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 "tree-log.h"
  40 #include "locking.h"
  41 #include "compat.h"
  42
  43
  44 static int noinline btrfs_copy_from_user(loff_t pos, int num_pages,
  45                                          int write_bytes,
  46                                          struct page **prepared_pages,
  47                                          const char __user * buf)
  48 {
  49         long page_fault = 0;
  50         int i;
  51         int offset = pos & (PAGE_CACHE_SIZE - 1);
  52
  53         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  54                 size_t count = min_t(size_t,
  55                                      PAGE_CACHE_SIZE - offset, write_bytes);
  56                 struct page *page = prepared_pages[i];
  57                 fault_in_pages_readable(buf, count);
  58
  59                 /* Copy data from userspace to the current page */
  60                 kmap(page);
  61                 page_fault = __copy_from_user(page_address(page) + offset,
  62                                               buf, count);
  63                 /* Flush processor's dcache for this page */
  64                 flush_dcache_page(page);
  65                 kunmap(page);
  66                 buf += count;
  67                 write_bytes -= count;
  68
  69                 if (page_fault)
  70                         break;
  71         }
  72         return page_fault ? -EFAULT : 0;
  73 }
  74
  75 static void noinline btrfs_drop_pages(struct page **pages, size_t num_pages)
  76 {
  77         size_t i;
  78         for (i = 0; i < num_pages; i++) {
  79                 if (!pages[i])
  80                         break;
  81                 ClearPageChecked(pages[i]);
  82                 unlock_page(pages[i]);
  83                 mark_page_accessed(pages[i]);
  84                 page_cache_release(pages[i]);
  85         }
  86 }
  87
  88 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
  89                                 struct btrfs_root *root, struct inode *inode,
  90                                 u64 offset, size_t size,
  91                                 struct page **pages, size_t page_offset,
  92                                 int num_pages)
  93 {
  94         struct btrfs_key key;
  95         struct btrfs_path *path;
  96         struct extent_buffer *leaf;
  97         char *kaddr;
  98         unsigned long ptr;
  99         struct btrfs_file_extent_item *ei;
 100         struct page *page;
 101         u32 datasize;
 102         int err = 0;
 103         int ret;
 104         int i;
 105         ssize_t cur_size;
 106
 107         path = btrfs_alloc_path();
 108         if (!path)
 109                 return -ENOMEM;
 110
 111         btrfs_set_trans_block_group(trans, inode);
 112
 113         key.objectid = inode->i_ino;
 114         key.offset = offset;
 115         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 116
 117         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 118         if (ret < 0) {
 119                 err = ret;
 120                 goto fail;
 121         }
 122         if (ret == 1) {
 123                 struct btrfs_key found_key;
 124
 125                 if (path->slots[0] == 0)
 126                         goto insert;
 127
 128                 path->slots[0]--;
 129                 leaf = path->nodes[0];
 130                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 131
 132                 if (found_key.objectid != inode->i_ino)
 133                         goto insert;
 134
 135                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 136                         goto insert;
 137                 ei = btrfs_item_ptr(leaf, path->slots[0],
 138                                     struct btrfs_file_extent_item);
 139
 140                 if (btrfs_file_extent_type(leaf, ei) !=
 141                     BTRFS_FILE_EXTENT_INLINE) {
 142                         goto insert;
 143                 }
 144                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 145                 ret = 0;
 146         }
 147         if (ret == 0) {
 148                 u32 found_size;
 149                 u64 found_end;
 150
 151                 leaf = path->nodes[0];
 152                 ei = btrfs_item_ptr(leaf, path->slots[0],
 153                                     struct btrfs_file_extent_item);
 154
 155                 if (btrfs_file_extent_type(leaf, ei) !=
 156                     BTRFS_FILE_EXTENT_INLINE) {
 157                         err = ret;
 158                         btrfs_print_leaf(root, leaf);
 159                         printk("found wasn't inline offset %Lu inode %lu\n",
 160                                offset, inode->i_ino);
 161                         goto fail;
 162                 }
 163                 found_size = btrfs_file_extent_inline_len(leaf,
 164                                           btrfs_item_nr(leaf, path->slots[0]));
 165                 found_end = key.offset + found_size;
 166
 167                 if (found_end < offset + size) {
 168                         btrfs_release_path(root, path);
 169                         ret = btrfs_search_slot(trans, root, &key, path,
 170                                                 offset + size - found_end, 1);
 171                         BUG_ON(ret != 0);
 172
 173                         ret = btrfs_extend_item(trans, root, path,
 174                                                 offset + size - found_end);
 175                         if (ret) {
 176                                 err = ret;
 177                                 goto fail;
 178                         }
 179                         leaf = path->nodes[0];
 180                         ei = btrfs_item_ptr(leaf, path->slots[0],
 181                                             struct btrfs_file_extent_item);
 182                         inode->i_blocks += (offset + size - found_end) >> 9;
 183                 }
 184                 if (found_end < offset) {
 185                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 186                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 187                 }
 188         } else {
 189 insert:
 190                 btrfs_release_path(root, path);
 191                 datasize = offset + size - key.offset;
 192                 inode->i_blocks += datasize >> 9;
 193                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 194                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 195                                               datasize);
 196                 if (ret) {
 197                         err = ret;
 198                         printk("got bad ret %d\n", ret);
 199                         goto fail;
 200                 }
 201                 leaf = path->nodes[0];
 202                 ei = btrfs_item_ptr(leaf, path->slots[0],
 203                                     struct btrfs_file_extent_item);
 204                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 205                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 206         }
 207         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 208
 209         cur_size = size;
 210         i = 0;
 211         while (size > 0) {
 212                 page = pages[i];
 213                 kaddr = kmap_atomic(page, KM_USER0);
 214                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 215                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 216                 kunmap_atomic(kaddr, KM_USER0);
 217                 page_offset = 0;
 218                 ptr += cur_size;
 219                 size -= cur_size;
 220                 if (i >= num_pages) {
 221                         printk("i %d num_pages %d\n", i, num_pages);
 222                 }
 223                 i++;
 224         }
 225         btrfs_mark_buffer_dirty(leaf);
 226 fail:
 227         btrfs_free_path(path);
 228         return err;
 229 }
 230
 231 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 232                                    struct btrfs_root *root,
 233                                    struct file *file,
 234                                    struct page **pages,
 235                                    size_t num_pages,
 236                                    loff_t pos,
 237                                    size_t write_bytes)
 238 {
 239         int err = 0;
 240         int i;
 241         struct inode *inode = fdentry(file)->d_inode;
 242         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 243         u64 hint_byte;
 244         u64 num_bytes;
 245         u64 start_pos;
 246         u64 end_of_last_block;
 247         u64 end_pos = pos + write_bytes;
 248         u64 inline_size;
 249         int did_inline = 0;
 250         loff_t isize = i_size_read(inode);
 251
 252         start_pos = pos & ~((u64)root->sectorsize - 1);
 253         num_bytes = (write_bytes + pos - start_pos +
 254                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 255
 256         end_of_last_block = start_pos + num_bytes - 1;
 257
 258         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 259         trans = btrfs_join_transaction(root, 1);
 260         if (!trans) {
 261                 err = -ENOMEM;
 262                 goto out_unlock;
 263         }
 264         btrfs_set_trans_block_group(trans, inode);
 265         hint_byte = 0;
 266
 267         if ((end_of_last_block & 4095) == 0) {
 268                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 269         }
 270         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 271
 272         /* FIXME...EIEIO, ENOSPC and more */
 273         /* insert any holes we need to create */
 274         if (isize < start_pos) {
 275                 u64 last_pos_in_file;
 276                 u64 hole_size;
 277                 u64 mask = root->sectorsize - 1;
 278                 last_pos_in_file = (isize + mask) & ~mask;
 279                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 280                 if (hole_size > 0) {
 281                         btrfs_wait_ordered_range(inode, last_pos_in_file,
 282                                                  last_pos_in_file + hole_size);
 283                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 284                         err = btrfs_drop_extents(trans, root, inode,
 285                                                  last_pos_in_file,
 286                                                  last_pos_in_file + hole_size,
 287                                                  last_pos_in_file,
 288                                                  &hint_byte);
 289                         if (err)
 290                                 goto failed;
 291
 292                         err = btrfs_insert_file_extent(trans, root,
 293                                                        inode->i_ino,
 294                                                        last_pos_in_file,
 295                                                        0, 0, hole_size, 0);
 296                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 297                                         last_pos_in_file + hole_size -1);
 298                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 299                         btrfs_check_file(root, inode);
 300                 }
 301                 if (err)
 302                         goto failed;
 303         }
 304
 305         /*
 306          * either allocate an extent for the new bytes or setup the key
 307          * to show we are doing inline data in the extent
 308          */
 309         inline_size = end_pos;
 310         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 311             inline_size > root->fs_info->max_inline ||
 312             (inline_size & (root->sectorsize -1)) == 0 ||
 313             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 314                 /* check for reserved extents on each page, we don't want
 315                  * to reset the delalloc bit on things that already have
 316                  * extents reserved.
 317                  */
 318                 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 319                 for (i = 0; i < num_pages; i++) {
 320                         struct page *p = pages[i];
 321                         SetPageUptodate(p);
 322                         ClearPageChecked(p);
 323                         set_page_dirty(p);
 324                 }
 325         } else {
 326                 u64 aligned_end;
 327                 /* step one, delete the existing extents in this range */
 328                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 329                         ~((u64)root->sectorsize - 1);
 330                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 331                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 332                                          aligned_end, aligned_end, &hint_byte);
 333                 if (err)
 334                         goto failed;
 335                 if (isize > inline_size)
 336                         inline_size = min_t(u64, isize, aligned_end);
 337                 inline_size -= start_pos;
 338                 err = insert_inline_extent(trans, root, inode, start_pos,
 339                                            inline_size, pages, 0, num_pages);
 340                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 341                 BUG_ON(err);
 342                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 343
 344                 /*
 345                  * an ugly way to do all the prop accounting around
 346                  * the page bits and mapping tags
 347                  */
 348                 set_page_writeback(pages[0]);
 349                 end_page_writeback(pages[0]);
 350                 did_inline = 1;
 351         }
 352         if (end_pos > isize) {
 353                 i_size_write(inode, end_pos);
 354                 if (did_inline)
 355                         BTRFS_I(inode)->disk_i_size = end_pos;
 356                 btrfs_update_inode(trans, root, inode);
 357         }
 358 failed:
 359         err = btrfs_end_transaction(trans, root);
 360 out_unlock:
 361         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 362         return err;
 363 }
 364
 365 int noinline btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 366 {
 367         struct extent_map *em;
 368         struct extent_map *split = NULL;
 369         struct extent_map *split2 = NULL;
 370         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 371         u64 len = end - start + 1;
 372         int ret;
 373         int testend = 1;
 374
 375         WARN_ON(end < start);
 376         if (end == (u64)-1) {
 377                 len = (u64)-1;
 378                 testend = 0;
 379         }
 380         while(1) {
 381                 if (!split)
 382                         split = alloc_extent_map(GFP_NOFS);
 383                 if (!split2)
 384                         split2 = alloc_extent_map(GFP_NOFS);
 385
 386                 spin_lock(&em_tree->lock);
 387                 em = lookup_extent_mapping(em_tree, start, len);
 388                 if (!em) {
 389                         spin_unlock(&em_tree->lock);
 390                         break;
 391                 }
 392                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 393                 remove_extent_mapping(em_tree, em);
 394
 395                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 396                     em->start < start) {
 397                         split->start = em->start;
 398                         split->len = start - em->start;
 399                         split->block_start = em->block_start;
 400                         split->bdev = em->bdev;
 401                         split->flags = em->flags;
 402                         ret = add_extent_mapping(em_tree, split);
 403                         BUG_ON(ret);
 404                         free_extent_map(split);
 405                         split = split2;
 406                         split2 = NULL;
 407                 }
 408                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 409                     testend && em->start + em->len > start + len) {
 410                         u64 diff = start + len - em->start;
 411
 412                         split->start = start + len;
 413                         split->len = em->start + em->len - (start + len);
 414                         split->bdev = em->bdev;
 415                         split->flags = em->flags;
 416
 417                         split->block_start = em->block_start + diff;
 418
 419                         ret = add_extent_mapping(em_tree, split);
 420                         BUG_ON(ret);
 421                         free_extent_map(split);
 422                         split = NULL;
 423                 }
 424                 spin_unlock(&em_tree->lock);
 425
 426                 /* once for us */
 427                 free_extent_map(em);
 428                 /* once for the tree*/
 429                 free_extent_map(em);
 430         }
 431         if (split)
 432                 free_extent_map(split);
 433         if (split2)
 434                 free_extent_map(split2);
 435         return 0;
 436 }
 437
 438 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 439 {
 440         return 0;
 441 #if 0
 442         struct btrfs_path *path;
 443         struct btrfs_key found_key;
 444         struct extent_buffer *leaf;
 445         struct btrfs_file_extent_item *extent;
 446         u64 last_offset = 0;
 447         int nritems;
 448         int slot;
 449         int found_type;
 450         int ret;
 451         int err = 0;
 452         u64 extent_end = 0;
 453
 454         path = btrfs_alloc_path();
 455         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 456                                        last_offset, 0);
 457         while(1) {
 458                 nritems = btrfs_header_nritems(path->nodes[0]);
 459                 if (path->slots[0] >= nritems) {
 460                         ret = btrfs_next_leaf(root, path);
 461                         if (ret)
 462                                 goto out;
 463                         nritems = btrfs_header_nritems(path->nodes[0]);
 464                 }
 465                 slot = path->slots[0];
 466                 leaf = path->nodes[0];
 467                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 468                 if (found_key.objectid != inode->i_ino)
 469                         break;
 470                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 471                         goto out;
 472
 473                 if (found_key.offset < last_offset) {
 474                         WARN_ON(1);
 475                         btrfs_print_leaf(root, leaf);
 476                         printk("inode %lu found offset %Lu expected %Lu\n",
 477                                inode->i_ino, found_key.offset, last_offset);
 478                         err = 1;
 479                         goto out;
 480                 }
 481                 extent = btrfs_item_ptr(leaf, slot,
 482                                         struct btrfs_file_extent_item);
 483                 found_type = btrfs_file_extent_type(leaf, extent);
 484                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 485                         extent_end = found_key.offset +
 486                              btrfs_file_extent_num_bytes(leaf, extent);
 487                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 488                         struct btrfs_item *item;
 489                         item = btrfs_item_nr(leaf, slot);
 490                         extent_end = found_key.offset +
 491                              btrfs_file_extent_inline_len(leaf, item);
 492                         extent_end = (extent_end + root->sectorsize - 1) &
 493                                 ~((u64)root->sectorsize -1 );
 494                 }
 495                 last_offset = extent_end;
 496                 path->slots[0]++;
 497         }
 498         if (0 && last_offset < inode->i_size) {
 499                 WARN_ON(1);
 500                 btrfs_print_leaf(root, leaf);
 501                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 502                        last_offset, inode->i_size);
 503                 err = 1;
 504
 505         }
 506 out:
 507         btrfs_free_path(path);
 508         return err;
 509 #endif
 510 }
 511
 512 /*
 513  * this is very complex, but the basic idea is to drop all extents
 514  * in the range start - end.  hint_block is filled in with a block number
 515  * that would be a good hint to the block allocator for this file.
 516  *
 517  * If an extent intersects the range but is not entirely inside the range
 518  * it is either truncated or split.  Anything entirely inside the range
 519  * is deleted from the tree.
 520  */
 521 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 522                        struct btrfs_root *root, struct inode *inode,
 523                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 524 {
 525         u64 extent_end = 0;
 526         u64 search_start = start;
 527         struct extent_buffer *leaf;
 528         struct btrfs_file_extent_item *extent;
 529         struct btrfs_path *path;
 530         struct btrfs_key key;
 531         struct btrfs_file_extent_item old;
 532         int keep;
 533         int slot;
 534         int bookend;
 535         int found_type;
 536         int found_extent;
 537         int found_inline;
 538         int recow;
 539         int ret;
 540
 541         btrfs_drop_extent_cache(inode, start, end - 1);
 542
 543         path = btrfs_alloc_path();
 544         if (!path)
 545                 return -ENOMEM;
 546         while(1) {
 547                 recow = 0;
 548                 btrfs_release_path(root, path);
 549                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 550                                                search_start, -1);
 551                 if (ret < 0)
 552                         goto out;
 553                 if (ret > 0) {
 554                         if (path->slots[0] == 0) {
 555                                 ret = 0;
 556                                 goto out;
 557                         }
 558                         path->slots[0]--;
 559                 }
 560 next_slot:
 561                 keep = 0;
 562                 bookend = 0;
 563                 found_extent = 0;
 564                 found_inline = 0;
 565                 extent = NULL;
 566                 leaf = path->nodes[0];
 567                 slot = path->slots[0];
 568                 ret = 0;
 569                 btrfs_item_key_to_cpu(leaf, &key, slot);
 570                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 571                     key.offset >= end) {
 572                         goto out;
 573                 }
 574                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 575                     key.objectid != inode->i_ino) {
 576                         goto out;
 577                 }
 578                 if (recow) {
 579                         search_start = key.offset;
 580                         continue;
 581                 }
 582                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 583                         extent = btrfs_item_ptr(leaf, slot,
 584                                                 struct btrfs_file_extent_item);
 585                         found_type = btrfs_file_extent_type(leaf, extent);
 586                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 587                                 extent_end =
 588                                      btrfs_file_extent_disk_bytenr(leaf,
 589                                                                    extent);
 590                                 if (extent_end)
 591                                         *hint_byte = extent_end;
 592
 593                                 extent_end = key.offset +
 594                                      btrfs_file_extent_num_bytes(leaf, extent);
 595                                 found_extent = 1;
 596                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 597                                 struct btrfs_item *item;
 598                                 item = btrfs_item_nr(leaf, slot);
 599                                 found_inline = 1;
 600                                 extent_end = key.offset +
 601                                      btrfs_file_extent_inline_len(leaf, item);
 602                         }
 603                 } else {
 604                         extent_end = search_start;
 605                 }
 606
 607                 /* we found nothing we can drop */
 608                 if ((!found_extent && !found_inline) ||
 609                     search_start >= extent_end) {
 610                         int nextret;
 611                         u32 nritems;
 612                         nritems = btrfs_header_nritems(leaf);
 613                         if (slot >= nritems - 1) {
 614                                 nextret = btrfs_next_leaf(root, path);
 615                                 if (nextret)
 616                                         goto out;
 617                                 recow = 1;
 618                         } else {
 619                                 path->slots[0]++;
 620                         }
 621                         goto next_slot;
 622                 }
 623
 624                 if (found_inline) {
 625                         u64 mask = root->sectorsize - 1;
 626                         search_start = (extent_end + mask) & ~mask;
 627                 } else
 628                         search_start = extent_end;
 629                 if (end <= extent_end && start >= key.offset && found_inline) {
 630                         *hint_byte = EXTENT_MAP_INLINE;
 631                         continue;
 632                 }
 633                 if (end < extent_end && end >= key.offset) {
 634                         if (found_extent) {
 635                                 u64 disk_bytenr =
 636                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 637                                 u64 disk_num_bytes =
 638                                     btrfs_file_extent_disk_num_bytes(leaf,
 639                                                                       extent);
 640                                 read_extent_buffer(leaf, &old,
 641                                                    (unsigned long)extent,
 642                                                    sizeof(old));
 643                                 if (disk_bytenr != 0) {
 644                                         ret = btrfs_inc_extent_ref(trans, root,
 645                                                  disk_bytenr, disk_num_bytes,
 646                                                  root->root_key.objectid,
 647                                                  trans->transid,
 648                                                  key.objectid, end);
 649                                         BUG_ON(ret);
 650                                 }
 651                         }
 652                         bookend = 1;
 653                         if (found_inline && start <= key.offset)
 654                                 keep = 1;
 655                 }
 656                 /* truncate existing extent */
 657                 if (start > key.offset) {
 658                         u64 new_num;
 659                         u64 old_num;
 660                         keep = 1;
 661                         WARN_ON(start & (root->sectorsize - 1));
 662                         if (found_extent) {
 663                                 new_num = start - key.offset;
 664                                 old_num = btrfs_file_extent_num_bytes(leaf,
 665                                                                       extent);
 666                                 *hint_byte =
 667                                         btrfs_file_extent_disk_bytenr(leaf,
 668                                                                       extent);
 669                                 if (btrfs_file_extent_disk_bytenr(leaf,
 670                                                                   extent)) {
 671                                         dec_i_blocks(inode, old_num - new_num);
 672                                 }
 673                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 674                                                                 new_num);
 675                                 btrfs_mark_buffer_dirty(leaf);
 676                         } else if (key.offset < inline_limit &&
 677                                    (end > extent_end) &&
 678                                    (inline_limit < extent_end)) {
 679                                 u32 new_size;
 680                                 new_size = btrfs_file_extent_calc_inline_size(
 681                                                    inline_limit - key.offset);
 682                                 dec_i_blocks(inode, (extent_end - key.offset) -
 683                                         (inline_limit - key.offset));
 684                                 btrfs_truncate_item(trans, root, path,
 685                                                     new_size, 1);
 686                         }
 687                 }
 688                 /* delete the entire extent */
 689                 if (!keep) {
 690                         u64 disk_bytenr = 0;
 691                         u64 disk_num_bytes = 0;
 692                         u64 extent_num_bytes = 0;
 693                         u64 root_gen;
 694                         u64 root_owner;
 695
 696                         root_gen = btrfs_header_generation(leaf);
 697                         root_owner = btrfs_header_owner(leaf);
 698                         if (found_extent) {
 699                                 disk_bytenr =
 700                                       btrfs_file_extent_disk_bytenr(leaf,
 701                                                                      extent);
 702                                 disk_num_bytes =
 703                                       btrfs_file_extent_disk_num_bytes(leaf,
 704                                                                        extent);
 705                                 extent_num_bytes =
 706                                       btrfs_file_extent_num_bytes(leaf, extent);
 707                                 *hint_byte =
 708                                         btrfs_file_extent_disk_bytenr(leaf,
 709                                                                       extent);
 710                         }
 711                         ret = btrfs_del_item(trans, root, path);
 712                         /* TODO update progress marker and return */
 713                         BUG_ON(ret);
 714                         btrfs_release_path(root, path);
 715                         extent = NULL;
 716                         if (found_extent && disk_bytenr != 0) {
 717                                 dec_i_blocks(inode, extent_num_bytes);
 718                                 ret = btrfs_free_extent(trans, root,
 719                                                 disk_bytenr,
 720                                                 disk_num_bytes,
 721                                                 root_owner,
 722                                                 root_gen, inode->i_ino,
 723                                                 key.offset, 0);
 724                         }
 725
 726                         BUG_ON(ret);
 727                         if (!bookend && search_start >= end) {
 728                                 ret = 0;
 729                                 goto out;
 730                         }
 731                         if (!bookend)
 732                                 continue;
 733                 }
 734                 if (bookend && found_inline && start <= key.offset) {
 735                         u32 new_size;
 736                         new_size = btrfs_file_extent_calc_inline_size(
 737                                                    extent_end - end);
 738                         dec_i_blocks(inode, (extent_end - key.offset) -
 739                                         (extent_end - end));
 740                         btrfs_truncate_item(trans, root, path, new_size, 0);
 741                 }
 742                 /* create bookend, splitting the extent in two */
 743                 if (bookend && found_extent) {
 744                         struct btrfs_key ins;
 745                         ins.objectid = inode->i_ino;
 746                         ins.offset = end;
 747                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 748                         btrfs_release_path(root, path);
 749                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 750                                                       sizeof(*extent));
 751
 752                         leaf = path->nodes[0];
 753                         if (ret) {
 754                                 btrfs_print_leaf(root, leaf);
 755                                 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);
 756                         }
 757                         BUG_ON(ret);
 758                         extent = btrfs_item_ptr(leaf, path->slots[0],
 759                                                 struct btrfs_file_extent_item);
 760                         write_extent_buffer(leaf, &old,
 761                                             (unsigned long)extent, sizeof(old));
 762
 763                         btrfs_set_file_extent_offset(leaf, extent,
 764                                     le64_to_cpu(old.offset) + end - key.offset);
 765                         WARN_ON(le64_to_cpu(old.num_bytes) <
 766                                 (extent_end - end));
 767                         btrfs_set_file_extent_num_bytes(leaf, extent,
 768                                                         extent_end - end);
 769                         btrfs_set_file_extent_type(leaf, extent,
 770                                                    BTRFS_FILE_EXTENT_REG);
 771
 772                         btrfs_mark_buffer_dirty(path->nodes[0]);
 773                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 774                                 inode->i_blocks +=
 775                                       btrfs_file_extent_num_bytes(leaf,
 776                                                                   extent) >> 9;
 777                         }
 778                         ret = 0;
 779                         goto out;
 780                 }
 781         }
 782 out:
 783         btrfs_free_path(path);
 784         btrfs_check_file(root, inode);
 785         return ret;
 786 }
 787
 788 /*
 789  * this gets pages into the page cache and locks them down
 790  */
 791 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
 792                          struct page **pages, size_t num_pages,
 793                          loff_t pos, unsigned long first_index,
 794                          unsigned long last_index, size_t write_bytes)
 795 {
 796         int i;
 797         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 798         struct inode *inode = fdentry(file)->d_inode;
 799         int err = 0;
 800         u64 start_pos;
 801         u64 last_pos;
 802
 803         start_pos = pos & ~((u64)root->sectorsize - 1);
 804         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 805
 806         memset(pages, 0, num_pages * sizeof(struct page *));
 807 again:
 808         for (i = 0; i < num_pages; i++) {
 809                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 810                 if (!pages[i]) {
 811                         err = -ENOMEM;
 812                         BUG_ON(1);
 813                 }
 814                 wait_on_page_writeback(pages[i]);
 815         }
 816         if (start_pos < inode->i_size) {
 817                 struct btrfs_ordered_extent *ordered;
 818                 lock_extent(&BTRFS_I(inode)->io_tree,
 819                             start_pos, last_pos - 1, GFP_NOFS);
 820                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 821                 if (ordered &&
 822                     ordered->file_offset + ordered->len > start_pos &&
 823                     ordered->file_offset < last_pos) {
 824                         btrfs_put_ordered_extent(ordered);
 825                         unlock_extent(&BTRFS_I(inode)->io_tree,
 826                                       start_pos, last_pos - 1, GFP_NOFS);
 827                         for (i = 0; i < num_pages; i++) {
 828                                 unlock_page(pages[i]);
 829                                 page_cache_release(pages[i]);
 830                         }
 831                         btrfs_wait_ordered_range(inode, start_pos,
 832                                                  last_pos - start_pos);
 833                         goto again;
 834                 }
 835                 if (ordered)
 836                         btrfs_put_ordered_extent(ordered);
 837
 838                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 839                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 840                                   GFP_NOFS);
 841                 unlock_extent(&BTRFS_I(inode)->io_tree,
 842                               start_pos, last_pos - 1, GFP_NOFS);
 843         }
 844         for (i = 0; i < num_pages; i++) {
 845                 clear_page_dirty_for_io(pages[i]);
 846                 set_page_extent_mapped(pages[i]);
 847                 WARN_ON(!PageLocked(pages[i]));
 848         }
 849         return 0;
 850 }
 851
 852 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 853                                 size_t count, loff_t *ppos)
 854 {
 855         loff_t pos;
 856         loff_t start_pos;
 857         ssize_t num_written = 0;
 858         ssize_t err = 0;
 859         int ret = 0;
 860         struct inode *inode = fdentry(file)->d_inode;
 861         struct btrfs_root *root = BTRFS_I(inode)->root;
 862         struct page **pages = NULL;
 863         int nrptrs;
 864         struct page *pinned[2];
 865         unsigned long first_index;
 866         unsigned long last_index;
 867
 868         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 869                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 870         pinned[0] = NULL;
 871         pinned[1] = NULL;
 872
 873         pos = *ppos;
 874         start_pos = pos;
 875
 876         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 877         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 878         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 879         if (err)
 880                 goto out_nolock;
 881         if (count == 0)
 882                 goto out_nolock;
 883 #ifdef REMOVE_SUID_PATH
 884         err = remove_suid(&file->f_path);
 885 #else
 886 # if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
 887         err = file_remove_suid(file);
 888 # else
 889         err = remove_suid(fdentry(file));
 890 # endif
 891 #endif
 892         if (err)
 893                 goto out_nolock;
 894         file_update_time(file);
 895
 896         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 897
 898         mutex_lock(&inode->i_mutex);
 899         first_index = pos >> PAGE_CACHE_SHIFT;
 900         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 901
 902         /*
 903          * if this is a nodatasum mount, force summing off for the inode
 904          * all the time.  That way a later mount with summing on won't
 905          * get confused
 906          */
 907         if (btrfs_test_opt(root, NODATASUM))
 908                 btrfs_set_flag(inode, NODATASUM);
 909
 910         /*
 911          * there are lots of better ways to do this, but this code
 912          * makes sure the first and last page in the file range are
 913          * up to date and ready for cow
 914          */
 915         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 916                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 917                 if (!PageUptodate(pinned[0])) {
 918                         ret = btrfs_readpage(NULL, pinned[0]);
 919                         BUG_ON(ret);
 920                         wait_on_page_locked(pinned[0]);
 921                 } else {
 922                         unlock_page(pinned[0]);
 923                 }
 924         }
 925         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 926                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 927                 if (!PageUptodate(pinned[1])) {
 928                         ret = btrfs_readpage(NULL, pinned[1]);
 929                         BUG_ON(ret);
 930                         wait_on_page_locked(pinned[1]);
 931                 } else {
 932                         unlock_page(pinned[1]);
 933                 }
 934         }
 935
 936         while(count > 0) {
 937                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 938                 size_t write_bytes = min(count, nrptrs *
 939                                         (size_t)PAGE_CACHE_SIZE -
 940                                          offset);
 941                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 942                                         PAGE_CACHE_SHIFT;
 943
 944                 WARN_ON(num_pages > nrptrs);
 945                 memset(pages, 0, sizeof(pages));
 946
 947                 ret = btrfs_check_free_space(root, write_bytes, 0);
 948                 if (ret)
 949                         goto out;
 950
 951                 ret = prepare_pages(root, file, pages, num_pages,
 952                                     pos, first_index, last_index,
 953                                     write_bytes);
 954                 if (ret)
 955                         goto out;
 956
 957                 ret = btrfs_copy_from_user(pos, num_pages,
 958                                            write_bytes, pages, buf);
 959                 if (ret) {
 960                         btrfs_drop_pages(pages, num_pages);
 961                         goto out;
 962                 }
 963
 964                 ret = dirty_and_release_pages(NULL, root, file, pages,
 965                                               num_pages, pos, write_bytes);
 966                 btrfs_drop_pages(pages, num_pages);
 967                 if (ret)
 968                         goto out;
 969
 970                 buf += write_bytes;
 971                 count -= write_bytes;
 972                 pos += write_bytes;
 973                 num_written += write_bytes;
 974
 975                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 976                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 977                         btrfs_btree_balance_dirty(root, 1);
 978                 btrfs_throttle(root);
 979                 cond_resched();
 980         }
 981 out:
 982         mutex_unlock(&inode->i_mutex);
 983
 984 out_nolock:
 985         kfree(pages);
 986         if (pinned[0])
 987                 page_cache_release(pinned[0]);
 988         if (pinned[1])
 989                 page_cache_release(pinned[1]);
 990         *ppos = pos;
 991
 992         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 993                 struct btrfs_trans_handle *trans;
 994
 995                 err = btrfs_fdatawrite_range(inode->i_mapping, start_pos,
 996                                              start_pos + num_written -1,
 997                                              WB_SYNC_NONE);
 998                 if (err < 0)
 999                         num_written = err;
1000
1001                 err = btrfs_wait_on_page_writeback_range(inode->i_mapping,
1002                                  start_pos, start_pos + num_written - 1);
1003                 if (err < 0)
1004                         num_written = err;
1005
1006                 trans = btrfs_start_transaction(root, 1);
1007                 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1008                 if (ret == 0) {
1009                         btrfs_sync_log(trans, root);
1010                         btrfs_end_transaction(trans, root);
1011                 } else {
1012                         btrfs_commit_transaction(trans, root);
1013                 }
1014         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
1015 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
1016                 do_sync_file_range(file, start_pos,
1017                                       start_pos + num_written - 1,
1018                                       SYNC_FILE_RANGE_WRITE |
1019                                       SYNC_FILE_RANGE_WAIT_AFTER);
1020 #else
1021                 do_sync_mapping_range(inode->i_mapping, start_pos,
1022                                       start_pos + num_written - 1,
1023                                       SYNC_FILE_RANGE_WRITE |
1024                                       SYNC_FILE_RANGE_WAIT_AFTER);
1025 #endif
1026                 invalidate_mapping_pages(inode->i_mapping,
1027                       start_pos >> PAGE_CACHE_SHIFT,
1028                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1029         }
1030         current->backing_dev_info = NULL;
1031         return num_written ? num_written : err;
1032 }
1033
1034 int btrfs_release_file(struct inode * inode, struct file * filp)
1035 {
1036         if (filp->private_data)
1037                 btrfs_ioctl_trans_end(filp);
1038         return 0;
1039 }
1040
1041 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1042 {
1043         struct inode *inode = dentry->d_inode;
1044         struct btrfs_root *root = BTRFS_I(inode)->root;
1045         int ret = 0;
1046         struct btrfs_trans_handle *trans;
1047
1048         /*
1049          * check the transaction that last modified this inode
1050          * and see if its already been committed
1051          */
1052         if (!BTRFS_I(inode)->last_trans)
1053                 goto out;
1054
1055         mutex_lock(&root->fs_info->trans_mutex);
1056         if (BTRFS_I(inode)->last_trans <=
1057             root->fs_info->last_trans_committed) {
1058                 BTRFS_I(inode)->last_trans = 0;
1059                 mutex_unlock(&root->fs_info->trans_mutex);
1060                 goto out;
1061         }
1062         mutex_unlock(&root->fs_info->trans_mutex);
1063
1064         filemap_fdatawait(inode->i_mapping);
1065
1066         /*
1067          * ok we haven't committed the transaction yet, lets do a commit
1068          */
1069         if (file->private_data)
1070                 btrfs_ioctl_trans_end(file);
1071
1072         trans = btrfs_start_transaction(root, 1);
1073         if (!trans) {
1074                 ret = -ENOMEM;
1075                 goto out;
1076         }
1077
1078         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1079         if (ret < 0)
1080                 goto out;
1081         if (ret > 0) {
1082                 ret = btrfs_commit_transaction(trans, root);
1083         } else {
1084                 btrfs_sync_log(trans, root);
1085                 ret = btrfs_end_transaction(trans, root);
1086         }
1087 out:
1088         return ret > 0 ? EIO : ret;
1089 }
1090
1091 static struct vm_operations_struct btrfs_file_vm_ops = {
1092 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1093         .nopage         = filemap_nopage,
1094         .populate       = filemap_populate,
1095 #else
1096         .fault          = filemap_fault,
1097 #endif
1098         .page_mkwrite   = btrfs_page_mkwrite,
1099 };
1100
1101 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1102 {
1103         vma->vm_ops = &btrfs_file_vm_ops;
1104         file_accessed(filp);
1105         return 0;
1106 }
1107
1108 struct file_operations btrfs_file_operations = {
1109         .llseek         = generic_file_llseek,
1110         .read           = do_sync_read,
1111         .aio_read       = generic_file_aio_read,
1112         .splice_read    = generic_file_splice_read,
1113 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1114         .sendfile       = generic_file_sendfile,
1115 #endif
1116         .write          = btrfs_file_write,
1117         .mmap           = btrfs_file_mmap,
1118         .open           = generic_file_open,
1119         .release        = btrfs_release_file,
1120         .fsync          = btrfs_sync_file,
1121         .unlocked_ioctl = btrfs_ioctl,
1122 #ifdef CONFIG_COMPAT
1123         .compat_ioctl   = btrfs_ioctl,
1124 #endif
1125 };