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
  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_trans_handle *trans,
  84                                 struct btrfs_root *root, struct inode *inode,
  85                                 u64 offset, size_t size,
  86                                 struct page **pages, size_t page_offset,
  87                                 int num_pages)
  88 {
  89         struct btrfs_key key;
  90         struct btrfs_path *path;
  91         struct extent_buffer *leaf;
  92         char *kaddr;
  93         unsigned long ptr;
  94         struct btrfs_file_extent_item *ei;
  95         struct page *page;
  96         u32 datasize;
  97         int err = 0;
  98         int ret;
  99         int i;
 100         ssize_t cur_size;
 101
 102         path = btrfs_alloc_path();
 103         if (!path)
 104                 return -ENOMEM;
 105
 106         btrfs_set_trans_block_group(trans, inode);
 107
 108         key.objectid = inode->i_ino;
 109         key.offset = offset;
 110         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 111
 112         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 113         if (ret < 0) {
 114                 err = ret;
 115                 goto fail;
 116         }
 117         if (ret == 1) {
 118                 struct btrfs_key found_key;
 119
 120                 if (path->slots[0] == 0)
 121                         goto insert;
 122
 123                 path->slots[0]--;
 124                 leaf = path->nodes[0];
 125                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 126
 127                 if (found_key.objectid != inode->i_ino)
 128                         goto insert;
 129
 130                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 131                         goto insert;
 132                 ei = btrfs_item_ptr(leaf, path->slots[0],
 133                                     struct btrfs_file_extent_item);
 134
 135                 if (btrfs_file_extent_type(leaf, ei) !=
 136                     BTRFS_FILE_EXTENT_INLINE) {
 137                         goto insert;
 138                 }
 139                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 140                 ret = 0;
 141         }
 142         if (ret == 0) {
 143                 u32 found_size;
 144                 u64 found_end;
 145
 146                 leaf = path->nodes[0];
 147                 ei = btrfs_item_ptr(leaf, path->slots[0],
 148                                     struct btrfs_file_extent_item);
 149
 150                 if (btrfs_file_extent_type(leaf, ei) !=
 151                     BTRFS_FILE_EXTENT_INLINE) {
 152                         err = ret;
 153                         btrfs_print_leaf(root, leaf);
 154                         printk("found wasn't inline offset %Lu inode %lu\n",
 155                                offset, inode->i_ino);
 156                         goto fail;
 157                 }
 158                 found_size = btrfs_file_extent_inline_len(leaf,
 159                                           btrfs_item_nr(leaf, path->slots[0]));
 160                 found_end = key.offset + found_size;
 161
 162                 if (found_end < offset + size) {
 163                         btrfs_release_path(root, path);
 164                         ret = btrfs_search_slot(trans, root, &key, path,
 165                                                 offset + size - found_end, 1);
 166                         BUG_ON(ret != 0);
 167
 168                         ret = btrfs_extend_item(trans, root, path,
 169                                                 offset + size - found_end);
 170                         if (ret) {
 171                                 err = ret;
 172                                 goto fail;
 173                         }
 174                         leaf = path->nodes[0];
 175                         ei = btrfs_item_ptr(leaf, path->slots[0],
 176                                             struct btrfs_file_extent_item);
 177                 }
 178                 if (found_end < offset) {
 179                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 180                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 181                 }
 182         } else {
 183 insert:
 184                 btrfs_release_path(root, path);
 185                 datasize = offset + size - key.offset;
 186                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 187                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 188                                               datasize);
 189                 if (ret) {
 190                         err = ret;
 191                         printk("got bad ret %d\n", ret);
 192                         goto fail;
 193                 }
 194                 leaf = path->nodes[0];
 195                 ei = btrfs_item_ptr(leaf, path->slots[0],
 196                                     struct btrfs_file_extent_item);
 197                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 198                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 199         }
 200         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 201
 202         cur_size = size;
 203         i = 0;
 204         while (size > 0) {
 205                 page = pages[i];
 206                 kaddr = kmap_atomic(page, KM_USER0);
 207                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 208                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 209                 kunmap_atomic(kaddr, KM_USER0);
 210                 page_offset = 0;
 211                 ptr += cur_size;
 212                 size -= cur_size;
 213                 if (i >= num_pages) {
 214                         printk("i %d num_pages %d\n", i, num_pages);
 215                 }
 216                 i++;
 217         }
 218         btrfs_mark_buffer_dirty(leaf);
 219 fail:
 220         btrfs_free_path(path);
 221         return err;
 222 }
 223
 224 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 225                                    struct btrfs_root *root,
 226                                    struct file *file,
 227                                    struct page **pages,
 228                                    size_t num_pages,
 229                                    loff_t pos,
 230                                    size_t write_bytes)
 231 {
 232         int err = 0;
 233         int i;
 234         struct inode *inode = file->f_path.dentry->d_inode;
 235         struct extent_map *em;
 236         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 237         u64 hint_byte;
 238         u64 num_bytes;
 239         u64 start_pos;
 240         u64 end_of_last_block;
 241         u64 end_pos = pos + write_bytes;
 242         u64 inline_size;
 243         loff_t isize = i_size_read(inode);
 244         em = alloc_extent_map(GFP_NOFS);
 245         if (!em)
 246                 return -ENOMEM;
 247
 248         em->bdev = inode->i_sb->s_bdev;
 249
 250         start_pos = pos & ~((u64)root->sectorsize - 1);
 251         num_bytes = (write_bytes + pos - start_pos +
 252                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 253
 254         down_read(&BTRFS_I(inode)->root->snap_sem);
 255         end_of_last_block = start_pos + num_bytes - 1;
 256
 257         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 258         mutex_lock(&root->fs_info->fs_mutex);
 259         trans = btrfs_start_transaction(root, 1);
 260         if (!trans) {
 261                 err = -ENOMEM;
 262                 goto out_unlock;
 263         }
 264         btrfs_set_trans_block_group(trans, inode);
 265         inode->i_blocks += num_bytes >> 9;
 266         hint_byte = 0;
 267
 268         if ((end_of_last_block & 4095) == 0) {
 269                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 270         }
 271         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 272
 273         /* FIXME...EIEIO, ENOSPC and more */
 274
 275         /* insert any holes we need to create */
 276         if (inode->i_size < start_pos) {
 277                 u64 last_pos_in_file;
 278                 u64 hole_size;
 279                 u64 mask = root->sectorsize - 1;
 280                 last_pos_in_file = (isize + mask) & ~mask;
 281                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 282
 283                 if (last_pos_in_file < start_pos) {
 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);
 296                 }
 297                 if (err)
 298                         goto failed;
 299         }
 300
 301         /*
 302          * either allocate an extent for the new bytes or setup the key
 303          * to show we are doing inline data in the extent
 304          */
 305         inline_size = end_pos;
 306         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 307             inline_size > 32768 ||
 308             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 309                 u64 last_end;
 310
 311                 for (i = 0; i < num_pages; i++) {
 312                         struct page *p = pages[i];
 313                         SetPageUptodate(p);
 314                         set_page_dirty(p);
 315                 }
 316                 last_end = (u64)(pages[num_pages -1]->index) <<
 317                                 PAGE_CACHE_SHIFT;
 318                 last_end += PAGE_CACHE_SIZE - 1;
 319                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 320                                  GFP_NOFS);
 321         } else {
 322                 u64 aligned_end;
 323                 /* step one, delete the existing extents in this range */
 324                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 325                         ~((u64)root->sectorsize - 1);
 326                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 327                                          aligned_end, aligned_end, &hint_byte);
 328                 if (err)
 329                         goto failed;
 330                 if (isize > inline_size)
 331                         inline_size = min_t(u64, isize, aligned_end);
 332                 inline_size -= start_pos;
 333                 err = insert_inline_extent(trans, root, inode, start_pos,
 334                                            inline_size, pages, 0, num_pages);
 335                 BUG_ON(err);
 336         }
 337         if (end_pos > isize) {
 338                 i_size_write(inode, end_pos);
 339                 btrfs_update_inode(trans, root, inode);
 340         }
 341 failed:
 342         err = btrfs_end_transaction(trans, root);
 343 out_unlock:
 344         mutex_unlock(&root->fs_info->fs_mutex);
 345         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 346         free_extent_map(em);
 347         up_read(&BTRFS_I(inode)->root->snap_sem);
 348         return err;
 349 }
 350
 351 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 352 {
 353         struct extent_map *em;
 354         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 355
 356         while(1) {
 357                 em = lookup_extent_mapping(em_tree, start, end);
 358                 if (!em)
 359                         break;
 360                 remove_extent_mapping(em_tree, em);
 361                 /* once for us */
 362                 free_extent_map(em);
 363                 /* once for the tree*/
 364                 free_extent_map(em);
 365         }
 366         return 0;
 367 }
 368
 369 /*
 370  * this is very complex, but the basic idea is to drop all extents
 371  * in the range start - end.  hint_block is filled in with a block number
 372  * that would be a good hint to the block allocator for this file.
 373  *
 374  * If an extent intersects the range but is not entirely inside the range
 375  * it is either truncated or split.  Anything entirely inside the range
 376  * is deleted from the tree.
 377  */
 378 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 379                        struct btrfs_root *root, struct inode *inode,
 380                        u64 start, u64 end, u64 inline_end, u64 *hint_byte)
 381 {
 382         int ret;
 383         struct btrfs_key key;
 384         struct extent_buffer *leaf;
 385         int slot;
 386         struct btrfs_file_extent_item *extent;
 387         u64 extent_end = 0;
 388         int keep;
 389         struct btrfs_file_extent_item old;
 390         struct btrfs_path *path;
 391         u64 search_start = start;
 392         int bookend;
 393         int found_type;
 394         int found_extent;
 395         int found_inline;
 396         int recow;
 397
 398         btrfs_drop_extent_cache(inode, start, end - 1);
 399
 400         path = btrfs_alloc_path();
 401         if (!path)
 402                 return -ENOMEM;
 403         while(1) {
 404                 recow = 0;
 405                 btrfs_release_path(root, path);
 406                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 407                                                search_start, -1);
 408                 if (ret < 0)
 409                         goto out;
 410                 if (ret > 0) {
 411                         if (path->slots[0] == 0) {
 412                                 ret = 0;
 413                                 goto out;
 414                         }
 415                         path->slots[0]--;
 416                 }
 417 next_slot:
 418                 keep = 0;
 419                 bookend = 0;
 420                 found_extent = 0;
 421                 found_inline = 0;
 422                 extent = NULL;
 423                 leaf = path->nodes[0];
 424                 slot = path->slots[0];
 425                 ret = 0;
 426                 btrfs_item_key_to_cpu(leaf, &key, slot);
 427                 if (key.offset >= end || key.objectid != inode->i_ino) {
 428                         goto out;
 429                 }
 430                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 431                         goto out;
 432                 }
 433                 if (recow) {
 434                         search_start = key.offset;
 435                         continue;
 436                 }
 437                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 438                         extent = btrfs_item_ptr(leaf, slot,
 439                                                 struct btrfs_file_extent_item);
 440                         found_type = btrfs_file_extent_type(leaf, extent);
 441                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 442                                 extent_end = key.offset +
 443                                      btrfs_file_extent_num_bytes(leaf, extent);
 444                                 found_extent = 1;
 445                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 446                                 struct btrfs_item *item;
 447                                 item = btrfs_item_nr(leaf, slot);
 448                                 found_inline = 1;
 449                                 extent_end = key.offset +
 450                                      btrfs_file_extent_inline_len(leaf, item);
 451                         }
 452                 } else {
 453                         extent_end = search_start;
 454                 }
 455
 456                 /* we found nothing we can drop */
 457                 if ((!found_extent && !found_inline) ||
 458                     search_start >= extent_end) {
 459                         int nextret;
 460                         u32 nritems;
 461                         nritems = btrfs_header_nritems(leaf);
 462                         if (slot >= nritems - 1) {
 463                                 nextret = btrfs_next_leaf(root, path);
 464                                 if (nextret)
 465                                         goto out;
 466                                 recow = 1;
 467                         } else {
 468                                 path->slots[0]++;
 469                         }
 470                         goto next_slot;
 471                 }
 472
 473                 if (found_inline) {
 474                         u64 mask = root->sectorsize - 1;
 475                         search_start = (extent_end + mask) & ~mask;
 476                 } else
 477                         search_start = extent_end;
 478                 if (end < extent_end && start > key.offset && found_inline) {
 479                         *hint_byte = EXTENT_MAP_INLINE;
 480                 }
 481                 if (end < extent_end && end >= key.offset) {
 482                         if (found_extent) {
 483                                 u64 disk_bytenr =
 484                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 485                                 u64 disk_num_bytes =
 486                                     btrfs_file_extent_disk_num_bytes(leaf,
 487                                                                       extent);
 488                                 read_extent_buffer(leaf, &old,
 489                                                    (unsigned long)extent,
 490                                                    sizeof(old));
 491                                 if (disk_bytenr != 0) {
 492                                         ret = btrfs_inc_extent_ref(trans, root,
 493                                                  disk_bytenr, disk_num_bytes);
 494                                         BUG_ON(ret);
 495                                 }
 496                         }
 497                         bookend = 1;
 498                         if (found_inline && start <= key.offset &&
 499                             inline_end < extent_end)
 500                                 keep = 1;
 501                 }
 502                 /* truncate existing extent */
 503                 if (start > key.offset) {
 504                         u64 new_num;
 505                         u64 old_num;
 506                         keep = 1;
 507                         WARN_ON(start & (root->sectorsize - 1));
 508                         if (found_extent) {
 509                                 new_num = start - key.offset;
 510                                 old_num = btrfs_file_extent_num_bytes(leaf,
 511                                                                       extent);
 512                                 *hint_byte =
 513                                         btrfs_file_extent_disk_bytenr(leaf,
 514                                                                       extent);
 515                                 if (btrfs_file_extent_disk_bytenr(leaf,
 516                                                                   extent)) {
 517                                         inode->i_blocks -=
 518                                                 (old_num - new_num) >> 9;
 519                                 }
 520                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 521                                                                 new_num);
 522                                 btrfs_mark_buffer_dirty(leaf);
 523                         } else if (end > extent_end &&
 524                                    key.offset < inline_end &&
 525                                    inline_end < extent_end) {
 526                                 u32 new_size;
 527                                 new_size = btrfs_file_extent_calc_inline_size(
 528                                                    inline_end - key.offset);
 529                                 btrfs_truncate_item(trans, root, path,
 530                                                     new_size, 1);
 531                         }
 532                 }
 533                 /* delete the entire extent */
 534                 if (!keep) {
 535                         u64 disk_bytenr = 0;
 536                         u64 disk_num_bytes = 0;
 537                         u64 extent_num_bytes = 0;
 538                         if (found_extent) {
 539                                 disk_bytenr =
 540                                       btrfs_file_extent_disk_bytenr(leaf,
 541                                                                      extent);
 542                                 disk_num_bytes =
 543                                       btrfs_file_extent_disk_num_bytes(leaf,
 544                                                                        extent);
 545                                 extent_num_bytes =
 546                                       btrfs_file_extent_num_bytes(leaf, extent);
 547                                 *hint_byte =
 548                                         btrfs_file_extent_disk_bytenr(leaf,
 549                                                                       extent);
 550                         }
 551                         ret = btrfs_del_item(trans, root, path);
 552                         /* TODO update progress marker and return */
 553                         BUG_ON(ret);
 554                         btrfs_release_path(root, path);
 555                         extent = NULL;
 556                         if (found_extent && disk_bytenr != 0) {
 557                                 inode->i_blocks -= extent_num_bytes >> 9;
 558                                 ret = btrfs_free_extent(trans, root,
 559                                                         disk_bytenr,
 560                                                         disk_num_bytes, 0);
 561                         }
 562
 563                         BUG_ON(ret);
 564                         if (!bookend && search_start >= end) {
 565                                 ret = 0;
 566                                 goto out;
 567                         }
 568                         if (!bookend)
 569                                 continue;
 570                 }
 571                 if (bookend && found_inline && start <= key.offset &&
 572                     inline_end < extent_end) {
 573                         u32 new_size;
 574                         new_size = btrfs_file_extent_calc_inline_size(
 575                                                    extent_end - inline_end);
 576                         btrfs_truncate_item(trans, root, path, new_size, 0);
 577                 }
 578                 /* create bookend, splitting the extent in two */
 579                 if (bookend && found_extent) {
 580                         struct btrfs_key ins;
 581                         ins.objectid = inode->i_ino;
 582                         ins.offset = end;
 583                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 584                         btrfs_release_path(root, path);
 585                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 586                                                       sizeof(*extent));
 587
 588                         leaf = path->nodes[0];
 589                         if (ret) {
 590                                 btrfs_print_leaf(root, leaf);
 591                                 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.type, ins.offset, start, end, key.offset, extent_end, keep);
 592                         }
 593                         BUG_ON(ret);
 594                         extent = btrfs_item_ptr(leaf, path->slots[0],
 595                                                 struct btrfs_file_extent_item);
 596                         write_extent_buffer(leaf, &old,
 597                                             (unsigned long)extent, sizeof(old));
 598
 599                         btrfs_set_file_extent_offset(leaf, extent,
 600                                     le64_to_cpu(old.offset) + end - key.offset);
 601                         WARN_ON(le64_to_cpu(old.num_bytes) <
 602                                 (extent_end - end));
 603                         btrfs_set_file_extent_num_bytes(leaf, extent,
 604                                                         extent_end - end);
 605                         btrfs_set_file_extent_type(leaf, extent,
 606                                                    BTRFS_FILE_EXTENT_REG);
 607
 608                         btrfs_mark_buffer_dirty(path->nodes[0]);
 609                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 610                                 inode->i_blocks +=
 611                                       btrfs_file_extent_num_bytes(leaf,
 612                                                                   extent) >> 9;
 613                         }
 614                         ret = 0;
 615                         goto out;
 616                 }
 617         }
 618 out:
 619         btrfs_free_path(path);
 620         return ret;
 621 }
 622
 623 /*
 624  * this gets pages into the page cache and locks them down
 625  */
 626 static int prepare_pages(struct btrfs_root *root,
 627                          struct file *file,
 628                          struct page **pages,
 629                          size_t num_pages,
 630                          loff_t pos,
 631                          unsigned long first_index,
 632                          unsigned long last_index,
 633                          size_t write_bytes)
 634 {
 635         int i;
 636         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 637         struct inode *inode = file->f_path.dentry->d_inode;
 638         int err = 0;
 639         u64 start_pos;
 640
 641         start_pos = pos & ~((u64)root->sectorsize - 1);
 642
 643         memset(pages, 0, num_pages * sizeof(struct page *));
 644
 645         for (i = 0; i < num_pages; i++) {
 646                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 647                 if (!pages[i]) {
 648                         err = -ENOMEM;
 649                         BUG_ON(1);
 650                 }
 651                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 652                 wait_on_page_writeback(pages[i]);
 653                 set_page_extent_mapped(pages[i]);
 654                 WARN_ON(!PageLocked(pages[i]));
 655         }
 656         return 0;
 657 }
 658
 659 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 660                                 size_t count, loff_t *ppos)
 661 {
 662         loff_t pos;
 663         loff_t start_pos;
 664         ssize_t num_written = 0;
 665         ssize_t err = 0;
 666         int ret = 0;
 667         struct inode *inode = file->f_path.dentry->d_inode;
 668         struct btrfs_root *root = BTRFS_I(inode)->root;
 669         struct page **pages = NULL;
 670         int nrptrs;
 671         struct page *pinned[2];
 672         unsigned long first_index;
 673         unsigned long last_index;
 674
 675         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 676                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 677         pinned[0] = NULL;
 678         pinned[1] = NULL;
 679         if (file->f_flags & O_DIRECT)
 680                 return -EINVAL;
 681
 682         pos = *ppos;
 683         start_pos = pos;
 684
 685         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 686         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 687         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 688         if (err)
 689                 goto out;
 690         if (count == 0)
 691                 goto out;
 692         err = remove_suid(file->f_path.dentry);
 693         if (err)
 694                 goto out;
 695         file_update_time(file);
 696
 697         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 698
 699         mutex_lock(&inode->i_mutex);
 700         first_index = pos >> PAGE_CACHE_SHIFT;
 701         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 702
 703         /*
 704          * there are lots of better ways to do this, but this code
 705          * makes sure the first and last page in the file range are
 706          * up to date and ready for cow
 707          */
 708         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 709                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 710                 if (!PageUptodate(pinned[0])) {
 711                         ret = btrfs_readpage(NULL, pinned[0]);
 712                         BUG_ON(ret);
 713                         wait_on_page_locked(pinned[0]);
 714                 } else {
 715                         unlock_page(pinned[0]);
 716                 }
 717         }
 718         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 719                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 720                 if (!PageUptodate(pinned[1])) {
 721                         ret = btrfs_readpage(NULL, pinned[1]);
 722                         BUG_ON(ret);
 723                         wait_on_page_locked(pinned[1]);
 724                 } else {
 725                         unlock_page(pinned[1]);
 726                 }
 727         }
 728
 729         while(count > 0) {
 730                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 731                 size_t write_bytes = min(count, nrptrs *
 732                                         (size_t)PAGE_CACHE_SIZE -
 733                                          offset);
 734                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 735                                         PAGE_CACHE_SHIFT;
 736
 737                 WARN_ON(num_pages > nrptrs);
 738                 memset(pages, 0, sizeof(pages));
 739                 ret = prepare_pages(root, file, pages, num_pages,
 740                                     pos, first_index, last_index,
 741                                     write_bytes);
 742                 if (ret)
 743                         goto out;
 744
 745                 ret = btrfs_copy_from_user(pos, num_pages,
 746                                            write_bytes, pages, buf);
 747                 if (ret) {
 748                         btrfs_drop_pages(pages, num_pages);
 749                         goto out;
 750                 }
 751
 752                 ret = dirty_and_release_pages(NULL, root, file, pages,
 753                                               num_pages, pos, write_bytes);
 754                 btrfs_drop_pages(pages, num_pages);
 755                 if (ret)
 756                         goto out;
 757
 758                 buf += write_bytes;
 759                 count -= write_bytes;
 760                 pos += write_bytes;
 761                 num_written += write_bytes;
 762
 763                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 764                 btrfs_btree_balance_dirty(root, 1);
 765                 cond_resched();
 766         }
 767         mutex_unlock(&inode->i_mutex);
 768 out:
 769         kfree(pages);
 770         if (pinned[0])
 771                 page_cache_release(pinned[0]);
 772         if (pinned[1])
 773                 page_cache_release(pinned[1]);
 774         *ppos = pos;
 775
 776         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 777                 err = sync_page_range(inode, inode->i_mapping,
 778                                       start_pos, num_written);
 779                 if (err < 0)
 780                         num_written = err;
 781         }
 782         current->backing_dev_info = NULL;
 783         return num_written ? num_written : err;
 784 }
 785
 786 static int btrfs_sync_file(struct file *file,
 787                            struct dentry *dentry, int datasync)
 788 {
 789         struct inode *inode = dentry->d_inode;
 790         struct btrfs_root *root = BTRFS_I(inode)->root;
 791         int ret = 0;
 792         struct btrfs_trans_handle *trans;
 793
 794         /*
 795          * check the transaction that last modified this inode
 796          * and see if its already been committed
 797          */
 798         mutex_lock(&root->fs_info->fs_mutex);
 799         if (!BTRFS_I(inode)->last_trans)
 800                 goto out;
 801         mutex_lock(&root->fs_info->trans_mutex);
 802         if (BTRFS_I(inode)->last_trans <=
 803             root->fs_info->last_trans_committed) {
 804                 BTRFS_I(inode)->last_trans = 0;
 805                 mutex_unlock(&root->fs_info->trans_mutex);
 806                 goto out;
 807         }
 808         mutex_unlock(&root->fs_info->trans_mutex);
 809
 810         /*
 811          * ok we haven't committed the transaction yet, lets do a commit
 812          */
 813         trans = btrfs_start_transaction(root, 1);
 814         if (!trans) {
 815                 ret = -ENOMEM;
 816                 goto out;
 817         }
 818         ret = btrfs_commit_transaction(trans, root);
 819 out:
 820         mutex_unlock(&root->fs_info->fs_mutex);
 821         return ret > 0 ? EIO : ret;
 822 }
 823
 824 static struct vm_operations_struct btrfs_file_vm_ops = {
 825 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 826         .nopage         = filemap_nopage,
 827         .populate       = filemap_populate,
 828 #else
 829         .fault          = filemap_fault,
 830 #endif
 831         .page_mkwrite   = btrfs_page_mkwrite,
 832 };
 833
 834 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 835 {
 836         vma->vm_ops = &btrfs_file_vm_ops;
 837         file_accessed(filp);
 838         return 0;
 839 }
 840
 841 struct file_operations btrfs_file_operations = {
 842         .llseek         = generic_file_llseek,
 843         .read           = do_sync_read,
 844         .aio_read       = generic_file_aio_read,
 845         .write          = btrfs_file_write,
 846         .mmap           = btrfs_file_mmap,
 847         .open           = generic_file_open,
 848         .fsync          = btrfs_sync_file,
 849         .unlocked_ioctl = btrfs_ioctl,
 850 #ifdef CONFIG_COMPAT
 851         .compat_ioctl   = btrfs_ioctl,
 852 #endif
 853 };
 854