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                 ClearPageChecked(pages[i]);
  79                 unlock_page(pages[i]);
  80                 mark_page_accessed(pages[i]);
  81                 page_cache_release(pages[i]);
  82         }
  83 }
  84
  85 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
  86                                 struct btrfs_root *root, struct inode *inode,
  87                                 u64 offset, size_t size,
  88                                 struct page **pages, size_t page_offset,
  89                                 int num_pages)
  90 {
  91         struct btrfs_key key;
  92         struct btrfs_path *path;
  93         struct extent_buffer *leaf;
  94         char *kaddr;
  95         unsigned long ptr;
  96         struct btrfs_file_extent_item *ei;
  97         struct page *page;
  98         u32 datasize;
  99         int err = 0;
 100         int ret;
 101         int i;
 102         ssize_t cur_size;
 103
 104         path = btrfs_alloc_path();
 105         if (!path)
 106                 return -ENOMEM;
 107
 108         btrfs_set_trans_block_group(trans, inode);
 109
 110         key.objectid = inode->i_ino;
 111         key.offset = offset;
 112         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 113
 114         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 115         if (ret < 0) {
 116                 err = ret;
 117                 goto fail;
 118         }
 119         if (ret == 1) {
 120                 struct btrfs_key found_key;
 121
 122                 if (path->slots[0] == 0)
 123                         goto insert;
 124
 125                 path->slots[0]--;
 126                 leaf = path->nodes[0];
 127                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 128
 129                 if (found_key.objectid != inode->i_ino)
 130                         goto insert;
 131
 132                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 133                         goto insert;
 134                 ei = btrfs_item_ptr(leaf, path->slots[0],
 135                                     struct btrfs_file_extent_item);
 136
 137                 if (btrfs_file_extent_type(leaf, ei) !=
 138                     BTRFS_FILE_EXTENT_INLINE) {
 139                         goto insert;
 140                 }
 141                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 142                 ret = 0;
 143         }
 144         if (ret == 0) {
 145                 u32 found_size;
 146                 u64 found_end;
 147
 148                 leaf = path->nodes[0];
 149                 ei = btrfs_item_ptr(leaf, path->slots[0],
 150                                     struct btrfs_file_extent_item);
 151
 152                 if (btrfs_file_extent_type(leaf, ei) !=
 153                     BTRFS_FILE_EXTENT_INLINE) {
 154                         err = ret;
 155                         btrfs_print_leaf(root, leaf);
 156                         printk("found wasn't inline offset %Lu inode %lu\n",
 157                                offset, inode->i_ino);
 158                         goto fail;
 159                 }
 160                 found_size = btrfs_file_extent_inline_len(leaf,
 161                                           btrfs_item_nr(leaf, path->slots[0]));
 162                 found_end = key.offset + found_size;
 163
 164                 if (found_end < offset + size) {
 165                         btrfs_release_path(root, path);
 166                         ret = btrfs_search_slot(trans, root, &key, path,
 167                                                 offset + size - found_end, 1);
 168                         BUG_ON(ret != 0);
 169
 170                         ret = btrfs_extend_item(trans, root, path,
 171                                                 offset + size - found_end);
 172                         if (ret) {
 173                                 err = ret;
 174                                 goto fail;
 175                         }
 176                         leaf = path->nodes[0];
 177                         ei = btrfs_item_ptr(leaf, path->slots[0],
 178                                             struct btrfs_file_extent_item);
 179                         inode->i_blocks += (offset + size - found_end) >> 9;
 180                 }
 181                 if (found_end < offset) {
 182                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 183                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 184                 }
 185         } else {
 186 insert:
 187                 btrfs_release_path(root, path);
 188                 datasize = offset + size - key.offset;
 189                 inode->i_blocks += datasize >> 9;
 190                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 191                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 192                                               datasize);
 193                 if (ret) {
 194                         err = ret;
 195                         printk("got bad ret %d\n", ret);
 196                         goto fail;
 197                 }
 198                 leaf = path->nodes[0];
 199                 ei = btrfs_item_ptr(leaf, path->slots[0],
 200                                     struct btrfs_file_extent_item);
 201                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 202                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 203         }
 204         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 205
 206         cur_size = size;
 207         i = 0;
 208         while (size > 0) {
 209                 page = pages[i];
 210                 kaddr = kmap_atomic(page, KM_USER0);
 211                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 212                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 213                 kunmap_atomic(kaddr, KM_USER0);
 214                 page_offset = 0;
 215                 ptr += cur_size;
 216                 size -= cur_size;
 217                 if (i >= num_pages) {
 218                         printk("i %d num_pages %d\n", i, num_pages);
 219                 }
 220                 i++;
 221         }
 222         btrfs_mark_buffer_dirty(leaf);
 223 fail:
 224         btrfs_free_path(path);
 225         return err;
 226 }
 227
 228 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 229                                    struct btrfs_root *root,
 230                                    struct file *file,
 231                                    struct page **pages,
 232                                    size_t num_pages,
 233                                    loff_t pos,
 234                                    size_t write_bytes)
 235 {
 236         int err = 0;
 237         int i;
 238         struct inode *inode = fdentry(file)->d_inode;
 239         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 240         u64 hint_byte;
 241         u64 num_bytes;
 242         u64 start_pos;
 243         u64 end_of_last_block;
 244         u64 end_pos = pos + write_bytes;
 245         u64 inline_size;
 246         int did_inline = 0;
 247         loff_t isize = i_size_read(inode);
 248
 249         start_pos = pos & ~((u64)root->sectorsize - 1);
 250         num_bytes = (write_bytes + pos - start_pos +
 251                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 252
 253         end_of_last_block = start_pos + num_bytes - 1;
 254
 255         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 256         trans = btrfs_join_transaction(root, 1);
 257         if (!trans) {
 258                 err = -ENOMEM;
 259                 goto out_unlock;
 260         }
 261         btrfs_set_trans_block_group(trans, inode);
 262         hint_byte = 0;
 263
 264         if ((end_of_last_block & 4095) == 0) {
 265                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 266         }
 267         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 268
 269         /* FIXME...EIEIO, ENOSPC and more */
 270         /* insert any holes we need to create */
 271         if (isize < start_pos) {
 272                 u64 last_pos_in_file;
 273                 u64 hole_size;
 274                 u64 mask = root->sectorsize - 1;
 275                 last_pos_in_file = (isize + mask) & ~mask;
 276                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 277                 if (hole_size > 0) {
 278                         btrfs_wait_ordered_range(inode, last_pos_in_file,
 279                                                  last_pos_in_file + hole_size);
 280                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 281                         err = btrfs_drop_extents(trans, root, inode,
 282                                                  last_pos_in_file,
 283                                                  last_pos_in_file + hole_size,
 284                                                  last_pos_in_file,
 285                                                  &hint_byte);
 286                         if (err)
 287                                 goto failed;
 288
 289                         err = btrfs_insert_file_extent(trans, root,
 290                                                        inode->i_ino,
 291                                                        last_pos_in_file,
 292                                                        0, 0, hole_size, 0);
 293                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 294                                         last_pos_in_file + hole_size -1);
 295                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 296                         btrfs_check_file(root, inode);
 297                 }
 298                 if (err)
 299                         goto failed;
 300         }
 301
 302         /*
 303          * either allocate an extent for the new bytes or setup the key
 304          * to show we are doing inline data in the extent
 305          */
 306         inline_size = end_pos;
 307         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 308             inline_size > root->fs_info->max_inline ||
 309             (inline_size & (root->sectorsize -1)) == 0 ||
 310             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 311                 /* check for reserved extents on each page, we don't want
 312                  * to reset the delalloc bit on things that already have
 313                  * extents reserved.
 314                  */
 315                 set_extent_delalloc(io_tree, start_pos,
 316                                     end_of_last_block, GFP_NOFS);
 317                 for (i = 0; i < num_pages; i++) {
 318                         struct page *p = pages[i];
 319                         SetPageUptodate(p);
 320                         ClearPageChecked(p);
 321                         set_page_dirty(p);
 322                 }
 323         } else {
 324                 u64 aligned_end;
 325                 /* step one, delete the existing extents in this range */
 326                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 327                         ~((u64)root->sectorsize - 1);
 328                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 329                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 330                                          aligned_end, aligned_end, &hint_byte);
 331                 if (err)
 332                         goto failed;
 333                 if (isize > inline_size)
 334                         inline_size = min_t(u64, isize, aligned_end);
 335                 inline_size -= start_pos;
 336                 err = insert_inline_extent(trans, root, inode, start_pos,
 337                                            inline_size, pages, 0, num_pages);
 338                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 339                 BUG_ON(err);
 340                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 341                 did_inline = 1;
 342         }
 343         if (end_pos > isize) {
 344                 i_size_write(inode, end_pos);
 345                 if (did_inline)
 346                         BTRFS_I(inode)->disk_i_size = end_pos;
 347                 btrfs_update_inode(trans, root, inode);
 348         }
 349 failed:
 350         err = btrfs_end_transaction(trans, root);
 351 out_unlock:
 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 *split = NULL;
 360         struct extent_map *split2 = NULL;
 361         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 362         u64 len = end - start + 1;
 363         int ret;
 364         int testend = 1;
 365
 366         WARN_ON(end < start);
 367         if (end == (u64)-1) {
 368                 len = (u64)-1;
 369                 testend = 0;
 370         }
 371         while(1) {
 372                 if (!split)
 373                         split = alloc_extent_map(GFP_NOFS);
 374                 if (!split2)
 375                         split2 = alloc_extent_map(GFP_NOFS);
 376
 377                 spin_lock(&em_tree->lock);
 378                 em = lookup_extent_mapping(em_tree, start, len);
 379                 if (!em) {
 380                         spin_unlock(&em_tree->lock);
 381                         break;
 382                 }
 383                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 384                 remove_extent_mapping(em_tree, em);
 385
 386                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 387                     em->start < start) {
 388                         split->start = em->start;
 389                         split->len = start - em->start;
 390                         split->block_start = em->block_start;
 391                         split->bdev = em->bdev;
 392                         split->flags = em->flags;
 393                         ret = add_extent_mapping(em_tree, split);
 394                         BUG_ON(ret);
 395                         free_extent_map(split);
 396                         split = split2;
 397                         split2 = NULL;
 398                 }
 399                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 400                     testend && em->start + em->len > start + len) {
 401                         u64 diff = start + len - em->start;
 402
 403                         split->start = start + len;
 404                         split->len = em->start + em->len - (start + len);
 405                         split->bdev = em->bdev;
 406                         split->flags = em->flags;
 407
 408                         split->block_start = em->block_start + diff;
 409
 410                         ret = add_extent_mapping(em_tree, split);
 411                         BUG_ON(ret);
 412                         free_extent_map(split);
 413                         split = NULL;
 414                 }
 415                 spin_unlock(&em_tree->lock);
 416
 417                 /* once for us */
 418                 free_extent_map(em);
 419                 /* once for the tree*/
 420                 free_extent_map(em);
 421         }
 422         if (split)
 423                 free_extent_map(split);
 424         if (split2)
 425                 free_extent_map(split2);
 426         return 0;
 427 }
 428
 429 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 430 {
 431         return 0;
 432 #if 0
 433         struct btrfs_path *path;
 434         struct btrfs_key found_key;
 435         struct extent_buffer *leaf;
 436         struct btrfs_file_extent_item *extent;
 437         u64 last_offset = 0;
 438         int nritems;
 439         int slot;
 440         int found_type;
 441         int ret;
 442         int err = 0;
 443         u64 extent_end = 0;
 444
 445         path = btrfs_alloc_path();
 446         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 447                                        last_offset, 0);
 448         while(1) {
 449                 nritems = btrfs_header_nritems(path->nodes[0]);
 450                 if (path->slots[0] >= nritems) {
 451                         ret = btrfs_next_leaf(root, path);
 452                         if (ret)
 453                                 goto out;
 454                         nritems = btrfs_header_nritems(path->nodes[0]);
 455                 }
 456                 slot = path->slots[0];
 457                 leaf = path->nodes[0];
 458                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 459                 if (found_key.objectid != inode->i_ino)
 460                         break;
 461                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 462                         goto out;
 463
 464                 if (found_key.offset < last_offset) {
 465                         WARN_ON(1);
 466                         btrfs_print_leaf(root, leaf);
 467                         printk("inode %lu found offset %Lu expected %Lu\n",
 468                                inode->i_ino, found_key.offset, last_offset);
 469                         err = 1;
 470                         goto out;
 471                 }
 472                 extent = btrfs_item_ptr(leaf, slot,
 473                                         struct btrfs_file_extent_item);
 474                 found_type = btrfs_file_extent_type(leaf, extent);
 475                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 476                         extent_end = found_key.offset +
 477                              btrfs_file_extent_num_bytes(leaf, extent);
 478                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 479                         struct btrfs_item *item;
 480                         item = btrfs_item_nr(leaf, slot);
 481                         extent_end = found_key.offset +
 482                              btrfs_file_extent_inline_len(leaf, item);
 483                         extent_end = (extent_end + root->sectorsize - 1) &
 484                                 ~((u64)root->sectorsize -1 );
 485                 }
 486                 last_offset = extent_end;
 487                 path->slots[0]++;
 488         }
 489         if (0 && last_offset < inode->i_size) {
 490                 WARN_ON(1);
 491                 btrfs_print_leaf(root, leaf);
 492                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 493                        last_offset, inode->i_size);
 494                 err = 1;
 495
 496         }
 497 out:
 498         btrfs_free_path(path);
 499         return err;
 500 #endif
 501 }
 502
 503 /*
 504  * this is very complex, but the basic idea is to drop all extents
 505  * in the range start - end.  hint_block is filled in with a block number
 506  * that would be a good hint to the block allocator for this file.
 507  *
 508  * If an extent intersects the range but is not entirely inside the range
 509  * it is either truncated or split.  Anything entirely inside the range
 510  * is deleted from the tree.
 511  */
 512 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 513                        struct btrfs_root *root, struct inode *inode,
 514                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 515 {
 516         u64 extent_end = 0;
 517         u64 search_start = start;
 518         struct extent_buffer *leaf;
 519         struct btrfs_file_extent_item *extent;
 520         struct btrfs_path *path;
 521         struct btrfs_key key;
 522         struct btrfs_file_extent_item old;
 523         int keep;
 524         int slot;
 525         int bookend;
 526         int found_type;
 527         int found_extent;
 528         int found_inline;
 529         int recow;
 530         int ret;
 531
 532         btrfs_drop_extent_cache(inode, start, end - 1);
 533
 534         path = btrfs_alloc_path();
 535         if (!path)
 536                 return -ENOMEM;
 537         while(1) {
 538                 recow = 0;
 539                 btrfs_release_path(root, path);
 540                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 541                                                search_start, -1);
 542                 if (ret < 0)
 543                         goto out;
 544                 if (ret > 0) {
 545                         if (path->slots[0] == 0) {
 546                                 ret = 0;
 547                                 goto out;
 548                         }
 549                         path->slots[0]--;
 550                 }
 551 next_slot:
 552                 keep = 0;
 553                 bookend = 0;
 554                 found_extent = 0;
 555                 found_inline = 0;
 556                 extent = NULL;
 557                 leaf = path->nodes[0];
 558                 slot = path->slots[0];
 559                 ret = 0;
 560                 btrfs_item_key_to_cpu(leaf, &key, slot);
 561                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 562                     key.offset >= end) {
 563                         goto out;
 564                 }
 565                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 566                     key.objectid != inode->i_ino) {
 567                         goto out;
 568                 }
 569                 if (recow) {
 570                         search_start = key.offset;
 571                         continue;
 572                 }
 573                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 574                         extent = btrfs_item_ptr(leaf, slot,
 575                                                 struct btrfs_file_extent_item);
 576                         found_type = btrfs_file_extent_type(leaf, extent);
 577                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 578                                 extent_end =
 579                                      btrfs_file_extent_disk_bytenr(leaf,
 580                                                                    extent);
 581                                 if (extent_end)
 582                                         *hint_byte = extent_end;
 583
 584                                 extent_end = key.offset +
 585                                      btrfs_file_extent_num_bytes(leaf, extent);
 586                                 found_extent = 1;
 587                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 588                                 struct btrfs_item *item;
 589                                 item = btrfs_item_nr(leaf, slot);
 590                                 found_inline = 1;
 591                                 extent_end = key.offset +
 592                                      btrfs_file_extent_inline_len(leaf, item);
 593                         }
 594                 } else {
 595                         extent_end = search_start;
 596                 }
 597
 598                 /* we found nothing we can drop */
 599                 if ((!found_extent && !found_inline) ||
 600                     search_start >= extent_end) {
 601                         int nextret;
 602                         u32 nritems;
 603                         nritems = btrfs_header_nritems(leaf);
 604                         if (slot >= nritems - 1) {
 605                                 nextret = btrfs_next_leaf(root, path);
 606                                 if (nextret)
 607                                         goto out;
 608                                 recow = 1;
 609                         } else {
 610                                 path->slots[0]++;
 611                         }
 612                         goto next_slot;
 613                 }
 614
 615                 if (found_inline) {
 616                         u64 mask = root->sectorsize - 1;
 617                         search_start = (extent_end + mask) & ~mask;
 618                 } else
 619                         search_start = extent_end;
 620                 if (end <= extent_end && start >= key.offset && found_inline) {
 621                         *hint_byte = EXTENT_MAP_INLINE;
 622                         continue;
 623                 }
 624                 if (end < extent_end && end >= key.offset) {
 625                         if (found_extent) {
 626                                 u64 disk_bytenr =
 627                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 628                                 u64 disk_num_bytes =
 629                                     btrfs_file_extent_disk_num_bytes(leaf,
 630                                                                       extent);
 631                                 read_extent_buffer(leaf, &old,
 632                                                    (unsigned long)extent,
 633                                                    sizeof(old));
 634                                 if (disk_bytenr != 0) {
 635                                         ret = btrfs_inc_extent_ref(trans, root,
 636                                                  disk_bytenr, disk_num_bytes,
 637                                                  root->root_key.objectid,
 638                                                  trans->transid,
 639                                                  key.objectid, end);
 640                                         BUG_ON(ret);
 641                                 }
 642                         }
 643                         bookend = 1;
 644                         if (found_inline && start <= key.offset)
 645                                 keep = 1;
 646                 }
 647                 /* truncate existing extent */
 648                 if (start > key.offset) {
 649                         u64 new_num;
 650                         u64 old_num;
 651                         keep = 1;
 652                         WARN_ON(start & (root->sectorsize - 1));
 653                         if (found_extent) {
 654                                 new_num = start - key.offset;
 655                                 old_num = btrfs_file_extent_num_bytes(leaf,
 656                                                                       extent);
 657                                 *hint_byte =
 658                                         btrfs_file_extent_disk_bytenr(leaf,
 659                                                                       extent);
 660                                 if (btrfs_file_extent_disk_bytenr(leaf,
 661                                                                   extent)) {
 662                                         dec_i_blocks(inode, old_num - new_num);
 663                                 }
 664                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 665                                                                 new_num);
 666                                 btrfs_mark_buffer_dirty(leaf);
 667                         } else if (key.offset < inline_limit &&
 668                                    (end > extent_end) &&
 669                                    (inline_limit < extent_end)) {
 670                                 u32 new_size;
 671                                 new_size = btrfs_file_extent_calc_inline_size(
 672                                                    inline_limit - key.offset);
 673                                 dec_i_blocks(inode, (extent_end - key.offset) -
 674                                         (inline_limit - key.offset));
 675                                 btrfs_truncate_item(trans, root, path,
 676                                                     new_size, 1);
 677                         }
 678                 }
 679                 /* delete the entire extent */
 680                 if (!keep) {
 681                         u64 disk_bytenr = 0;
 682                         u64 disk_num_bytes = 0;
 683                         u64 extent_num_bytes = 0;
 684                         u64 root_gen;
 685                         u64 root_owner;
 686
 687                         root_gen = btrfs_header_generation(leaf);
 688                         root_owner = btrfs_header_owner(leaf);
 689                         if (found_extent) {
 690                                 disk_bytenr =
 691                                       btrfs_file_extent_disk_bytenr(leaf,
 692                                                                      extent);
 693                                 disk_num_bytes =
 694                                       btrfs_file_extent_disk_num_bytes(leaf,
 695                                                                        extent);
 696                                 extent_num_bytes =
 697                                       btrfs_file_extent_num_bytes(leaf, extent);
 698                                 *hint_byte =
 699                                         btrfs_file_extent_disk_bytenr(leaf,
 700                                                                       extent);
 701                         }
 702                         ret = btrfs_del_item(trans, root, path);
 703                         /* TODO update progress marker and return */
 704                         BUG_ON(ret);
 705                         btrfs_release_path(root, path);
 706                         extent = NULL;
 707                         if (found_extent && disk_bytenr != 0) {
 708                                 dec_i_blocks(inode, extent_num_bytes);
 709                                 ret = btrfs_free_extent(trans, root,
 710                                                 disk_bytenr,
 711                                                 disk_num_bytes,
 712                                                 root_owner,
 713                                                 root_gen, inode->i_ino,
 714                                                 key.offset, 0);
 715                         }
 716
 717                         BUG_ON(ret);
 718                         if (!bookend && search_start >= end) {
 719                                 ret = 0;
 720                                 goto out;
 721                         }
 722                         if (!bookend)
 723                                 continue;
 724                 }
 725                 if (bookend && found_inline && start <= key.offset) {
 726                         u32 new_size;
 727                         new_size = btrfs_file_extent_calc_inline_size(
 728                                                    extent_end - end);
 729                         dec_i_blocks(inode, (extent_end - key.offset) -
 730                                         (extent_end - end));
 731                         btrfs_truncate_item(trans, root, path, new_size, 0);
 732                 }
 733                 /* create bookend, splitting the extent in two */
 734                 if (bookend && found_extent) {
 735                         struct btrfs_key ins;
 736                         ins.objectid = inode->i_ino;
 737                         ins.offset = end;
 738                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 739                         btrfs_release_path(root, path);
 740                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 741                                                       sizeof(*extent));
 742
 743                         leaf = path->nodes[0];
 744                         if (ret) {
 745                                 btrfs_print_leaf(root, leaf);
 746                                 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);
 747                         }
 748                         BUG_ON(ret);
 749                         extent = btrfs_item_ptr(leaf, path->slots[0],
 750                                                 struct btrfs_file_extent_item);
 751                         write_extent_buffer(leaf, &old,
 752                                             (unsigned long)extent, sizeof(old));
 753
 754                         btrfs_set_file_extent_offset(leaf, extent,
 755                                     le64_to_cpu(old.offset) + end - key.offset);
 756                         WARN_ON(le64_to_cpu(old.num_bytes) <
 757                                 (extent_end - end));
 758                         btrfs_set_file_extent_num_bytes(leaf, extent,
 759                                                         extent_end - end);
 760                         btrfs_set_file_extent_type(leaf, extent,
 761                                                    BTRFS_FILE_EXTENT_REG);
 762
 763                         btrfs_mark_buffer_dirty(path->nodes[0]);
 764                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 765                                 inode->i_blocks +=
 766                                       btrfs_file_extent_num_bytes(leaf,
 767                                                                   extent) >> 9;
 768                         }
 769                         ret = 0;
 770                         goto out;
 771                 }
 772         }
 773 out:
 774         btrfs_free_path(path);
 775         btrfs_check_file(root, inode);
 776         return ret;
 777 }
 778
 779 /*
 780  * this gets pages into the page cache and locks them down
 781  */
 782 static int prepare_pages(struct btrfs_root *root, struct file *file,
 783                          struct page **pages, size_t num_pages,
 784                          loff_t pos, unsigned long first_index,
 785                          unsigned long last_index, size_t write_bytes)
 786 {
 787         int i;
 788         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 789         struct inode *inode = fdentry(file)->d_inode;
 790         int err = 0;
 791         u64 start_pos;
 792         u64 last_pos;
 793
 794         start_pos = pos & ~((u64)root->sectorsize - 1);
 795         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 796
 797         memset(pages, 0, num_pages * sizeof(struct page *));
 798 again:
 799         for (i = 0; i < num_pages; i++) {
 800                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 801                 if (!pages[i]) {
 802                         err = -ENOMEM;
 803                         BUG_ON(1);
 804                 }
 805                 wait_on_page_writeback(pages[i]);
 806         }
 807         if (start_pos < inode->i_size) {
 808                 struct btrfs_ordered_extent *ordered;
 809                 lock_extent(&BTRFS_I(inode)->io_tree,
 810                             start_pos, last_pos - 1, GFP_NOFS);
 811                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 812                 if (ordered &&
 813                     ordered->file_offset + ordered->len > start_pos &&
 814                     ordered->file_offset < last_pos) {
 815                         btrfs_put_ordered_extent(ordered);
 816                         unlock_extent(&BTRFS_I(inode)->io_tree,
 817                                       start_pos, last_pos - 1, GFP_NOFS);
 818                         for (i = 0; i < num_pages; i++) {
 819                                 unlock_page(pages[i]);
 820                                 page_cache_release(pages[i]);
 821                         }
 822                         btrfs_wait_ordered_range(inode, start_pos,
 823                                                  last_pos - start_pos);
 824                         goto again;
 825                 }
 826                 if (ordered)
 827                         btrfs_put_ordered_extent(ordered);
 828
 829                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 830                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 831                                   GFP_NOFS);
 832                 unlock_extent(&BTRFS_I(inode)->io_tree,
 833                               start_pos, last_pos - 1, GFP_NOFS);
 834         }
 835         for (i = 0; i < num_pages; i++) {
 836 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 837                 ClearPageDirty(pages[i]);
 838 #else
 839                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 840 #endif
 841                 set_page_extent_mapped(pages[i]);
 842                 WARN_ON(!PageLocked(pages[i]));
 843         }
 844         return 0;
 845 }
 846
 847 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 848                                 size_t count, loff_t *ppos)
 849 {
 850         loff_t pos;
 851         loff_t start_pos;
 852         ssize_t num_written = 0;
 853         ssize_t err = 0;
 854         int ret = 0;
 855         struct inode *inode = fdentry(file)->d_inode;
 856         struct btrfs_root *root = BTRFS_I(inode)->root;
 857         struct page **pages = NULL;
 858         int nrptrs;
 859         struct page *pinned[2];
 860         unsigned long first_index;
 861         unsigned long last_index;
 862
 863         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 864                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 865         pinned[0] = NULL;
 866         pinned[1] = NULL;
 867
 868         pos = *ppos;
 869         start_pos = pos;
 870
 871         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 872         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 873         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 874         if (err)
 875                 goto out_nolock;
 876         if (count == 0)
 877                 goto out_nolock;
 878 #ifdef REMOVE_SUID_PATH
 879         err = remove_suid(&file->f_path);
 880 #else
 881 # if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
 882         err = file_remove_suid(file);
 883 # else
 884         err = remove_suid(fdentry(file));
 885 # endif
 886 #endif
 887         if (err)
 888                 goto out_nolock;
 889         file_update_time(file);
 890
 891         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 892
 893         mutex_lock(&inode->i_mutex);
 894         first_index = pos >> PAGE_CACHE_SHIFT;
 895         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 896
 897         /*
 898          * if this is a nodatasum mount, force summing off for the inode
 899          * all the time.  That way a later mount with summing on won't
 900          * get confused
 901          */
 902         if (btrfs_test_opt(root, NODATASUM))
 903                 btrfs_set_flag(inode, NODATASUM);
 904
 905         /*
 906          * there are lots of better ways to do this, but this code
 907          * makes sure the first and last page in the file range are
 908          * up to date and ready for cow
 909          */
 910         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 911                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 912                 if (!PageUptodate(pinned[0])) {
 913                         ret = btrfs_readpage(NULL, pinned[0]);
 914                         BUG_ON(ret);
 915                         wait_on_page_locked(pinned[0]);
 916                 } else {
 917                         unlock_page(pinned[0]);
 918                 }
 919         }
 920         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 921                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 922                 if (!PageUptodate(pinned[1])) {
 923                         ret = btrfs_readpage(NULL, pinned[1]);
 924                         BUG_ON(ret);
 925                         wait_on_page_locked(pinned[1]);
 926                 } else {
 927                         unlock_page(pinned[1]);
 928                 }
 929         }
 930
 931         while(count > 0) {
 932                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 933                 size_t write_bytes = min(count, nrptrs *
 934                                         (size_t)PAGE_CACHE_SIZE -
 935                                          offset);
 936                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 937                                         PAGE_CACHE_SHIFT;
 938
 939                 WARN_ON(num_pages > nrptrs);
 940                 memset(pages, 0, sizeof(pages));
 941
 942                 ret = btrfs_check_free_space(root, write_bytes, 0);
 943                 if (ret)
 944                         goto out;
 945
 946                 ret = prepare_pages(root, file, pages, num_pages,
 947                                     pos, first_index, last_index,
 948                                     write_bytes);
 949                 if (ret)
 950                         goto out;
 951
 952                 ret = btrfs_copy_from_user(pos, num_pages,
 953                                            write_bytes, pages, buf);
 954                 if (ret) {
 955                         btrfs_drop_pages(pages, num_pages);
 956                         goto out;
 957                 }
 958
 959                 ret = dirty_and_release_pages(NULL, root, file, pages,
 960                                               num_pages, pos, write_bytes);
 961                 btrfs_drop_pages(pages, num_pages);
 962                 if (ret)
 963                         goto out;
 964
 965                 buf += write_bytes;
 966                 count -= write_bytes;
 967                 pos += write_bytes;
 968                 num_written += write_bytes;
 969
 970                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 971                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 972                         btrfs_btree_balance_dirty(root, 1);
 973                 btrfs_throttle(root);
 974                 cond_resched();
 975         }
 976 out:
 977         mutex_unlock(&inode->i_mutex);
 978
 979 out_nolock:
 980         kfree(pages);
 981         if (pinned[0])
 982                 page_cache_release(pinned[0]);
 983         if (pinned[1])
 984                 page_cache_release(pinned[1]);
 985         *ppos = pos;
 986
 987         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 988                 err = sync_page_range(inode, inode->i_mapping,
 989                                       start_pos, num_written);
 990                 if (err < 0)
 991                         num_written = err;
 992         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
 993 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
 994                 do_sync_file_range(file, start_pos,
 995                                       start_pos + num_written - 1,
 996                                       SYNC_FILE_RANGE_WRITE |
 997                                       SYNC_FILE_RANGE_WAIT_AFTER);
 998 #else
 999                 do_sync_mapping_range(inode->i_mapping, start_pos,
1000                                       start_pos + num_written - 1,
1001                                       SYNC_FILE_RANGE_WRITE |
1002                                       SYNC_FILE_RANGE_WAIT_AFTER);
1003 #endif
1004                 invalidate_mapping_pages(inode->i_mapping,
1005                       start_pos >> PAGE_CACHE_SHIFT,
1006                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1007         }
1008         current->backing_dev_info = NULL;
1009         return num_written ? num_written : err;
1010 }
1011
1012 int btrfs_release_file(struct inode * inode, struct file * filp)
1013 {
1014         if (filp->private_data)
1015                 btrfs_ioctl_trans_end(filp);
1016         return 0;
1017 }
1018
1019 static int btrfs_sync_file(struct file *file,
1020                            struct dentry *dentry, int datasync)
1021 {
1022         struct inode *inode = dentry->d_inode;
1023         struct btrfs_root *root = BTRFS_I(inode)->root;
1024         int ret = 0;
1025         struct btrfs_trans_handle *trans;
1026
1027         /*
1028          * check the transaction that last modified this inode
1029          * and see if its already been committed
1030          */
1031         if (!BTRFS_I(inode)->last_trans)
1032                 goto out;
1033
1034         mutex_lock(&root->fs_info->trans_mutex);
1035         if (BTRFS_I(inode)->last_trans <=
1036             root->fs_info->last_trans_committed) {
1037                 BTRFS_I(inode)->last_trans = 0;
1038                 mutex_unlock(&root->fs_info->trans_mutex);
1039                 goto out;
1040         }
1041         mutex_unlock(&root->fs_info->trans_mutex);
1042
1043         /*
1044          * ok we haven't committed the transaction yet, lets do a commit
1045          */
1046         if (file->private_data)
1047                 btrfs_ioctl_trans_end(file);
1048
1049         trans = btrfs_start_transaction(root, 1);
1050         if (!trans) {
1051                 ret = -ENOMEM;
1052                 goto out;
1053         }
1054         ret = btrfs_commit_transaction(trans, root);
1055 out:
1056         return ret > 0 ? EIO : ret;
1057 }
1058
1059 static struct vm_operations_struct btrfs_file_vm_ops = {
1060 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1061         .nopage         = filemap_nopage,
1062         .populate       = filemap_populate,
1063 #else
1064         .fault          = filemap_fault,
1065 #endif
1066         .page_mkwrite   = btrfs_page_mkwrite,
1067 };
1068
1069 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1070 {
1071         vma->vm_ops = &btrfs_file_vm_ops;
1072         file_accessed(filp);
1073         return 0;
1074 }
1075
1076 struct file_operations btrfs_file_operations = {
1077         .llseek         = generic_file_llseek,
1078         .read           = do_sync_read,
1079         .aio_read       = generic_file_aio_read,
1080         .splice_read    = generic_file_splice_read,
1081 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1082         .sendfile       = generic_file_sendfile,
1083 #endif
1084         .write          = btrfs_file_write,
1085         .mmap           = btrfs_file_mmap,
1086         .open           = generic_file_open,
1087         .release        = btrfs_release_file,
1088         .fsync          = btrfs_sync_file,
1089         .unlocked_ioctl = btrfs_ioctl,
1090 #ifdef CONFIG_COMPAT
1091         .compat_ioctl   = btrfs_ioctl,
1092 #endif
1093 };