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         /* FIXME...EIEIO, ENOSPC and more */
 146         /* insert any holes we need to create */
 147         if (isize < start_pos) {
 148                 u64 last_pos_in_file;
 149                 u64 hole_size;
 150                 u64 mask = root->sectorsize - 1;
 151                 last_pos_in_file = (isize + mask) & ~mask;
 152                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 153                 if (hole_size > 0) {
 154                         btrfs_wait_ordered_range(inode, last_pos_in_file,
 155                                                  last_pos_in_file + hole_size);
 156                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 157                         err = btrfs_drop_extents(trans, root, inode,
 158                                                  last_pos_in_file,
 159                                                  last_pos_in_file + hole_size,
 160                                                  last_pos_in_file,
 161                                                  &hint_byte);
 162                         if (err)
 163                                 goto failed;
 164
 165                         err = btrfs_insert_file_extent(trans, root,
 166                                                        inode->i_ino,
 167                                                        last_pos_in_file,
 168                                                        0, 0, hole_size, 0,
 169                                                        hole_size, 0, 0, 0);
 170                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 171                                         last_pos_in_file + hole_size - 1, 0);
 172                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 173                         btrfs_check_file(root, inode);
 174                 }
 175                 if (err)
 176                         goto failed;
 177         }
 178
 179         /* check for reserved extents on each page, we don't want
 180          * to reset the delalloc bit on things that already have
 181          * extents reserved.
 182          */
 183         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 184         for (i = 0; i < num_pages; i++) {
 185                 struct page *p = pages[i];
 186                 SetPageUptodate(p);
 187                 ClearPageChecked(p);
 188                 set_page_dirty(p);
 189         }
 190         if (end_pos > isize) {
 191                 i_size_write(inode, end_pos);
 192                 btrfs_update_inode(trans, root, inode);
 193         }
 194 failed:
 195         err = btrfs_end_transaction(trans, root);
 196 out_unlock:
 197         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 198         return err;
 199 }
 200
 201 /*
 202  * this drops all the extents in the cache that intersect the range
 203  * [start, end].  Existing extents are split as required.
 204  */
 205 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 206                             int skip_pinned)
 207 {
 208         struct extent_map *em;
 209         struct extent_map *split = NULL;
 210         struct extent_map *split2 = NULL;
 211         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 212         u64 len = end - start + 1;
 213         int ret;
 214         int testend = 1;
 215         unsigned long flags;
 216         int compressed = 0;
 217
 218         WARN_ON(end < start);
 219         if (end == (u64)-1) {
 220                 len = (u64)-1;
 221                 testend = 0;
 222         }
 223         while(1) {
 224                 if (!split)
 225                         split = alloc_extent_map(GFP_NOFS);
 226                 if (!split2)
 227                         split2 = alloc_extent_map(GFP_NOFS);
 228
 229                 spin_lock(&em_tree->lock);
 230                 em = lookup_extent_mapping(em_tree, start, len);
 231                 if (!em) {
 232                         spin_unlock(&em_tree->lock);
 233                         break;
 234                 }
 235                 flags = em->flags;
 236                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 237                         spin_unlock(&em_tree->lock);
 238                         if (em->start <= start &&
 239                             (!testend || em->start + em->len >= start + len)) {
 240                                 free_extent_map(em);
 241                                 break;
 242                         }
 243                         if (start < em->start) {
 244                                 len = em->start - start;
 245                         } else {
 246                                 len = start + len - (em->start + em->len);
 247                                 start = em->start + em->len;
 248                         }
 249                         free_extent_map(em);
 250                         continue;
 251                 }
 252                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 253                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 254                 remove_extent_mapping(em_tree, em);
 255
 256                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 257                     em->start < start) {
 258                         split->start = em->start;
 259                         split->len = start - em->start;
 260                         split->block_start = em->block_start;
 261
 262                         if (compressed)
 263                                 split->block_len = em->block_len;
 264                         else
 265                                 split->block_len = split->len;
 266
 267                         split->bdev = em->bdev;
 268                         split->flags = flags;
 269                         ret = add_extent_mapping(em_tree, split);
 270                         BUG_ON(ret);
 271                         free_extent_map(split);
 272                         split = split2;
 273                         split2 = NULL;
 274                 }
 275                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 276                     testend && em->start + em->len > start + len) {
 277                         u64 diff = start + len - em->start;
 278
 279                         split->start = start + len;
 280                         split->len = em->start + em->len - (start + len);
 281                         split->bdev = em->bdev;
 282                         split->flags = flags;
 283
 284                         if (compressed) {
 285                                 split->block_len = em->block_len;
 286                                 split->block_start = em->block_start;
 287                         } else {
 288                                 split->block_len = split->len;
 289                                 split->block_start = em->block_start + diff;
 290                         }
 291
 292                         ret = add_extent_mapping(em_tree, split);
 293                         BUG_ON(ret);
 294                         free_extent_map(split);
 295                         split = NULL;
 296                 }
 297                 spin_unlock(&em_tree->lock);
 298
 299                 /* once for us */
 300                 free_extent_map(em);
 301                 /* once for the tree*/
 302                 free_extent_map(em);
 303         }
 304         if (split)
 305                 free_extent_map(split);
 306         if (split2)
 307                 free_extent_map(split2);
 308         return 0;
 309 }
 310
 311 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 312 {
 313         return 0;
 314 #if 0
 315         struct btrfs_path *path;
 316         struct btrfs_key found_key;
 317         struct extent_buffer *leaf;
 318         struct btrfs_file_extent_item *extent;
 319         u64 last_offset = 0;
 320         int nritems;
 321         int slot;
 322         int found_type;
 323         int ret;
 324         int err = 0;
 325         u64 extent_end = 0;
 326
 327         path = btrfs_alloc_path();
 328         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 329                                        last_offset, 0);
 330         while(1) {
 331                 nritems = btrfs_header_nritems(path->nodes[0]);
 332                 if (path->slots[0] >= nritems) {
 333                         ret = btrfs_next_leaf(root, path);
 334                         if (ret)
 335                                 goto out;
 336                         nritems = btrfs_header_nritems(path->nodes[0]);
 337                 }
 338                 slot = path->slots[0];
 339                 leaf = path->nodes[0];
 340                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 341                 if (found_key.objectid != inode->i_ino)
 342                         break;
 343                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 344                         goto out;
 345
 346                 if (found_key.offset < last_offset) {
 347                         WARN_ON(1);
 348                         btrfs_print_leaf(root, leaf);
 349                         printk("inode %lu found offset %Lu expected %Lu\n",
 350                                inode->i_ino, found_key.offset, last_offset);
 351                         err = 1;
 352                         goto out;
 353                 }
 354                 extent = btrfs_item_ptr(leaf, slot,
 355                                         struct btrfs_file_extent_item);
 356                 found_type = btrfs_file_extent_type(leaf, extent);
 357                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 358                         extent_end = found_key.offset +
 359                              btrfs_file_extent_num_bytes(leaf, extent);
 360                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 361                         struct btrfs_item *item;
 362                         item = btrfs_item_nr(leaf, slot);
 363                         extent_end = found_key.offset +
 364                              btrfs_file_extent_inline_len(leaf, extent);
 365                         extent_end = (extent_end + root->sectorsize - 1) &
 366                                 ~((u64)root->sectorsize -1 );
 367                 }
 368                 last_offset = extent_end;
 369                 path->slots[0]++;
 370         }
 371         if (0 && last_offset < inode->i_size) {
 372                 WARN_ON(1);
 373                 btrfs_print_leaf(root, leaf);
 374                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 375                        last_offset, inode->i_size);
 376                 err = 1;
 377
 378         }
 379 out:
 380         btrfs_free_path(path);
 381         return err;
 382 #endif
 383 }
 384
 385 /*
 386  * this is very complex, but the basic idea is to drop all extents
 387  * in the range start - end.  hint_block is filled in with a block number
 388  * that would be a good hint to the block allocator for this file.
 389  *
 390  * If an extent intersects the range but is not entirely inside the range
 391  * it is either truncated or split.  Anything entirely inside the range
 392  * is deleted from the tree.
 393  *
 394  * inline_limit is used to tell this code which offsets in the file to keep
 395  * if they contain inline extents.
 396  */
 397 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 398                        struct btrfs_root *root, struct inode *inode,
 399                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 400 {
 401         u64 extent_end = 0;
 402         u64 search_start = start;
 403         u64 leaf_start;
 404         u64 ram_bytes = 0;
 405         u8 compression = 0;
 406         u8 encryption = 0;
 407         u16 other_encoding = 0;
 408         u64 root_gen;
 409         u64 root_owner;
 410         struct extent_buffer *leaf;
 411         struct btrfs_file_extent_item *extent;
 412         struct btrfs_path *path;
 413         struct btrfs_key key;
 414         struct btrfs_file_extent_item old;
 415         int keep;
 416         int slot;
 417         int bookend;
 418         int found_type;
 419         int found_extent;
 420         int found_inline;
 421         int recow;
 422         int ret;
 423
 424         inline_limit = 0;
 425         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 426
 427         path = btrfs_alloc_path();
 428         if (!path)
 429                 return -ENOMEM;
 430         while(1) {
 431                 recow = 0;
 432                 btrfs_release_path(root, path);
 433                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 434                                                search_start, -1);
 435                 if (ret < 0)
 436                         goto out;
 437                 if (ret > 0) {
 438                         if (path->slots[0] == 0) {
 439                                 ret = 0;
 440                                 goto out;
 441                         }
 442                         path->slots[0]--;
 443                 }
 444 next_slot:
 445                 keep = 0;
 446                 bookend = 0;
 447                 found_extent = 0;
 448                 found_inline = 0;
 449                 leaf_start = 0;
 450                 root_gen = 0;
 451                 root_owner = 0;
 452                 extent = NULL;
 453                 leaf = path->nodes[0];
 454                 slot = path->slots[0];
 455                 ret = 0;
 456                 btrfs_item_key_to_cpu(leaf, &key, slot);
 457                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 458                     key.offset >= end) {
 459                         goto out;
 460                 }
 461                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 462                     key.objectid != inode->i_ino) {
 463                         goto out;
 464                 }
 465                 if (recow) {
 466                         search_start = key.offset;
 467                         continue;
 468                 }
 469                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 470                         extent = btrfs_item_ptr(leaf, slot,
 471                                                 struct btrfs_file_extent_item);
 472                         found_type = btrfs_file_extent_type(leaf, extent);
 473                         compression = btrfs_file_extent_compression(leaf,
 474                                                                     extent);
 475                         encryption = btrfs_file_extent_encryption(leaf,
 476                                                                   extent);
 477                         other_encoding = btrfs_file_extent_other_encoding(leaf,
 478                                                                   extent);
 479                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 480                                 extent_end =
 481                                      btrfs_file_extent_disk_bytenr(leaf,
 482                                                                    extent);
 483                                 if (extent_end)
 484                                         *hint_byte = extent_end;
 485
 486                                 extent_end = key.offset +
 487                                      btrfs_file_extent_num_bytes(leaf, extent);
 488                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
 489                                                                 extent);
 490                                 found_extent = 1;
 491                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 492                                 found_inline = 1;
 493                                 extent_end = key.offset +
 494                                      btrfs_file_extent_inline_len(leaf, extent);
 495                         }
 496                 } else {
 497                         extent_end = search_start;
 498                 }
 499
 500                 /* we found nothing we can drop */
 501                 if ((!found_extent && !found_inline) ||
 502                     search_start >= extent_end) {
 503                         int nextret;
 504                         u32 nritems;
 505                         nritems = btrfs_header_nritems(leaf);
 506                         if (slot >= nritems - 1) {
 507                                 nextret = btrfs_next_leaf(root, path);
 508                                 if (nextret)
 509                                         goto out;
 510                                 recow = 1;
 511                         } else {
 512                                 path->slots[0]++;
 513                         }
 514                         goto next_slot;
 515                 }
 516
 517                 if (found_inline) {
 518                         u64 mask = root->sectorsize - 1;
 519                         search_start = (extent_end + mask) & ~mask;
 520                 } else
 521                         search_start = extent_end;
 522
 523                 if (end <= extent_end && start >= key.offset && found_inline)
 524                         *hint_byte = EXTENT_MAP_INLINE;
 525
 526                 if (found_extent) {
 527                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 528                                            sizeof(old));
 529                         root_gen = btrfs_header_generation(leaf);
 530                         root_owner = btrfs_header_owner(leaf);
 531                         leaf_start = leaf->start;
 532                 }
 533
 534                 if (end < extent_end && end >= key.offset) {
 535                         bookend = 1;
 536                         if (found_inline && start <= key.offset)
 537                                 keep = 1;
 538                 }
 539                 /* truncate existing extent */
 540                 if (start > key.offset) {
 541                         u64 new_num;
 542                         u64 old_num;
 543                         keep = 1;
 544                         WARN_ON(start & (root->sectorsize - 1));
 545                         if (found_extent) {
 546                                 new_num = start - key.offset;
 547                                 old_num = btrfs_file_extent_num_bytes(leaf,
 548                                                                       extent);
 549                                 *hint_byte =
 550                                         btrfs_file_extent_disk_bytenr(leaf,
 551                                                                       extent);
 552                                 if (btrfs_file_extent_disk_bytenr(leaf,
 553                                                                   extent)) {
 554                                         inode_sub_bytes(inode, old_num -
 555                                                         new_num);
 556                                 }
 557                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 558                                                                 new_num);
 559                                 btrfs_mark_buffer_dirty(leaf);
 560                         } else if (key.offset < inline_limit &&
 561                                    (end > extent_end) &&
 562                                    (inline_limit < extent_end)) {
 563                                 u32 new_size;
 564                                 new_size = btrfs_file_extent_calc_inline_size(
 565                                                    inline_limit - key.offset);
 566                                 inode_sub_bytes(inode, extent_end -
 567                                                 inline_limit);
 568                                 btrfs_truncate_item(trans, root, path,
 569                                                     new_size, 1);
 570                         }
 571                 }
 572                 /* delete the entire extent */
 573                 if (!keep) {
 574                         if (found_inline)
 575                                 inode_sub_bytes(inode, extent_end -
 576                                                 key.offset);
 577                         ret = btrfs_del_item(trans, root, path);
 578                         /* TODO update progress marker and return */
 579                         BUG_ON(ret);
 580                         extent = NULL;
 581                         btrfs_release_path(root, path);
 582                         /* the extent will be freed later */
 583                 }
 584                 if (bookend && found_inline && start <= key.offset) {
 585                         u32 new_size;
 586                         new_size = btrfs_file_extent_calc_inline_size(
 587                                                    extent_end - end);
 588                         inode_sub_bytes(inode, end - key.offset);
 589                         ret = btrfs_truncate_item(trans, root, path,
 590                                                   new_size, 0);
 591                         BUG_ON(ret);
 592                 }
 593                 /* create bookend, splitting the extent in two */
 594                 if (bookend && found_extent) {
 595                         u64 disk_bytenr;
 596                         struct btrfs_key ins;
 597                         ins.objectid = inode->i_ino;
 598                         ins.offset = end;
 599                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 600                         btrfs_release_path(root, path);
 601                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 602                                                       sizeof(*extent));
 603                         BUG_ON(ret);
 604
 605                         leaf = path->nodes[0];
 606                         extent = btrfs_item_ptr(leaf, path->slots[0],
 607                                                 struct btrfs_file_extent_item);
 608                         write_extent_buffer(leaf, &old,
 609                                             (unsigned long)extent, sizeof(old));
 610
 611                         btrfs_set_file_extent_compression(leaf, extent,
 612                                                           compression);
 613                         btrfs_set_file_extent_encryption(leaf, extent,
 614                                                          encryption);
 615                         btrfs_set_file_extent_other_encoding(leaf, extent,
 616                                                              other_encoding);
 617                         btrfs_set_file_extent_offset(leaf, extent,
 618                                     le64_to_cpu(old.offset) + end - key.offset);
 619                         WARN_ON(le64_to_cpu(old.num_bytes) <
 620                                 (extent_end - end));
 621                         btrfs_set_file_extent_num_bytes(leaf, extent,
 622                                                         extent_end - end);
 623
 624                         /*
 625                          * set the ram bytes to the size of the full extent
 626                          * before splitting.  This is a worst case flag,
 627                          * but its the best we can do because we don't know
 628                          * how splitting affects compression
 629                          */
 630                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 631                                                         ram_bytes);
 632                         btrfs_set_file_extent_type(leaf, extent,
 633                                                    BTRFS_FILE_EXTENT_REG);
 634
 635                         btrfs_mark_buffer_dirty(path->nodes[0]);
 636
 637                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 638                         if (disk_bytenr != 0) {
 639                                 ret = btrfs_inc_extent_ref(trans, root,
 640                                                 disk_bytenr,
 641                                                 le64_to_cpu(old.disk_num_bytes),
 642                                                 leaf->start,
 643                                                 root->root_key.objectid,
 644                                                 trans->transid, ins.objectid);
 645                                 BUG_ON(ret);
 646                         }
 647                         btrfs_release_path(root, path);
 648                         if (disk_bytenr != 0) {
 649                                 inode_add_bytes(inode, extent_end - end);
 650                         }
 651                 }
 652
 653                 if (found_extent && !keep) {
 654                         u64 disk_bytenr = le64_to_cpu(old.disk_bytenr);
 655
 656                         if (disk_bytenr != 0) {
 657                                 inode_sub_bytes(inode,
 658                                                 le64_to_cpu(old.num_bytes));
 659                                 ret = btrfs_free_extent(trans, root,
 660                                                 disk_bytenr,
 661                                                 le64_to_cpu(old.disk_num_bytes),
 662                                                 leaf_start, root_owner,
 663                                                 root_gen, key.objectid, 0);
 664                                 BUG_ON(ret);
 665                                 *hint_byte = disk_bytenr;
 666                         }
 667                 }
 668
 669                 if (search_start >= end) {
 670                         ret = 0;
 671                         goto out;
 672                 }
 673         }
 674 out:
 675         btrfs_free_path(path);
 676         btrfs_check_file(root, inode);
 677         return ret;
 678 }
 679
 680 /*
 681  * this gets pages into the page cache and locks them down, it also properly
 682  * waits for data=ordered extents to finish before allowing the pages to be
 683  * modified.
 684  */
 685 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
 686                          struct page **pages, size_t num_pages,
 687                          loff_t pos, unsigned long first_index,
 688                          unsigned long last_index, size_t write_bytes)
 689 {
 690         int i;
 691         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 692         struct inode *inode = fdentry(file)->d_inode;
 693         int err = 0;
 694         u64 start_pos;
 695         u64 last_pos;
 696
 697         start_pos = pos & ~((u64)root->sectorsize - 1);
 698         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 699
 700         memset(pages, 0, num_pages * sizeof(struct page *));
 701 again:
 702         for (i = 0; i < num_pages; i++) {
 703                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 704                 if (!pages[i]) {
 705                         err = -ENOMEM;
 706                         BUG_ON(1);
 707                 }
 708                 wait_on_page_writeback(pages[i]);
 709         }
 710         if (start_pos < inode->i_size) {
 711                 struct btrfs_ordered_extent *ordered;
 712                 lock_extent(&BTRFS_I(inode)->io_tree,
 713                             start_pos, last_pos - 1, GFP_NOFS);
 714                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 715                 if (ordered &&
 716                     ordered->file_offset + ordered->len > start_pos &&
 717                     ordered->file_offset < last_pos) {
 718                         btrfs_put_ordered_extent(ordered);
 719                         unlock_extent(&BTRFS_I(inode)->io_tree,
 720                                       start_pos, last_pos - 1, GFP_NOFS);
 721                         for (i = 0; i < num_pages; i++) {
 722                                 unlock_page(pages[i]);
 723                                 page_cache_release(pages[i]);
 724                         }
 725                         btrfs_wait_ordered_range(inode, start_pos,
 726                                                  last_pos - start_pos);
 727                         goto again;
 728                 }
 729                 if (ordered)
 730                         btrfs_put_ordered_extent(ordered);
 731
 732                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 733                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 734                                   GFP_NOFS);
 735                 unlock_extent(&BTRFS_I(inode)->io_tree,
 736                               start_pos, last_pos - 1, GFP_NOFS);
 737         }
 738         for (i = 0; i < num_pages; i++) {
 739                 clear_page_dirty_for_io(pages[i]);
 740                 set_page_extent_mapped(pages[i]);
 741                 WARN_ON(!PageLocked(pages[i]));
 742         }
 743         return 0;
 744 }
 745
 746 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 747                                 size_t count, loff_t *ppos)
 748 {
 749         loff_t pos;
 750         loff_t start_pos;
 751         ssize_t num_written = 0;
 752         ssize_t err = 0;
 753         int ret = 0;
 754         struct inode *inode = fdentry(file)->d_inode;
 755         struct btrfs_root *root = BTRFS_I(inode)->root;
 756         struct page **pages = NULL;
 757         int nrptrs;
 758         struct page *pinned[2];
 759         unsigned long first_index;
 760         unsigned long last_index;
 761         int will_write;
 762
 763         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 764                       (file->f_flags & O_DIRECT));
 765
 766         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 767                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 768         pinned[0] = NULL;
 769         pinned[1] = NULL;
 770
 771         pos = *ppos;
 772         start_pos = pos;
 773
 774         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 775         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 776         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 777         if (err)
 778                 goto out_nolock;
 779         if (count == 0)
 780                 goto out_nolock;
 781
 782         err = file_remove_suid(file);
 783         if (err)
 784                 goto out_nolock;
 785         file_update_time(file);
 786
 787         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 788
 789         mutex_lock(&inode->i_mutex);
 790         first_index = pos >> PAGE_CACHE_SHIFT;
 791         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 792
 793         /*
 794          * if this is a nodatasum mount, force summing off for the inode
 795          * all the time.  That way a later mount with summing on won't
 796          * get confused
 797          */
 798         if (btrfs_test_opt(root, NODATASUM))
 799                 btrfs_set_flag(inode, NODATASUM);
 800
 801         /*
 802          * there are lots of better ways to do this, but this code
 803          * makes sure the first and last page in the file range are
 804          * up to date and ready for cow
 805          */
 806         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 807                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 808                 if (!PageUptodate(pinned[0])) {
 809                         ret = btrfs_readpage(NULL, pinned[0]);
 810                         BUG_ON(ret);
 811                         wait_on_page_locked(pinned[0]);
 812                 } else {
 813                         unlock_page(pinned[0]);
 814                 }
 815         }
 816         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 817                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 818                 if (!PageUptodate(pinned[1])) {
 819                         ret = btrfs_readpage(NULL, pinned[1]);
 820                         BUG_ON(ret);
 821                         wait_on_page_locked(pinned[1]);
 822                 } else {
 823                         unlock_page(pinned[1]);
 824                 }
 825         }
 826
 827         while(count > 0) {
 828                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 829                 size_t write_bytes = min(count, nrptrs *
 830                                         (size_t)PAGE_CACHE_SIZE -
 831                                          offset);
 832                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 833                                         PAGE_CACHE_SHIFT;
 834
 835                 WARN_ON(num_pages > nrptrs);
 836                 memset(pages, 0, sizeof(pages));
 837
 838                 ret = btrfs_check_free_space(root, write_bytes, 0);
 839                 if (ret)
 840                         goto out;
 841
 842                 ret = prepare_pages(root, file, pages, num_pages,
 843                                     pos, first_index, last_index,
 844                                     write_bytes);
 845                 if (ret)
 846                         goto out;
 847
 848                 ret = btrfs_copy_from_user(pos, num_pages,
 849                                            write_bytes, pages, buf);
 850                 if (ret) {
 851                         btrfs_drop_pages(pages, num_pages);
 852                         goto out;
 853                 }
 854
 855                 ret = dirty_and_release_pages(NULL, root, file, pages,
 856                                               num_pages, pos, write_bytes);
 857                 btrfs_drop_pages(pages, num_pages);
 858                 if (ret)
 859                         goto out;
 860
 861                 if (will_write) {
 862                         btrfs_fdatawrite_range(inode->i_mapping, pos,
 863                                                pos + write_bytes - 1,
 864                                                WB_SYNC_NONE);
 865                 } else {
 866                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
 867                                                            num_pages);
 868                         if (num_pages <
 869                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 870                                 btrfs_btree_balance_dirty(root, 1);
 871                         btrfs_throttle(root);
 872                 }
 873
 874                 buf += write_bytes;
 875                 count -= write_bytes;
 876                 pos += write_bytes;
 877                 num_written += write_bytes;
 878
 879                 cond_resched();
 880         }
 881 out:
 882         mutex_unlock(&inode->i_mutex);
 883
 884 out_nolock:
 885         kfree(pages);
 886         if (pinned[0])
 887                 page_cache_release(pinned[0]);
 888         if (pinned[1])
 889                 page_cache_release(pinned[1]);
 890         *ppos = pos;
 891
 892         if (num_written > 0 && will_write) {
 893                 struct btrfs_trans_handle *trans;
 894
 895                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
 896                 if (err)
 897                         num_written = err;
 898
 899                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
 900                         trans = btrfs_start_transaction(root, 1);
 901                         ret = btrfs_log_dentry_safe(trans, root,
 902                                                     file->f_dentry);
 903                         if (ret == 0) {
 904                                 btrfs_sync_log(trans, root);
 905                                 btrfs_end_transaction(trans, root);
 906                         } else {
 907                                 btrfs_commit_transaction(trans, root);
 908                         }
 909                 }
 910                 if (file->f_flags & O_DIRECT) {
 911                         invalidate_mapping_pages(inode->i_mapping,
 912                               start_pos >> PAGE_CACHE_SHIFT,
 913                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 914                 }
 915         }
 916         current->backing_dev_info = NULL;
 917         return num_written ? num_written : err;
 918 }
 919
 920 int btrfs_release_file(struct inode * inode, struct file * filp)
 921 {
 922         if (filp->private_data)
 923                 btrfs_ioctl_trans_end(filp);
 924         return 0;
 925 }
 926
 927 /*
 928  * fsync call for both files and directories.  This logs the inode into
 929  * the tree log instead of forcing full commits whenever possible.
 930  *
 931  * It needs to call filemap_fdatawait so that all ordered extent updates are
 932  * in the metadata btree are up to date for copying to the log.
 933  *
 934  * It drops the inode mutex before doing the tree log commit.  This is an
 935  * important optimization for directories because holding the mutex prevents
 936  * new operations on the dir while we write to disk.
 937  */
 938 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
 939 {
 940         struct inode *inode = dentry->d_inode;
 941         struct btrfs_root *root = BTRFS_I(inode)->root;
 942         int ret = 0;
 943         struct btrfs_trans_handle *trans;
 944
 945         /*
 946          * check the transaction that last modified this inode
 947          * and see if its already been committed
 948          */
 949         if (!BTRFS_I(inode)->last_trans)
 950                 goto out;
 951
 952         mutex_lock(&root->fs_info->trans_mutex);
 953         if (BTRFS_I(inode)->last_trans <=
 954             root->fs_info->last_trans_committed) {
 955                 BTRFS_I(inode)->last_trans = 0;
 956                 mutex_unlock(&root->fs_info->trans_mutex);
 957                 goto out;
 958         }
 959         mutex_unlock(&root->fs_info->trans_mutex);
 960
 961         root->fs_info->tree_log_batch++;
 962         filemap_fdatawait(inode->i_mapping);
 963         root->fs_info->tree_log_batch++;
 964
 965         /*
 966          * ok we haven't committed the transaction yet, lets do a commit
 967          */
 968         if (file->private_data)
 969                 btrfs_ioctl_trans_end(file);
 970
 971         trans = btrfs_start_transaction(root, 1);
 972         if (!trans) {
 973                 ret = -ENOMEM;
 974                 goto out;
 975         }
 976
 977         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
 978         if (ret < 0) {
 979                 goto out;
 980         }
 981
 982         /* we've logged all the items and now have a consistent
 983          * version of the file in the log.  It is possible that
 984          * someone will come in and modify the file, but that's
 985          * fine because the log is consistent on disk, and we
 986          * have references to all of the file's extents
 987          *
 988          * It is possible that someone will come in and log the
 989          * file again, but that will end up using the synchronization
 990          * inside btrfs_sync_log to keep things safe.
 991          */
 992         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 993
 994         if (ret > 0) {
 995                 ret = btrfs_commit_transaction(trans, root);
 996         } else {
 997                 btrfs_sync_log(trans, root);
 998                 ret = btrfs_end_transaction(trans, root);
 999         }
1000         mutex_lock(&file->f_dentry->d_inode->i_mutex);
1001 out:
1002         return ret > 0 ? EIO : ret;
1003 }
1004
1005 static struct vm_operations_struct btrfs_file_vm_ops = {
1006         .fault          = filemap_fault,
1007         .page_mkwrite   = btrfs_page_mkwrite,
1008 };
1009
1010 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1011 {
1012         vma->vm_ops = &btrfs_file_vm_ops;
1013         file_accessed(filp);
1014         return 0;
1015 }
1016
1017 struct file_operations btrfs_file_operations = {
1018         .llseek         = generic_file_llseek,
1019         .read           = do_sync_read,
1020         .aio_read       = generic_file_aio_read,
1021         .splice_read    = generic_file_splice_read,
1022         .write          = btrfs_file_write,
1023         .mmap           = btrfs_file_mmap,
1024         .open           = generic_file_open,
1025         .release        = btrfs_release_file,
1026         .fsync          = btrfs_sync_file,
1027         .unlocked_ioctl = btrfs_ioctl,
1028 #ifdef CONFIG_COMPAT
1029         .compat_ioctl   = btrfs_ioctl,
1030 #endif
1031 };