                    inline_limit - key.offset);
 666                                 dec_i_blocks(inode, (extent_end - key.offset) -
 667                                         (inline_limit - key.offset));
 668                                 btrfs_truncate_item(trans, root, path,
 669                                                     new_size, 1);
 670                         }
 671                 }
 672                 /* delete the entire extent */
 673                 if (!keep) {
 674                         u64 disk_bytenr = 0;
 675                         u64 disk_num_bytes = 0;
 676                         u64 extent_num_bytes = 0;
 677                         u64 root_gen;
 678                         u64 root_owner;
 679
 680                         root_gen = btrfs_header_generation(leaf);
 681                         root_owner = btrfs_header_owner(leaf);
 682                         if (found_extent) {
 683                                 disk_bytenr =
 684                                       btrfs_file_extent_disk_bytenr(leaf,
 685                                                                      extent);
 686                                 disk_num_bytes =
 687                                       btrfs_file_extent_disk_num_bytes(leaf,
 688                                                                        extent);
 689                                 extent_num_bytes =
 690                                       btrfs_file_extent_num_bytes(leaf, extent);
 691                                 *hint_byte =
 692                                         btrfs_file_extent_disk_bytenr(leaf,
 693                                                                       extent);
 694                         }
 695                         ret = btrfs_del_item(trans, root, path);
 696                         /* TODO update progress marker and return */
 697                         BUG_ON(ret);
 698                         btrfs_release_path(root, path);
 699                         extent = NULL;
 700                         if (found_extent && disk_bytenr != 0) {
 701                                 dec_i_blocks(inode, extent_num_bytes);
 702                                 ret = btrfs_free_extent(trans, root,
 703                                                 disk_bytenr,
 704                                                 disk_num_bytes,
 705                                                 root_owner,
 706                                                 root_gen, inode->i_ino,
 707                                                 key.offset, 0);
 708                         }
 709
 710                         BUG_ON(ret);
 711                         if (!bookend && search_start >= end) {
 712                                 ret = 0;
 713                                 goto out;
 714                         }
 715                         if (!bookend)
 716                                 continue;
 717                 }
 718                 if (bookend && found_inline && start <= key.offset) {
 719                         u32 new_size;
 720                         new_size = btrfs_file_extent_calc_inline_size(
 721                                                    extent_end - end);
 722                         dec_i_blocks(inode, (extent_end - key.offset) -
 723                                         (extent_end - end));
 724                         btrfs_truncate_item(trans, root, path, new_size, 0);
 725                 }
 726                 /* create bookend, splitting the extent in two */
 727                 if (bookend && found_extent) {
 728                         struct btrfs_key ins;
 729                         ins.objectid = inode->i_ino;
 730                         ins.offset = end;
 731                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 732                         btrfs_release_path(root, path);
 733                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 734                                                       sizeof(*extent));
 735
 736                         leaf = path->nodes[0];
 737                         if (ret) {
 738                                 btrfs_print_leaf(root, leaf);
 739                                 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);
 740                         }
 741                         BUG_ON(ret);
 742                         extent = btrfs_item_ptr(leaf, path->slots[0],
 743                                                 struct btrfs_file_extent_item);
 744                         write_extent_buffer(leaf, &old,
 745                                             (unsigned long)extent, sizeof(old));
 746
 747                         btrfs_set_file_extent_offset(leaf, extent,
 748                                     le64_to_cpu(old.offset) + end - key.offset);
 749                         WARN_ON(le64_to_cpu(old.num_bytes) <
 750                                 (extent_end - end));
 751                         btrfs_set_file_extent_num_bytes(leaf, extent,
 752                                                         extent_end - end);
 753                         btrfs_set_file_extent_type(leaf, extent,
 754                                                    BTRFS_FILE_EXTENT_REG);
 755
 756                         btrfs_mark_buffer_dirty(path->nodes[0]);
 757                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 758                                 inode->i_blocks +=
 759                                       btrfs_file_extent_num_bytes(leaf,
 760                                                                   extent) >> 9;
 761                         }
 762                         ret = 0;
 763                         goto out;
 764                 }
 765         }
 766 out:
 767         btrfs_free_path(path);
 768         btrfs_check_file(root, inode);
 769         return ret;
 770 }
 771
 772 /*
 773  * this gets pages into the page cache and locks them down
 774  */
 775 static int prepare_pages(struct btrfs_root *root, struct file *file,
 776                          struct page **pages, size_t num_pages,
 777                          loff_t pos, unsigned long first_index,
 778                          unsigned long last_index, size_t write_bytes)
 779 {
 780         int i;
 781         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 782         struct inode *inode = fdentry(file)->d_inode;
 783         int err = 0;
 784         u64 start_pos;
 785         u64 last_pos;
 786
 787         start_pos = pos & ~((u64)root->sectorsize - 1);
 788         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 789
 790         memset(pages, 0, num_pages * sizeof(struct page *));
 791 again:
 792         for (i = 0; i < num_pages; i++) {
 793                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 794                 if (!pages[i]) {
 795                         err = -ENOMEM;
 796                         BUG_ON(1);
 797                 }
 798                 wait_on_page_writeback(pages[i]);
 799         }
 800         if (start_pos < inode->i_size) {
 801                 struct btrfs_ordered_extent *ordered;
 802                 lock_extent(&BTRFS_I(inode)->io_tree,
 803                             start_pos, last_pos - 1, GFP_NOFS);
 804                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 805                 if (ordered &&
 806                     ordered->file_offset + ordered->len > start_pos &&
 807                     ordered->file_offset < last_pos) {
 808                         btrfs_put_ordered_extent(ordered);
 809                         unlock_extent(&BTRFS_I(inode)->io_tree,
 810                                       start_pos, last_pos - 1, GFP_NOFS);
 811                         for (i = 0; i < num_pages; i++) {
 812                                 unlock_page(pages[i]);
 813                                 page_cache_release(pages[i]);
 814                         }
 815                         btrfs_wait_ordered_range(inode, start_pos,
 816                                                  last_pos - start_pos);
 817                         goto again;
 818                 }
 819                 if (ordered)
 820                         btrfs_put_ordered_extent(ordered);
 821
 822                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 823                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 824                                   GFP_NOFS);
 825                 unlock_extent(&BTRFS_I(inode)->io_tree,
 826                               start_pos, last_pos - 1, GFP_NOFS);
 827         }
 828         for (i = 0; i < num_pages; i++) {
 829 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 830                 ClearPageDirty(pages[i]);
 831 #else
 832                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 833 #endif
 834                 set_page_extent_mapped(pages[i]);
 835                 WARN_ON(!PageLocked(pages[i]));
 836         }
 837         return 0;
 838 }
 839
 840 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 841                                 size_t count, loff_t *ppos)
 842 {
 843         loff_t pos;
 844         loff_t start_pos;
 845         ssize_t num_written = 0;
 846         ssize_t err = 0;
 847         int ret = 0;
 848         struct inode *inode = fdentry(file)->d_inode;
 849         struct btrfs_root *root = BTRFS_I(inode)->root;
 850         struct page **pages = NULL;
 851         int nrptrs;
 852         struct page *pinned[2];
 853         unsigned long first_index;
 854         unsigned long last_index;
 855
 856         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 857                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 858         pinned[0] = NULL;
 859         pinned[1] = NULL;
 860
 861         pos = *ppos;
 862         start_pos = pos;
 863
 864         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 865         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 866         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 867         if (err)
 868                 goto out_nolock;
 869         if (count == 0)
 870                 goto out_nolock;
 871 #ifdef REMOVE_SUID_PATH
 872         err = remove_suid(&file->f_path);
 873 #else
 874         err = remove_suid(fdentry(file));
 875 #endif
 876         if (err)
 877                 goto out_nolock;
 878         file_update_time(file);
 879
 880         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 881
 882         mutex_lock(&inode->i_mutex);
 883         first_index = pos >> PAGE_CACHE_SHIFT;
 884         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 885
 886         /*
 887          * if this is a nodatasum mount, force summing off for the inode
 888          * all the time.  That way a later mount with summing on won't
 889          * get confused
 890          */
 891         if (btrfs_test_opt(root, NODATASUM))
 892                 btrfs_set_flag(inode, NODATASUM);
 893
 894         /*
 895          * there are lots of better ways to do this, but this code
 896          * makes sure the first and last page in the file range are
 897          * up to date and ready for cow
 898          */
 899         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 900                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 901                 if (!PageUptodate(pinned[0])) {
 902                         ret = btrfs_readpage(NULL, pinned[0]);
 903                         BUG_ON(ret);
 904                         wait_on_page_locked(pinned[0]);
 905                 } else {
 906                         unlock_page(pinned[0]);
 907                 }
 908         }
 909         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 910                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 911                 if (!PageUptodate(pinned[1])) {
 912                         ret = btrfs_readpage(NULL, pinned[1]);
 913                         BUG_ON(ret);
 914                         wait_on_page_locked(pinned[1]);
 915                 } else {
 916                         unlock_page(pinned[1]);
 917                 }
 918         }
 919
 920         while(count > 0) {
 921                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 922                 size_t write_bytes = min(count, nrptrs *
 923                                         (size_t)PAGE_CACHE_SIZE -
 924                                          offset);
 925                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 926                                         PAGE_CACHE_SHIFT;
 927
 928                 WARN_ON(num_pages > nrptrs);
 929                 memset(pages, 0, sizeof(pages));
 930
 931                 ret = btrfs_check_free_space(root, write_bytes, 0);
 932                 if (ret)
 933                         goto out;
 934
 935                 ret = prepare_pages(root, file, pages, num_pages,
 936                                     pos, first_index, last_index,
 937                                     write_bytes);
 938                 if (ret)
 939                         goto out;
 940
 941                 ret = btrfs_copy_from_user(pos, num_pages,
 942                                            write_bytes, pages, buf);
 943                 if (ret) {
 944                         btrfs_drop_pages(pages, num_pages);
 945                         goto out;
 946                 }
 947
 948                 ret = dirty_and_release_pages(NULL, root, file, pages,
 949                                               num_pages, pos, write_bytes);
 950                 btrfs_drop_pages(pages, num_pages);
 951                 if (ret)
 952                         goto out;
 953
 954                 buf += write_bytes;
 955                 count -= write_bytes;
 956                 pos += write_bytes;
 957                 num_written += write_bytes;
 958
 959                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 960                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 961                         btrfs_btree_balance_dirty(root, 1);
 962                 cond_resched();
 963         }
 964 out:
 965         mutex_unlock(&inode->i_mutex);
 966
 967 out_nolock:
 968         kfree(pages);
 969         if (pinned[0])
 970                 page_cache_release(pinned[0]);
 971         if (pinned[1])
 972                 page_cache_release(pinned[1]);
 973         *ppos = pos;
 974
 975         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 976                 err = sync_page_range(inode, inode->i_mapping,
 977                                       start_pos, num_written);
 978                 if (err < 0)
 979                         num_written = err;
 980         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
 981 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
 982                 do_sync_file_range(file, start_pos,
 983                                       start_pos + num_written - 1,
 984                                       SYNC_FILE_RANGE_WRITE |
 985                                       SYNC_FILE_RANGE_WAIT_AFTER);
 986 #else
 987                 do_sync_mapping_range(inode->i_mapping, start_pos,
 988                                       start_pos + num_written - 1,
 989                                       SYNC_FILE_RANGE_WRITE |
 990                                       SYNC_FILE_RANGE_WAIT_AFTER);
 991 #endif
 992                 invalidate_mapping_pages(inode->i_mapping,
 993                       start_pos >> PAGE_CACHE_SHIFT,
 994                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 995         }
 996         current->backing_dev_info = NULL;
 997         return num_written ? num_written : err;
 998 }
 999
1000 int btrfs_release_file(struct inode * inode, struct file * filp)
1001 {
1002         if (filp->private_data)
1003                 btrfs_ioctl_trans_end(filp);
1004         return 0;
1005 }
1006
1007 static int btrfs_sync_file(struct file *file,
1008                            struct dentry *dentry, int datasync)
1009 {
1010         struct inode *inode = dentry->d_inode;
1011         struct btrfs_root *root = BTRFS_I(inode)->root;
1012         int ret = 0;
1013         struct btrfs_trans_handle *trans;
1014
1015         /*
1016          * check the transaction that last modified this inode
1017          * and see if its already been committed
1018          */
1019         if (!BTRFS_I(inode)->last_trans)
1020                 goto out;
1021
1022         mutex_lock(&root->fs_info->trans_mutex);
1023         if (BTRFS_I(inode)->last_trans <=
1024             root->fs_info->last_trans_committed) {
1025                 BTRFS_I(inode)->last_trans = 0;
1026                 mutex_unlock(&root->fs_info->trans_mutex);
1027                 goto out;
1028         }
1029         mutex_unlock(&root->fs_info->trans_mutex);
1030
1031         /*
1032          * ok we haven't committed the transaction yet, lets do a commit
1033          */
1034         if (file->private_data)
1035                 btrfs_ioctl_trans_end(file);
1036
1037         trans = btrfs_start_transaction(root, 1);
1038         if (!trans) {
1039                 ret = -ENOMEM;
1040                 goto out;
1041         }
1042         ret = btrfs_commit_transaction(trans, root);
1043 out:
1044         return ret > 0 ? EIO : ret;
1045 }
1046
1047 static struct vm_operations_struct btrfs_file_vm_ops = {
1048 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1049         .nopage         = filemap_nopage,
1050         .populate       = filemap_populate,
1051 #else
1052         .fault          = filemap_fault,
1053 #endif
1054         .page_mkwrite   = btrfs_page_mkwrite,
1055 };
1056
1057 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1058 {
1059         vma->vm_ops = &btrfs_file_vm_ops;
1060         file_accessed(filp);
1061         return 0;
1062 }
1063
1064 struct file_operations btrfs_file_operations = {
1065         .llseek         = generic_file_llseek,
1066         .read           = do_sync_read,
1067         .aio_read       = generic_file_aio_read,
1068         .splice_read    = generic_file_splice_read,
1069 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1070         .sendfile       = generic_file_sendfile,
1071 #endif
1072         .write          = btrfs_file_write,
1073         .mmap           = btrfs_file_mmap,
1074         .open           = generic_file_open,
1075         .release        = btrfs_release_file,
1076         .fsync          = btrfs_sync_file,
1077         .unlocked_ioctl = btrfs_ioctl,
1078 #ifdef CONFIG_COMPAT
1079         .compat_ioctl   = btrfs_ioctl,
1080 #endif
1081 };
1082