fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/fs.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/highmem.h>
  24 #include <linux/time.h>
  25 #include <linux/init.h>
  26 #include <linux/string.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/mpage.h>
  30 #include <linux/swap.h>
  31 #include <linux/writeback.h>
  32 #include <linux/statfs.h>
  33 #include <linux/compat.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "transaction.h"
  37 #include "btrfs_inode.h"
  38 #include "ioctl.h"
  39 #include "print-tree.h"
  40
  41
  42 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  43                                 struct page **prepared_pages,
  44                                 const char __user * buf)
  45 {
  46         long page_fault = 0;
  47         int i;
  48         int offset = pos & (PAGE_CACHE_SIZE - 1);
  49
  50         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  51                 size_t count = min_t(size_t,
  52                                      PAGE_CACHE_SIZE - offset, write_bytes);
  53                 struct page *page = prepared_pages[i];
  54                 fault_in_pages_readable(buf, count);
  55
  56                 /* Copy data from userspace to the current page */
  57                 kmap(page);
  58                 page_fault = __copy_from_user(page_address(page) + offset,
  59                                               buf, count);
  60                 /* Flush processor's dcache for this page */
  61                 flush_dcache_page(page);
  62                 kunmap(page);
  63                 buf += count;
  64                 write_bytes -= count;
  65
  66                 if (page_fault)
  67                         break;
  68         }
  69         return page_fault ? -EFAULT : 0;
  70 }
  71
  72 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  73 {
  74         size_t i;
  75         for (i = 0; i < num_pages; i++) {
  76                 if (!pages[i])
  77                         break;
  78                 unlock_page(pages[i]);
  79                 mark_page_accessed(pages[i]);
  80                 page_cache_release(pages[i]);
  81         }
  82 }
  83
  84 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
  85                                    struct btrfs_root *root,
  86                                    struct file *file,
  87                                    struct page **pages,
  88                                    size_t num_pages,
  89                                    loff_t pos,
  90                                    size_t write_bytes)
  91 {
  92         int i;
  93         int offset;
  94         int err = 0;
  95         int ret;
  96         int this_write;
  97         struct inode *inode = file->f_path.dentry->d_inode;
  98         struct buffer_head *bh;
  99         struct btrfs_file_extent_item *ei;
 100
 101         for (i = 0; i < num_pages; i++) {
 102                 offset = pos & (PAGE_CACHE_SIZE -1);
 103                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 104                 /* FIXME, one block at a time */
 105
 106                 mutex_lock(&root->fs_info->fs_mutex);
 107                 trans = btrfs_start_transaction(root, 1);
 108                 btrfs_set_trans_block_group(trans, inode);
 109
 110                 bh = page_buffers(pages[i]);
 111
 112                 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 113                         struct btrfs_key key;
 114                         struct btrfs_path *path;
 115                         char *ptr, *kaddr;
 116                         u32 datasize;
 117
 118                         /* create an inline extent, and copy the data in */
 119                         path = btrfs_alloc_path();
 120                         BUG_ON(!path);
 121                         key.objectid = inode->i_ino;
 122                         key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
 123                         key.flags = 0;
 124                         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 125                         BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
 126                         datasize = offset +
 127                                 btrfs_file_extent_calc_inline_size(write_bytes);
 128
 129                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 130                                                       datasize);
 131                         BUG_ON(ret);
 132                         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 133                                path->slots[0], struct btrfs_file_extent_item);
 134                         btrfs_set_file_extent_generation(ei, trans->transid);
 135                         btrfs_set_file_extent_type(ei,
 136                                                    BTRFS_FILE_EXTENT_INLINE);
 137                         ptr = btrfs_file_extent_inline_start(ei);
 138                         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 139                         btrfs_memcpy(root, path->nodes[0]->b_data,
 140                                      ptr, kaddr + bh_offset(bh),
 141                                      offset + write_bytes);
 142                         kunmap_atomic(kaddr, KM_USER0);
 143                         mark_buffer_dirty(path->nodes[0]);
 144                         btrfs_free_path(path);
 145                 } else if (buffer_mapped(bh)) {
 146                         /* csum the file data */
 147                         btrfs_csum_file_block(trans, root, inode->i_ino,
 148                                       pages[i]->index << PAGE_CACHE_SHIFT,
 149                                       kmap(pages[i]), PAGE_CACHE_SIZE);
 150                         kunmap(pages[i]);
 151                 }
 152                 SetPageChecked(pages[i]);
 153                 ret = btrfs_end_transaction(trans, root);
 154                 BUG_ON(ret);
 155                 mutex_unlock(&root->fs_info->fs_mutex);
 156
 157                 ret = btrfs_commit_write(file, pages[i], offset,
 158                                          offset + this_write);
 159                 pos += this_write;
 160                 if (ret) {
 161                         err = ret;
 162                         goto failed;
 163                 }
 164                 WARN_ON(this_write > write_bytes);
 165                 write_bytes -= this_write;
 166         }
 167 failed:
 168         return err;
 169 }
 170
 171 /*
 172  * this is very complex, but the basic idea is to drop all extents
 173  * in the range start - end.  hint_block is filled in with a block number
 174  * that would be a good hint to the block allocator for this file.
 175  *
 176  * If an extent intersects the range but is not entirely inside the range
 177  * it is either truncated or split.  Anything entirely inside the range
 178  * is deleted from the tree.
 179  */
 180 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 181                        struct btrfs_root *root, struct inode *inode,
 182                        u64 start, u64 end, u64 *hint_block)
 183 {
 184         int ret;
 185         struct btrfs_key key;
 186         struct btrfs_leaf *leaf;
 187         int slot;
 188         struct btrfs_file_extent_item *extent;
 189         u64 extent_end = 0;
 190         int keep;
 191         struct btrfs_file_extent_item old;
 192         struct btrfs_path *path;
 193         u64 search_start = start;
 194         int bookend;
 195         int found_type;
 196         int found_extent;
 197         int found_inline;
 198
 199         path = btrfs_alloc_path();
 200         if (!path)
 201                 return -ENOMEM;
 202         while(1) {
 203                 btrfs_release_path(root, path);
 204                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 205                                                search_start, -1);
 206                 if (ret < 0)
 207                         goto out;
 208                 if (ret > 0) {
 209                         if (path->slots[0] == 0) {
 210                                 ret = 0;
 211                                 goto out;
 212                         }
 213                         path->slots[0]--;
 214                 }
 215                 keep = 0;
 216                 bookend = 0;
 217                 found_extent = 0;
 218                 found_inline = 0;
 219                 extent = NULL;
 220                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 221                 slot = path->slots[0];
 222                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
 223                 if (key.offset >= end || key.objectid != inode->i_ino) {
 224                         ret = 0;
 225                         goto out;
 226                 }
 227                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
 228                         ret = 0;
 229                         goto out;
 230                 }
 231                 extent = btrfs_item_ptr(leaf, slot,
 232                                         struct btrfs_file_extent_item);
 233                 found_type = btrfs_file_extent_type(extent);
 234                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 235                         extent_end = key.offset +
 236                                 (btrfs_file_extent_num_blocks(extent) <<
 237                                  inode->i_blkbits);
 238                         found_extent = 1;
 239                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 240                         found_inline = 1;
 241                         extent_end = key.offset +
 242                              btrfs_file_extent_inline_len(leaf->items + slot);
 243                 }
 244
 245                 /* we found nothing we can drop */
 246                 if (!found_extent && !found_inline) {
 247                         ret = 0;
 248                         goto out;
 249                 }
 250
 251                 /* we found nothing inside the range */
 252                 if (search_start >= extent_end) {
 253                         ret = 0;
 254                         goto out;
 255                 }
 256
 257                 /* FIXME, there's only one inline extent allowed right now */
 258                 if (found_inline) {
 259                         u64 mask = root->blocksize - 1;
 260                         search_start = (extent_end + mask) & ~mask;
 261                 } else
 262                         search_start = extent_end;
 263
 264                 if (end < extent_end && end >= key.offset) {
 265                         if (found_extent) {
 266                                 u64 disk_blocknr =
 267                                         btrfs_file_extent_disk_blocknr(extent);
 268                                 u64 disk_num_blocks =
 269                                       btrfs_file_extent_disk_num_blocks(extent);
 270                                 memcpy(&old, extent, sizeof(old));
 271                                 if (disk_blocknr != 0) {
 272                                         ret = btrfs_inc_extent_ref(trans, root,
 273                                                  disk_blocknr, disk_num_blocks);
 274                                         BUG_ON(ret);
 275                                 }
 276                         }
 277                         WARN_ON(found_inline);
 278                         bookend = 1;
 279                 }
 280
 281                 /* truncate existing extent */
 282                 if (start > key.offset) {
 283                         u64 new_num;
 284                         u64 old_num;
 285                         keep = 1;
 286                         WARN_ON(start & (root->blocksize - 1));
 287                         if (found_extent) {
 288                                 new_num = (start - key.offset) >>
 289                                         inode->i_blkbits;
 290                                 old_num = btrfs_file_extent_num_blocks(extent);
 291                                 *hint_block =
 292                                         btrfs_file_extent_disk_blocknr(extent);
 293                                 if (btrfs_file_extent_disk_blocknr(extent)) {
 294                                         inode->i_blocks -=
 295                                                 (old_num - new_num) << 3;
 296                                 }
 297                                 btrfs_set_file_extent_num_blocks(extent,
 298                                                                  new_num);
 299                                 mark_buffer_dirty(path->nodes[0]);
 300                         } else {
 301                                 WARN_ON(1);
 302                         }
 303                 }
 304                 /* delete the entire extent */
 305                 if (!keep) {
 306                         u64 disk_blocknr = 0;
 307                         u64 disk_num_blocks = 0;
 308                         u64 extent_num_blocks = 0;
 309                         if (found_extent) {
 310                                 disk_blocknr =
 311                                       btrfs_file_extent_disk_blocknr(extent);
 312                                 disk_num_blocks =
 313                                       btrfs_file_extent_disk_num_blocks(extent);
 314                                 extent_num_blocks =
 315                                       btrfs_file_extent_num_blocks(extent);
 316                                 *hint_block =
 317                                         btrfs_file_extent_disk_blocknr(extent);
 318                         }
 319                         ret = btrfs_del_item(trans, root, path);
 320                         BUG_ON(ret);
 321                         btrfs_release_path(root, path);
 322                         extent = NULL;
 323                         if (found_extent && disk_blocknr != 0) {
 324                                 inode->i_blocks -= extent_num_blocks << 3;
 325                                 ret = btrfs_free_extent(trans, root,
 326                                                         disk_blocknr,
 327                                                         disk_num_blocks, 0);
 328                         }
 329
 330                         BUG_ON(ret);
 331                         if (!bookend && search_start >= end) {
 332                                 ret = 0;
 333                                 goto out;
 334                         }
 335                         if (!bookend)
 336                                 continue;
 337                 }
 338                 /* create bookend, splitting the extent in two */
 339                 if (bookend && found_extent) {
 340                         struct btrfs_key ins;
 341                         ins.objectid = inode->i_ino;
 342                         ins.offset = end;
 343                         ins.flags = 0;
 344                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 345
 346                         btrfs_release_path(root, path);
 347                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 348                                                       sizeof(*extent));
 349                         BUG_ON(ret);
 350                         extent = btrfs_item_ptr(
 351                                     btrfs_buffer_leaf(path->nodes[0]),
 352                                     path->slots[0],
 353                                     struct btrfs_file_extent_item);
 354                         btrfs_set_file_extent_disk_blocknr(extent,
 355                                     btrfs_file_extent_disk_blocknr(&old));
 356                         btrfs_set_file_extent_disk_num_blocks(extent,
 357                                     btrfs_file_extent_disk_num_blocks(&old));
 358
 359                         btrfs_set_file_extent_offset(extent,
 360                                     btrfs_file_extent_offset(&old) +
 361                                     ((end - key.offset) >> inode->i_blkbits));
 362                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
 363                                 (extent_end - end) >> inode->i_blkbits);
 364                         btrfs_set_file_extent_num_blocks(extent,
 365                                     (extent_end - end) >> inode->i_blkbits);
 366
 367                         btrfs_set_file_extent_type(extent,
 368                                                    BTRFS_FILE_EXTENT_REG);
 369                         btrfs_set_file_extent_generation(extent,
 370                                     btrfs_file_extent_generation(&old));
 371                         btrfs_mark_buffer_dirty(path->nodes[0]);
 372                         if (btrfs_file_extent_disk_blocknr(&old) != 0) {
 373                                 inode->i_blocks +=
 374                                       btrfs_file_extent_num_blocks(extent) << 3;
 375                         }
 376                         ret = 0;
 377                         goto out;
 378                 }
 379         }
 380 out:
 381         btrfs_free_path(path);
 382         return ret;
 383 }
 384
 385 /*
 386  * this gets pages into the page cache and locks them down
 387  */
 388 static int prepare_pages(struct btrfs_root *root,
 389                          struct file *file,
 390                          struct page **pages,
 391                          size_t num_pages,
 392                          loff_t pos,
 393                          unsigned long first_index,
 394                          unsigned long last_index,
 395                          size_t write_bytes,
 396                          u64 alloc_extent_start)
 397 {
 398         int i;
 399         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 400         struct inode *inode = file->f_path.dentry->d_inode;
 401         int offset;
 402         int err = 0;
 403         int this_write;
 404         struct buffer_head *bh;
 405         struct buffer_head *head;
 406         loff_t isize = i_size_read(inode);
 407
 408         memset(pages, 0, num_pages * sizeof(struct page *));
 409
 410         for (i = 0; i < num_pages; i++) {
 411                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 412                 if (!pages[i]) {
 413                         err = -ENOMEM;
 414                         goto failed_release;
 415                 }
 416                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
 417                 wait_on_page_writeback(pages[i]);
 418                 offset = pos & (PAGE_CACHE_SIZE -1);
 419                 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
 420                 if (!page_has_buffers(pages[i])) {
 421                         create_empty_buffers(pages[i],
 422                                              root->fs_info->sb->s_blocksize,
 423                                              (1 << BH_Uptodate));
 424                 }
 425                 head = page_buffers(pages[i]);
 426                 bh = head;
 427                 do {
 428                         err = btrfs_map_bh_to_logical(root, bh,
 429                                                       alloc_extent_start);
 430                         BUG_ON(err);
 431                         if (err)
 432                                 goto failed_truncate;
 433                         bh = bh->b_this_page;
 434                         if (alloc_extent_start)
 435                                 alloc_extent_start++;
 436                 } while (bh != head);
 437                 pos += this_write;
 438                 WARN_ON(this_write > write_bytes);
 439                 write_bytes -= this_write;
 440         }
 441         return 0;
 442
 443 failed_release:
 444         btrfs_drop_pages(pages, num_pages);
 445         return err;
 446
 447 failed_truncate:
 448         btrfs_drop_pages(pages, num_pages);
 449         if (pos > isize)
 450                 vmtruncate(inode, isize);
 451         return err;
 452 }
 453
 454 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 455                                 size_t count, loff_t *ppos)
 456 {
 457         loff_t pos;
 458         size_t num_written = 0;
 459         int err = 0;
 460         int ret = 0;
 461         struct inode *inode = file->f_path.dentry->d_inode;
 462         struct btrfs_root *root = BTRFS_I(inode)->root;
 463         struct page *pages[8];
 464         struct page *pinned[2];
 465         unsigned long first_index;
 466         unsigned long last_index;
 467         u64 start_pos;
 468         u64 num_blocks;
 469         u64 alloc_extent_start;
 470         u64 hint_block;
 471         struct btrfs_trans_handle *trans;
 472         struct btrfs_key ins;
 473         pinned[0] = NULL;
 474         pinned[1] = NULL;
 475         if (file->f_flags & O_DIRECT)
 476                 return -EINVAL;
 477         pos = *ppos;
 478         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 479         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 480         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 481         if (err)
 482                 goto out;
 483         if (count == 0)
 484                 goto out;
 485         err = remove_suid(file->f_path.dentry);
 486         if (err)
 487                 goto out;
 488         file_update_time(file);
 489
 490         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
 491         num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
 492                         inode->i_blkbits;
 493
 494         mutex_lock(&inode->i_mutex);
 495         first_index = pos >> PAGE_CACHE_SHIFT;
 496         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 497
 498         /*
 499          * there are lots of better ways to do this, but this code
 500          * makes sure the first and last page in the file range are
 501          * up to date and ready for cow
 502          */
 503         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 504                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 505                 if (!PageUptodate(pinned[0])) {
 506                         ret = mpage_readpage(pinned[0], btrfs_get_block);
 507                         BUG_ON(ret);
 508                         wait_on_page_locked(pinned[0]);
 509                 } else {
 510                         unlock_page(pinned[0]);
 511                 }
 512         }
 513         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 514                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 515                 if (!PageUptodate(pinned[1])) {
 516                         ret = mpage_readpage(pinned[1], btrfs_get_block);
 517                         BUG_ON(ret);
 518                         wait_on_page_locked(pinned[1]);
 519                 } else {
 520                         unlock_page(pinned[1]);
 521                 }
 522         }
 523
 524         mutex_lock(&root->fs_info->fs_mutex);
 525         trans = btrfs_start_transaction(root, 1);
 526         if (!trans) {
 527                 err = -ENOMEM;
 528                 mutex_unlock(&root->fs_info->fs_mutex);
 529                 goto out_unlock;
 530         }
 531         btrfs_set_trans_block_group(trans, inode);
 532         /* FIXME blocksize != 4096 */
 533         inode->i_blocks += num_blocks << 3;
 534         hint_block = 0;
 535
 536         /* FIXME...EIEIO, ENOSPC and more */
 537
 538         /* step one, delete the existing extents in this range */
 539         if (start_pos < inode->i_size) {
 540                 /* FIXME blocksize != pagesize */
 541                 ret = btrfs_drop_extents(trans, root, inode,
 542                                          start_pos,
 543                                          (pos + count + root->blocksize -1) &
 544                                          ~((u64)root->blocksize - 1),
 545                                          &hint_block);
 546                 BUG_ON(ret);
 547         }
 548
 549         /* insert any holes we need to create */
 550         if (inode->i_size < start_pos) {
 551                 u64 last_pos_in_file;
 552                 u64 hole_size;
 553                 u64 mask = root->blocksize - 1;
 554                 last_pos_in_file = (inode->i_size + mask) & ~mask;
 555                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 556                 hole_size >>= inode->i_blkbits;
 557                 if (last_pos_in_file < start_pos) {
 558                         ret = btrfs_insert_file_extent(trans, root,
 559                                                        inode->i_ino,
 560                                                        last_pos_in_file,
 561                                                        0, 0, hole_size);
 562                 }
 563                 BUG_ON(ret);
 564         }
 565
 566         /*
 567          * either allocate an extent for the new bytes or setup the key
 568          * to show we are doing inline data in the extent
 569          */
 570         if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
 571             pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 572                 ret = btrfs_alloc_extent(trans, root, inode->i_ino,
 573                                          num_blocks, hint_block, (u64)-1,
 574                                          &ins, 1);
 575                 BUG_ON(ret);
 576                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 577                                        start_pos, ins.objectid, ins.offset,
 578                                        ins.offset);
 579                 BUG_ON(ret);
 580         } else {
 581                 ins.offset = 0;
 582                 ins.objectid = 0;
 583         }
 584         BUG_ON(ret);
 585         alloc_extent_start = ins.objectid;
 586         ret = btrfs_end_transaction(trans, root);
 587         mutex_unlock(&root->fs_info->fs_mutex);
 588
 589         while(count > 0) {
 590                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 591                 size_t write_bytes = min(count,
 592                                          (size_t)PAGE_CACHE_SIZE - offset);
 593                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 594                                         PAGE_CACHE_SHIFT;
 595
 596                 memset(pages, 0, sizeof(pages));
 597                 ret = prepare_pages(root, file, pages, num_pages,
 598                                     pos, first_index, last_index,
 599                                     write_bytes, alloc_extent_start);
 600                 BUG_ON(ret);
 601
 602                 /* FIXME blocks != pagesize */
 603                 if (alloc_extent_start)
 604                         alloc_extent_start += num_pages;
 605                 ret = btrfs_copy_from_user(pos, num_pages,
 606                                            write_bytes, pages, buf);
 607                 BUG_ON(ret);
 608
 609                 ret = dirty_and_release_pages(NULL, root, file, pages,
 610                                               num_pages, pos, write_bytes);
 611                 BUG_ON(ret);
 612                 btrfs_drop_pages(pages, num_pages);
 613
 614                 buf += write_bytes;
 615                 count -= write_bytes;
 616                 pos += write_bytes;
 617                 num_written += write_bytes;
 618
 619                 balance_dirty_pages_ratelimited(inode->i_mapping);
 620                 btrfs_btree_balance_dirty(root);
 621                 cond_resched();
 622         }
 623 out_unlock:
 624         mutex_unlock(&inode->i_mutex);
 625 out:
 626         if (pinned[0])
 627                 page_cache_release(pinned[0]);
 628         if (pinned[1])
 629                 page_cache_release(pinned[1]);
 630         *ppos = pos;
 631         current->backing_dev_info = NULL;
 632         mark_inode_dirty(inode);
 633         return num_written ? num_written : err;
 634 }
 635
 636 /*
 637  * FIXME, do this by stuffing the csum we want in the info hanging off
 638  * page->private.  For now, verify file csums on read
 639  */
 640 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
 641                         unsigned long offset, unsigned long size)
 642 {
 643         char *kaddr;
 644         unsigned long left, count = desc->count;
 645         struct inode *inode = page->mapping->host;
 646
 647         if (size > count)
 648                 size = count;
 649
 650         if (!PageChecked(page)) {
 651                 /* FIXME, do it per block */
 652                 struct btrfs_root *root = BTRFS_I(inode)->root;
 653                 int ret;
 654                 struct buffer_head *bh;
 655
 656                 if (page_has_buffers(page)) {
 657                         bh = page_buffers(page);
 658                         if (!buffer_mapped(bh)) {
 659                                 SetPageChecked(page);
 660                                 goto checked;
 661                         }
 662                 }
 663
 664                 ret = btrfs_csum_verify_file_block(root,
 665                                   page->mapping->host->i_ino,
 666                                   page->index << PAGE_CACHE_SHIFT,
 667                                   kmap(page), PAGE_CACHE_SIZE);
 668                 if (ret) {
 669                         if (ret != -ENOENT) {
 670                                 printk("failed to verify ino %lu page %lu ret %d\n",
 671                                        page->mapping->host->i_ino,
 672                                        page->index, ret);
 673                                 memset(page_address(page), 1, PAGE_CACHE_SIZE);
 674                                 flush_dcache_page(page);
 675                         }
 676                 }
 677                 SetPageChecked(page);
 678                 kunmap(page);
 679         }
 680 checked:
 681         /*
 682          * Faults on the destination of a read are common, so do it before
 683          * taking the kmap.
 684          */
 685         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
 686                 kaddr = kmap_atomic(page, KM_USER0);
 687                 left = __copy_to_user_inatomic(desc->arg.buf,
 688                                                 kaddr + offset, size);
 689                 kunmap_atomic(kaddr, KM_USER0);
 690                 if (left == 0)
 691                         goto success;
 692         }
 693
 694         /* Do it the slow way */
 695         kaddr = kmap(page);
 696         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
 697         kunmap(page);
 698
 699         if (left) {
 700                 size -= left;
 701                 desc->error = -EFAULT;
 702         }
 703 success:
 704         desc->count = count - size;
 705         desc->written += size;
 706         desc->arg.buf += size;
 707         return size;
 708 }
 709
 710 /**
 711  * btrfs_file_aio_read - filesystem read routine, with a mod to csum verify
 712  * @iocb:       kernel I/O control block
 713  * @iov:        io vector request
 714  * @nr_segs:    number of segments in the iovec
 715  * @pos:        current file position
 716  */
 717 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
 718                                    unsigned long nr_segs, loff_t pos)
 719 {
 720         struct file *filp = iocb->ki_filp;
 721         ssize_t retval;
 722         unsigned long seg;
 723         size_t count;
 724         loff_t *ppos = &iocb->ki_pos;
 725
 726         count = 0;
 727         for (seg = 0; seg < nr_segs; seg++) {
 728                 const struct iovec *iv = &iov[seg];
 729
 730                 /*
 731                  * If any segment has a negative length, or the cumulative
 732                  * length ever wraps negative then return -EINVAL.
 733                  */
 734                 count += iv->iov_len;
 735                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
 736                         return -EINVAL;
 737                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
 738                         continue;
 739                 if (seg == 0)
 740                         return -EFAULT;
 741                 nr_segs = seg;
 742                 count -= iv->iov_len;   /* This segment is no good */
 743                 break;
 744         }
 745         retval = 0;
 746         if (count) {
 747                 for (seg = 0; seg < nr_segs; seg++) {
 748                         read_descriptor_t desc;
 749
 750                         desc.written = 0;
 751                         desc.arg.buf = iov[seg].iov_base;
 752                         desc.count = iov[seg].iov_len;
 753                         if (desc.count == 0)
 754                                 continue;
 755                         desc.error = 0;
 756                         do_generic_file_read(filp, ppos, &desc,
 757                                              btrfs_read_actor);
 758                         retval += desc.written;
 759                         if (desc.error) {
 760                                 retval = retval ?: desc.error;
 761                                 break;
 762                         }
 763                 }
 764         }
 765         return retval;
 766 }
 767
 768 static int btrfs_sync_file(struct file *file,
 769                            struct dentry *dentry, int datasync)
 770 {
 771         struct inode *inode = dentry->d_inode;
 772         struct btrfs_root *root = BTRFS_I(inode)->root;
 773         int ret;
 774         struct btrfs_trans_handle *trans;
 775
 776         /*
 777          * FIXME, use inode generation number to check if we can skip the
 778          * commit
 779          */
 780         mutex_lock(&root->fs_info->fs_mutex);
 781         trans = btrfs_start_transaction(root, 1);
 782         if (!trans) {
 783                 ret = -ENOMEM;
 784                 goto out;
 785         }
 786         ret = btrfs_commit_transaction(trans, root);
 787         mutex_unlock(&root->fs_info->fs_mutex);
 788 out:
 789         return ret > 0 ? EIO : ret;
 790 }
 791
 792 struct file_operations btrfs_file_operations = {
 793         .llseek         = generic_file_llseek,
 794         .read           = do_sync_read,
 795         .aio_read       = btrfs_file_aio_read,
 796         .write          = btrfs_file_write,
 797         .mmap           = generic_file_mmap,
 798         .open           = generic_file_open,
 799         .ioctl          = btrfs_ioctl,
 800         .fsync          = btrfs_sync_file,
 801 #ifdef CONFIG_COMPAT
 802         .compat_ioctl   = btrfs_compat_ioctl,
 803 #endif
 804 };
 805