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, 0);
 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, 0);
 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 btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 366                             int skip_pinned)
 367 {
 368         struct extent_map *em;
 369         struct extent_map *split = NULL;
 370         struct extent_map *split2 = NULL;
 371         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 372         u64 len = end - start + 1;
 373         int ret;
 374         int testend = 1;
 375         unsigned long flags;
 376
 377         WARN_ON(end < start);
 378         if (end == (u64)-1) {
 379                 len = (u64)-1;
 380                 testend = 0;
 381         }
 382         while(1) {
 383                 if (!split)
 384                         split = alloc_extent_map(GFP_NOFS);
 385                 if (!split2)
 386                         split2 = alloc_extent_map(GFP_NOFS);
 387
 388                 spin_lock(&em_tree->lock);
 389                 em = lookup_extent_mapping(em_tree, start, len);
 390                 if (!em) {
 391                         spin_unlock(&em_tree->lock);
 392                         break;
 393                 }
 394                 flags = em->flags;
 395                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 396                         spin_unlock(&em_tree->lock);
 397                         if (em->start <= start &&
 398                             (!testend || em->start + em->len >= start + len)) {
 399                                 free_extent_map(em);
 400                                 break;
 401                         }
 402                         if (start < em->start) {
 403                                 len = em->start - start;
 404                         } else {
 405                                 len = start + len - (em->start + em->len);
 406                                 start = em->start + em->len;
 407                         }
 408                         free_extent_map(em);
 409                         continue;
 410                 }
 411                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 412                 remove_extent_mapping(em_tree, em);
 413
 414                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 415                     em->start < start) {
 416                         split->start = em->start;
 417                         split->len = start - em->start;
 418                         split->block_start = em->block_start;
 419                         split->bdev = em->bdev;
 420                         split->flags = flags;
 421                         ret = add_extent_mapping(em_tree, split);
 422                         BUG_ON(ret);
 423                         free_extent_map(split);
 424                         split = split2;
 425                         split2 = NULL;
 426                 }
 427                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 428                     testend && em->start + em->len > start + len) {
 429                         u64 diff = start + len - em->start;
 430
 431                         split->start = start + len;
 432                         split->len = em->start + em->len - (start + len);
 433                         split->bdev = em->bdev;
 434                         split->flags = flags;
 435
 436                         split->block_start = em->block_start + diff;
 437
 438                         ret = add_extent_mapping(em_tree, split);
 439                         BUG_ON(ret);
 440                         free_extent_map(split);
 441                         split = NULL;
 442                 }
 443                 spin_unlock(&em_tree->lock);
 444
 445                 /* once for us */
 446                 free_extent_map(em);
 447                 /* once for the tree*/
 448                 free_extent_map(em);
 449         }
 450         if (split)
 451                 free_extent_map(split);
 452         if (split2)
 453                 free_extent_map(split2);
 454         return 0;
 455 }
 456
 457 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 458 {
 459         return 0;
 460 #if 0
 461         struct btrfs_path *path;
 462         struct btrfs_key found_key;
 463         struct extent_buffer *leaf;
 464         struct btrfs_file_extent_item *extent;
 465         u64 last_offset = 0;
 466         int nritems;
 467         int slot;
 468         int found_type;
 469         int ret;
 470         int err = 0;
 471         u64 extent_end = 0;
 472
 473         path = btrfs_alloc_path();
 474         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 475                                        last_offset, 0);
 476         while(1) {
 477                 nritems = btrfs_header_nritems(path->nodes[0]);
 478                 if (path->slots[0] >= nritems) {
 479                         ret = btrfs_next_leaf(root, path);
 480                         if (ret)
 481                                 goto out;
 482                         nritems = btrfs_header_nritems(path->nodes[0]);
 483                 }
 484                 slot = path->slots[0];
 485                 leaf = path->nodes[0];
 486                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 487                 if (found_key.objectid != inode->i_ino)
 488                         break;
 489                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 490                         goto out;
 491
 492                 if (found_key.offset < last_offset) {
 493                         WARN_ON(1);
 494                         btrfs_print_leaf(root, leaf);
 495                         printk("inode %lu found offset %Lu expected %Lu\n",
 496                                inode->i_ino, found_key.offset, last_offset);
 497                         err = 1;
 498                         goto out;
 499                 }
 500                 extent = btrfs_item_ptr(leaf, slot,
 501                                         struct btrfs_file_extent_item);
 502                 found_type = btrfs_file_extent_type(leaf, extent);
 503                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 504                         extent_end = found_key.offset +
 505                              btrfs_file_extent_num_bytes(leaf, extent);
 506                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 507                         struct btrfs_item *item;
 508                         item = btrfs_item_nr(leaf, slot);
 509                         extent_end = found_key.offset +
 510                              btrfs_file_extent_inline_len(leaf, item);
 511                         extent_end = (extent_end + root->sectorsize - 1) &
 512                                 ~((u64)root->sectorsize -1 );
 513                 }
 514                 last_offset = extent_end;
 515                 path->slots[0]++;
 516         }
 517         if (0 && last_offset < inode->i_size) {
 518                 WARN_ON(1);
 519                 btrfs_print_leaf(root, leaf);
 520                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 521                        last_offset, inode->i_size);
 522                 err = 1;
 523
 524         }
 525 out:
 526         btrfs_free_path(path);
 527         return err;
 528 #endif
 529 }
 530
 531 /*
 532  * this is very complex, but the basic idea is to drop all extents
 533  * in the range start - end.  hint_block is filled in with a block number
 534  * that would be a good hint to the block allocator for this file.
 535  *
 536  * If an extent intersects the range but is not entirely inside the range
 537  * it is either truncated or split.  Anything entirely inside the range
 538  * is deleted from the tree.
 539  */
 540 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 541                        struct btrfs_root *root, struct inode *inode,
 542                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 543 {
 544         u64 extent_end = 0;
 545         u64 search_start = start;
 546         u64 leaf_start;
 547         u64 root_gen;
 548         u64 root_owner;
 549         struct extent_buffer *leaf;
 550         struct btrfs_file_extent_item *extent;
 551         struct btrfs_path *path;
 552         struct btrfs_key key;
 553         struct btrfs_file_extent_item old;
 554         int keep;
 555         int slot;
 556         int bookend;
 557         int found_type;
 558         int found_extent;
 559         int found_inline;
 560         int recow;
 561         int ret;
 562
 563         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 564
 565         path = btrfs_alloc_path();
 566         if (!path)
 567                 return -ENOMEM;
 568         while(1) {
 569                 recow = 0;
 570                 btrfs_release_path(root, path);
 571                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 572                                                search_start, -1);
 573                 if (ret < 0)
 574                         goto out;
 575                 if (ret > 0) {
 576                         if (path->slots[0] == 0) {
 577                                 ret = 0;
 578                                 goto out;
 579                         }
 580                         path->slots[0]--;
 581                 }
 582 next_slot:
 583                 keep = 0;
 584                 bookend = 0;
 585                 found_extent = 0;
 586                 found_inline = 0;
 587                 leaf_start = 0;
 588                 root_gen = 0;
 589                 root_owner = 0;
 590                 extent = NULL;
 591                 leaf = path->nodes[0];
 592                 slot = path->slots[0];
 593                 ret = 0;
 594                 btrfs_item_key_to_cpu(leaf, &key, slot);
 595                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 596                     key.offset >= end) {
 597                         goto out;
 598                 }
 599                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 600                     key.objectid != inode->i_ino) {
 601                         goto out;
 602                 }
 603                 if (recow) {
 604                         search_start = key.offset;
 605                         continue;
 606                 }
 607                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 608                         extent = btrfs_item_ptr(leaf, slot,
 609                                                 struct btrfs_file_extent_item);
 610                         found_type = btrfs_file_extent_type(leaf, extent);
 611                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 612                                 extent_end =
 613                                      btrfs_file_extent_disk_bytenr(leaf,
 614                                                                    extent);
 615                                 if (extent_end)
 616                                         *hint_byte = extent_end;
 617
 618                                 extent_end = key.offset +
 619                                      btrfs_file_extent_num_bytes(leaf, extent);
 620                                 found_extent = 1;
 621                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 622                                 struct btrfs_item *item;
 623                                 item = btrfs_item_nr(leaf, slot);
 624                                 found_inline = 1;
 625                                 extent_end = key.offset +
 626                                      btrfs_file_extent_inline_len(leaf, item);
 627                         }
 628                 } else {
 629                         extent_end = search_start;
 630                 }
 631
 632                 /* we found nothing we can drop */
 633                 if ((!found_extent && !found_inline) ||
 634                     search_start >= extent_end) {
 635                         int nextret;
 636                         u32 nritems;
 637                         nritems = btrfs_header_nritems(leaf);
 638                         if (slot >= nritems - 1) {
 639                                 nextret = btrfs_next_leaf(root, path);
 640                                 if (nextret)
 641                                         goto out;
 642                                 recow = 1;
 643                         } else {
 644                                 path->slots[0]++;
 645                         }
 646                         goto next_slot;
 647                 }
 648
 649                 if (found_inline) {
 650                         u64 mask = root->sectorsize - 1;
 651                         search_start = (extent_end + mask) & ~mask;
 652                 } else
 653                         search_start = extent_end;
 654                 if (end <= extent_end && start >= key.offset && found_inline) {
 655                         *hint_byte = EXTENT_MAP_INLINE;
 656                         goto out;
 657                 }
 658
 659                 if (found_extent) {
 660                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 661                                            sizeof(old));
 662                         root_gen = btrfs_header_generation(leaf);
 663                         root_owner = btrfs_header_owner(leaf);
 664                         leaf_start = leaf->start;
 665                 }
 666
 667                 if (end < extent_end && end >= key.offset) {
 668                         bookend = 1;
 669                         if (found_inline && start <= key.offset)
 670                                 keep = 1;
 671                 }
 672                 /* truncate existing extent */
 673                 if (start > key.offset) {
 674                         u64 new_num;
 675                         u64 old_num;
 676                         keep = 1;
 677                         WARN_ON(start & (root->sectorsize - 1));
 678                         if (found_extent) {
 679                                 new_num = start - key.offset;
 680                                 old_num = btrfs_file_extent_num_bytes(leaf,
 681                                                                       extent);
 682                                 *hint_byte =
 683                                         btrfs_file_extent_disk_bytenr(leaf,
 684                                                                       extent);
 685                                 if (btrfs_file_extent_disk_bytenr(leaf,
 686                                                                   extent)) {
 687                                         dec_i_blocks(inode, old_num - new_num);
 688                                 }
 689                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 690                                                                 new_num);
 691                                 btrfs_mark_buffer_dirty(leaf);
 692                         } else if (key.offset < inline_limit &&
 693                                    (end > extent_end) &&
 694                                    (inline_limit < extent_end)) {
 695                                 u32 new_size;
 696                                 new_size = btrfs_file_extent_calc_inline_size(
 697                                                    inline_limit - key.offset);
 698                                 dec_i_blocks(inode, (extent_end - key.offset) -
 699                                         (inline_limit - key.offset));
 700                                 btrfs_truncate_item(trans, root, path,
 701                                                     new_size, 1);
 702                         }
 703                 }
 704                 /* delete the entire extent */
 705                 if (!keep) {
 706                         ret = btrfs_del_item(trans, root, path);
 707                         /* TODO update progress marker and return */
 708                         BUG_ON(ret);
 709                         extent = NULL;
 710                         btrfs_release_path(root, path);
 711                         /* the extent will be freed later */
 712                 }
 713                 if (bookend && found_inline && start <= key.offset) {
 714                         u32 new_size;
 715                         new_size = btrfs_file_extent_calc_inline_size(
 716                                                    extent_end - end);
 717                         dec_i_blocks(inode, (extent_end - key.offset) -
 718                                         (extent_end - end));
 719                         ret = btrfs_truncate_item(trans, root, path,
 720                                                   new_size, 0);
 721                         BUG_ON(ret);
 722                 }
 723                 /* create bookend, splitting the extent in two */
 724                 if (bookend && found_extent) {
 725                         u64 disk_bytenr;
 726                         struct btrfs_key ins;
 727                         ins.objectid = inode->i_ino;
 728                         ins.offset = end;
 729                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 730                         btrfs_release_path(root, path);
 731                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 732                                                       sizeof(*extent));
 733                         BUG_ON(ret);
 734
 735                         leaf = path->nodes[0];
 736                         extent = btrfs_item_ptr(leaf, path->slots[0],
 737                                                 struct btrfs_file_extent_item);
 738                         write_extent_buffer(leaf, &old,
 739                                             (unsigned long)extent, sizeof(old));
 740
 741                         btrfs_set_file_extent_offset(leaf, extent,
 742                                     le64_to_cpu(old.offset) + end - key.offset);
 743                         WARN_ON(le64_to_cpu(old.num_bytes) <
 744                                 (extent_end - end));
 745                         btrfs_set_file_extent_num_bytes(leaf, extent,
 746                                                         extent_end - end);
 747                         btrfs_set_file_extent_type(leaf, extent,
 748                                                    BTRFS_FILE_EXTENT_REG);
 749
 750                         btrfs_mark_buffer_dirty(path->nodes[0]);
 751
 752                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 753                         if (disk_bytenr != 0) {
 754                                 ret = btrfs_inc_extent_ref(trans, root,
 755                                                 disk_bytenr,
 756                                                 le64_to_cpu(old.disk_num_bytes),
 757                                                 leaf->start,
 758                                                 root->root_key.objectid,
 759                                                 trans->transid,
 760                                                 ins.objectid, ins.offset);
 761                                 BUG_ON(ret);
 762                         }
 763                         btrfs_release_path(root, path);
 764                         if (disk_bytenr != 0) {
 765                                 inode->i_blocks +=
 766                                       btrfs_file_extent_num_bytes(leaf,
 767                                                                   extent) >> 9;
 768                         }
 769                 }
 770
 771                 if (found_extent && !keep) {
 772                         u64 disk_bytenr = le64_to_cpu(old.disk_bytenr);
 773
 774                         if (disk_bytenr != 0) {
 775                                 dec_i_blocks(inode, le64_to_cpu(old.num_bytes));
 776                                 ret = btrfs_free_extent(trans, root,
 777                                                 disk_bytenr,
 778                                                 le64_to_cpu(old.disk_num_bytes),
 779                                                 leaf_start, root_owner,
 780                                                 root_gen, key.objectid,
 781                                                 key.offset, 0);
 782                                 BUG_ON(ret);
 783                                 *hint_byte = disk_bytenr;
 784                         }
 785                 }
 786
 787                 if (search_start >= end) {
 788                         ret = 0;
 789                         goto out;
 790                 }
 791         }
 792 out:
 793         btrfs_free_path(path);
 794         btrfs_check_file(root, inode);
 795         return ret;
 796 }
 797
 798 /*
 799  * this gets pages into the page cache and locks them down
 800  */
 801 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
 802                          struct page **pages, size_t num_pages,
 803                          loff_t pos, unsigned long first_index,
 804                          unsigned long last_index, size_t write_bytes)
 805 {
 806         int i;
 807         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 808         struct inode *inode = fdentry(file)->d_inode;
 809         int err = 0;
 810         u64 start_pos;
 811         u64 last_pos;
 812
 813         start_pos = pos & ~((u64)root->sectorsize - 1);
 814         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 815
 816         memset(pages, 0, num_pages * sizeof(struct page *));
 817 again:
 818         for (i = 0; i < num_pages; i++) {
 819                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 820                 if (!pages[i]) {
 821                         err = -ENOMEM;
 822                         BUG_ON(1);
 823                 }
 824                 wait_on_page_writeback(pages[i]);
 825         }
 826         if (start_pos < inode->i_size) {
 827                 struct btrfs_ordered_extent *ordered;
 828                 lock_extent(&BTRFS_I(inode)->io_tree,
 829                             start_pos, last_pos - 1, GFP_NOFS);
 830                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 831                 if (ordered &&
 832                     ordered->file_offset + ordered->len > start_pos &&
 833                     ordered->file_offset < last_pos) {
 834                         btrfs_put_ordered_extent(ordered);
 835                         unlock_extent(&BTRFS_I(inode)->io_tree,
 836                                       start_pos, last_pos - 1, GFP_NOFS);
 837                         for (i = 0; i < num_pages; i++) {
 838                                 unlock_page(pages[i]);
 839                                 page_cache_release(pages[i]);
 840                         }
 841                         btrfs_wait_ordered_range(inode, start_pos,
 842                                                  last_pos - start_pos);
 843                         goto again;
 844                 }
 845                 if (ordered)
 846                         btrfs_put_ordered_extent(ordered);
 847
 848                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 849                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 850                                   GFP_NOFS);
 851                 unlock_extent(&BTRFS_I(inode)->io_tree,
 852                               start_pos, last_pos - 1, GFP_NOFS);
 853         }
 854         for (i = 0; i < num_pages; i++) {
 855                 clear_page_dirty_for_io(pages[i]);
 856                 set_page_extent_mapped(pages[i]);
 857                 WARN_ON(!PageLocked(pages[i]));
 858         }
 859         return 0;
 860 }
 861
 862 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 863                                 size_t count, loff_t *ppos)
 864 {
 865         loff_t pos;
 866         loff_t start_pos;
 867         ssize_t num_written = 0;
 868         ssize_t err = 0;
 869         int ret = 0;
 870         struct inode *inode = fdentry(file)->d_inode;
 871         struct btrfs_root *root = BTRFS_I(inode)->root;
 872         struct page **pages = NULL;
 873         int nrptrs;
 874         struct page *pinned[2];
 875         unsigned long first_index;
 876         unsigned long last_index;
 877
 878         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 879                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 880         pinned[0] = NULL;
 881         pinned[1] = NULL;
 882
 883         pos = *ppos;
 884         start_pos = pos;
 885
 886         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 887         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 888         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 889         if (err)
 890                 goto out_nolock;
 891         if (count == 0)
 892                 goto out_nolock;
 893
 894         err = file_remove_suid(file);
 895         if (err)
 896                 goto out_nolock;
 897         file_update_time(file);
 898
 899         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 900
 901         mutex_lock(&inode->i_mutex);
 902         first_index = pos >> PAGE_CACHE_SHIFT;
 903         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 904
 905         /*
 906          * if this is a nodatasum mount, force summing off for the inode
 907          * all the time.  That way a later mount with summing on won't
 908          * get confused
 909          */
 910         if (btrfs_test_opt(root, NODATASUM))
 911                 btrfs_set_flag(inode, NODATASUM);
 912
 913         /*
 914          * there are lots of better ways to do this, but this code
 915          * makes sure the first and last page in the file range are
 916          * up to date and ready for cow
 917          */
 918         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 919                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 920                 if (!PageUptodate(pinned[0])) {
 921                         ret = btrfs_readpage(NULL, pinned[0]);
 922                         BUG_ON(ret);
 923                         wait_on_page_locked(pinned[0]);
 924                 } else {
 925                         unlock_page(pinned[0]);
 926                 }
 927         }
 928         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 929                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 930                 if (!PageUptodate(pinned[1])) {
 931                         ret = btrfs_readpage(NULL, pinned[1]);
 932                         BUG_ON(ret);
 933                         wait_on_page_locked(pinned[1]);
 934                 } else {
 935                         unlock_page(pinned[1]);
 936                 }
 937         }
 938
 939         while(count > 0) {
 940                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 941                 size_t write_bytes = min(count, nrptrs *
 942                                         (size_t)PAGE_CACHE_SIZE -
 943                                          offset);
 944                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 945                                         PAGE_CACHE_SHIFT;
 946
 947                 WARN_ON(num_pages > nrptrs);
 948                 memset(pages, 0, sizeof(pages));
 949
 950                 ret = btrfs_check_free_space(root, write_bytes, 0);
 951                 if (ret)
 952                         goto out;
 953
 954                 ret = prepare_pages(root, file, pages, num_pages,
 955                                     pos, first_index, last_index,
 956                                     write_bytes);
 957                 if (ret)
 958                         goto out;
 959
 960                 ret = btrfs_copy_from_user(pos, num_pages,
 961                                            write_bytes, pages, buf);
 962                 if (ret) {
 963                         btrfs_drop_pages(pages, num_pages);
 964                         goto out;
 965                 }
 966
 967                 ret = dirty_and_release_pages(NULL, root, file, pages,
 968                                               num_pages, pos, write_bytes);
 969                 btrfs_drop_pages(pages, num_pages);
 970                 if (ret)
 971                         goto out;
 972
 973                 buf += write_bytes;
 974                 count -= write_bytes;
 975                 pos += write_bytes;
 976                 num_written += write_bytes;
 977
 978                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 979                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 980                         btrfs_btree_balance_dirty(root, 1);
 981                 btrfs_throttle(root);
 982                 cond_resched();
 983         }
 984 out:
 985         mutex_unlock(&inode->i_mutex);
 986
 987 out_nolock:
 988         kfree(pages);
 989         if (pinned[0])
 990                 page_cache_release(pinned[0]);
 991         if (pinned[1])
 992                 page_cache_release(pinned[1]);
 993         *ppos = pos;
 994
 995         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 996                 struct btrfs_trans_handle *trans;
 997
 998                 err = btrfs_fdatawrite_range(inode->i_mapping, start_pos,
 999                                              start_pos + num_written -1,
1000                                              WB_SYNC_NONE);
1001                 if (err < 0)
1002                         num_written = err;
1003
1004                 err = btrfs_wait_on_page_writeback_range(inode->i_mapping,
1005                                  start_pos, start_pos + num_written - 1);
1006                 if (err < 0)
1007                         num_written = err;
1008
1009                 trans = btrfs_start_transaction(root, 1);
1010                 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1011                 if (ret == 0) {
1012                         btrfs_sync_log(trans, root);
1013                         btrfs_end_transaction(trans, root);
1014                 } else {
1015                         btrfs_commit_transaction(trans, root);
1016                 }
1017         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
1018                 do_sync_mapping_range(inode->i_mapping, start_pos,
1019                                       start_pos + num_written - 1,
1020                                       SYNC_FILE_RANGE_WRITE |
1021                                       SYNC_FILE_RANGE_WAIT_AFTER);
1022                 invalidate_mapping_pages(inode->i_mapping,
1023                       start_pos >> PAGE_CACHE_SHIFT,
1024                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1025         }
1026         current->backing_dev_info = NULL;
1027         return num_written ? num_written : err;
1028 }
1029
1030 int btrfs_release_file(struct inode * inode, struct file * filp)
1031 {
1032         if (filp->private_data)
1033                 btrfs_ioctl_trans_end(filp);
1034         return 0;
1035 }
1036
1037 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1038 {
1039         struct inode *inode = dentry->d_inode;
1040         struct btrfs_root *root = BTRFS_I(inode)->root;
1041         int ret = 0;
1042         struct btrfs_trans_handle *trans;
1043
1044         /*
1045          * check the transaction that last modified this inode
1046          * and see if its already been committed
1047          */
1048         if (!BTRFS_I(inode)->last_trans)
1049                 goto out;
1050
1051         mutex_lock(&root->fs_info->trans_mutex);
1052         if (BTRFS_I(inode)->last_trans <=
1053             root->fs_info->last_trans_committed) {
1054                 BTRFS_I(inode)->last_trans = 0;
1055                 mutex_unlock(&root->fs_info->trans_mutex);
1056                 goto out;
1057         }
1058         mutex_unlock(&root->fs_info->trans_mutex);
1059
1060         root->fs_info->tree_log_batch++;
1061         filemap_fdatawait(inode->i_mapping);
1062         root->fs_info->tree_log_batch++;
1063
1064         /*
1065          * ok we haven't committed the transaction yet, lets do a commit
1066          */
1067         if (file->private_data)
1068                 btrfs_ioctl_trans_end(file);
1069
1070         trans = btrfs_start_transaction(root, 1);
1071         if (!trans) {
1072                 ret = -ENOMEM;
1073                 goto out;
1074         }
1075
1076         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1077         if (ret < 0) {
1078                 goto out;
1079         }
1080
1081         /* we've logged all the items and now have a consistent
1082          * version of the file in the log.  It is possible that
1083          * someone will come in and modify the file, but that's
1084          * fine because the log is consistent on disk, and we
1085          * have references to all of the file's extents
1086          *
1087          * It is possible that someone will come in and log the
1088          * file again, but that will end up using the synchronization
1089          * inside btrfs_sync_log to keep things safe.
1090          */
1091         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1092
1093         if (ret > 0) {
1094                 ret = btrfs_commit_transaction(trans, root);
1095         } else {
1096                 btrfs_sync_log(trans, root);
1097                 ret = btrfs_end_transaction(trans, root);
1098         }
1099         mutex_lock(&file->f_dentry->d_inode->i_mutex);
1100 out:
1101         return ret > 0 ? EIO : ret;
1102 }
1103
1104 static struct vm_operations_struct btrfs_file_vm_ops = {
1105         .fault          = filemap_fault,
1106         .page_mkwrite   = btrfs_page_mkwrite,
1107 };
1108
1109 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1110 {
1111         vma->vm_ops = &btrfs_file_vm_ops;
1112         file_accessed(filp);
1113         return 0;
1114 }
1115
1116 struct file_operations btrfs_file_operations = {
1117         .llseek         = generic_file_llseek,
1118         .read           = do_sync_read,
1119         .aio_read       = generic_file_aio_read,
1120         .splice_read    = generic_file_splice_read,
1121         .write          = btrfs_file_write,
1122         .mmap           = btrfs_file_mmap,
1123         .open           = generic_file_open,
1124         .release        = btrfs_release_file,
1125         .fsync          = btrfs_sync_file,
1126         .unlocked_ioctl = btrfs_ioctl,
1127 #ifdef CONFIG_COMPAT
1128         .compat_ioctl   = btrfs_ioctl,
1129 #endif
1130 };