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/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.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
  40
  41 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  42                                 struct page **prepared_pages,
  43                                 const char __user * buf)
  44 {
  45         long page_fault = 0;
  46         int i;
  47         int offset = pos & (PAGE_CACHE_SIZE - 1);
  48
  49         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  50                 size_t count = min_t(size_t,
  51                                      PAGE_CACHE_SIZE - offset, write_bytes);
  52                 struct page *page = prepared_pages[i];
  53                 fault_in_pages_readable(buf, count);
  54
  55                 /* Copy data from userspace to the current page */
  56                 kmap(page);
  57                 page_fault = __copy_from_user(page_address(page) + offset,
  58                                               buf, count);
  59                 /* Flush processor's dcache for this page */
  60                 flush_dcache_page(page);
  61                 kunmap(page);
  62                 buf += count;
  63                 write_bytes -= count;
  64
  65                 if (page_fault)
  66                         break;
  67         }
  68         return page_fault ? -EFAULT : 0;
  69 }
  70
  71 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  72 {
  73         size_t i;
  74         for (i = 0; i < num_pages; i++) {
  75                 if (!pages[i])
  76                         break;
  77                 unlock_page(pages[i]);
  78                 mark_page_accessed(pages[i]);
  79                 page_cache_release(pages[i]);
  80         }
  81 }
  82
  83 static int insert_inline_extent(struct btrfs_root *root, struct inode *inode,
  84                                 u64 offset, ssize_t size,
  85                                 struct buffer_head *bh)
  86 {
  87         struct btrfs_key key;
  88         struct btrfs_path *path;
  89         char *ptr, *kaddr;
  90         struct btrfs_trans_handle *trans;
  91         struct btrfs_file_extent_item *ei;
  92         u32 datasize;
  93         int err = 0;
  94         int ret;
  95
  96         path = btrfs_alloc_path();
  97         if (!path)
  98                 return -ENOMEM;
  99
 100         mutex_lock(&root->fs_info->fs_mutex);
 101         trans = btrfs_start_transaction(root, 1);
 102         btrfs_set_trans_block_group(trans, inode);
 103
 104         key.objectid = inode->i_ino;
 105         key.offset = offset;
 106         key.flags = 0;
 107         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 108         BUG_ON(size >= PAGE_CACHE_SIZE);
 109         datasize = btrfs_file_extent_calc_inline_size(size);
 110
 111         ret = btrfs_insert_empty_item(trans, root, path, &key,
 112                                       datasize);
 113         if (ret) {
 114                 err = ret;
 115                 goto fail;
 116         }
 117         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 118                path->slots[0], struct btrfs_file_extent_item);
 119         btrfs_set_file_extent_generation(ei, trans->transid);
 120         btrfs_set_file_extent_type(ei,
 121                                    BTRFS_FILE_EXTENT_INLINE);
 122         ptr = btrfs_file_extent_inline_start(ei);
 123
 124         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 125         btrfs_memcpy(root, path->nodes[0]->b_data,
 126                      ptr, kaddr + bh_offset(bh),
 127                      size);
 128         kunmap_atomic(kaddr, KM_USER0);
 129         btrfs_mark_buffer_dirty(path->nodes[0]);
 130 fail:
 131         btrfs_free_path(path);
 132         ret = btrfs_end_transaction(trans, root);
 133         if (ret && !err)
 134                 err = ret;
 135         mutex_unlock(&root->fs_info->fs_mutex);
 136         return err;
 137 }
 138
 139 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 140                                    struct btrfs_root *root,
 141                                    struct file *file,
 142                                    struct page **pages,
 143                                    size_t num_pages,
 144                                    loff_t pos,
 145                                    size_t write_bytes)
 146 {
 147         int i;
 148         int offset;
 149         int err = 0;
 150         int ret;
 151         int this_write;
 152         struct inode *inode = file->f_path.dentry->d_inode;
 153         struct buffer_head *bh;
 154
 155         for (i = 0; i < num_pages; i++) {
 156                 offset = pos & (PAGE_CACHE_SIZE -1);
 157                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 158
 159                 /* FIXME, one block at a time */
 160                 bh = page_buffers(pages[i]);
 161
 162                 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 163                         ret = insert_inline_extent(root, inode,
 164                                         pages[i]->index << PAGE_CACHE_SHIFT,
 165                                         offset + this_write, bh);
 166                         if (ret) {
 167                                 err = ret;
 168                                 goto failed;
 169                         }
 170                 }
 171
 172                 ret = btrfs_commit_write(file, pages[i], offset,
 173                                          offset + this_write);
 174                 pos += this_write;
 175                 if (ret) {
 176                         err = ret;
 177                         goto failed;
 178                 }
 179                 WARN_ON(this_write > write_bytes);
 180                 write_bytes -= this_write;
 181         }
 182 failed:
 183         return err;
 184 }
 185
 186 /*
 187  * this is very complex, but the basic idea is to drop all extents
 188  * in the range start - end.  hint_block is filled in with a block number
 189  * that would be a good hint to the block allocator for this file.
 190  *
 191  * If an extent intersects the range but is not entirely inside the range
 192  * it is either truncated or split.  Anything entirely inside the range
 193  * is deleted from the tree.
 194  */
 195 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 196                        struct btrfs_root *root, struct inode *inode,
 197                        u64 start, u64 end, u64 *hint_block)
 198 {
 199         int ret;
 200         struct btrfs_key key;
 201         struct btrfs_leaf *leaf;
 202         int slot;
 203         struct btrfs_file_extent_item *extent;
 204         u64 extent_end = 0;
 205         int keep;
 206         struct btrfs_file_extent_item old;
 207         struct btrfs_path *path;
 208         u64 search_start = start;
 209         int bookend;
 210         int found_type;
 211         int found_extent;
 212         int found_inline;
 213         int recow;
 214
 215         path = btrfs_alloc_path();
 216         if (!path)
 217                 return -ENOMEM;
 218         while(1) {
 219                 recow = 0;
 220                 btrfs_release_path(root, path);
 221                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 222                                                search_start, -1);
 223                 if (ret < 0)
 224                         goto out;
 225                 if (ret > 0) {
 226                         if (path->slots[0] == 0) {
 227                                 ret = 0;
 228                                 goto out;
 229                         }
 230                         path->slots[0]--;
 231                 }
 232 next_slot:
 233                 keep = 0;
 234                 bookend = 0;
 235                 found_extent = 0;
 236                 found_inline = 0;
 237                 extent = NULL;
 238                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 239                 slot = path->slots[0];
 240                 ret = 0;
 241                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
 242                 if (key.offset >= end || key.objectid != inode->i_ino) {
 243                         goto out;
 244                 }
 245                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 246                         goto out;
 247                 }
 248                 if (recow) {
 249                         search_start = key.offset;
 250                         continue;
 251                 }
 252                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 253                         extent = btrfs_item_ptr(leaf, slot,
 254                                                 struct btrfs_file_extent_item);
 255                         found_type = btrfs_file_extent_type(extent);
 256                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 257                                 extent_end = key.offset +
 258                                         (btrfs_file_extent_num_blocks(extent) <<
 259                                          inode->i_blkbits);
 260                                 found_extent = 1;
 261                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 262                                 found_inline = 1;
 263                                 extent_end = key.offset +
 264                                      btrfs_file_extent_inline_len(leaf->items +
 265                                                                   slot);
 266                         }
 267                 } else {
 268                         extent_end = search_start;
 269                 }
 270
 271                 /* we found nothing we can drop */
 272                 if ((!found_extent && !found_inline) ||
 273                     search_start >= extent_end) {
 274                         int nextret;
 275                         u32 nritems;
 276                         nritems = btrfs_header_nritems(
 277                                         btrfs_buffer_header(path->nodes[0]));
 278                         if (slot >= nritems - 1) {
 279                                 nextret = btrfs_next_leaf(root, path);
 280                                 if (nextret)
 281                                         goto out;
 282                                 recow = 1;
 283                         } else {
 284                                 path->slots[0]++;
 285                         }
 286                         goto next_slot;
 287                 }
 288
 289                 /* FIXME, there's only one inline extent allowed right now */
 290                 if (found_inline) {
 291                         u64 mask = root->blocksize - 1;
 292                         search_start = (extent_end + mask) & ~mask;
 293                 } else
 294                         search_start = extent_end;
 295
 296                 if (end < extent_end && end >= key.offset) {
 297                         if (found_extent) {
 298                                 u64 disk_blocknr =
 299                                         btrfs_file_extent_disk_blocknr(extent);
 300                                 u64 disk_num_blocks =
 301                                       btrfs_file_extent_disk_num_blocks(extent);
 302                                 memcpy(&old, extent, sizeof(old));
 303                                 if (disk_blocknr != 0) {
 304                                         ret = btrfs_inc_extent_ref(trans, root,
 305                                                  disk_blocknr, disk_num_blocks);
 306                                         BUG_ON(ret);
 307                                 }
 308                         }
 309                         WARN_ON(found_inline);
 310                         bookend = 1;
 311                 }
 312                 /* truncate existing extent */
 313                 if (start > key.offset) {
 314                         u64 new_num;
 315                         u64 old_num;
 316                         keep = 1;
 317                         WARN_ON(start & (root->blocksize - 1));
 318                         if (found_extent) {
 319                                 new_num = (start - key.offset) >>
 320                                         inode->i_blkbits;
 321                                 old_num = btrfs_file_extent_num_blocks(extent);
 322                                 *hint_block =
 323                                         btrfs_file_extent_disk_blocknr(extent);
 324                                 if (btrfs_file_extent_disk_blocknr(extent)) {
 325                                         inode->i_blocks -=
 326                                                 (old_num - new_num) << 3;
 327                                 }
 328                                 btrfs_set_file_extent_num_blocks(extent,
 329                                                                  new_num);
 330                                 btrfs_mark_buffer_dirty(path->nodes[0]);
 331                         } else {
 332                                 WARN_ON(1);
 333                         }
 334                 }
 335                 /* delete the entire extent */
 336                 if (!keep) {
 337                         u64 disk_blocknr = 0;
 338                         u64 disk_num_blocks = 0;
 339                         u64 extent_num_blocks = 0;
 340                         if (found_extent) {
 341                                 disk_blocknr =
 342                                       btrfs_file_extent_disk_blocknr(extent);
 343                                 disk_num_blocks =
 344                                       btrfs_file_extent_disk_num_blocks(extent);
 345                                 extent_num_blocks =
 346                                       btrfs_file_extent_num_blocks(extent);
 347                                 *hint_block =
 348                                         btrfs_file_extent_disk_blocknr(extent);
 349                         }
 350                         ret = btrfs_del_item(trans, root, path);
 351                         /* TODO update progress marker and return */
 352                         BUG_ON(ret);
 353                         btrfs_release_path(root, path);
 354                         extent = NULL;
 355                         if (found_extent && disk_blocknr != 0) {
 356                                 inode->i_blocks -= extent_num_blocks << 3;
 357                                 ret = btrfs_free_extent(trans, root,
 358                                                         disk_blocknr,
 359                                                         disk_num_blocks, 0);
 360                         }
 361
 362                         BUG_ON(ret);
 363                         if (!bookend && search_start >= end) {
 364                                 ret = 0;
 365                                 goto out;
 366                         }
 367                         if (!bookend)
 368                                 continue;
 369                 }
 370                 /* create bookend, splitting the extent in two */
 371                 if (bookend && found_extent) {
 372                         struct btrfs_key ins;
 373                         ins.objectid = inode->i_ino;
 374                         ins.offset = end;
 375                         ins.flags = 0;
 376                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 377                         btrfs_release_path(root, path);
 378                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 379                                                       sizeof(*extent));
 380
 381                         if (ret) {
 382                                 btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
 383                                 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);
 384                         }
 385                         BUG_ON(ret);
 386                         extent = btrfs_item_ptr(
 387                                     btrfs_buffer_leaf(path->nodes[0]),
 388                                     path->slots[0],
 389                                     struct btrfs_file_extent_item);
 390                         btrfs_set_file_extent_disk_blocknr(extent,
 391                                     btrfs_file_extent_disk_blocknr(&old));
 392                         btrfs_set_file_extent_disk_num_blocks(extent,
 393                                     btrfs_file_extent_disk_num_blocks(&old));
 394
 395                         btrfs_set_file_extent_offset(extent,
 396                                     btrfs_file_extent_offset(&old) +
 397                                     ((end - key.offset) >> inode->i_blkbits));
 398                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
 399                                 (extent_end - end) >> inode->i_blkbits);
 400                         btrfs_set_file_extent_num_blocks(extent,
 401                                     (extent_end - end) >> inode->i_blkbits);
 402
 403                         btrfs_set_file_extent_type(extent,
 404                                                    BTRFS_FILE_EXTENT_REG);
 405                         btrfs_set_file_extent_generation(extent,
 406                                     btrfs_file_extent_generation(&old));
 407                         btrfs_mark_buffer_dirty(path->nodes[0]);
 408                         if (btrfs_file_extent_disk_blocknr(&old) != 0) {
 409                                 inode->i_blocks +=
 410                                       btrfs_file_extent_num_blocks(extent) << 3;
 411                         }
 412                         ret = 0;
 413                         goto out;
 414                 }
 415         }
 416 out:
 417         btrfs_free_path(path);
 418         return ret;
 419 }
 420
 421 /*
 422  * this gets pages into the page cache and locks them down
 423  */
 424 static int prepare_pages(struct btrfs_root *root,
 425                          struct file *file,
 426                          struct page **pages,
 427                          size_t num_pages,
 428                          loff_t pos,
 429                          unsigned long first_index,
 430                          unsigned long last_index,
 431                          size_t write_bytes)
 432 {
 433         int i;
 434         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 435         struct inode *inode = file->f_path.dentry->d_inode;
 436         int offset;
 437         int err = 0;
 438         int this_write;
 439         struct buffer_head *bh;
 440         struct buffer_head *head;
 441         loff_t isize = i_size_read(inode);
 442         struct btrfs_trans_handle *trans;
 443         u64 hint_block;
 444         u64 num_blocks;
 445         u64 alloc_extent_start;
 446         u64 start_pos;
 447         struct btrfs_key ins;
 448
 449         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
 450         num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
 451                         inode->i_blkbits;
 452
 453         memset(pages, 0, num_pages * sizeof(struct page *));
 454
 455         for (i = 0; i < num_pages; i++) {
 456                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 457                 if (!pages[i]) {
 458                         err = -ENOMEM;
 459                         goto failed_release;
 460                 }
 461                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 462                 wait_on_page_writeback(pages[i]);
 463         }
 464
 465         mutex_lock(&root->fs_info->fs_mutex);
 466         trans = btrfs_start_transaction(root, 1);
 467         if (!trans) {
 468                 err = -ENOMEM;
 469                 mutex_unlock(&root->fs_info->fs_mutex);
 470                 goto out_unlock;
 471         }
 472         btrfs_set_trans_block_group(trans, inode);
 473         /* FIXME blocksize != 4096 */
 474         inode->i_blocks += num_blocks << 3;
 475         hint_block = 0;
 476
 477         /* FIXME...EIEIO, ENOSPC and more */
 478
 479         /* step one, delete the existing extents in this range */
 480         /* FIXME blocksize != pagesize */
 481         if (start_pos < inode->i_size) {
 482                 err = btrfs_drop_extents(trans, root, inode,
 483                          start_pos, (pos + write_bytes + root->blocksize -1) &
 484                          ~((u64)root->blocksize - 1), &hint_block);
 485                 if (err)
 486                         goto failed_release;
 487         }
 488
 489         /* insert any holes we need to create */
 490         if (inode->i_size < start_pos) {
 491                 u64 last_pos_in_file;
 492                 u64 hole_size;
 493                 u64 mask = root->blocksize - 1;
 494                 last_pos_in_file = (isize + mask) & ~mask;
 495                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 496                 hole_size >>= inode->i_blkbits;
 497                 if (last_pos_in_file < start_pos) {
 498                         err = btrfs_insert_file_extent(trans, root,
 499                                                        inode->i_ino,
 500                                                        last_pos_in_file,
 501                                                        0, 0, hole_size);
 502                 }
 503                 if (err)
 504                         goto failed_release;
 505         }
 506
 507         /*
 508          * either allocate an extent for the new bytes or setup the key
 509          * to show we are doing inline data in the extent
 510          */
 511         if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
 512             pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 513                 err = btrfs_alloc_extent(trans, root, inode->i_ino,
 514                                          num_blocks, hint_block, (u64)-1,
 515                                          &ins, 1);
 516                 if (err)
 517                         goto failed_truncate;
 518                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 519                                        start_pos, ins.objectid, ins.offset,
 520                                        ins.offset);
 521                 if (err)
 522                         goto failed_truncate;
 523         } else {
 524                 ins.offset = 0;
 525                 ins.objectid = 0;
 526         }
 527         BUG_ON(err);
 528         alloc_extent_start = ins.objectid;
 529         err = btrfs_end_transaction(trans, root);
 530         mutex_unlock(&root->fs_info->fs_mutex);
 531
 532         for (i = 0; i < num_pages; i++) {
 533                 offset = pos & (PAGE_CACHE_SIZE -1);
 534                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 535                 if (!page_has_buffers(pages[i])) {
 536                         create_empty_buffers(pages[i],
 537                                              root->fs_info->sb->s_blocksize,
 538                                              (1 << BH_Uptodate));
 539                 }
 540                 head = page_buffers(pages[i]);
 541                 bh = head;
 542                 do {
 543                         err = btrfs_map_bh_to_logical(root, bh,
 544                                                       alloc_extent_start);
 545                         BUG_ON(err);
 546                         if (err)
 547                                 goto failed_truncate;
 548                         bh = bh->b_this_page;
 549                         if (alloc_extent_start)
 550                                 alloc_extent_start++;
 551                 } while (bh != head);
 552                 pos += this_write;
 553                 WARN_ON(this_write > write_bytes);
 554                 write_bytes -= this_write;
 555         }
 556         return 0;
 557
 558 failed_release:
 559         btrfs_drop_pages(pages, num_pages);
 560         return err;
 561
 562 failed_truncate:
 563         btrfs_drop_pages(pages, num_pages);
 564         if (pos > isize)
 565                 vmtruncate(inode, isize);
 566         return err;
 567
 568 out_unlock:
 569         mutex_unlock(&root->fs_info->fs_mutex);
 570         goto failed_release;
 571
 572 }
 573
 574 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 575                                 size_t count, loff_t *ppos)
 576 {
 577         loff_t pos;
 578         size_t num_written = 0;
 579         int err = 0;
 580         int ret = 0;
 581         struct inode *inode = file->f_path.dentry->d_inode;
 582         struct btrfs_root *root = BTRFS_I(inode)->root;
 583         struct page **pages = NULL;
 584         int nrptrs;
 585         struct page *pinned[2];
 586         unsigned long first_index;
 587         unsigned long last_index;
 588
 589         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 590                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 591         pinned[0] = NULL;
 592         pinned[1] = NULL;
 593         if (file->f_flags & O_DIRECT)
 594                 return -EINVAL;
 595         pos = *ppos;
 596         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 597         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 598         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 599         if (err)
 600                 goto out;
 601         if (count == 0)
 602                 goto out;
 603         err = remove_suid(file->f_path.dentry);
 604         if (err)
 605                 goto out;
 606         file_update_time(file);
 607
 608         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 609
 610         mutex_lock(&inode->i_mutex);
 611         first_index = pos >> PAGE_CACHE_SHIFT;
 612         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 613
 614         /*
 615          * there are lots of better ways to do this, but this code
 616          * makes sure the first and last page in the file range are
 617          * up to date and ready for cow
 618          */
 619         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 620                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 621                 if (!PageUptodate(pinned[0])) {
 622                         ret = btrfs_readpage(NULL, pinned[0]);
 623                         BUG_ON(ret);
 624                         wait_on_page_locked(pinned[0]);
 625                 } else {
 626                         unlock_page(pinned[0]);
 627                 }
 628         }
 629         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 630                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 631                 if (!PageUptodate(pinned[1])) {
 632                         ret = btrfs_readpage(NULL, pinned[1]);
 633                         BUG_ON(ret);
 634                         wait_on_page_locked(pinned[1]);
 635                 } else {
 636                         unlock_page(pinned[1]);
 637                 }
 638         }
 639
 640         while(count > 0) {
 641                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 642                 size_t write_bytes = min(count, nrptrs *
 643                                         (size_t)PAGE_CACHE_SIZE -
 644                                          offset);
 645                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 646                                         PAGE_CACHE_SHIFT;
 647
 648                 WARN_ON(num_pages > nrptrs);
 649                 memset(pages, 0, sizeof(pages));
 650                 ret = prepare_pages(root, file, pages, num_pages,
 651                                     pos, first_index, last_index,
 652                                     write_bytes);
 653                 if (ret)
 654                         goto out;
 655
 656                 ret = btrfs_copy_from_user(pos, num_pages,
 657                                            write_bytes, pages, buf);
 658                 if (ret) {
 659                         btrfs_drop_pages(pages, num_pages);
 660                         goto out;
 661                 }
 662
 663                 ret = dirty_and_release_pages(NULL, root, file, pages,
 664                                               num_pages, pos, write_bytes);
 665                 btrfs_drop_pages(pages, num_pages);
 666                 if (ret)
 667                         goto out;
 668
 669                 buf += write_bytes;
 670                 count -= write_bytes;
 671                 pos += write_bytes;
 672                 num_written += write_bytes;
 673
 674                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 675                 btrfs_btree_balance_dirty(root);
 676                 cond_resched();
 677         }
 678         mutex_unlock(&inode->i_mutex);
 679 out:
 680         kfree(pages);
 681         if (pinned[0])
 682                 page_cache_release(pinned[0]);
 683         if (pinned[1])
 684                 page_cache_release(pinned[1]);
 685         *ppos = pos;
 686         current->backing_dev_info = NULL;
 687         mark_inode_dirty(inode);
 688         return num_written ? num_written : err;
 689 }
 690
 691 static int btrfs_sync_file(struct file *file,
 692                            struct dentry *dentry, int datasync)
 693 {
 694         struct inode *inode = dentry->d_inode;
 695         struct btrfs_root *root = BTRFS_I(inode)->root;
 696         int ret;
 697         struct btrfs_trans_handle *trans;
 698
 699         /*
 700          * FIXME, use inode generation number to check if we can skip the
 701          * commit
 702          */
 703         mutex_lock(&root->fs_info->fs_mutex);
 704         trans = btrfs_start_transaction(root, 1);
 705         if (!trans) {
 706                 ret = -ENOMEM;
 707                 goto out;
 708         }
 709         ret = btrfs_commit_transaction(trans, root);
 710         mutex_unlock(&root->fs_info->fs_mutex);
 711 out:
 712         return ret > 0 ? EIO : ret;
 713 }
 714
 715 static struct vm_operations_struct btrfs_file_vm_ops = {
 716         .nopage         = filemap_nopage,
 717         .populate       = filemap_populate,
 718         .page_mkwrite   = btrfs_page_mkwrite,
 719 };
 720
 721 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 722 {
 723         vma->vm_ops = &btrfs_file_vm_ops;
 724         file_accessed(filp);
 725         return 0;
 726 }
 727
 728 struct file_operations btrfs_file_operations = {
 729         .llseek         = generic_file_llseek,
 730         .read           = do_sync_read,
 731         .aio_read       = generic_file_aio_read,
 732         .write          = btrfs_file_write,
 733         .mmap           = btrfs_file_mmap,
 734         .open           = generic_file_open,
 735         .ioctl          = btrfs_ioctl,
 736         .fsync          = btrfs_sync_file,
 737 #ifdef CONFIG_COMPAT
 738         .compat_ioctl   = btrfs_compat_ioctl,
 739 #endif
 740 };
 741