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/module.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/fs.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/highmem.h>
  24 #include <linux/time.h>
  25 #include <linux/init.h>
  26 #include <linux/string.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/mpage.h>
  30 #include <linux/swap.h>
  31 #include <linux/writeback.h>
  32 #include <linux/statfs.h>
  33 #include <linux/compat.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "transaction.h"
  37 #include "btrfs_inode.h"
  38 #include "ioctl.h"
  39 #include "print-tree.h"
  40
  41
  42 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  43                                 struct page **prepared_pages,
  44                                 const char __user * buf)
  45 {
  46         long page_fault = 0;
  47         int i;
  48         int offset = pos & (PAGE_CACHE_SIZE - 1);
  49
  50         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  51                 size_t count = min_t(size_t,
  52                                      PAGE_CACHE_SIZE - offset, write_bytes);
  53                 struct page *page = prepared_pages[i];
  54                 fault_in_pages_readable(buf, count);
  55
  56                 /* Copy data from userspace to the current page */
  57                 kmap(page);
  58                 page_fault = __copy_from_user(page_address(page) + offset,
  59                                               buf, count);
  60                 /* Flush processor's dcache for this page */
  61                 flush_dcache_page(page);
  62                 kunmap(page);
  63                 buf += count;
  64                 write_bytes -= count;
  65
  66                 if (page_fault)
  67                         break;
  68         }
  69         return page_fault ? -EFAULT : 0;
  70 }
  71
  72 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  73 {
  74         size_t i;
  75         for (i = 0; i < num_pages; i++) {
  76                 if (!pages[i])
  77                         break;
  78                 unlock_page(pages[i]);
  79                 mark_page_accessed(pages[i]);
  80                 page_cache_release(pages[i]);
  81         }
  82 }
  83
  84 static int insert_inline_extent(struct btrfs_root *root, struct inode *inode,
  85                                 u64 offset, ssize_t size,
  86                                 struct buffer_head *bh)
  87 {
  88         struct btrfs_key key;
  89         struct btrfs_path *path;
  90         char *ptr, *kaddr;
  91         struct btrfs_trans_handle *trans;
  92         struct btrfs_file_extent_item *ei;
  93         u32 datasize;
  94         int err = 0;
  95         int ret;
  96
  97         path = btrfs_alloc_path();
  98         if (!path)
  99                 return -ENOMEM;
 100
 101         mutex_lock(&root->fs_info->fs_mutex);
 102         trans = btrfs_start_transaction(root, 1);
 103         btrfs_set_trans_block_group(trans, inode);
 104
 105         key.objectid = inode->i_ino;
 106         key.offset = offset;
 107         key.flags = 0;
 108         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 109         BUG_ON(size >= PAGE_CACHE_SIZE);
 110         datasize = btrfs_file_extent_calc_inline_size(size);
 111
 112         ret = btrfs_insert_empty_item(trans, root, path, &key,
 113                                       datasize);
 114         if (ret) {
 115                 err = ret;
 116                 goto fail;
 117         }
 118         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 119                path->slots[0], struct btrfs_file_extent_item);
 120         btrfs_set_file_extent_generation(ei, trans->transid);
 121         btrfs_set_file_extent_type(ei,
 122                                    BTRFS_FILE_EXTENT_INLINE);
 123         ptr = btrfs_file_extent_inline_start(ei);
 124
 125         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 126         btrfs_memcpy(root, path->nodes[0]->b_data,
 127                      ptr, kaddr + bh_offset(bh),
 128                      size);
 129         kunmap_atomic(kaddr, KM_USER0);
 130         mark_buffer_dirty(path->nodes[0]);
 131 fail:
 132         btrfs_free_path(path);
 133         ret = btrfs_end_transaction(trans, root);
 134         if (ret && !err)
 135                 err = ret;
 136         mutex_unlock(&root->fs_info->fs_mutex);
 137         return err;
 138 }
 139
 140 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 141                                    struct btrfs_root *root,
 142                                    struct file *file,
 143                                    struct page **pages,
 144                                    size_t num_pages,
 145                                    loff_t pos,
 146                                    size_t write_bytes)
 147 {
 148         int i;
 149         int offset;
 150         int err = 0;
 151         int ret;
 152         int this_write;
 153         struct inode *inode = file->f_path.dentry->d_inode;
 154         struct buffer_head *bh;
 155
 156         for (i = 0; i < num_pages; i++) {
 157                 offset = pos & (PAGE_CACHE_SIZE -1);
 158                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 159
 160                 /* FIXME, one block at a time */
 161                 bh = page_buffers(pages[i]);
 162
 163                 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 164                         ret = insert_inline_extent(root, inode,
 165                                         pages[i]->index << PAGE_CACHE_SHIFT,
 166                                         offset + this_write, bh);
 167                         if (ret) {
 168                                 err = ret;
 169                                 goto failed;
 170                         }
 171                 }
 172
 173                 ret = btrfs_commit_write(file, pages[i], offset,
 174                                          offset + this_write);
 175                 pos += this_write;
 176                 if (ret) {
 177                         err = ret;
 178                         goto failed;
 179                 }
 180                 WARN_ON(this_write > write_bytes);
 181                 write_bytes -= this_write;
 182         }
 183 failed:
 184         return err;
 185 }
 186
 187 /*
 188  * this is very complex, but the basic idea is to drop all extents
 189  * in the range start - end.  hint_block is filled in with a block number
 190  * that would be a good hint to the block allocator for this file.
 191  *
 192  * If an extent intersects the range but is not entirely inside the range
 193  * it is either truncated or split.  Anything entirely inside the range
 194  * is deleted from the tree.
 195  */
 196 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 197                        struct btrfs_root *root, struct inode *inode,
 198                        u64 start, u64 end, u64 *hint_block)
 199 {
 200         int ret;
 201         struct btrfs_key key;
 202         struct btrfs_leaf *leaf;
 203         int slot;
 204         struct btrfs_file_extent_item *extent;
 205         u64 extent_end = 0;
 206         int keep;
 207         struct btrfs_file_extent_item old;
 208         struct btrfs_path *path;
 209         u64 search_start = start;
 210         int bookend;
 211         int found_type;
 212         int found_extent;
 213         int found_inline;
 214
 215         path = btrfs_alloc_path();
 216         if (!path)
 217                 return -ENOMEM;
 218         while(1) {
 219                 btrfs_release_path(root, path);
 220                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 221                                                search_start, -1);
 222                 if (ret < 0)
 223                         goto out;
 224                 if (ret > 0) {
 225                         if (path->slots[0] == 0) {
 226                                 ret = 0;
 227                                 goto out;
 228                         }
 229                         path->slots[0]--;
 230                 }
 231 next_slot:
 232                 keep = 0;
 233                 bookend = 0;
 234                 found_extent = 0;
 235                 found_inline = 0;
 236                 extent = NULL;
 237                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 238                 slot = path->slots[0];
 239                 ret = 0;
 240                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
 241                 if (key.offset >= end || key.objectid != inode->i_ino) {
 242                         goto out;
 243                 }
 244                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 245                         goto out;
 246                 }
 247                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 248                         extent = btrfs_item_ptr(leaf, slot,
 249                                                 struct btrfs_file_extent_item);
 250                         found_type = btrfs_file_extent_type(extent);
 251                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 252                                 extent_end = key.offset +
 253                                         (btrfs_file_extent_num_blocks(extent) <<
 254                                          inode->i_blkbits);
 255                                 found_extent = 1;
 256                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 257                                 found_inline = 1;
 258                                 extent_end = key.offset +
 259                                      btrfs_file_extent_inline_len(leaf->items +
 260                                                                   slot);
 261                         }
 262                 } else {
 263                         extent_end = search_start;
 264                 }
 265
 266                 /* we found nothing we can drop */
 267                 if ((!found_extent && !found_inline) ||
 268                     search_start >= extent_end) {
 269                         int nextret;
 270                         u32 nritems;
 271                         nritems = btrfs_header_nritems(
 272                                         btrfs_buffer_header(path->nodes[0]));
 273                         if (slot >= nritems - 1) {
 274                                 nextret = btrfs_next_leaf(root, path);
 275                                 if (nextret)
 276                                         goto out;
 277                         } else {
 278                                 path->slots[0]++;
 279                         }
 280                         goto next_slot;
 281                 }
 282
 283                 /* FIXME, there's only one inline extent allowed right now */
 284                 if (found_inline) {
 285                         u64 mask = root->blocksize - 1;
 286                         search_start = (extent_end + mask) & ~mask;
 287                 } else
 288                         search_start = extent_end;
 289
 290                 if (end < extent_end && end >= key.offset) {
 291                         if (found_extent) {
 292                                 u64 disk_blocknr =
 293                                         btrfs_file_extent_disk_blocknr(extent);
 294                                 u64 disk_num_blocks =
 295                                       btrfs_file_extent_disk_num_blocks(extent);
 296                                 memcpy(&old, extent, sizeof(old));
 297                                 if (disk_blocknr != 0) {
 298                                         ret = btrfs_inc_extent_ref(trans, root,
 299                                                  disk_blocknr, disk_num_blocks);
 300                                         BUG_ON(ret);
 301                                 }
 302                         }
 303                         WARN_ON(found_inline);
 304                         bookend = 1;
 305                 }
 306                 /* truncate existing extent */
 307                 if (start > key.offset) {
 308                         u64 new_num;
 309                         u64 old_num;
 310                         keep = 1;
 311                         WARN_ON(start & (root->blocksize - 1));
 312                         if (found_extent) {
 313                                 new_num = (start - key.offset) >>
 314                                         inode->i_blkbits;
 315                                 old_num = btrfs_file_extent_num_blocks(extent);
 316                                 *hint_block =
 317                                         btrfs_file_extent_disk_blocknr(extent);
 318                                 if (btrfs_file_extent_disk_blocknr(extent)) {
 319                                         inode->i_blocks -=
 320                                                 (old_num - new_num) << 3;
 321                                 }
 322                                 btrfs_set_file_extent_num_blocks(extent,
 323                                                                  new_num);
 324                                 mark_buffer_dirty(path->nodes[0]);
 325                         } else {
 326                                 WARN_ON(1);
 327                         }
 328                 }
 329                 /* delete the entire extent */
 330                 if (!keep) {
 331                         u64 disk_blocknr = 0;
 332                         u64 disk_num_blocks = 0;
 333                         u64 extent_num_blocks = 0;
 334                         if (found_extent) {
 335                                 disk_blocknr =
 336                                       btrfs_file_extent_disk_blocknr(extent);
 337                                 disk_num_blocks =
 338                                       btrfs_file_extent_disk_num_blocks(extent);
 339                                 extent_num_blocks =
 340                                       btrfs_file_extent_num_blocks(extent);
 341                                 *hint_block =
 342                                         btrfs_file_extent_disk_blocknr(extent);
 343                         }
 344                         ret = btrfs_del_item(trans, root, path);
 345                         /* TODO update progress marker and return */
 346                         BUG_ON(ret);
 347                         btrfs_release_path(root, path);
 348                         extent = NULL;
 349                         if (found_extent && disk_blocknr != 0) {
 350                                 inode->i_blocks -= extent_num_blocks << 3;
 351                                 ret = btrfs_free_extent(trans, root,
 352                                                         disk_blocknr,
 353                                                         disk_num_blocks, 0);
 354                         }
 355
 356                         BUG_ON(ret);
 357                         if (!bookend && search_start >= end) {
 358                                 ret = 0;
 359                                 goto out;
 360                         }
 361                         if (!bookend)
 362                                 continue;
 363                 }
 364                 /* create bookend, splitting the extent in two */
 365                 if (bookend && found_extent) {
 366                         struct btrfs_key ins;
 367                         ins.objectid = inode->i_ino;
 368                         ins.offset = end;
 369                         ins.flags = 0;
 370                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 371                         btrfs_release_path(root, path);
 372                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 373                                                       sizeof(*extent));
 374
 375                         if (ret) {
 376                                 btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
 377                                 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu\n", ret , ins.objectid, ins.flags, ins.offset, start, end, key.offset, extent_end);
 378                         }
 379                         BUG_ON(ret);
 380                         extent = btrfs_item_ptr(
 381                                     btrfs_buffer_leaf(path->nodes[0]),
 382                                     path->slots[0],
 383                                     struct btrfs_file_extent_item);
 384                         btrfs_set_file_extent_disk_blocknr(extent,
 385                                     btrfs_file_extent_disk_blocknr(&old));
 386                         btrfs_set_file_extent_disk_num_blocks(extent,
 387                                     btrfs_file_extent_disk_num_blocks(&old));
 388
 389                         btrfs_set_file_extent_offset(extent,
 390                                     btrfs_file_extent_offset(&old) +
 391                                     ((end - key.offset) >> inode->i_blkbits));
 392                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
 393                                 (extent_end - end) >> inode->i_blkbits);
 394                         btrfs_set_file_extent_num_blocks(extent,
 395                                     (extent_end - end) >> inode->i_blkbits);
 396
 397                         btrfs_set_file_extent_type(extent,
 398                                                    BTRFS_FILE_EXTENT_REG);
 399                         btrfs_set_file_extent_generation(extent,
 400                                     btrfs_file_extent_generation(&old));
 401                         btrfs_mark_buffer_dirty(path->nodes[0]);
 402                         if (btrfs_file_extent_disk_blocknr(&old) != 0) {
 403                                 inode->i_blocks +=
 404                                       btrfs_file_extent_num_blocks(extent) << 3;
 405                         }
 406                         ret = 0;
 407                         goto out;
 408                 }
 409         }
 410 out:
 411         btrfs_free_path(path);
 412         return ret;
 413 }
 414
 415 /*
 416  * this gets pages into the page cache and locks them down
 417  */
 418 static int prepare_pages(struct btrfs_root *root,
 419                          struct file *file,
 420                          struct page **pages,
 421                          size_t num_pages,
 422                          loff_t pos,
 423                          unsigned long first_index,
 424                          unsigned long last_index,
 425                          size_t write_bytes)
 426 {
 427         int i;
 428         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 429         struct inode *inode = file->f_path.dentry->d_inode;
 430         int offset;
 431         int err = 0;
 432         int this_write;
 433         struct buffer_head *bh;
 434         struct buffer_head *head;
 435         loff_t isize = i_size_read(inode);
 436         struct btrfs_trans_handle *trans;
 437         u64 hint_block;
 438         u64 num_blocks;
 439         u64 alloc_extent_start;
 440         u64 start_pos;
 441         struct btrfs_key ins;
 442
 443         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
 444         num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
 445                         inode->i_blkbits;
 446
 447         memset(pages, 0, num_pages * sizeof(struct page *));
 448
 449         for (i = 0; i < num_pages; i++) {
 450                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 451                 if (!pages[i]) {
 452                         err = -ENOMEM;
 453                         goto failed_release;
 454                 }
 455         }
 456
 457         mutex_lock(&root->fs_info->fs_mutex);
 458         trans = btrfs_start_transaction(root, 1);
 459         if (!trans) {
 460                 err = -ENOMEM;
 461                 mutex_unlock(&root->fs_info->fs_mutex);
 462                 goto out_unlock;
 463         }
 464         btrfs_set_trans_block_group(trans, inode);
 465         /* FIXME blocksize != 4096 */
 466         inode->i_blocks += num_blocks << 3;
 467         hint_block = 0;
 468
 469         /* FIXME...EIEIO, ENOSPC and more */
 470
 471         /* step one, delete the existing extents in this range */
 472         /* FIXME blocksize != pagesize */
 473         if (start_pos < inode->i_size) {
 474                 err = btrfs_drop_extents(trans, root, inode,
 475                          start_pos, (pos + write_bytes + root->blocksize -1) &
 476                          ~((u64)root->blocksize - 1), &hint_block);
 477                 if (err)
 478                         goto failed_release;
 479         }
 480
 481         /* insert any holes we need to create */
 482         if (inode->i_size < start_pos) {
 483                 u64 last_pos_in_file;
 484                 u64 hole_size;
 485                 u64 mask = root->blocksize - 1;
 486                 last_pos_in_file = (isize + mask) & ~mask;
 487                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 488                 hole_size >>= inode->i_blkbits;
 489                 if (last_pos_in_file < start_pos) {
 490                         err = btrfs_insert_file_extent(trans, root,
 491                                                        inode->i_ino,
 492                                                        last_pos_in_file,
 493                                                        0, 0, hole_size);
 494                 }
 495                 if (err)
 496                         goto failed_release;
 497         }
 498
 499         /*
 500          * either allocate an extent for the new bytes or setup the key
 501          * to show we are doing inline data in the extent
 502          */
 503         if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
 504             pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 505                 err = btrfs_alloc_extent(trans, root, inode->i_ino,
 506                                          num_blocks, hint_block, (u64)-1,
 507                                          &ins, 1);
 508                 if (err)
 509                         goto failed_truncate;
 510                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 511                                        start_pos, ins.objectid, ins.offset,
 512                                        ins.offset);
 513                 if (err)
 514                         goto failed_truncate;
 515         } else {
 516                 ins.offset = 0;
 517                 ins.objectid = 0;
 518         }
 519         BUG_ON(err);
 520         alloc_extent_start = ins.objectid;
 521         err = btrfs_end_transaction(trans, root);
 522         mutex_unlock(&root->fs_info->fs_mutex);
 523
 524         for (i = 0; i < num_pages; i++) {
 525                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 526                 wait_on_page_writeback(pages[i]);
 527                 offset = pos & (PAGE_CACHE_SIZE -1);
 528                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 529                 if (!page_has_buffers(pages[i])) {
 530                         create_empty_buffers(pages[i],
 531                                              root->fs_info->sb->s_blocksize,
 532                                              (1 << BH_Uptodate));
 533                 }
 534                 head = page_buffers(pages[i]);
 535                 bh = head;
 536                 do {
 537                         err = btrfs_map_bh_to_logical(root, bh,
 538                                                       alloc_extent_start);
 539                         BUG_ON(err);
 540                         if (err)
 541                                 goto failed_truncate;
 542                         bh = bh->b_this_page;
 543                         if (alloc_extent_start)
 544                                 alloc_extent_start++;
 545                 } while (bh != head);
 546                 pos += this_write;
 547                 WARN_ON(this_write > write_bytes);
 548                 write_bytes -= this_write;
 549         }
 550         return 0;
 551
 552 failed_release:
 553         btrfs_drop_pages(pages, num_pages);
 554         return err;
 555
 556 failed_truncate:
 557         btrfs_drop_pages(pages, num_pages);
 558         if (pos > isize)
 559                 vmtruncate(inode, isize);
 560         return err;
 561
 562 out_unlock:
 563         mutex_unlock(&root->fs_info->fs_mutex);
 564         goto failed_release;
 565
 566 }
 567
 568 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 569                                 size_t count, loff_t *ppos)
 570 {
 571         loff_t pos;
 572         size_t num_written = 0;
 573         int err = 0;
 574         int ret = 0;
 575         struct inode *inode = file->f_path.dentry->d_inode;
 576         struct btrfs_root *root = BTRFS_I(inode)->root;
 577         struct page **pages = NULL;
 578         int nrptrs;
 579         struct page *pinned[2];
 580         unsigned long first_index;
 581         unsigned long last_index;
 582
 583         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 584                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 585         pinned[0] = NULL;
 586         pinned[1] = NULL;
 587         if (file->f_flags & O_DIRECT)
 588                 return -EINVAL;
 589         pos = *ppos;
 590         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 591         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 592         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 593         if (err)
 594                 goto out;
 595         if (count == 0)
 596                 goto out;
 597         err = remove_suid(file->f_path.dentry);
 598         if (err)
 599                 goto out;
 600         file_update_time(file);
 601
 602         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 603
 604         mutex_lock(&inode->i_mutex);
 605         first_index = pos >> PAGE_CACHE_SHIFT;
 606         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 607
 608         /*
 609          * there are lots of better ways to do this, but this code
 610          * makes sure the first and last page in the file range are
 611          * up to date and ready for cow
 612          */
 613         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 614                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 615                 if (!PageUptodate(pinned[0])) {
 616                         ret = btrfs_readpage(NULL, pinned[0]);
 617                         BUG_ON(ret);
 618                         wait_on_page_locked(pinned[0]);
 619                 } else {
 620                         unlock_page(pinned[0]);
 621                 }
 622         }
 623         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 624                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 625                 if (!PageUptodate(pinned[1])) {
 626                         ret = btrfs_readpage(NULL, pinned[1]);
 627                         BUG_ON(ret);
 628                         wait_on_page_locked(pinned[1]);
 629                 } else {
 630                         unlock_page(pinned[1]);
 631                 }
 632         }
 633
 634         while(count > 0) {
 635                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 636                 size_t write_bytes = min(count, nrptrs *
 637                                         (size_t)PAGE_CACHE_SIZE -
 638                                          offset);
 639                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 640                                         PAGE_CACHE_SHIFT;
 641
 642                 WARN_ON(num_pages > nrptrs);
 643                 memset(pages, 0, sizeof(pages));
 644                 ret = prepare_pages(root, file, pages, num_pages,
 645                                     pos, first_index, last_index,
 646                                     write_bytes);
 647                 if (ret)
 648                         goto out;
 649
 650                 ret = btrfs_copy_from_user(pos, num_pages,
 651                                            write_bytes, pages, buf);
 652                 if (ret) {
 653                         btrfs_drop_pages(pages, num_pages);
 654                         goto out;
 655                 }
 656
 657                 ret = dirty_and_release_pages(NULL, root, file, pages,
 658                                               num_pages, pos, write_bytes);
 659                 btrfs_drop_pages(pages, num_pages);
 660                 if (ret)
 661                         goto out;
 662
 663                 buf += write_bytes;
 664                 count -= write_bytes;
 665                 pos += write_bytes;
 666                 num_written += write_bytes;
 667
 668                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 669                 btrfs_btree_balance_dirty(root);
 670                 cond_resched();
 671         }
 672         mutex_unlock(&inode->i_mutex);
 673 out:
 674         kfree(pages);
 675         if (pinned[0])
 676                 page_cache_release(pinned[0]);
 677         if (pinned[1])
 678                 page_cache_release(pinned[1]);
 679         *ppos = pos;
 680         current->backing_dev_info = NULL;
 681         mark_inode_dirty(inode);
 682         return num_written ? num_written : err;
 683 }
 684
 685 static int btrfs_sync_file(struct file *file,
 686                            struct dentry *dentry, int datasync)
 687 {
 688         struct inode *inode = dentry->d_inode;
 689         struct btrfs_root *root = BTRFS_I(inode)->root;
 690         int ret;
 691         struct btrfs_trans_handle *trans;
 692
 693         /*
 694          * FIXME, use inode generation number to check if we can skip the
 695          * commit
 696          */
 697         mutex_lock(&root->fs_info->fs_mutex);
 698         trans = btrfs_start_transaction(root, 1);
 699         if (!trans) {
 700                 ret = -ENOMEM;
 701                 goto out;
 702         }
 703         ret = btrfs_commit_transaction(trans, root);
 704         mutex_unlock(&root->fs_info->fs_mutex);
 705 out:
 706         return ret > 0 ? EIO : ret;
 707 }
 708
 709 static struct vm_operations_struct btrfs_file_vm_ops = {
 710         .nopage         = filemap_nopage,
 711         .populate       = filemap_populate,
 712         .page_mkwrite   = btrfs_page_mkwrite,
 713 };
 714
 715 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 716 {
 717         vma->vm_ops = &btrfs_file_vm_ops;
 718         file_accessed(filp);
 719         return 0;
 720 }
 721
 722 struct file_operations btrfs_file_operations = {
 723         .llseek         = generic_file_llseek,
 724         .read           = do_sync_read,
 725         .aio_read       = generic_file_aio_read,
 726         .write          = btrfs_file_write,
 727         .mmap           = btrfs_file_mmap,
 728         .open           = generic_file_open,
 729         .ioctl          = btrfs_ioctl,
 730         .fsync          = btrfs_sync_file,
 731 #ifdef CONFIG_COMPAT
 732         .compat_ioctl   = btrfs_compat_ioctl,
 733 #endif
 734 };
 735