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

   1 #include <linux/module.h>
   2 #include <linux/buffer_head.h>
   3 #include <linux/fs.h>
   4 #include <linux/pagemap.h>
   5 #include <linux/highmem.h>
   6 #include <linux/time.h>
   7 #include <linux/init.h>
   8 #include <linux/string.h>
   9 #include <linux/smp_lock.h>
  10 #include <linux/backing-dev.h>
  11 #include <linux/mpage.h>
  12 #include <linux/swap.h>
  13 #include <linux/writeback.h>
  14 #include <linux/statfs.h>
  15 #include <linux/compat.h>
  16 #include "ctree.h"
  17 #include "disk-io.h"
  18 #include "transaction.h"
  19 #include "btrfs_inode.h"
  20 #include "ioctl.h"
  21 #include "print-tree.h"
  22
  23
  24 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  25                                 struct page **prepared_pages,
  26                                 const char __user * buf)
  27 {
  28         long page_fault = 0;
  29         int i;
  30         int offset = pos & (PAGE_CACHE_SIZE - 1);
  31
  32         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  33                 size_t count = min_t(size_t,
  34                                      PAGE_CACHE_SIZE - offset, write_bytes);
  35                 struct page *page = prepared_pages[i];
  36                 fault_in_pages_readable(buf, count);
  37
  38                 /* Copy data from userspace to the current page */
  39                 kmap(page);
  40                 page_fault = __copy_from_user(page_address(page) + offset,
  41                                               buf, count);
  42                 /* Flush processor's dcache for this page */
  43                 flush_dcache_page(page);
  44                 kunmap(page);
  45                 buf += count;
  46                 write_bytes -= count;
  47
  48                 if (page_fault)
  49                         break;
  50         }
  51         return page_fault ? -EFAULT : 0;
  52 }
  53
  54 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  55 {
  56         size_t i;
  57         for (i = 0; i < num_pages; i++) {
  58                 if (!pages[i])
  59                         break;
  60                 unlock_page(pages[i]);
  61                 mark_page_accessed(pages[i]);
  62                 page_cache_release(pages[i]);
  63         }
  64 }
  65
  66 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
  67                                    struct btrfs_root *root,
  68                                    struct file *file,
  69                                    struct page **pages,
  70                                    size_t num_pages,
  71                                    loff_t pos,
  72                                    size_t write_bytes)
  73 {
  74         int i;
  75         int offset;
  76         int err = 0;
  77         int ret;
  78         int this_write;
  79         struct inode *inode = file->f_path.dentry->d_inode;
  80         struct buffer_head *bh;
  81         struct btrfs_file_extent_item *ei;
  82
  83         for (i = 0; i < num_pages; i++) {
  84                 offset = pos & (PAGE_CACHE_SIZE -1);
  85                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
  86                 /* FIXME, one block at a time */
  87
  88                 mutex_lock(&root->fs_info->fs_mutex);
  89                 trans = btrfs_start_transaction(root, 1);
  90                 btrfs_set_trans_block_group(trans, inode);
  91
  92                 bh = page_buffers(pages[i]);
  93
  94                 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
  95                         struct btrfs_key key;
  96                         struct btrfs_path *path;
  97                         char *ptr;
  98                         u32 datasize;
  99
 100                         /* create an inline extent, and copy the data in */
 101                         path = btrfs_alloc_path();
 102                         BUG_ON(!path);
 103                         key.objectid = inode->i_ino;
 104                         key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
 105                         key.flags = 0;
 106                         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 107                         BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
 108                         datasize = offset +
 109                                 btrfs_file_extent_calc_inline_size(write_bytes);
 110
 111                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 112                                                       datasize);
 113                         BUG_ON(ret);
 114                         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 115                                path->slots[0], struct btrfs_file_extent_item);
 116                         btrfs_set_file_extent_generation(ei, trans->transid);
 117                         btrfs_set_file_extent_type(ei,
 118                                                    BTRFS_FILE_EXTENT_INLINE);
 119                         ptr = btrfs_file_extent_inline_start(ei);
 120                         btrfs_memcpy(root, path->nodes[0]->b_data,
 121                                      ptr, bh->b_data, offset + write_bytes);
 122                         mark_buffer_dirty(path->nodes[0]);
 123                         btrfs_free_path(path);
 124                 } else if (buffer_mapped(bh)) {
 125                         /* csum the file data */
 126                         btrfs_csum_file_block(trans, root, inode->i_ino,
 127                                       pages[i]->index << PAGE_CACHE_SHIFT,
 128                                       kmap(pages[i]), PAGE_CACHE_SIZE);
 129                         kunmap(pages[i]);
 130                 }
 131                 SetPageChecked(pages[i]);
 132                 ret = btrfs_end_transaction(trans, root);
 133                 BUG_ON(ret);
 134                 mutex_unlock(&root->fs_info->fs_mutex);
 135
 136                 ret = btrfs_commit_write(file, pages[i], offset,
 137                                          offset + this_write);
 138                 pos += this_write;
 139                 if (ret) {
 140                         err = ret;
 141                         goto failed;
 142                 }
 143                 WARN_ON(this_write > write_bytes);
 144                 write_bytes -= this_write;
 145         }
 146 failed:
 147         return err;
 148 }
 149
 150 /*
 151  * this is very complex, but the basic idea is to drop all extents
 152  * in the range start - end.  hint_block is filled in with a block number
 153  * that would be a good hint to the block allocator for this file.
 154  *
 155  * If an extent intersects the range but is not entirely inside the range
 156  * it is either truncated or split.  Anything entirely inside the range
 157  * is deleted from the tree.
 158  */
 159 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 160                        struct btrfs_root *root, struct inode *inode,
 161                        u64 start, u64 end, u64 *hint_block)
 162 {
 163         int ret;
 164         struct btrfs_key key;
 165         struct btrfs_leaf *leaf;
 166         int slot;
 167         struct btrfs_file_extent_item *extent;
 168         u64 extent_end = 0;
 169         int keep;
 170         struct btrfs_file_extent_item old;
 171         struct btrfs_path *path;
 172         u64 search_start = start;
 173         int bookend;
 174         int found_type;
 175         int found_extent;
 176         int found_inline;
 177
 178         path = btrfs_alloc_path();
 179         if (!path)
 180                 return -ENOMEM;
 181         while(1) {
 182                 btrfs_release_path(root, path);
 183                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 184                                                search_start, -1);
 185                 if (ret < 0)
 186                         goto out;
 187                 if (ret > 0) {
 188                         if (path->slots[0] == 0) {
 189                                 ret = 0;
 190                                 goto out;
 191                         }
 192                         path->slots[0]--;
 193                 }
 194                 keep = 0;
 195                 bookend = 0;
 196                 found_extent = 0;
 197                 found_inline = 0;
 198                 extent = NULL;
 199                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 200                 slot = path->slots[0];
 201                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
 202                 if (key.offset >= end || key.objectid != inode->i_ino) {
 203                         ret = 0;
 204                         goto out;
 205                 }
 206                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
 207                         ret = 0;
 208                         goto out;
 209                 }
 210                 extent = btrfs_item_ptr(leaf, slot,
 211                                         struct btrfs_file_extent_item);
 212                 found_type = btrfs_file_extent_type(extent);
 213                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 214                         extent_end = key.offset +
 215                                 (btrfs_file_extent_num_blocks(extent) <<
 216                                  inode->i_blkbits);
 217                         found_extent = 1;
 218                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 219                         found_inline = 1;
 220                         extent_end = key.offset +
 221                              btrfs_file_extent_inline_len(leaf->items + slot);
 222                 }
 223
 224                 /* we found nothing we can drop */
 225                 if (!found_extent && !found_inline) {
 226                         ret = 0;
 227                         goto out;
 228                 }
 229
 230                 /* we found nothing inside the range */
 231                 if (search_start >= extent_end) {
 232                         ret = 0;
 233                         goto out;
 234                 }
 235
 236                 /* FIXME, there's only one inline extent allowed right now */
 237                 if (found_inline) {
 238                         u64 mask = root->blocksize - 1;
 239                         search_start = (extent_end + mask) & ~mask;
 240                 } else
 241                         search_start = extent_end;
 242
 243                 if (end < extent_end && end >= key.offset) {
 244                         if (found_extent) {
 245                                 u64 disk_blocknr =
 246                                         btrfs_file_extent_disk_blocknr(extent);
 247                                 u64 disk_num_blocks =
 248                                       btrfs_file_extent_disk_num_blocks(extent);
 249                                 memcpy(&old, extent, sizeof(old));
 250                                 if (disk_blocknr != 0) {
 251                                         ret = btrfs_inc_extent_ref(trans, root,
 252                                                  disk_blocknr, disk_num_blocks);
 253                                         BUG_ON(ret);
 254                                 }
 255                         }
 256                         WARN_ON(found_inline);
 257                         bookend = 1;
 258                 }
 259
 260                 /* truncate existing extent */
 261                 if (start > key.offset) {
 262                         u64 new_num;
 263                         u64 old_num;
 264                         keep = 1;
 265                         WARN_ON(start & (root->blocksize - 1));
 266                         if (found_extent) {
 267                                 new_num = (start - key.offset) >>
 268                                         inode->i_blkbits;
 269                                 old_num = btrfs_file_extent_num_blocks(extent);
 270                                 *hint_block =
 271                                         btrfs_file_extent_disk_blocknr(extent);
 272                                 if (btrfs_file_extent_disk_blocknr(extent)) {
 273                                         inode->i_blocks -=
 274                                                 (old_num - new_num) << 3;
 275                                 }
 276                                 btrfs_set_file_extent_num_blocks(extent,
 277                                                                  new_num);
 278                                 mark_buffer_dirty(path->nodes[0]);
 279                         } else {
 280                                 WARN_ON(1);
 281                         }
 282                 }
 283                 /* delete the entire extent */
 284                 if (!keep) {
 285                         u64 disk_blocknr = 0;
 286                         u64 disk_num_blocks = 0;
 287                         u64 extent_num_blocks = 0;
 288                         if (found_extent) {
 289                                 disk_blocknr =
 290                                       btrfs_file_extent_disk_blocknr(extent);
 291                                 disk_num_blocks =
 292                                       btrfs_file_extent_disk_num_blocks(extent);
 293                                 extent_num_blocks =
 294                                       btrfs_file_extent_num_blocks(extent);
 295                                 *hint_block =
 296                                         btrfs_file_extent_disk_blocknr(extent);
 297                         }
 298                         ret = btrfs_del_item(trans, root, path);
 299                         BUG_ON(ret);
 300                         btrfs_release_path(root, path);
 301                         extent = NULL;
 302                         if (found_extent && disk_blocknr != 0) {
 303                                 inode->i_blocks -= extent_num_blocks << 3;
 304                                 ret = btrfs_free_extent(trans, root,
 305                                                         disk_blocknr,
 306                                                         disk_num_blocks, 0);
 307                         }
 308
 309                         BUG_ON(ret);
 310                         if (!bookend && search_start >= end) {
 311                                 ret = 0;
 312                                 goto out;
 313                         }
 314                         if (!bookend)
 315                                 continue;
 316                 }
 317                 /* create bookend, splitting the extent in two */
 318                 if (bookend && found_extent) {
 319                         struct btrfs_key ins;
 320                         ins.objectid = inode->i_ino;
 321                         ins.offset = end;
 322                         ins.flags = 0;
 323                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 324
 325                         btrfs_release_path(root, path);
 326                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 327                                                       sizeof(*extent));
 328                         BUG_ON(ret);
 329                         extent = btrfs_item_ptr(
 330                                     btrfs_buffer_leaf(path->nodes[0]),
 331                                     path->slots[0],
 332                                     struct btrfs_file_extent_item);
 333                         btrfs_set_file_extent_disk_blocknr(extent,
 334                                     btrfs_file_extent_disk_blocknr(&old));
 335                         btrfs_set_file_extent_disk_num_blocks(extent,
 336                                     btrfs_file_extent_disk_num_blocks(&old));
 337
 338                         btrfs_set_file_extent_offset(extent,
 339                                     btrfs_file_extent_offset(&old) +
 340                                     ((end - key.offset) >> inode->i_blkbits));
 341                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
 342                                 (extent_end - end) >> inode->i_blkbits);
 343                         btrfs_set_file_extent_num_blocks(extent,
 344                                     (extent_end - end) >> inode->i_blkbits);
 345
 346                         btrfs_set_file_extent_type(extent,
 347                                                    BTRFS_FILE_EXTENT_REG);
 348                         btrfs_set_file_extent_generation(extent,
 349                                     btrfs_file_extent_generation(&old));
 350                         btrfs_mark_buffer_dirty(path->nodes[0]);
 351                         if (btrfs_file_extent_disk_blocknr(&old) != 0) {
 352                                 inode->i_blocks +=
 353                                       btrfs_file_extent_num_blocks(extent) << 3;
 354                         }
 355                         ret = 0;
 356                         goto out;
 357                 }
 358         }
 359 out:
 360         btrfs_free_path(path);
 361         return ret;
 362 }
 363
 364 /*
 365  * this gets pages into the page cache and locks them down
 366  */
 367 static int prepare_pages(struct btrfs_root *root,
 368                          struct file *file,
 369                          struct page **pages,
 370                          size_t num_pages,
 371                          loff_t pos,
 372                          unsigned long first_index,
 373                          unsigned long last_index,
 374                          size_t write_bytes,
 375                          u64 alloc_extent_start)
 376 {
 377         int i;
 378         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 379         struct inode *inode = file->f_path.dentry->d_inode;
 380         int offset;
 381         int err = 0;
 382         int this_write;
 383         struct buffer_head *bh;
 384         struct buffer_head *head;
 385         loff_t isize = i_size_read(inode);
 386
 387         memset(pages, 0, num_pages * sizeof(struct page *));
 388
 389         for (i = 0; i < num_pages; i++) {
 390                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 391                 if (!pages[i]) {
 392                         err = -ENOMEM;
 393                         goto failed_release;
 394                 }
 395                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 396                 wait_on_page_writeback(pages[i]);
 397                 offset = pos & (PAGE_CACHE_SIZE -1);
 398                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 399                 if (!page_has_buffers(pages[i])) {
 400                         create_empty_buffers(pages[i],
 401                                              root->fs_info->sb->s_blocksize,
 402                                              (1 << BH_Uptodate));
 403                 }
 404                 head = page_buffers(pages[i]);
 405                 bh = head;
 406                 do {
 407                         err = btrfs_map_bh_to_logical(root, bh,
 408                                                       alloc_extent_start);
 409                         BUG_ON(err);
 410                         if (err)
 411                                 goto failed_truncate;
 412                         bh = bh->b_this_page;
 413                         if (alloc_extent_start)
 414                                 alloc_extent_start++;
 415                 } while (bh != head);
 416                 pos += this_write;
 417                 WARN_ON(this_write > write_bytes);
 418                 write_bytes -= this_write;
 419         }
 420         return 0;
 421
 422 failed_release:
 423         btrfs_drop_pages(pages, num_pages);
 424         return err;
 425
 426 failed_truncate:
 427         btrfs_drop_pages(pages, num_pages);
 428         if (pos > isize)
 429                 vmtruncate(inode, isize);
 430         return err;
 431 }
 432
 433 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 434                                 size_t count, loff_t *ppos)
 435 {
 436         loff_t pos;
 437         size_t num_written = 0;
 438         int err = 0;
 439         int ret = 0;
 440         struct inode *inode = file->f_path.dentry->d_inode;
 441         struct btrfs_root *root = BTRFS_I(inode)->root;
 442         struct page *pages[8];
 443         struct page *pinned[2];
 444         unsigned long first_index;
 445         unsigned long last_index;
 446         u64 start_pos;
 447         u64 num_blocks;
 448         u64 alloc_extent_start;
 449         u64 hint_block;
 450         struct btrfs_trans_handle *trans;
 451         struct btrfs_key ins;
 452         pinned[0] = NULL;
 453         pinned[1] = NULL;
 454         if (file->f_flags & O_DIRECT)
 455                 return -EINVAL;
 456         pos = *ppos;
 457         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 458         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 459         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 460         if (err)
 461                 goto out;
 462         if (count == 0)
 463                 goto out;
 464         err = remove_suid(file->f_path.dentry);
 465         if (err)
 466                 goto out;
 467         file_update_time(file);
 468
 469         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
 470         num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
 471                         inode->i_blkbits;
 472
 473         mutex_lock(&inode->i_mutex);
 474         first_index = pos >> PAGE_CACHE_SHIFT;
 475         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 476
 477         /*
 478          * there are lots of better ways to do this, but this code
 479          * makes sure the first and last page in the file range are
 480          * up to date and ready for cow
 481          */
 482         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 483                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 484                 if (!PageUptodate(pinned[0])) {
 485                         ret = mpage_readpage(pinned[0], btrfs_get_block);
 486                         BUG_ON(ret);
 487                         wait_on_page_locked(pinned[0]);
 488                 } else {
 489                         unlock_page(pinned[0]);
 490                 }
 491         }
 492         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 493                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 494                 if (!PageUptodate(pinned[1])) {
 495                         ret = mpage_readpage(pinned[1], btrfs_get_block);
 496                         BUG_ON(ret);
 497                         wait_on_page_locked(pinned[1]);
 498                 } else {
 499                         unlock_page(pinned[1]);
 500                 }
 501         }
 502
 503         mutex_lock(&root->fs_info->fs_mutex);
 504         trans = btrfs_start_transaction(root, 1);
 505         if (!trans) {
 506                 err = -ENOMEM;
 507                 mutex_unlock(&root->fs_info->fs_mutex);
 508                 goto out_unlock;
 509         }
 510         btrfs_set_trans_block_group(trans, inode);
 511         /* FIXME blocksize != 4096 */
 512         inode->i_blocks += num_blocks << 3;
 513         hint_block = 0;
 514
 515         /* FIXME...EIEIO, ENOSPC and more */
 516
 517         /* step one, delete the existing extents in this range */
 518         if (start_pos < inode->i_size) {
 519                 /* FIXME blocksize != pagesize */
 520                 ret = btrfs_drop_extents(trans, root, inode,
 521                                          start_pos,
 522                                          (pos + count + root->blocksize -1) &
 523                                          ~((u64)root->blocksize - 1),
 524                                          &hint_block);
 525                 BUG_ON(ret);
 526         }
 527
 528         /* insert any holes we need to create */
 529         if (inode->i_size < start_pos) {
 530                 u64 last_pos_in_file;
 531                 u64 hole_size;
 532                 u64 mask = root->blocksize - 1;
 533                 last_pos_in_file = (inode->i_size + mask) & ~mask;
 534                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 535                 hole_size >>= inode->i_blkbits;
 536                 if (last_pos_in_file < start_pos) {
 537                         ret = btrfs_insert_file_extent(trans, root,
 538                                                        inode->i_ino,
 539                                                        last_pos_in_file,
 540                                                        0, 0, hole_size);
 541                 }
 542                 BUG_ON(ret);
 543         }
 544
 545         /*
 546          * either allocate an extent for the new bytes or setup the key
 547          * to show we are doing inline data in the extent
 548          */
 549         if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
 550             pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 551                 ret = btrfs_alloc_extent(trans, root, inode->i_ino,
 552                                          num_blocks, hint_block, (u64)-1,
 553                                          &ins, 1);
 554                 BUG_ON(ret);
 555                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 556                                        start_pos, ins.objectid, ins.offset,
 557                                        ins.offset);
 558                 BUG_ON(ret);
 559         } else {
 560                 ins.offset = 0;
 561                 ins.objectid = 0;
 562         }
 563         BUG_ON(ret);
 564         alloc_extent_start = ins.objectid;
 565         ret = btrfs_end_transaction(trans, root);
 566         mutex_unlock(&root->fs_info->fs_mutex);
 567
 568         while(count > 0) {
 569                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 570                 size_t write_bytes = min(count,
 571                                          (size_t)PAGE_CACHE_SIZE - offset);
 572                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 573                                         PAGE_CACHE_SHIFT;
 574
 575                 memset(pages, 0, sizeof(pages));
 576                 ret = prepare_pages(root, file, pages, num_pages,
 577                                     pos, first_index, last_index,
 578                                     write_bytes, alloc_extent_start);
 579                 BUG_ON(ret);
 580
 581                 /* FIXME blocks != pagesize */
 582                 if (alloc_extent_start)
 583                         alloc_extent_start += num_pages;
 584                 ret = btrfs_copy_from_user(pos, num_pages,
 585                                            write_bytes, pages, buf);
 586                 BUG_ON(ret);
 587
 588                 ret = dirty_and_release_pages(NULL, root, file, pages,
 589                                               num_pages, pos, write_bytes);
 590                 BUG_ON(ret);
 591                 btrfs_drop_pages(pages, num_pages);
 592
 593                 buf += write_bytes;
 594                 count -= write_bytes;
 595                 pos += write_bytes;
 596                 num_written += write_bytes;
 597
 598                 balance_dirty_pages_ratelimited(inode->i_mapping);
 599                 btrfs_btree_balance_dirty(root);
 600                 cond_resched();
 601         }
 602 out_unlock:
 603         mutex_unlock(&inode->i_mutex);
 604 out:
 605         if (pinned[0])
 606                 page_cache_release(pinned[0]);
 607         if (pinned[1])
 608                 page_cache_release(pinned[1]);
 609         *ppos = pos;
 610         current->backing_dev_info = NULL;
 611         mark_inode_dirty(inode);
 612         return num_written ? num_written : err;
 613 }
 614
 615 /*
 616  * FIXME, do this by stuffing the csum we want in the info hanging off
 617  * page->private.  For now, verify file csums on read
 618  */
 619 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
 620                         unsigned long offset, unsigned long size)
 621 {
 622         char *kaddr;
 623         unsigned long left, count = desc->count;
 624         struct inode *inode = page->mapping->host;
 625
 626         if (size > count)
 627                 size = count;
 628
 629         if (!PageChecked(page)) {
 630                 /* FIXME, do it per block */
 631                 struct btrfs_root *root = BTRFS_I(inode)->root;
 632                 int ret;
 633                 struct buffer_head *bh;
 634
 635                 if (page_has_buffers(page)) {
 636                         bh = page_buffers(page);
 637                         if (!buffer_mapped(bh)) {
 638                                 SetPageChecked(page);
 639                                 goto checked;
 640                         }
 641                 }
 642
 643                 ret = btrfs_csum_verify_file_block(root,
 644                                   page->mapping->host->i_ino,
 645                                   page->index << PAGE_CACHE_SHIFT,
 646                                   kmap(page), PAGE_CACHE_SIZE);
 647                 if (ret) {
 648                         if (ret != -ENOENT) {
 649                                 printk("failed to verify ino %lu page %lu ret %d\n",
 650                                        page->mapping->host->i_ino,
 651                                        page->index, ret);
 652                                 memset(page_address(page), 1, PAGE_CACHE_SIZE);
 653                                 flush_dcache_page(page);
 654                         }
 655                 }
 656                 SetPageChecked(page);
 657                 kunmap(page);
 658         }
 659 checked:
 660         /*
 661          * Faults on the destination of a read are common, so do it before
 662          * taking the kmap.
 663          */
 664         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
 665                 kaddr = kmap_atomic(page, KM_USER0);
 666                 left = __copy_to_user_inatomic(desc->arg.buf,
 667                                                 kaddr + offset, size);
 668                 kunmap_atomic(kaddr, KM_USER0);
 669                 if (left == 0)
 670                         goto success;
 671         }
 672
 673         /* Do it the slow way */
 674         kaddr = kmap(page);
 675         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
 676         kunmap(page);
 677
 678         if (left) {
 679                 size -= left;
 680                 desc->error = -EFAULT;
 681         }
 682 success:
 683         desc->count = count - size;
 684         desc->written += size;
 685         desc->arg.buf += size;
 686         return size;
 687 }
 688
 689 /**
 690  * btrfs_file_aio_read - filesystem read routine, with a mod to csum verify
 691  * @iocb:       kernel I/O control block
 692  * @iov:        io vector request
 693  * @nr_segs:    number of segments in the iovec
 694  * @pos:        current file position
 695  */
 696 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
 697                                    unsigned long nr_segs, loff_t pos)
 698 {
 699         struct file *filp = iocb->ki_filp;
 700         ssize_t retval;
 701         unsigned long seg;
 702         size_t count;
 703         loff_t *ppos = &iocb->ki_pos;
 704
 705         count = 0;
 706         for (seg = 0; seg < nr_segs; seg++) {
 707                 const struct iovec *iv = &iov[seg];
 708
 709                 /*
 710                  * If any segment has a negative length, or the cumulative
 711                  * length ever wraps negative then return -EINVAL.
 712                  */
 713                 count += iv->iov_len;
 714                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
 715                         return -EINVAL;
 716                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
 717                         continue;
 718                 if (seg == 0)
 719                         return -EFAULT;
 720                 nr_segs = seg;
 721                 count -= iv->iov_len;   /* This segment is no good */
 722                 break;
 723         }
 724         retval = 0;
 725         if (count) {
 726                 for (seg = 0; seg < nr_segs; seg++) {
 727                         read_descriptor_t desc;
 728
 729                         desc.written = 0;
 730                         desc.arg.buf = iov[seg].iov_base;
 731                         desc.count = iov[seg].iov_len;
 732                         if (desc.count == 0)
 733                                 continue;
 734                         desc.error = 0;
 735                         do_generic_file_read(filp, ppos, &desc,
 736                                              btrfs_read_actor);
 737                         retval += desc.written;
 738                         if (desc.error) {
 739                                 retval = retval ?: desc.error;
 740                                 break;
 741                         }
 742                 }
 743         }
 744         return retval;
 745 }
 746
 747 static int btrfs_sync_file(struct file *file,
 748                            struct dentry *dentry, int datasync)
 749 {
 750         struct inode *inode = dentry->d_inode;
 751         struct btrfs_root *root = BTRFS_I(inode)->root;
 752         int ret;
 753         struct btrfs_trans_handle *trans;
 754
 755         /*
 756          * FIXME, use inode generation number to check if we can skip the
 757          * commit
 758          */
 759         mutex_lock(&root->fs_info->fs_mutex);
 760         trans = btrfs_start_transaction(root, 1);
 761         if (!trans) {
 762                 ret = -ENOMEM;
 763                 goto out;
 764         }
 765         ret = btrfs_commit_transaction(trans, root);
 766         mutex_unlock(&root->fs_info->fs_mutex);
 767 out:
 768         return ret > 0 ? EIO : ret;
 769 }
 770
 771 struct file_operations btrfs_file_operations = {
 772         .llseek         = generic_file_llseek,
 773         .read           = do_sync_read,
 774         .aio_read       = btrfs_file_aio_read,
 775         .write          = btrfs_file_write,
 776         .mmap           = generic_file_mmap,
 777         .open           = generic_file_open,
 778         .ioctl          = btrfs_ioctl,
 779         .fsync          = btrfs_sync_file,
 780 #ifdef CONFIG_COMPAT
 781         .compat_ioctl   = btrfs_compat_ioctl,
 782 #endif
 783 };
 784