SAFE public projects git trees. - safe/jmp/linux-2.6/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 #include "tree-log.h"
  40 #include "locking.h"
  41 #include "compat.h"
  42
  43
  44 /* simple helper to fault in pages and copy.  This should go away
  45  * and be replaced with calls into generic code.
  46  */
  47 static int noinline btrfs_copy_from_user(loff_t pos, int num_pages,
  48                                          int write_bytes,
  49                                          struct page **prepared_pages,
  50                                          const char __user * buf)
  51 {
  52         long page_fault = 0;
  53         int i;
  54         int offset = pos & (PAGE_CACHE_SIZE - 1);
  55
  56         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  57                 size_t count = min_t(size_t,
  58                                      PAGE_CACHE_SIZE - offset, write_bytes);
  59                 struct page *page = prepared_pages[i];
  60                 fault_in_pages_readable(buf, count);
  61
  62                 /* Copy data from userspace to the current page */
  63                 kmap(page);
  64                 page_fault = __copy_from_user(page_address(page) + offset,
  65                                               buf, count);
  66                 /* Flush processor's dcache for this page */
  67                 flush_dcache_page(page);
  68                 kunmap(page);
  69                 buf += count;
  70                 write_bytes -= count;
  71
  72                 if (page_fault)
  73                         break;
  74         }
  75         return page_fault ? -EFAULT : 0;
  76 }
  77
  78 /*
  79  * unlocks pages after btrfs_file_write is done with them
  80  */
  81 static void noinline btrfs_drop_pages(struct page **pages, size_t num_pages)
  82 {
  83         size_t i;
  84         for (i = 0; i < num_pages; i++) {
  85                 if (!pages[i])
  86                         break;
  87                 /* page checked is some magic around finding pages that
  88                  * have been modified without going through btrfs_set_page_dirty
  89                  * clear it here
  90                  */
  91                 ClearPageChecked(pages[i]);
  92                 unlock_page(pages[i]);
  93                 mark_page_accessed(pages[i]);
  94                 page_cache_release(pages[i]);
  95         }
  96 }
  97
  98 /*
  99  * after copy_from_user, pages need to be dirtied and we need to make
 100  * sure holes are created between the current EOF and the start of
 101  * any next extents (if required).
 102  *
 103  * this also makes the decision about creating an inline extent vs
 104  * doing real data extents, marking pages dirty and delalloc as required.
 105  */
 106 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 107                                    struct btrfs_root *root,
 108                                    struct file *file,
 109                                    struct page **pages,
 110                                    size_t num_pages,
 111                                    loff_t pos,
 112                                    size_t write_bytes)
 113 {
 114         int err = 0;
 115         int i;
 116         struct inode *inode = fdentry(file)->d_inode;
 117         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 118         u64 hint_byte;
 119         u64 num_bytes;
 120         u64 start_pos;
 121         u64 end_of_last_block;
 122         u64 end_pos = pos + write_bytes;
 123         loff_t isize = i_size_read(inode);
 124
 125         start_pos = pos & ~((u64)root->sectorsize - 1);
 126         num_bytes = (write_bytes + pos - start_pos +
 127                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 128
 129         end_of_last_block = start_pos + num_bytes - 1;
 130
 131         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 132         trans = btrfs_join_transaction(root, 1);
 133         if (!trans) {
 134                 err = -ENOMEM;
 135                 goto out_unlock;
 136         }
 137         btrfs_set_trans_block_group(trans, inode);
 138         hint_byte = 0;
 139
 140         if ((end_of_last_block & 4095) == 0) {
 141                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 142         }
 143         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 144
 145         /* check for reserved extents on each page, we don't want
 146          * to reset the delalloc bit on things that already have
 147          * extents reserved.
 148          */
 149         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 150         for (i = 0; i < num_pages; i++) {
 151                 struct page *p = pages[i];
 152                 SetPageUptodate(p);
 153                 ClearPageChecked(p);
 154                 set_page_dirty(p);
 155         }
 156         if (end_pos > isize) {
 157                 i_size_write(inode, end_pos);
 158                 btrfs_update_inode(trans, root, inode);
 159         }
 160         err = btrfs_end_transaction(trans, root);
 161 out_unlock:
 162         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 163         return err;
 164 }
 165
 166 /*
 167  * this drops all the extents in the cache that intersect the range
 168  * [start, end].  Existing extents are split as required.
 169  */
 170 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 171                             int skip_pinned)
 172 {
 173         struct extent_map *em;
 174         struct extent_map *split = NULL;
 175         struct extent_map *split2 = NULL;
 176         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 177         u64 len = end - start + 1;
 178         int ret;
 179         int testend = 1;
 180         unsigned long flags;
 181         int compressed = 0;
 182
 183         WARN_ON(end < start);
 184         if (end == (u64)-1) {
 185                 len = (u64)-1;
 186                 testend = 0;
 187         }
 188         while(1) {
 189                 if (!split)
 190                         split = alloc_extent_map(GFP_NOFS);
 191                 if (!split2)
 192                         split2 = alloc_extent_map(GFP_NOFS);
 193
 194                 spin_lock(&em_tree->lock);
 195                 em = lookup_extent_mapping(em_tree, start, len);
 196                 if (!em) {
 197                         spin_unlock(&em_tree->lock);
 198                         break;
 199                 }
 200                 flags = em->flags;
 201                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 202                         spin_unlock(&em_tree->lock);
 203                         if (em->start <= start &&
 204                             (!testend || em->start + em->len >= start + len)) {
 205                                 free_extent_map(em);
 206                                 break;
 207                         }
 208                         if (start < em->start) {
 209                                 len = em->start - start;
 210                         } else {
 211                                 len = start + len - (em->start + em->len);
 212                                 start = em->start + em->len;
 213                         }
 214                         free_extent_map(em);
 215                         continue;
 216                 }
 217                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 218                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 219                 remove_extent_mapping(em_tree, em);
 220
 221                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 222                     em->start < start) {
 223                         split->start = em->start;
 224                         split->len = start - em->start;
 225                         split->block_start = em->block_start;
 226
 227                         if (compressed)
 228                                 split->block_len = em->block_len;
 229                         else
 230                                 split->block_len = split->len;
 231
 232                         split->bdev = em->bdev;
 233                         split->flags = flags;
 234                         ret = add_extent_mapping(em_tree, split);
 235                         BUG_ON(ret);
 236                         free_extent_map(split);
 237                         split = split2;
 238                         split2 = NULL;
 239                 }
 240                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 241                     testend && em->start + em->len > start + len) {
 242                         u64 diff = start + len - em->start;
 243
 244                         split->start = start + len;
 245                         split->len = em->start + em->len - (start + len);
 246                         split->bdev = em->bdev;
 247                         split->flags = flags;
 248
 249                         if (compressed) {
 250                                 split->block_len = em->block_len;
 251                                 split->block_start = em->block_start;
 252                         } else {
 253                                 split->block_len = split->len;
 254                                 split->block_start = em->block_start + diff;
 255                         }
 256
 257                         ret = add_extent_mapping(em_tree, split);
 258                         BUG_ON(ret);
 259                         free_extent_map(split);
 260                         split = NULL;
 261                 }
 262                 spin_unlock(&em_tree->lock);
 263
 264                 /* once for us */
 265                 free_extent_map(em);
 266                 /* once for the tree*/
 267                 free_extent_map(em);
 268         }
 269         if (split)
 270                 free_extent_map(split);
 271         if (split2)
 272                 free_extent_map(split2);
 273         return 0;
 274 }
 275
 276 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 277 {
 278         return 0;
 279 #if 0
 280         struct btrfs_path *path;
 281         struct btrfs_key found_key;
 282         struct extent_buffer *leaf;
 283         struct btrfs_file_extent_item *extent;
 284         u64 last_offset = 0;
 285         int nritems;
 286         int slot;
 287         int found_type;
 288         int ret;
 289         int err = 0;
 290         u64 extent_end = 0;
 291
 292         path = btrfs_alloc_path();
 293         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 294                                        last_offset, 0);
 295         while(1) {
 296                 nritems = btrfs_header_nritems(path->nodes[0]);
 297                 if (path->slots[0] >= nritems) {
 298                         ret = btrfs_next_leaf(root, path);
 299                         if (ret)
 300                                 goto out;
 301                         nritems = btrfs_header_nritems(path->nodes[0]);
 302                 }
 303                 slot = path->slots[0];
 304                 leaf = path->nodes[0];
 305                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 306                 if (found_key.objectid != inode->i_ino)
 307                         break;
 308                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 309                         goto out;
 310
 311                 if (found_key.offset < last_offset) {
 312                         WARN_ON(1);
 313                         btrfs_print_leaf(root, leaf);
 314                         printk("inode %lu found offset %Lu expected %Lu\n",
 315                                inode->i_ino, found_key.offset, last_offset);
 316                         err = 1;
 317                         goto out;
 318                 }
 319                 extent = btrfs_item_ptr(leaf, slot,
 320                                         struct btrfs_file_extent_item);
 321                 found_type = btrfs_file_extent_type(leaf, extent);
 322                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 323                         extent_end = found_key.offset +
 324                              btrfs_file_extent_num_bytes(leaf, extent);
 325                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 326                         struct btrfs_item *item;
 327                         item = btrfs_item_nr(leaf, slot);
 328                         extent_end = found_key.offset +
 329                              btrfs_file_extent_inline_len(leaf, extent);
 330                         extent_end = (extent_end + root->sectorsize - 1) &
 331                                 ~((u64)root->sectorsize -1 );
 332                 }
 333                 last_offset = extent_end;
 334                 path->slots[0]++;
 335         }
 336         if (0 && last_offset < inode->i_size) {
 337                 WARN_ON(1);
 338                 btrfs_print_leaf(root, leaf);
 339                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 340                        last_offset, inode->i_size);
 341                 err = 1;
 342
 343         }
 344 out:
 345         btrfs_free_path(path);
 346         return err;
 347 #endif
 348 }
 349
 350 /*
 351  * this is very complex, but the basic idea is to drop all extents
 352  * in the range start - end.  hint_block is filled in with a block number
 353  * that would be a good hint to the block allocator for this file.
 354  *
 355  * If an extent intersects the range but is not entirely inside the range
 356  * it is either truncated or split.  Anything entirely inside the range
 357  * is deleted from the tree.
 358  *
 359  * inline_limit is used to tell this code which offsets in the file to keep
 360  * if they contain inline extents.
 361  */
 362 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 363                        struct btrfs_root *root, struct inode *inode,
 364                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 365 {
 366         u64 extent_end = 0;
 367         u64 locked_end = end;
 368         u64 search_start = start;
 369         u64 leaf_start;
 370         u64 ram_bytes = 0;
 371         u8 compression = 0;
 372         u8 encryption = 0;
 373         u16 other_encoding = 0;
 374         u64 root_gen;
 375         u64 root_owner;
 376         struct extent_buffer *leaf;
 377         struct btrfs_file_extent_item *extent;
 378         struct btrfs_path *path;
 379         struct btrfs_key key;
 380         struct btrfs_file_extent_item old;
 381         int keep;
 382         int slot;
 383         int bookend;
 384         int found_type;
 385         int found_extent;
 386         int found_inline;
 387         int recow;
 388         int ret;
 389
 390         inline_limit = 0;
 391         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 392
 393         path = btrfs_alloc_path();
 394         if (!path)
 395                 return -ENOMEM;
 396         while(1) {
 397                 recow = 0;
 398                 btrfs_release_path(root, path);
 399                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 400                                                search_start, -1);
 401                 if (ret < 0)
 402                         goto out;
 403                 if (ret > 0) {
 404                         if (path->slots[0] == 0) {
 405                                 ret = 0;
 406                                 goto out;
 407                         }
 408                         path->slots[0]--;
 409                 }
 410 next_slot:
 411                 keep = 0;
 412                 bookend = 0;
 413                 found_extent = 0;
 414                 found_inline = 0;
 415                 leaf_start = 0;
 416                 root_gen = 0;
 417                 root_owner = 0;
 418                 extent = NULL;
 419                 leaf = path->nodes[0];
 420                 slot = path->slots[0];
 421                 ret = 0;
 422                 btrfs_item_key_to_cpu(leaf, &key, slot);
 423                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 424                     key.offset >= end) {
 425                         goto out;
 426                 }
 427                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 428                     key.objectid != inode->i_ino) {
 429                         goto out;
 430                 }
 431                 if (recow) {
 432                         search_start = key.offset;
 433                         continue;
 434                 }
 435                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 436                         extent = btrfs_item_ptr(leaf, slot,
 437                                                 struct btrfs_file_extent_item);
 438                         found_type = btrfs_file_extent_type(leaf, extent);
 439                         compression = btrfs_file_extent_compression(leaf,
 440                                                                     extent);
 441                         encryption = btrfs_file_extent_encryption(leaf,
 442                                                                   extent);
 443                         other_encoding = btrfs_file_extent_other_encoding(leaf,
 444                                                                   extent);
 445                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 446                                 extent_end =
 447                                      btrfs_file_extent_disk_bytenr(leaf,
 448                                                                    extent);
 449                                 if (extent_end)
 450                                         *hint_byte = extent_end;
 451
 452                                 extent_end = key.offset +
 453                                      btrfs_file_extent_num_bytes(leaf, extent);
 454                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
 455                                                                 extent);
 456                                 found_extent = 1;
 457                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 458                                 found_inline = 1;
 459                                 extent_end = key.offset +
 460                                      btrfs_file_extent_inline_len(leaf, extent);
 461                         }
 462                 } else {
 463                         extent_end = search_start;
 464                 }
 465
 466                 /* we found nothing we can drop */
 467                 if ((!found_extent && !found_inline) ||
 468                     search_start >= extent_end) {
 469                         int nextret;
 470                         u32 nritems;
 471                         nritems = btrfs_header_nritems(leaf);
 472                         if (slot >= nritems - 1) {
 473                                 nextret = btrfs_next_leaf(root, path);
 474                                 if (nextret)
 475                                         goto out;
 476                                 recow = 1;
 477                         } else {
 478                                 path->slots[0]++;
 479                         }
 480                         goto next_slot;
 481                 }
 482
 483                 if (end <= extent_end && start >= key.offset && found_inline)
 484                         *hint_byte = EXTENT_MAP_INLINE;
 485
 486                 if (found_extent) {
 487                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 488                                            sizeof(old));
 489                         root_gen = btrfs_header_generation(leaf);
 490                         root_owner = btrfs_header_owner(leaf);
 491                         leaf_start = leaf->start;
 492                 }
 493
 494                 if (end < extent_end && end >= key.offset) {
 495                         bookend = 1;
 496                         if (found_inline && start <= key.offset)
 497                                 keep = 1;
 498                 }
 499
 500                 if (bookend && found_extent && locked_end < extent_end) {
 501                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 502                                         locked_end, extent_end - 1, GFP_NOFS);
 503                         if (!ret) {
 504                                 btrfs_release_path(root, path);
 505                                 lock_extent(&BTRFS_I(inode)->io_tree,
 506                                         locked_end, extent_end - 1, GFP_NOFS);
 507                                 locked_end = extent_end;
 508                                 continue;
 509                         }
 510                         locked_end = extent_end;
 511                 }
 512
 513                 if (found_inline) {
 514                         u64 mask = root->sectorsize - 1;
 515                         search_start = (extent_end + mask) & ~mask;
 516                 } else
 517                         search_start = extent_end;
 518
 519                 /* truncate existing extent */
 520                 if (start > key.offset) {
 521                         u64 new_num;
 522                         u64 old_num;
 523                         keep = 1;
 524                         WARN_ON(start & (root->sectorsize - 1));
 525                         if (found_extent) {
 526                                 new_num = start - key.offset;
 527                                 old_num = btrfs_file_extent_num_bytes(leaf,
 528                                                                       extent);
 529                                 *hint_byte =
 530                                         btrfs_file_extent_disk_bytenr(leaf,
 531                                                                       extent);
 532                                 if (btrfs_file_extent_disk_bytenr(leaf,
 533                                                                   extent)) {
 534                                         inode_sub_bytes(inode, old_num -
 535                                                         new_num);
 536                                 }
 537                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 538                                                                 new_num);
 539                                 btrfs_mark_buffer_dirty(leaf);
 540                         } else if (key.offset < inline_limit &&
 541                                    (end > extent_end) &&
 542                                    (inline_limit < extent_end)) {
 543                                 u32 new_size;
 544                                 new_size = btrfs_file_extent_calc_inline_size(
 545                                                    inline_limit - key.offset);
 546                                 inode_sub_bytes(inode, extent_end -
 547                                                 inline_limit);
 548                                 btrfs_truncate_item(trans, root, path,
 549                                                     new_size, 1);
 550                         }
 551                 }
 552                 /* delete the entire extent */
 553                 if (!keep) {
 554                         if (found_inline)
 555                                 inode_sub_bytes(inode, extent_end -
 556                                                 key.offset);
 557                         ret = btrfs_del_item(trans, root, path);
 558                         /* TODO update progress marker and return */
 559                         BUG_ON(ret);
 560                         extent = NULL;
 561                         btrfs_release_path(root, path);
 562                         /* the extent will be freed later */
 563                 }
 564                 if (bookend && found_inline && start <= key.offset) {
 565                         u32 new_size;
 566                         new_size = btrfs_file_extent_calc_inline_size(
 567                                                    extent_end - end);
 568                         inode_sub_bytes(inode, end - key.offset);
 569                         ret = btrfs_truncate_item(trans, root, path,
 570                                                   new_size, 0);
 571                         BUG_ON(ret);
 572                 }
 573                 /* create bookend, splitting the extent in two */
 574                 if (bookend && found_extent) {
 575                         u64 disk_bytenr;
 576                         struct btrfs_key ins;
 577                         ins.objectid = inode->i_ino;
 578                         ins.offset = end;
 579                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 580                         btrfs_release_path(root, path);
 581                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 582                                                       sizeof(*extent));
 583                         BUG_ON(ret);
 584
 585                         leaf = path->nodes[0];
 586                         extent = btrfs_item_ptr(leaf, path->slots[0],
 587                                                 struct btrfs_file_extent_item);
 588                         write_extent_buffer(leaf, &old,
 589                                             (unsigned long)extent, sizeof(old));
 590
 591                         btrfs_set_file_extent_compression(leaf, extent,
 592                                                           compression);
 593                         btrfs_set_file_extent_encryption(leaf, extent,
 594                                                          encryption);
 595                         btrfs_set_file_extent_other_encoding(leaf, extent,
 596                                                              other_encoding);
 597                         btrfs_set_file_extent_offset(leaf, extent,
 598                                     le64_to_cpu(old.offset) + end - key.offset);
 599                         WARN_ON(le64_to_cpu(old.num_bytes) <
 600                                 (extent_end - end));
 601                         btrfs_set_file_extent_num_bytes(leaf, extent,
 602                                                         extent_end - end);
 603
 604                         /*
 605                          * set the ram bytes to the size of the full extent
 606                          * before splitting.  This is a worst case flag,
 607                          * but its the best we can do because we don't know
 608                          * how splitting affects compression
 609                          */
 610                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 611                                                         ram_bytes);
 612                         btrfs_set_file_extent_type(leaf, extent,
 613                                                    BTRFS_FILE_EXTENT_REG);
 614
 615                         btrfs_mark_buffer_dirty(path->nodes[0]);
 616
 617                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 618                         if (disk_bytenr != 0) {
 619                                 ret = btrfs_inc_extent_ref(trans, root,
 620                                                 disk_bytenr,
 621                                                 le64_to_cpu(old.disk_num_bytes),
 622                                                 leaf->start,
 623                                                 root->root_key.objectid,
 624                                                 trans->transid, ins.objectid);
 625                                 BUG_ON(ret);
 626                         }
 627                         btrfs_release_path(root, path);
 628                         if (disk_bytenr != 0) {
 629                                 inode_add_bytes(inode, extent_end - end);
 630                         }
 631                 }
 632
 633                 if (found_extent && !keep) {
 634                         u64 disk_bytenr = le64_to_cpu(old.disk_bytenr);
 635
 636                         if (disk_bytenr != 0) {
 637                                 inode_sub_bytes(inode,
 638                                                 le64_to_cpu(old.num_bytes));
 639                                 ret = btrfs_free_extent(trans, root,
 640                                                 disk_bytenr,
 641                                                 le64_to_cpu(old.disk_num_bytes),
 642                                                 leaf_start, root_owner,
 643                                                 root_gen, key.objectid, 0);
 644                                 BUG_ON(ret);
 645                                 *hint_byte = disk_bytenr;
 646                         }
 647                 }
 648
 649                 if (search_start >= end) {
 650                         ret = 0;
 651                         goto out;
 652                 }
 653         }
 654 out:
 655         btrfs_free_path(path);
 656         if (locked_end > end) {
 657                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
 658                               GFP_NOFS);
 659         }
 660         btrfs_check_file(root, inode);
 661         return ret;
 662 }
 663
 664 /*
 665  * this gets pages into the page cache and locks them down, it also properly
 666  * waits for data=ordered extents to finish before allowing the pages to be
 667  * modified.
 668  */
 669 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
 670                          struct page **pages, size_t num_pages,
 671                          loff_t pos, unsigned long first_index,
 672                          unsigned long last_index, size_t write_bytes)
 673 {
 674         int i;
 675         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 676         struct inode *inode = fdentry(file)->d_inode;
 677         int err = 0;
 678         u64 start_pos;
 679         u64 last_pos;
 680
 681         start_pos = pos & ~((u64)root->sectorsize - 1);
 682         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 683
 684         if (start_pos > inode->i_size) {
 685                 err = btrfs_cont_expand(inode, start_pos);
 686                 if (err)
 687                         return err;
 688         }
 689
 690         memset(pages, 0, num_pages * sizeof(struct page *));
 691 again:
 692         for (i = 0; i < num_pages; i++) {
 693                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 694                 if (!pages[i]) {
 695                         err = -ENOMEM;
 696                         BUG_ON(1);
 697                 }
 698                 wait_on_page_writeback(pages[i]);
 699         }
 700         if (start_pos < inode->i_size) {
 701                 struct btrfs_ordered_extent *ordered;
 702                 lock_extent(&BTRFS_I(inode)->io_tree,
 703                             start_pos, last_pos - 1, GFP_NOFS);
 704                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 705                 if (ordered &&
 706                     ordered->file_offset + ordered->len > start_pos &&
 707                     ordered->file_offset < last_pos) {
 708                         btrfs_put_ordered_extent(ordered);
 709                         unlock_extent(&BTRFS_I(inode)->io_tree,
 710                                       start_pos, last_pos - 1, GFP_NOFS);
 711                         for (i = 0; i < num_pages; i++) {
 712                                 unlock_page(pages[i]);
 713                                 page_cache_release(pages[i]);
 714                         }
 715                         btrfs_wait_ordered_range(inode, start_pos,
 716                                                  last_pos - start_pos);
 717                         goto again;
 718                 }
 719                 if (ordered)
 720                         btrfs_put_ordered_extent(ordered);
 721
 722                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 723                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 724                                   GFP_NOFS);
 725                 unlock_extent(&BTRFS_I(inode)->io_tree,
 726                               start_pos, last_pos - 1, GFP_NOFS);
 727         }
 728         for (i = 0; i < num_pages; i++) {
 729                 clear_page_dirty_for_io(pages[i]);
 730                 set_page_extent_mapped(pages[i]);
 731                 WARN_ON(!PageLocked(pages[i]));
 732         }
 733         return 0;
 734 }
 735
 736 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 737                                 size_t count, loff_t *ppos)
 738 {
 739         loff_t pos;
 740         loff_t start_pos;
 741         ssize_t num_written = 0;
 742         ssize_t err = 0;
 743         int ret = 0;
 744         struct inode *inode = fdentry(file)->d_inode;
 745         struct btrfs_root *root = BTRFS_I(inode)->root;
 746         struct page **pages = NULL;
 747         int nrptrs;
 748         struct page *pinned[2];
 749         unsigned long first_index;
 750         unsigned long last_index;
 751         int will_write;
 752
 753         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 754                       (file->f_flags & O_DIRECT));
 755
 756         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 757                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 758         pinned[0] = NULL;
 759         pinned[1] = NULL;
 760
 761         pos = *ppos;
 762         start_pos = pos;
 763
 764         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 765         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 766         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 767         if (err)
 768                 goto out_nolock;
 769         if (count == 0)
 770                 goto out_nolock;
 771
 772         err = file_remove_suid(file);
 773         if (err)
 774                 goto out_nolock;
 775         file_update_time(file);
 776
 777         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 778
 779         mutex_lock(&inode->i_mutex);
 780         first_index = pos >> PAGE_CACHE_SHIFT;
 781         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 782
 783         /*
 784          * if this is a nodatasum mount, force summing off for the inode
 785          * all the time.  That way a later mount with summing on won't
 786          * get confused
 787          */
 788         if (btrfs_test_opt(root, NODATASUM))
 789                 btrfs_set_flag(inode, NODATASUM);
 790
 791         /*
 792          * there are lots of better ways to do this, but this code
 793          * makes sure the first and last page in the file range are
 794          * up to date and ready for cow
 795          */
 796         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 797                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 798                 if (!PageUptodate(pinned[0])) {
 799                         ret = btrfs_readpage(NULL, pinned[0]);
 800                         BUG_ON(ret);
 801                         wait_on_page_locked(pinned[0]);
 802                 } else {
 803                         unlock_page(pinned[0]);
 804                 }
 805         }
 806         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 807                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 808                 if (!PageUptodate(pinned[1])) {
 809                         ret = btrfs_readpage(NULL, pinned[1]);
 810                         BUG_ON(ret);
 811                         wait_on_page_locked(pinned[1]);
 812                 } else {
 813                         unlock_page(pinned[1]);
 814                 }
 815         }
 816
 817         while(count > 0) {
 818                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 819                 size_t write_bytes = min(count, nrptrs *
 820                                         (size_t)PAGE_CACHE_SIZE -
 821                                          offset);
 822                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 823                                         PAGE_CACHE_SHIFT;
 824
 825                 WARN_ON(num_pages > nrptrs);
 826                 memset(pages, 0, sizeof(pages));
 827
 828                 ret = btrfs_check_free_space(root, write_bytes, 0);
 829                 if (ret)
 830                         goto out;
 831
 832                 ret = prepare_pages(root, file, pages, num_pages,
 833                                     pos, first_index, last_index,
 834                                     write_bytes);
 835                 if (ret)
 836                         goto out;
 837
 838                 ret = btrfs_copy_from_user(pos, num_pages,
 839                                            write_bytes, pages, buf);
 840                 if (ret) {
 841                         btrfs_drop_pages(pages, num_pages);
 842                         goto out;
 843                 }
 844
 845                 ret = dirty_and_release_pages(NULL, root, file, pages,
 846                                               num_pages, pos, write_bytes);
 847                 btrfs_drop_pages(pages, num_pages);
 848                 if (ret)
 849                         goto out;
 850
 851                 if (will_write) {
 852                         btrfs_fdatawrite_range(inode->i_mapping, pos,
 853                                                pos + write_bytes - 1,
 854                                                WB_SYNC_NONE);
 855                 } else {
 856                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
 857                                                            num_pages);
 858                         if (num_pages <
 859                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 860                                 btrfs_btree_balance_dirty(root, 1);
 861                         btrfs_throttle(root);
 862                 }
 863
 864                 buf += write_bytes;
 865                 count -= write_bytes;
 866                 pos += write_bytes;
 867                 num_written += write_bytes;
 868
 869                 cond_resched();
 870         }
 871 out:
 872         mutex_unlock(&inode->i_mutex);
 873
 874 out_nolock:
 875         kfree(pages);
 876         if (pinned[0])
 877                 page_cache_release(pinned[0]);
 878         if (pinned[1])
 879                 page_cache_release(pinned[1]);
 880         *ppos = pos;
 881
 882         if (num_written > 0 && will_write) {
 883                 struct btrfs_trans_handle *trans;
 884
 885                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
 886                 if (err)
 887                         num_written = err;
 888
 889                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
 890                         trans = btrfs_start_transaction(root, 1);
 891                         ret = btrfs_log_dentry_safe(trans, root,
 892                                                     file->f_dentry);
 893                         if (ret == 0) {
 894                                 btrfs_sync_log(trans, root);
 895                                 btrfs_end_transaction(trans, root);
 896                         } else {
 897                                 btrfs_commit_transaction(trans, root);
 898                         }
 899                 }
 900                 if (file->f_flags & O_DIRECT) {
 901                         invalidate_mapping_pages(inode->i_mapping,
 902                               start_pos >> PAGE_CACHE_SHIFT,
 903                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 904                 }
 905         }
 906         current->backing_dev_info = NULL;
 907         return num_written ? num_written : err;
 908 }
 909
 910 int btrfs_release_file(struct inode * inode, struct file * filp)
 911 {
 912         if (filp->private_data)
 913                 btrfs_ioctl_trans_end(filp);
 914         return 0;
 915 }
 916
 917 /*
 918  * fsync call for both files and directories.  This logs the inode into
 919  * the tree log instead of forcing full commits whenever possible.
 920  *
 921  * It needs to call filemap_fdatawait so that all ordered extent updates are
 922  * in the metadata btree are up to date for copying to the log.
 923  *
 924  * It drops the inode mutex before doing the tree log commit.  This is an
 925  * important optimization for directories because holding the mutex prevents
 926  * new operations on the dir while we write to disk.
 927  */
 928 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 929 {
 930         struct inode *inode = dentry->d_inode;
 931         struct btrfs_root *root = BTRFS_I(inode)->root;
 932         int ret = 0;
 933         struct btrfs_trans_handle *trans;
 934
 935         /*
 936          * check the transaction that last modified this inode
 937          * and see if its already been committed
 938          */
 939         if (!BTRFS_I(inode)->last_trans)
 940                 goto out;
 941
 942         mutex_lock(&root->fs_info->trans_mutex);
 943         if (BTRFS_I(inode)->last_trans <=
 944             root->fs_info->last_trans_committed) {
 945                 BTRFS_I(inode)->last_trans = 0;
 946                 mutex_unlock(&root->fs_info->trans_mutex);
 947                 goto out;
 948         }
 949         mutex_unlock(&root->fs_info->trans_mutex);
 950
 951         root->fs_info->tree_log_batch++;
 952         filemap_fdatawait(inode->i_mapping);
 953         root->fs_info->tree_log_batch++;
 954
 955         /*
 956          * ok we haven't committed the transaction yet, lets do a commit
 957          */
 958         if (file->private_data)
 959                 btrfs_ioctl_trans_end(file);
 960
 961         trans = btrfs_start_transaction(root, 1);
 962         if (!trans) {
 963                 ret = -ENOMEM;
 964                 goto out;
 965         }
 966
 967         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
 968         if (ret < 0) {
 969                 goto out;
 970         }
 971
 972         /* we've logged all the items and now have a consistent
 973          * version of the file in the log.  It is possible that
 974          * someone will come in and modify the file, but that's
 975          * fine because the log is consistent on disk, and we
 976          * have references to all of the file's extents
 977          *
 978          * It is possible that someone will come in and log the
 979          * file again, but that will end up using the synchronization
 980          * inside btrfs_sync_log to keep things safe.
 981          */
 982         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 983
 984         if (ret > 0) {
 985                 ret = btrfs_commit_transaction(trans, root);
 986         } else {
 987                 btrfs_sync_log(trans, root);
 988                 ret = btrfs_end_transaction(trans, root);
 989         }
 990         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 991 out:
 992         return ret > 0 ? EIO : ret;
 993 }
 994
 995 static struct vm_operations_struct btrfs_file_vm_ops = {
 996         .fault          = filemap_fault,
 997         .page_mkwrite   = btrfs_page_mkwrite,
 998 };
 999
1000 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1001 {
1002         vma->vm_ops = &btrfs_file_vm_ops;
1003         file_accessed(filp);
1004         return 0;
1005 }
1006
1007 struct file_operations btrfs_file_operations = {
1008         .llseek         = generic_file_llseek,
1009         .read           = do_sync_read,
1010         .aio_read       = generic_file_aio_read,
1011         .splice_read    = generic_file_splice_read,
1012         .write          = btrfs_file_write,
1013         .mmap           = btrfs_file_mmap,
1014         .open           = generic_file_open,
1015         .release        = btrfs_release_file,
1016         .fsync          = btrfs_sync_file,
1017         .unlocked_ioctl = btrfs_ioctl,
1018 #ifdef CONFIG_COMPAT
1019         .compat_ioctl   = btrfs_ioctl,
1020 #endif
1021 };