Pileus Git - ~andy/linux/blob - fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/version.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39
  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                 path->slots[0]--;
 119                 leaf = path->nodes[0];
 120                 ei = btrfs_item_ptr(leaf, path->slots[0],
 121                                     struct btrfs_file_extent_item);
 122
 123                 if (btrfs_file_extent_type(leaf, ei) !=
 124                     BTRFS_FILE_EXTENT_INLINE) {
 125                         goto insert;
 126                 }
 127                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 128                 ret = 0;
 129         }
 130         if (ret == 0) {
 131                 u32 found_size;
 132                 u64 found_end;
 133
 134                 leaf = path->nodes[0];
 135                 ei = btrfs_item_ptr(leaf, path->slots[0],
 136                                     struct btrfs_file_extent_item);
 137
 138                 if (btrfs_file_extent_type(leaf, ei) !=
 139                     BTRFS_FILE_EXTENT_INLINE) {
 140                         err = ret;
 141                         btrfs_print_leaf(root, leaf);
 142                         printk("found wasn't inline offset %Lu inode %lu\n",
 143                                offset, inode->i_ino);
 144                         goto fail;
 145                 }
 146                 found_size = btrfs_file_extent_inline_len(leaf,
 147                                           btrfs_item_nr(leaf, path->slots[0]));
 148                 found_end = key.offset + found_size;
 149
 150                 if (found_end < offset + size) {
 151                         btrfs_release_path(root, path);
 152                         ret = btrfs_search_slot(trans, root, &key, path,
 153                                                 offset + size - found_end, 1);
 154                         BUG_ON(ret != 0);
 155                         ret = btrfs_extend_item(trans, root, path,
 156                                                 offset + size - found_end);
 157                         if (ret) {
 158                                 err = ret;
 159                                 goto fail;
 160                         }
 161                         leaf = path->nodes[0];
 162                         ei = btrfs_item_ptr(leaf, path->slots[0],
 163                                             struct btrfs_file_extent_item);
 164                 }
 165                 if (found_end < offset) {
 166                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 167                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 168                 }
 169         } else {
 170 insert:
 171                 btrfs_release_path(root, path);
 172                 datasize = offset + size - key.offset;
 173                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 174                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 175                                               datasize);
 176                 if (ret) {
 177                         err = ret;
 178                         printk("got bad ret %d\n", ret);
 179                         goto fail;
 180                 }
 181                 leaf = path->nodes[0];
 182                 ei = btrfs_item_ptr(leaf, path->slots[0],
 183                                     struct btrfs_file_extent_item);
 184                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 185                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 186         }
 187         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 188
 189         cur_size = size;
 190         i = 0;
 191         while (size > 0) {
 192                 page = pages[i];
 193                 kaddr = kmap_atomic(page, KM_USER0);
 194                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 195                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 196                 kunmap_atomic(kaddr, KM_USER0);
 197                 page_offset = 0;
 198                 ptr += cur_size;
 199                 size -= cur_size;
 200                 if (i >= num_pages) {
 201                         printk("i %d num_pages %d\n", i, num_pages);
 202                 }
 203                 i++;
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206 fail:
 207         btrfs_free_path(path);
 208         return err;
 209 }
 210
 211 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 212                                    struct btrfs_root *root,
 213                                    struct file *file,
 214                                    struct page **pages,
 215                                    size_t num_pages,
 216                                    loff_t pos,
 217                                    size_t write_bytes)
 218 {
 219         int err = 0;
 220         int i;
 221         struct inode *inode = file->f_path.dentry->d_inode;
 222         struct extent_map *em;
 223         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 224         u64 hint_byte;
 225         u64 num_bytes;
 226         u64 start_pos;
 227         u64 end_of_last_block;
 228         u64 end_pos = pos + write_bytes;
 229         u32 inline_size;
 230         loff_t isize = i_size_read(inode);
 231
 232         em = alloc_extent_map(GFP_NOFS);
 233         if (!em)
 234                 return -ENOMEM;
 235
 236         em->bdev = inode->i_sb->s_bdev;
 237
 238         start_pos = pos & ~((u64)root->sectorsize - 1);
 239         num_bytes = (write_bytes + pos - start_pos +
 240                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 241
 242         down_read(&BTRFS_I(inode)->root->snap_sem);
 243         end_of_last_block = start_pos + num_bytes - 1;
 244
 245         lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 246         mutex_lock(&root->fs_info->fs_mutex);
 247         trans = btrfs_start_transaction(root, 1);
 248         if (!trans) {
 249                 err = -ENOMEM;
 250                 goto out_unlock;
 251         }
 252         btrfs_set_trans_block_group(trans, inode);
 253         inode->i_blocks += num_bytes >> 9;
 254         hint_byte = 0;
 255
 256         if ((end_of_last_block & 4095) == 0) {
 257                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 258         }
 259         set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 260
 261         /* FIXME...EIEIO, ENOSPC and more */
 262
 263         /* insert any holes we need to create */
 264         if (inode->i_size < start_pos) {
 265                 u64 last_pos_in_file;
 266                 u64 hole_size;
 267                 u64 mask = root->sectorsize - 1;
 268                 last_pos_in_file = (isize + mask) & ~mask;
 269                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 270
 271                 if (last_pos_in_file < start_pos) {
 272                         err = btrfs_drop_extents(trans, root, inode,
 273                                                  last_pos_in_file,
 274                                                  last_pos_in_file + hole_size,
 275                                                  last_pos_in_file,
 276                                                  &hint_byte);
 277                         if (err)
 278                                 goto failed;
 279
 280                         err = btrfs_insert_file_extent(trans, root,
 281                                                        inode->i_ino,
 282                                                        last_pos_in_file,
 283                                                        0, 0, hole_size);
 284                 }
 285                 if (err)
 286                         goto failed;
 287         }
 288
 289         /*
 290          * either allocate an extent for the new bytes or setup the key
 291          * to show we are doing inline data in the extent
 292          */
 293         inline_size = end_pos;
 294         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 295             inline_size > 8192 ||
 296             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 297                 u64 last_end;
 298
 299                 for (i = 0; i < num_pages; i++) {
 300                         struct page *p = pages[i];
 301                         SetPageUptodate(p);
 302                         set_page_dirty(p);
 303                 }
 304                 last_end = (u64)(pages[num_pages -1]->index) <<
 305                                 PAGE_CACHE_SHIFT;
 306                 last_end += PAGE_CACHE_SIZE - 1;
 307                 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
 308                                  GFP_NOFS);
 309         } else {
 310                 u64 aligned_end;
 311                 /* step one, delete the existing extents in this range */
 312                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 313                         ~((u64)root->sectorsize - 1);
 314                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 315                                          aligned_end, end_pos, &hint_byte);
 316                 if (err)
 317                         goto failed;
 318                 err = insert_inline_extent(trans, root, inode, start_pos,
 319                                            end_pos - start_pos, pages, 0,
 320                                            num_pages);
 321                 BUG_ON(err);
 322         }
 323         if (end_pos > isize) {
 324                 i_size_write(inode, end_pos);
 325                 btrfs_update_inode(trans, root, inode);
 326         }
 327 failed:
 328         err = btrfs_end_transaction(trans, root);
 329 out_unlock:
 330         mutex_unlock(&root->fs_info->fs_mutex);
 331         unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 332         free_extent_map(em);
 333         up_read(&BTRFS_I(inode)->root->snap_sem);
 334         return err;
 335 }
 336
 337 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 338 {
 339         struct extent_map *em;
 340         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 341
 342         while(1) {
 343                 em = lookup_extent_mapping(em_tree, start, end);
 344                 if (!em)
 345                         break;
 346                 remove_extent_mapping(em_tree, em);
 347                 /* once for us */
 348                 free_extent_map(em);
 349                 /* once for the tree*/
 350                 free_extent_map(em);
 351         }
 352         return 0;
 353 }
 354
 355 /*
 356  * this is very complex, but the basic idea is to drop all extents
 357  * in the range start - end.  hint_block is filled in with a block number
 358  * that would be a good hint to the block allocator for this file.
 359  *
 360  * If an extent intersects the range but is not entirely inside the range
 361  * it is either truncated or split.  Anything entirely inside the range
 362  * is deleted from the tree.
 363  */
 364 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 365                        struct btrfs_root *root, struct inode *inode,
 366                        u64 start, u64 end, u64 inline_end, u64 *hint_byte)
 367 {
 368         int ret;
 369         struct btrfs_key key;
 370         struct extent_buffer *leaf;
 371         int slot;
 372         struct btrfs_file_extent_item *extent;
 373         u64 extent_end = 0;
 374         int keep;
 375         struct btrfs_file_extent_item old;
 376         struct btrfs_path *path;
 377         u64 search_start = start;
 378         int bookend;
 379         int found_type;
 380         int found_extent;
 381         int found_inline;
 382         int recow;
 383
 384         btrfs_drop_extent_cache(inode, start, end - 1);
 385
 386         path = btrfs_alloc_path();
 387         if (!path)
 388                 return -ENOMEM;
 389         while(1) {
 390                 recow = 0;
 391                 btrfs_release_path(root, path);
 392                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 393                                                search_start, -1);
 394                 if (ret < 0)
 395                         goto out;
 396                 if (ret > 0) {
 397                         if (path->slots[0] == 0) {
 398                                 ret = 0;
 399                                 goto out;
 400                         }
 401                         path->slots[0]--;
 402                 }
 403 next_slot:
 404                 keep = 0;
 405                 bookend = 0;
 406                 found_extent = 0;
 407                 found_inline = 0;
 408                 extent = NULL;
 409                 leaf = path->nodes[0];
 410                 slot = path->slots[0];
 411                 ret = 0;
 412                 btrfs_item_key_to_cpu(leaf, &key, slot);
 413                 if (key.offset >= end || key.objectid != inode->i_ino) {
 414                         goto out;
 415                 }
 416                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
 417                         goto out;
 418                 }
 419                 if (recow) {
 420                         search_start = key.offset;
 421                         continue;
 422                 }
 423                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 424                         extent = btrfs_item_ptr(leaf, slot,
 425                                                 struct btrfs_file_extent_item);
 426                         found_type = btrfs_file_extent_type(leaf, extent);
 427                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 428                                 extent_end = key.offset +
 429                                      btrfs_file_extent_num_bytes(leaf, extent);
 430                                 found_extent = 1;
 431                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 432                                 struct btrfs_item *item;
 433                                 item = btrfs_item_nr(leaf, slot);
 434                                 found_inline = 1;
 435                                 extent_end = key.offset +
 436                                      btrfs_file_extent_inline_len(leaf, item);
 437                         }
 438                 } else {
 439                         extent_end = search_start;
 440                 }
 441
 442                 /* we found nothing we can drop */
 443                 if ((!found_extent && !found_inline) ||
 444                     search_start >= extent_end) {
 445                         int nextret;
 446                         u32 nritems;
 447                         nritems = btrfs_header_nritems(leaf);
 448                         if (slot >= nritems - 1) {
 449                                 nextret = btrfs_next_leaf(root, path);
 450                                 if (nextret)
 451                                         goto out;
 452                                 recow = 1;
 453                         } else {
 454                                 path->slots[0]++;
 455                         }
 456                         goto next_slot;
 457                 }
 458
 459                 /* FIXME, there's only one inline extent allowed right now */
 460                 if (found_inline) {
 461                         u64 mask = root->sectorsize - 1;
 462                         search_start = (extent_end + mask) & ~mask;
 463                 } else
 464                         search_start = extent_end;
 465
 466                 if (end < extent_end && end >= key.offset) {
 467                         if (found_extent) {
 468                                 u64 disk_bytenr =
 469                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 470                                 u64 disk_num_bytes =
 471                                     btrfs_file_extent_disk_num_bytes(leaf,
 472                                                                       extent);
 473                                 read_extent_buffer(leaf, &old,
 474                                                    (unsigned long)extent,
 475                                                    sizeof(old));
 476                                 if (disk_bytenr != 0) {
 477                                         ret = btrfs_inc_extent_ref(trans, root,
 478                                                  disk_bytenr, disk_num_bytes);
 479                                         BUG_ON(ret);
 480                                 }
 481                         }
 482                         if (!found_inline)
 483                                 bookend = 1;
 484                 }
 485                 /* truncate existing extent */
 486                 if (start > key.offset) {
 487                         u64 new_num;
 488                         u64 old_num;
 489                         keep = 1;
 490                         WARN_ON(start & (root->sectorsize - 1));
 491                         if (found_extent) {
 492                                 new_num = start - key.offset;
 493                                 old_num = btrfs_file_extent_num_bytes(leaf,
 494                                                                       extent);
 495                                 *hint_byte =
 496                                         btrfs_file_extent_disk_bytenr(leaf,
 497                                                                       extent);
 498                                 if (btrfs_file_extent_disk_bytenr(leaf,
 499                                                                   extent)) {
 500                                         inode->i_blocks -=
 501                                                 (old_num - new_num) >> 9;
 502                                 }
 503                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 504                                                                 new_num);
 505                                 btrfs_mark_buffer_dirty(leaf);
 506                         } else if (end > extent_end &&
 507                                    key.offset < inline_end &&
 508                                    inline_end < extent_end) {
 509                                 u32 new_size;
 510                                 new_size = btrfs_file_extent_calc_inline_size(
 511                                                    inline_end - key.offset);
 512                                 btrfs_truncate_item(trans, root, path,
 513                                                     new_size);
 514                         }
 515                 }
 516                 /* delete the entire extent */
 517                 if (!keep) {
 518                         u64 disk_bytenr = 0;
 519                         u64 disk_num_bytes = 0;
 520                         u64 extent_num_bytes = 0;
 521                         if (found_extent) {
 522                                 disk_bytenr =
 523                                       btrfs_file_extent_disk_bytenr(leaf,
 524                                                                      extent);
 525                                 disk_num_bytes =
 526                                       btrfs_file_extent_disk_num_bytes(leaf,
 527                                                                        extent);
 528                                 extent_num_bytes =
 529                                       btrfs_file_extent_num_bytes(leaf, extent);
 530                                 *hint_byte =
 531                                         btrfs_file_extent_disk_bytenr(leaf,
 532                                                                       extent);
 533                         }
 534                         ret = btrfs_del_item(trans, root, path);
 535                         /* TODO update progress marker and return */
 536                         BUG_ON(ret);
 537                         btrfs_release_path(root, path);
 538                         extent = NULL;
 539                         if (found_extent && disk_bytenr != 0) {
 540                                 inode->i_blocks -= extent_num_bytes >> 9;
 541                                 ret = btrfs_free_extent(trans, root,
 542                                                         disk_bytenr,
 543                                                         disk_num_bytes, 0);
 544                         }
 545
 546                         BUG_ON(ret);
 547                         if (!bookend && search_start >= end) {
 548                                 ret = 0;
 549                                 goto out;
 550                         }
 551                         if (!bookend)
 552                                 continue;
 553                 }
 554                 /* create bookend, splitting the extent in two */
 555                 if (bookend && found_extent) {
 556                         struct btrfs_key ins;
 557                         ins.objectid = inode->i_ino;
 558                         ins.offset = end;
 559                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 560                         btrfs_release_path(root, path);
 561                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 562                                                       sizeof(*extent));
 563
 564                         leaf = path->nodes[0];
 565                         if (ret) {
 566                                 btrfs_print_leaf(root, leaf);
 567                                 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);
 568                         }
 569                         BUG_ON(ret);
 570                         extent = btrfs_item_ptr(leaf, path->slots[0],
 571                                                 struct btrfs_file_extent_item);
 572                         write_extent_buffer(leaf, &old,
 573                                             (unsigned long)extent, sizeof(old));
 574
 575                         btrfs_set_file_extent_offset(leaf, extent,
 576                                     le64_to_cpu(old.offset) + end - key.offset);
 577                         WARN_ON(le64_to_cpu(old.num_bytes) <
 578                                 (extent_end - end));
 579                         btrfs_set_file_extent_num_bytes(leaf, extent,
 580                                                         extent_end - end);
 581                         btrfs_set_file_extent_type(leaf, extent,
 582                                                    BTRFS_FILE_EXTENT_REG);
 583
 584                         btrfs_mark_buffer_dirty(path->nodes[0]);
 585                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 586                                 inode->i_blocks +=
 587                                       btrfs_file_extent_num_bytes(leaf,
 588                                                                   extent) >> 9;
 589                         }
 590                         ret = 0;
 591                         goto out;
 592                 }
 593         }
 594 out:
 595         btrfs_free_path(path);
 596         return ret;
 597 }
 598
 599 /*
 600  * this gets pages into the page cache and locks them down
 601  */
 602 static int prepare_pages(struct btrfs_root *root,
 603                          struct file *file,
 604                          struct page **pages,
 605                          size_t num_pages,
 606                          loff_t pos,
 607                          unsigned long first_index,
 608                          unsigned long last_index,
 609                          size_t write_bytes)
 610 {
 611         int i;
 612         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 613         struct inode *inode = file->f_path.dentry->d_inode;
 614         int err = 0;
 615         u64 start_pos;
 616
 617         start_pos = pos & ~((u64)root->sectorsize - 1);
 618
 619         memset(pages, 0, num_pages * sizeof(struct page *));
 620
 621         for (i = 0; i < num_pages; i++) {
 622                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 623                 if (!pages[i]) {
 624                         err = -ENOMEM;
 625                         BUG_ON(1);
 626                 }
 627                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 628                 wait_on_page_writeback(pages[i]);
 629                 set_page_extent_mapped(pages[i]);
 630                 WARN_ON(!PageLocked(pages[i]));
 631         }
 632         return 0;
 633 }
 634
 635 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 636                                 size_t count, loff_t *ppos)
 637 {
 638         loff_t pos;
 639         loff_t start_pos;
 640         ssize_t num_written = 0;
 641         ssize_t err = 0;
 642         int ret = 0;
 643         struct inode *inode = file->f_path.dentry->d_inode;
 644         struct btrfs_root *root = BTRFS_I(inode)->root;
 645         struct page **pages = NULL;
 646         int nrptrs;
 647         struct page *pinned[2];
 648         unsigned long first_index;
 649         unsigned long last_index;
 650
 651         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 652                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 653         pinned[0] = NULL;
 654         pinned[1] = NULL;
 655         if (file->f_flags & O_DIRECT)
 656                 return -EINVAL;
 657
 658         pos = *ppos;
 659         start_pos = pos;
 660
 661         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 662         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 663         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 664         if (err)
 665                 goto out;
 666         if (count == 0)
 667                 goto out;
 668         err = remove_suid(file->f_path.dentry);
 669         if (err)
 670                 goto out;
 671         file_update_time(file);
 672
 673         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 674
 675         mutex_lock(&inode->i_mutex);
 676         first_index = pos >> PAGE_CACHE_SHIFT;
 677         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 678
 679         /*
 680          * there are lots of better ways to do this, but this code
 681          * makes sure the first and last page in the file range are
 682          * up to date and ready for cow
 683          */
 684         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 685                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 686                 if (!PageUptodate(pinned[0])) {
 687                         ret = btrfs_readpage(NULL, pinned[0]);
 688                         BUG_ON(ret);
 689                         wait_on_page_locked(pinned[0]);
 690                 } else {
 691                         unlock_page(pinned[0]);
 692                 }
 693         }
 694         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 695                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 696                 if (!PageUptodate(pinned[1])) {
 697                         ret = btrfs_readpage(NULL, pinned[1]);
 698                         BUG_ON(ret);
 699                         wait_on_page_locked(pinned[1]);
 700                 } else {
 701                         unlock_page(pinned[1]);
 702                 }
 703         }
 704
 705         while(count > 0) {
 706                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 707                 size_t write_bytes = min(count, nrptrs *
 708                                         (size_t)PAGE_CACHE_SIZE -
 709                                          offset);
 710                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 711                                         PAGE_CACHE_SHIFT;
 712
 713                 WARN_ON(num_pages > nrptrs);
 714                 memset(pages, 0, sizeof(pages));
 715                 ret = prepare_pages(root, file, pages, num_pages,
 716                                     pos, first_index, last_index,
 717                                     write_bytes);
 718                 if (ret)
 719                         goto out;
 720
 721                 ret = btrfs_copy_from_user(pos, num_pages,
 722                                            write_bytes, pages, buf);
 723                 if (ret) {
 724                         btrfs_drop_pages(pages, num_pages);
 725                         goto out;
 726                 }
 727
 728                 ret = dirty_and_release_pages(NULL, root, file, pages,
 729                                               num_pages, pos, write_bytes);
 730                 btrfs_drop_pages(pages, num_pages);
 731                 if (ret)
 732                         goto out;
 733
 734                 buf += write_bytes;
 735                 count -= write_bytes;
 736                 pos += write_bytes;
 737                 num_written += write_bytes;
 738
 739                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 740                 btrfs_btree_balance_dirty(root, 1);
 741                 cond_resched();
 742         }
 743         mutex_unlock(&inode->i_mutex);
 744 out:
 745         kfree(pages);
 746         if (pinned[0])
 747                 page_cache_release(pinned[0]);
 748         if (pinned[1])
 749                 page_cache_release(pinned[1]);
 750         *ppos = pos;
 751
 752         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 753                 err = sync_page_range(inode, inode->i_mapping,
 754                                       start_pos, num_written);
 755                 if (err < 0)
 756                         num_written = err;
 757         }
 758         current->backing_dev_info = NULL;
 759         return num_written ? num_written : err;
 760 }
 761
 762 static int btrfs_sync_file(struct file *file,
 763                            struct dentry *dentry, int datasync)
 764 {
 765         struct inode *inode = dentry->d_inode;
 766         struct btrfs_root *root = BTRFS_I(inode)->root;
 767         int ret = 0;
 768         struct btrfs_trans_handle *trans;
 769
 770         /*
 771          * check the transaction that last modified this inode
 772          * and see if its already been committed
 773          */
 774         mutex_lock(&root->fs_info->fs_mutex);
 775         if (!BTRFS_I(inode)->last_trans)
 776                 goto out;
 777         mutex_lock(&root->fs_info->trans_mutex);
 778         if (BTRFS_I(inode)->last_trans <=
 779             root->fs_info->last_trans_committed) {
 780                 BTRFS_I(inode)->last_trans = 0;
 781                 mutex_unlock(&root->fs_info->trans_mutex);
 782                 goto out;
 783         }
 784         mutex_unlock(&root->fs_info->trans_mutex);
 785
 786         /*
 787          * ok we haven't committed the transaction yet, lets do a commit
 788          */
 789         trans = btrfs_start_transaction(root, 1);
 790         if (!trans) {
 791                 ret = -ENOMEM;
 792                 goto out;
 793         }
 794         ret = btrfs_commit_transaction(trans, root);
 795 out:
 796         mutex_unlock(&root->fs_info->fs_mutex);
 797         return ret > 0 ? EIO : ret;
 798 }
 799
 800 static struct vm_operations_struct btrfs_file_vm_ops = {
 801 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
 802         .nopage         = filemap_nopage,
 803         .populate       = filemap_populate,
 804 #else
 805         .fault          = filemap_fault,
 806 #endif
 807         .page_mkwrite   = btrfs_page_mkwrite,
 808 };
 809
 810 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
 811 {
 812         vma->vm_ops = &btrfs_file_vm_ops;
 813         file_accessed(filp);
 814         return 0;
 815 }
 816
 817 struct file_operations btrfs_file_operations = {
 818         .llseek         = generic_file_llseek,
 819         .read           = do_sync_read,
 820         .aio_read       = generic_file_aio_read,
 821         .write          = btrfs_file_write,
 822         .mmap           = btrfs_file_mmap,
 823         .open           = generic_file_open,
 824         .fsync          = btrfs_sync_file,
 825         .unlocked_ioctl = btrfs_ioctl,
 826 #ifdef CONFIG_COMPAT
 827         .compat_ioctl   = btrfs_ioctl,
 828 #endif
 829 };
 830