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 "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "ioctl.h"
  37 #include "print-tree.h"
  38 #include "tree-log.h"
  39 #include "locking.h"
  40 #include "compat.h"
  41
  42
  43 /* simple helper to fault in pages and copy.  This should go away
  44  * and be replaced with calls into generic code.
  45  */
  46 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  47                                          int write_bytes,
  48                                          struct page **prepared_pages,
  49                                          const char __user *buf)
  50 {
  51         long page_fault = 0;
  52         int i;
  53         int offset = pos & (PAGE_CACHE_SIZE - 1);
  54
  55         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  56                 size_t count = min_t(size_t,
  57                                      PAGE_CACHE_SIZE - offset, write_bytes);
  58                 struct page *page = prepared_pages[i];
  59                 fault_in_pages_readable(buf, count);
  60
  61                 /* Copy data from userspace to the current page */
  62                 kmap(page);
  63                 page_fault = __copy_from_user(page_address(page) + offset,
  64                                               buf, count);
  65                 /* Flush processor's dcache for this page */
  66                 flush_dcache_page(page);
  67                 kunmap(page);
  68                 buf += count;
  69                 write_bytes -= count;
  70
  71                 if (page_fault)
  72                         break;
  73         }
  74         return page_fault ? -EFAULT : 0;
  75 }
  76
  77 /*
  78  * unlocks pages after btrfs_file_write is done with them
  79  */
  80 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
  81 {
  82         size_t i;
  83         for (i = 0; i < num_pages; i++) {
  84                 if (!pages[i])
  85                         break;
  86                 /* page checked is some magic around finding pages that
  87                  * have been modified without going through btrfs_set_page_dirty
  88                  * clear it here
  89                  */
  90                 ClearPageChecked(pages[i]);
  91                 unlock_page(pages[i]);
  92                 mark_page_accessed(pages[i]);
  93                 page_cache_release(pages[i]);
  94         }
  95 }
  96
  97 /*
  98  * after copy_from_user, pages need to be dirtied and we need to make
  99  * sure holes are created between the current EOF and the start of
 100  * any next extents (if required).
 101  *
 102  * this also makes the decision about creating an inline extent vs
 103  * doing real data extents, marking pages dirty and delalloc as required.
 104  */
 105 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 106                                    struct btrfs_root *root,
 107                                    struct file *file,
 108                                    struct page **pages,
 109                                    size_t num_pages,
 110                                    loff_t pos,
 111                                    size_t write_bytes)
 112 {
 113         int err = 0;
 114         int i;
 115         struct inode *inode = fdentry(file)->d_inode;
 116         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 117         u64 hint_byte;
 118         u64 num_bytes;
 119         u64 start_pos;
 120         u64 end_of_last_block;
 121         u64 end_pos = pos + write_bytes;
 122         loff_t isize = i_size_read(inode);
 123
 124         start_pos = pos & ~((u64)root->sectorsize - 1);
 125         num_bytes = (write_bytes + pos - start_pos +
 126                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 127
 128         end_of_last_block = start_pos + num_bytes - 1;
 129
 130         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 131         trans = btrfs_join_transaction(root, 1);
 132         if (!trans) {
 133                 err = -ENOMEM;
 134                 goto out_unlock;
 135         }
 136         btrfs_set_trans_block_group(trans, inode);
 137         hint_byte = 0;
 138
 139         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 140
 141         /* check for reserved extents on each page, we don't want
 142          * to reset the delalloc bit on things that already have
 143          * extents reserved.
 144          */
 145         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 146         for (i = 0; i < num_pages; i++) {
 147                 struct page *p = pages[i];
 148                 SetPageUptodate(p);
 149                 ClearPageChecked(p);
 150                 set_page_dirty(p);
 151         }
 152         if (end_pos > isize) {
 153                 i_size_write(inode, end_pos);
 154                 /* we've only changed i_size in ram, and we haven't updated
 155                  * the disk i_size.  There is no need to log the inode
 156                  * at this time.
 157                  */
 158         }
 159         err = btrfs_end_transaction(trans, root);
 160 out_unlock:
 161         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 162         return err;
 163 }
 164
 165 /*
 166  * this drops all the extents in the cache that intersect the range
 167  * [start, end].  Existing extents are split as required.
 168  */
 169 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 170                             int skip_pinned)
 171 {
 172         struct extent_map *em;
 173         struct extent_map *split = NULL;
 174         struct extent_map *split2 = NULL;
 175         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 176         u64 len = end - start + 1;
 177         int ret;
 178         int testend = 1;
 179         unsigned long flags;
 180         int compressed = 0;
 181
 182         WARN_ON(end < start);
 183         if (end == (u64)-1) {
 184                 len = (u64)-1;
 185                 testend = 0;
 186         }
 187         while (1) {
 188                 if (!split)
 189                         split = alloc_extent_map(GFP_NOFS);
 190                 if (!split2)
 191                         split2 = alloc_extent_map(GFP_NOFS);
 192
 193                 spin_lock(&em_tree->lock);
 194                 em = lookup_extent_mapping(em_tree, start, len);
 195                 if (!em) {
 196                         spin_unlock(&em_tree->lock);
 197                         break;
 198                 }
 199                 flags = em->flags;
 200                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 201                         spin_unlock(&em_tree->lock);
 202                         if (em->start <= start &&
 203                             (!testend || em->start + em->len >= start + len)) {
 204                                 free_extent_map(em);
 205                                 break;
 206                         }
 207                         if (start < em->start) {
 208                                 len = em->start - start;
 209                         } else {
 210                                 len = start + len - (em->start + em->len);
 211                                 start = em->start + em->len;
 212                         }
 213                         free_extent_map(em);
 214                         continue;
 215                 }
 216                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 217                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 218                 remove_extent_mapping(em_tree, em);
 219
 220                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 221                     em->start < start) {
 222                         split->start = em->start;
 223                         split->len = start - em->start;
 224                         split->orig_start = em->orig_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                                 split->orig_start = em->orig_start;
 253                         } else {
 254                                 split->block_len = split->len;
 255                                 split->block_start = em->block_start + diff;
 256                                 split->orig_start = split->start;
 257                         }
 258
 259                         ret = add_extent_mapping(em_tree, split);
 260                         BUG_ON(ret);
 261                         free_extent_map(split);
 262                         split = NULL;
 263                 }
 264                 spin_unlock(&em_tree->lock);
 265
 266                 /* once for us */
 267                 free_extent_map(em);
 268                 /* once for the tree*/
 269                 free_extent_map(em);
 270         }
 271         if (split)
 272                 free_extent_map(split);
 273         if (split2)
 274                 free_extent_map(split2);
 275         return 0;
 276 }
 277
 278 /*
 279  * this is very complex, but the basic idea is to drop all extents
 280  * in the range start - end.  hint_block is filled in with a block number
 281  * that would be a good hint to the block allocator for this file.
 282  *
 283  * If an extent intersects the range but is not entirely inside the range
 284  * it is either truncated or split.  Anything entirely inside the range
 285  * is deleted from the tree.
 286  *
 287  * inline_limit is used to tell this code which offsets in the file to keep
 288  * if they contain inline extents.
 289  */
 290 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 291                        struct btrfs_root *root, struct inode *inode,
 292                        u64 start, u64 end, u64 locked_end,
 293                        u64 inline_limit, u64 *hint_byte)
 294 {
 295         u64 extent_end = 0;
 296         u64 search_start = start;
 297         u64 ram_bytes = 0;
 298         u64 disk_bytenr = 0;
 299         u64 orig_locked_end = locked_end;
 300         u8 compression;
 301         u8 encryption;
 302         u16 other_encoding = 0;
 303         struct extent_buffer *leaf;
 304         struct btrfs_file_extent_item *extent;
 305         struct btrfs_path *path;
 306         struct btrfs_key key;
 307         struct btrfs_file_extent_item old;
 308         int keep;
 309         int slot;
 310         int bookend;
 311         int found_type = 0;
 312         int found_extent;
 313         int found_inline;
 314         int recow;
 315         int ret;
 316
 317         inline_limit = 0;
 318         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 319
 320         path = btrfs_alloc_path();
 321         if (!path)
 322                 return -ENOMEM;
 323         while (1) {
 324                 recow = 0;
 325                 btrfs_release_path(root, path);
 326                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 327                                                search_start, -1);
 328                 if (ret < 0)
 329                         goto out;
 330                 if (ret > 0) {
 331                         if (path->slots[0] == 0) {
 332                                 ret = 0;
 333                                 goto out;
 334                         }
 335                         path->slots[0]--;
 336                 }
 337 next_slot:
 338                 keep = 0;
 339                 bookend = 0;
 340                 found_extent = 0;
 341                 found_inline = 0;
 342                 compression = 0;
 343                 encryption = 0;
 344                 extent = NULL;
 345                 leaf = path->nodes[0];
 346                 slot = path->slots[0];
 347                 ret = 0;
 348                 btrfs_item_key_to_cpu(leaf, &key, slot);
 349                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 350                     key.offset >= end) {
 351                         goto out;
 352                 }
 353                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 354                     key.objectid != inode->i_ino) {
 355                         goto out;
 356                 }
 357                 if (recow) {
 358                         search_start = max(key.offset, start);
 359                         continue;
 360                 }
 361                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 362                         extent = btrfs_item_ptr(leaf, slot,
 363                                                 struct btrfs_file_extent_item);
 364                         found_type = btrfs_file_extent_type(leaf, extent);
 365                         compression = btrfs_file_extent_compression(leaf,
 366                                                                     extent);
 367                         encryption = btrfs_file_extent_encryption(leaf,
 368                                                                   extent);
 369                         other_encoding = btrfs_file_extent_other_encoding(leaf,
 370                                                                   extent);
 371                         if (found_type == BTRFS_FILE_EXTENT_REG ||
 372                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 373                                 extent_end =
 374                                      btrfs_file_extent_disk_bytenr(leaf,
 375                                                                    extent);
 376                                 if (extent_end)
 377                                         *hint_byte = extent_end;
 378
 379                                 extent_end = key.offset +
 380                                      btrfs_file_extent_num_bytes(leaf, extent);
 381                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
 382                                                                 extent);
 383                                 found_extent = 1;
 384                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 385                                 found_inline = 1;
 386                                 extent_end = key.offset +
 387                                      btrfs_file_extent_inline_len(leaf, extent);
 388                         }
 389                 } else {
 390                         extent_end = search_start;
 391                 }
 392
 393                 /* we found nothing we can drop */
 394                 if ((!found_extent && !found_inline) ||
 395                     search_start >= extent_end) {
 396                         int nextret;
 397                         u32 nritems;
 398                         nritems = btrfs_header_nritems(leaf);
 399                         if (slot >= nritems - 1) {
 400                                 nextret = btrfs_next_leaf(root, path);
 401                                 if (nextret)
 402                                         goto out;
 403                                 recow = 1;
 404                         } else {
 405                                 path->slots[0]++;
 406                         }
 407                         goto next_slot;
 408                 }
 409
 410                 if (end <= extent_end && start >= key.offset && found_inline)
 411                         *hint_byte = EXTENT_MAP_INLINE;
 412
 413                 if (found_extent) {
 414                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 415                                            sizeof(old));
 416                 }
 417
 418                 if (end < extent_end && end >= key.offset) {
 419                         bookend = 1;
 420                         if (found_inline && start <= key.offset)
 421                                 keep = 1;
 422                 }
 423
 424                 if (bookend && found_extent) {
 425                         if (locked_end < extent_end) {
 426                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 427                                                 locked_end, extent_end - 1,
 428                                                 GFP_NOFS);
 429                                 if (!ret) {
 430                                         btrfs_release_path(root, path);
 431                                         lock_extent(&BTRFS_I(inode)->io_tree,
 432                                                 locked_end, extent_end - 1,
 433                                                 GFP_NOFS);
 434                                         locked_end = extent_end;
 435                                         continue;
 436                                 }
 437                                 locked_end = extent_end;
 438                         }
 439                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 440                         if (disk_bytenr != 0) {
 441                                 ret = btrfs_inc_extent_ref(trans, root,
 442                                            disk_bytenr,
 443                                            le64_to_cpu(old.disk_num_bytes), 0,
 444                                            root->root_key.objectid,
 445                                            key.objectid, key.offset -
 446                                            le64_to_cpu(old.offset));
 447                                 BUG_ON(ret);
 448                         }
 449                 }
 450
 451                 if (found_inline) {
 452                         u64 mask = root->sectorsize - 1;
 453                         search_start = (extent_end + mask) & ~mask;
 454                 } else
 455                         search_start = extent_end;
 456
 457                 /* truncate existing extent */
 458                 if (start > key.offset) {
 459                         u64 new_num;
 460                         u64 old_num;
 461                         keep = 1;
 462                         WARN_ON(start & (root->sectorsize - 1));
 463                         if (found_extent) {
 464                                 new_num = start - key.offset;
 465                                 old_num = btrfs_file_extent_num_bytes(leaf,
 466                                                                       extent);
 467                                 *hint_byte =
 468                                         btrfs_file_extent_disk_bytenr(leaf,
 469                                                                       extent);
 470                                 if (btrfs_file_extent_disk_bytenr(leaf,
 471                                                                   extent)) {
 472                                         inode_sub_bytes(inode, old_num -
 473                                                         new_num);
 474                                 }
 475                                 btrfs_set_file_extent_num_bytes(leaf,
 476                                                         extent, new_num);
 477                                 btrfs_mark_buffer_dirty(leaf);
 478                         } else if (key.offset < inline_limit &&
 479                                    (end > extent_end) &&
 480                                    (inline_limit < extent_end)) {
 481                                 u32 new_size;
 482                                 new_size = btrfs_file_extent_calc_inline_size(
 483                                                    inline_limit - key.offset);
 484                                 inode_sub_bytes(inode, extent_end -
 485                                                 inline_limit);
 486                                 btrfs_set_file_extent_ram_bytes(leaf, extent,
 487                                                         new_size);
 488                                 if (!compression && !encryption) {
 489                                         btrfs_truncate_item(trans, root, path,
 490                                                             new_size, 1);
 491                                 }
 492                         }
 493                 }
 494                 /* delete the entire extent */
 495                 if (!keep) {
 496                         if (found_inline)
 497                                 inode_sub_bytes(inode, extent_end -
 498                                                 key.offset);
 499                         ret = btrfs_del_item(trans, root, path);
 500                         /* TODO update progress marker and return */
 501                         BUG_ON(ret);
 502                         extent = NULL;
 503                         btrfs_release_path(root, path);
 504                         /* the extent will be freed later */
 505                 }
 506                 if (bookend && found_inline && start <= key.offset) {
 507                         u32 new_size;
 508                         new_size = btrfs_file_extent_calc_inline_size(
 509                                                    extent_end - end);
 510                         inode_sub_bytes(inode, end - key.offset);
 511                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 512                                                         new_size);
 513                         if (!compression && !encryption)
 514                                 ret = btrfs_truncate_item(trans, root, path,
 515                                                           new_size, 0);
 516                         BUG_ON(ret);
 517                 }
 518                 /* create bookend, splitting the extent in two */
 519                 if (bookend && found_extent) {
 520                         struct btrfs_key ins;
 521                         ins.objectid = inode->i_ino;
 522                         ins.offset = end;
 523                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 524
 525                         btrfs_release_path(root, path);
 526                         path->leave_spinning = 1;
 527                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 528                                                       sizeof(*extent));
 529                         BUG_ON(ret);
 530
 531                         leaf = path->nodes[0];
 532                         extent = btrfs_item_ptr(leaf, path->slots[0],
 533                                                 struct btrfs_file_extent_item);
 534                         write_extent_buffer(leaf, &old,
 535                                             (unsigned long)extent, sizeof(old));
 536
 537                         btrfs_set_file_extent_compression(leaf, extent,
 538                                                           compression);
 539                         btrfs_set_file_extent_encryption(leaf, extent,
 540                                                          encryption);
 541                         btrfs_set_file_extent_other_encoding(leaf, extent,
 542                                                              other_encoding);
 543                         btrfs_set_file_extent_offset(leaf, extent,
 544                                     le64_to_cpu(old.offset) + end - key.offset);
 545                         WARN_ON(le64_to_cpu(old.num_bytes) <
 546                                 (extent_end - end));
 547                         btrfs_set_file_extent_num_bytes(leaf, extent,
 548                                                         extent_end - end);
 549
 550                         /*
 551                          * set the ram bytes to the size of the full extent
 552                          * before splitting.  This is a worst case flag,
 553                          * but its the best we can do because we don't know
 554                          * how splitting affects compression
 555                          */
 556                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 557                                                         ram_bytes);
 558                         btrfs_set_file_extent_type(leaf, extent, found_type);
 559
 560                         btrfs_unlock_up_safe(path, 1);
 561                         btrfs_mark_buffer_dirty(path->nodes[0]);
 562                         btrfs_set_lock_blocking(path->nodes[0]);
 563
 564                         path->leave_spinning = 0;
 565                         btrfs_release_path(root, path);
 566                         if (disk_bytenr != 0)
 567                                 inode_add_bytes(inode, extent_end - end);
 568                 }
 569
 570                 if (found_extent && !keep) {
 571                         u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
 572
 573                         if (old_disk_bytenr != 0) {
 574                                 inode_sub_bytes(inode,
 575                                                 le64_to_cpu(old.num_bytes));
 576                                 ret = btrfs_free_extent(trans, root,
 577                                                 old_disk_bytenr,
 578                                                 le64_to_cpu(old.disk_num_bytes),
 579                                                 0, root->root_key.objectid,
 580                                                 key.objectid, key.offset -
 581                                                 le64_to_cpu(old.offset));
 582                                 BUG_ON(ret);
 583                                 *hint_byte = old_disk_bytenr;
 584                         }
 585                 }
 586
 587                 if (search_start >= end) {
 588                         ret = 0;
 589                         goto out;
 590                 }
 591         }
 592 out:
 593         btrfs_free_path(path);
 594         if (locked_end > orig_locked_end) {
 595                 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
 596                               locked_end - 1, GFP_NOFS);
 597         }
 598         return ret;
 599 }
 600
 601 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 602                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
 603 {
 604         struct btrfs_file_extent_item *fi;
 605         struct btrfs_key key;
 606         u64 extent_end;
 607
 608         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 609                 return 0;
 610
 611         btrfs_item_key_to_cpu(leaf, &key, slot);
 612         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 613                 return 0;
 614
 615         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 616         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 617             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 618             btrfs_file_extent_compression(leaf, fi) ||
 619             btrfs_file_extent_encryption(leaf, fi) ||
 620             btrfs_file_extent_other_encoding(leaf, fi))
 621                 return 0;
 622
 623         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 624         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 625                 return 0;
 626
 627         *start = key.offset;
 628         *end = extent_end;
 629         return 1;
 630 }
 631
 632 /*
 633  * Mark extent in the range start - end as written.
 634  *
 635  * This changes extent type from 'pre-allocated' to 'regular'. If only
 636  * part of extent is marked as written, the extent will be split into
 637  * two or three.
 638  */
 639 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 640                               struct btrfs_root *root,
 641                               struct inode *inode, u64 start, u64 end)
 642 {
 643         struct extent_buffer *leaf;
 644         struct btrfs_path *path;
 645         struct btrfs_file_extent_item *fi;
 646         struct btrfs_key key;
 647         u64 bytenr;
 648         u64 num_bytes;
 649         u64 extent_end;
 650         u64 orig_offset;
 651         u64 other_start;
 652         u64 other_end;
 653         u64 split = start;
 654         u64 locked_end = end;
 655         int extent_type;
 656         int split_end = 1;
 657         int ret;
 658
 659         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 660
 661         path = btrfs_alloc_path();
 662         BUG_ON(!path);
 663 again:
 664         key.objectid = inode->i_ino;
 665         key.type = BTRFS_EXTENT_DATA_KEY;
 666         if (split == start)
 667                 key.offset = split;
 668         else
 669                 key.offset = split - 1;
 670
 671         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 672         if (ret > 0 && path->slots[0] > 0)
 673                 path->slots[0]--;
 674
 675         leaf = path->nodes[0];
 676         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 677         BUG_ON(key.objectid != inode->i_ino ||
 678                key.type != BTRFS_EXTENT_DATA_KEY);
 679         fi = btrfs_item_ptr(leaf, path->slots[0],
 680                             struct btrfs_file_extent_item);
 681         extent_type = btrfs_file_extent_type(leaf, fi);
 682         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
 683         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 684         BUG_ON(key.offset > start || extent_end < end);
 685
 686         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 687         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 688         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 689
 690         if (key.offset == start)
 691                 split = end;
 692
 693         if (key.offset == start && extent_end == end) {
 694                 int del_nr = 0;
 695                 int del_slot = 0;
 696                 other_start = end;
 697                 other_end = 0;
 698                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 699                                      bytenr, &other_start, &other_end)) {
 700                         extent_end = other_end;
 701                         del_slot = path->slots[0] + 1;
 702                         del_nr++;
 703                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 704                                                 0, root->root_key.objectid,
 705                                                 inode->i_ino, orig_offset);
 706                         BUG_ON(ret);
 707                 }
 708                 other_start = 0;
 709                 other_end = start;
 710                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
 711                                      bytenr, &other_start, &other_end)) {
 712                         key.offset = other_start;
 713                         del_slot = path->slots[0];
 714                         del_nr++;
 715                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 716                                                 0, root->root_key.objectid,
 717                                                 inode->i_ino, orig_offset);
 718                         BUG_ON(ret);
 719                 }
 720                 split_end = 0;
 721                 if (del_nr == 0) {
 722                         btrfs_set_file_extent_type(leaf, fi,
 723                                                    BTRFS_FILE_EXTENT_REG);
 724                         goto done;
 725                 }
 726
 727                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 728                                     struct btrfs_file_extent_item);
 729                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 730                 btrfs_set_file_extent_num_bytes(leaf, fi,
 731                                                 extent_end - key.offset);
 732                 btrfs_mark_buffer_dirty(leaf);
 733
 734                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 735                 BUG_ON(ret);
 736                 goto release;
 737         } else if (split == start) {
 738                 if (locked_end < extent_end) {
 739                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 740                                         locked_end, extent_end - 1, GFP_NOFS);
 741                         if (!ret) {
 742                                 btrfs_release_path(root, path);
 743                                 lock_extent(&BTRFS_I(inode)->io_tree,
 744                                         locked_end, extent_end - 1, GFP_NOFS);
 745                                 locked_end = extent_end;
 746                                 goto again;
 747                         }
 748                         locked_end = extent_end;
 749                 }
 750                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
 751         } else  {
 752                 BUG_ON(key.offset != start);
 753                 key.offset = split;
 754                 btrfs_set_file_extent_offset(leaf, fi, key.offset -
 755                                              orig_offset);
 756                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
 757                 btrfs_set_item_key_safe(trans, root, path, &key);
 758                 extent_end = split;
 759         }
 760
 761         if (extent_end == end) {
 762                 split_end = 0;
 763                 extent_type = BTRFS_FILE_EXTENT_REG;
 764         }
 765         if (extent_end == end && split == start) {
 766                 other_start = end;
 767                 other_end = 0;
 768                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 769                                      bytenr, &other_start, &other_end)) {
 770                         path->slots[0]++;
 771                         fi = btrfs_item_ptr(leaf, path->slots[0],
 772                                             struct btrfs_file_extent_item);
 773                         key.offset = split;
 774                         btrfs_set_item_key_safe(trans, root, path, &key);
 775                         btrfs_set_file_extent_offset(leaf, fi, key.offset -
 776                                                      orig_offset);
 777                         btrfs_set_file_extent_num_bytes(leaf, fi,
 778                                                         other_end - split);
 779                         goto done;
 780                 }
 781         }
 782         if (extent_end == end && split == end) {
 783                 other_start = 0;
 784                 other_end = start;
 785                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
 786                                      bytenr, &other_start, &other_end)) {
 787                         path->slots[0]--;
 788                         fi = btrfs_item_ptr(leaf, path->slots[0],
 789                                             struct btrfs_file_extent_item);
 790                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
 791                                                         other_start);
 792                         goto done;
 793                 }
 794         }
 795
 796         btrfs_mark_buffer_dirty(leaf);
 797
 798         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 799                                    root->root_key.objectid,
 800                                    inode->i_ino, orig_offset);
 801         BUG_ON(ret);
 802         btrfs_release_path(root, path);
 803
 804         key.offset = start;
 805         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
 806         BUG_ON(ret);
 807
 808         leaf = path->nodes[0];
 809         fi = btrfs_item_ptr(leaf, path->slots[0],
 810                             struct btrfs_file_extent_item);
 811         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 812         btrfs_set_file_extent_type(leaf, fi, extent_type);
 813         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
 814         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
 815         btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
 816         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
 817         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 818         btrfs_set_file_extent_compression(leaf, fi, 0);
 819         btrfs_set_file_extent_encryption(leaf, fi, 0);
 820         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 821 done:
 822         btrfs_mark_buffer_dirty(leaf);
 823
 824 release:
 825         btrfs_release_path(root, path);
 826         if (split_end && split == start) {
 827                 split = end;
 828                 goto again;
 829         }
 830         if (locked_end > end) {
 831                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
 832                               GFP_NOFS);
 833         }
 834         btrfs_free_path(path);
 835         return 0;
 836 }
 837
 838 /*
 839  * this gets pages into the page cache and locks them down, it also properly
 840  * waits for data=ordered extents to finish before allowing the pages to be
 841  * modified.
 842  */
 843 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 844                          struct page **pages, size_t num_pages,
 845                          loff_t pos, unsigned long first_index,
 846                          unsigned long last_index, size_t write_bytes)
 847 {
 848         int i;
 849         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 850         struct inode *inode = fdentry(file)->d_inode;
 851         int err = 0;
 852         u64 start_pos;
 853         u64 last_pos;
 854
 855         start_pos = pos & ~((u64)root->sectorsize - 1);
 856         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 857
 858         if (start_pos > inode->i_size) {
 859                 err = btrfs_cont_expand(inode, start_pos);
 860                 if (err)
 861                         return err;
 862         }
 863
 864         memset(pages, 0, num_pages * sizeof(struct page *));
 865 again:
 866         for (i = 0; i < num_pages; i++) {
 867                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 868                 if (!pages[i]) {
 869                         err = -ENOMEM;
 870                         BUG_ON(1);
 871                 }
 872                 wait_on_page_writeback(pages[i]);
 873         }
 874         if (start_pos < inode->i_size) {
 875                 struct btrfs_ordered_extent *ordered;
 876                 lock_extent(&BTRFS_I(inode)->io_tree,
 877                             start_pos, last_pos - 1, GFP_NOFS);
 878                 ordered = btrfs_lookup_first_ordered_extent(inode,
 879                                                             last_pos - 1);
 880                 if (ordered &&
 881                     ordered->file_offset + ordered->len > start_pos &&
 882                     ordered->file_offset < last_pos) {
 883                         btrfs_put_ordered_extent(ordered);
 884                         unlock_extent(&BTRFS_I(inode)->io_tree,
 885                                       start_pos, last_pos - 1, GFP_NOFS);
 886                         for (i = 0; i < num_pages; i++) {
 887                                 unlock_page(pages[i]);
 888                                 page_cache_release(pages[i]);
 889                         }
 890                         btrfs_wait_ordered_range(inode, start_pos,
 891                                                  last_pos - start_pos);
 892                         goto again;
 893                 }
 894                 if (ordered)
 895                         btrfs_put_ordered_extent(ordered);
 896
 897                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 898                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 899                                   GFP_NOFS);
 900                 unlock_extent(&BTRFS_I(inode)->io_tree,
 901                               start_pos, last_pos - 1, GFP_NOFS);
 902         }
 903         for (i = 0; i < num_pages; i++) {
 904                 clear_page_dirty_for_io(pages[i]);
 905                 set_page_extent_mapped(pages[i]);
 906                 WARN_ON(!PageLocked(pages[i]));
 907         }
 908         return 0;
 909 }
 910
 911 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 912                                 size_t count, loff_t *ppos)
 913 {
 914         loff_t pos;
 915         loff_t start_pos;
 916         ssize_t num_written = 0;
 917         ssize_t err = 0;
 918         int ret = 0;
 919         struct inode *inode = fdentry(file)->d_inode;
 920         struct btrfs_root *root = BTRFS_I(inode)->root;
 921         struct page **pages = NULL;
 922         int nrptrs;
 923         struct page *pinned[2];
 924         unsigned long first_index;
 925         unsigned long last_index;
 926         int will_write;
 927
 928         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 929                       (file->f_flags & O_DIRECT));
 930
 931         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 932                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 933         pinned[0] = NULL;
 934         pinned[1] = NULL;
 935
 936         pos = *ppos;
 937         start_pos = pos;
 938
 939         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 940         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 941         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 942         if (err)
 943                 goto out_nolock;
 944         if (count == 0)
 945                 goto out_nolock;
 946
 947         err = file_remove_suid(file);
 948         if (err)
 949                 goto out_nolock;
 950         file_update_time(file);
 951
 952         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 953
 954         mutex_lock(&inode->i_mutex);
 955         BTRFS_I(inode)->sequence++;
 956         first_index = pos >> PAGE_CACHE_SHIFT;
 957         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 958
 959         /*
 960          * there are lots of better ways to do this, but this code
 961          * makes sure the first and last page in the file range are
 962          * up to date and ready for cow
 963          */
 964         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 965                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 966                 if (!PageUptodate(pinned[0])) {
 967                         ret = btrfs_readpage(NULL, pinned[0]);
 968                         BUG_ON(ret);
 969                         wait_on_page_locked(pinned[0]);
 970                 } else {
 971                         unlock_page(pinned[0]);
 972                 }
 973         }
 974         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 975                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 976                 if (!PageUptodate(pinned[1])) {
 977                         ret = btrfs_readpage(NULL, pinned[1]);
 978                         BUG_ON(ret);
 979                         wait_on_page_locked(pinned[1]);
 980                 } else {
 981                         unlock_page(pinned[1]);
 982                 }
 983         }
 984
 985         while (count > 0) {
 986                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 987                 size_t write_bytes = min(count, nrptrs *
 988                                         (size_t)PAGE_CACHE_SIZE -
 989                                          offset);
 990                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 991                                         PAGE_CACHE_SHIFT;
 992
 993                 WARN_ON(num_pages > nrptrs);
 994                 memset(pages, 0, sizeof(struct page *) * nrptrs);
 995
 996                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
 997                 if (ret)
 998                         goto out;
 999
1000                 ret = prepare_pages(root, file, pages, num_pages,
1001                                     pos, first_index, last_index,
1002                                     write_bytes);
1003                 if (ret) {
1004                         btrfs_free_reserved_data_space(root, inode,
1005                                                        write_bytes);
1006                         goto out;
1007                 }
1008
1009                 ret = btrfs_copy_from_user(pos, num_pages,
1010                                            write_bytes, pages, buf);
1011                 if (ret) {
1012                         btrfs_free_reserved_data_space(root, inode,
1013                                                        write_bytes);
1014                         btrfs_drop_pages(pages, num_pages);
1015                         goto out;
1016                 }
1017
1018                 ret = dirty_and_release_pages(NULL, root, file, pages,
1019                                               num_pages, pos, write_bytes);
1020                 btrfs_drop_pages(pages, num_pages);
1021                 if (ret) {
1022                         btrfs_free_reserved_data_space(root, inode,
1023                                                        write_bytes);
1024                         goto out;
1025                 }
1026
1027                 if (will_write) {
1028                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1029                                                pos + write_bytes - 1,
1030                                                WB_SYNC_ALL);
1031                 } else {
1032                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1033                                                            num_pages);
1034                         if (num_pages <
1035                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1036                                 btrfs_btree_balance_dirty(root, 1);
1037                         btrfs_throttle(root);
1038                 }
1039
1040                 buf += write_bytes;
1041                 count -= write_bytes;
1042                 pos += write_bytes;
1043                 num_written += write_bytes;
1044
1045                 cond_resched();
1046         }
1047 out:
1048         mutex_unlock(&inode->i_mutex);
1049         if (ret)
1050                 err = ret;
1051
1052 out_nolock:
1053         kfree(pages);
1054         if (pinned[0])
1055                 page_cache_release(pinned[0]);
1056         if (pinned[1])
1057                 page_cache_release(pinned[1]);
1058         *ppos = pos;
1059
1060         /*
1061          * we want to make sure fsync finds this change
1062          * but we haven't joined a transaction running right now.
1063          *
1064          * Later on, someone is sure to update the inode and get the
1065          * real transid recorded.
1066          *
1067          * We set last_trans now to the fs_info generation + 1,
1068          * this will either be one more than the running transaction
1069          * or the generation used for the next transaction if there isn't
1070          * one running right now.
1071          */
1072         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1073
1074         if (num_written > 0 && will_write) {
1075                 struct btrfs_trans_handle *trans;
1076
1077                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1078                 if (err)
1079                         num_written = err;
1080
1081                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1082                         trans = btrfs_start_transaction(root, 1);
1083                         ret = btrfs_log_dentry_safe(trans, root,
1084                                                     file->f_dentry);
1085                         if (ret == 0) {
1086                                 ret = btrfs_sync_log(trans, root);
1087                                 if (ret == 0)
1088                                         btrfs_end_transaction(trans, root);
1089                                 else
1090                                         btrfs_commit_transaction(trans, root);
1091                         } else {
1092                                 btrfs_commit_transaction(trans, root);
1093                         }
1094                 }
1095                 if (file->f_flags & O_DIRECT) {
1096                         invalidate_mapping_pages(inode->i_mapping,
1097                               start_pos >> PAGE_CACHE_SHIFT,
1098                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1099                 }
1100         }
1101         current->backing_dev_info = NULL;
1102         return num_written ? num_written : err;
1103 }
1104
1105 int btrfs_release_file(struct inode *inode, struct file *filp)
1106 {
1107         /*
1108          * ordered_data_close is set by settattr when we are about to truncate
1109          * a file from a non-zero size to a zero size.  This tries to
1110          * flush down new bytes that may have been written if the
1111          * application were using truncate to replace a file in place.
1112          */
1113         if (BTRFS_I(inode)->ordered_data_close) {
1114                 BTRFS_I(inode)->ordered_data_close = 0;
1115                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1116                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1117                         filemap_flush(inode->i_mapping);
1118         }
1119         if (filp->private_data)
1120                 btrfs_ioctl_trans_end(filp);
1121         return 0;
1122 }
1123
1124 /*
1125  * fsync call for both files and directories.  This logs the inode into
1126  * the tree log instead of forcing full commits whenever possible.
1127  *
1128  * It needs to call filemap_fdatawait so that all ordered extent updates are
1129  * in the metadata btree are up to date for copying to the log.
1130  *
1131  * It drops the inode mutex before doing the tree log commit.  This is an
1132  * important optimization for directories because holding the mutex prevents
1133  * new operations on the dir while we write to disk.
1134  */
1135 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1136 {
1137         struct inode *inode = dentry->d_inode;
1138         struct btrfs_root *root = BTRFS_I(inode)->root;
1139         int ret = 0;
1140         struct btrfs_trans_handle *trans;
1141
1142         /*
1143          * check the transaction that last modified this inode
1144          * and see if its already been committed
1145          */
1146         if (!BTRFS_I(inode)->last_trans)
1147                 goto out;
1148
1149         mutex_lock(&root->fs_info->trans_mutex);
1150         if (BTRFS_I(inode)->last_trans <=
1151             root->fs_info->last_trans_committed) {
1152                 BTRFS_I(inode)->last_trans = 0;
1153                 mutex_unlock(&root->fs_info->trans_mutex);
1154                 goto out;
1155         }
1156         mutex_unlock(&root->fs_info->trans_mutex);
1157
1158         root->log_batch++;
1159         filemap_fdatawrite(inode->i_mapping);
1160         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1161         root->log_batch++;
1162
1163         if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1164                 goto out;
1165         /*
1166          * ok we haven't committed the transaction yet, lets do a commit
1167          */
1168         if (file && file->private_data)
1169                 btrfs_ioctl_trans_end(file);
1170
1171         trans = btrfs_start_transaction(root, 1);
1172         if (!trans) {
1173                 ret = -ENOMEM;
1174                 goto out;
1175         }
1176
1177         ret = btrfs_log_dentry_safe(trans, root, dentry);
1178         if (ret < 0)
1179                 goto out;
1180
1181         /* we've logged all the items and now have a consistent
1182          * version of the file in the log.  It is possible that
1183          * someone will come in and modify the file, but that's
1184          * fine because the log is consistent on disk, and we
1185          * have references to all of the file's extents
1186          *
1187          * It is possible that someone will come in and log the
1188          * file again, but that will end up using the synchronization
1189          * inside btrfs_sync_log to keep things safe.
1190          */
1191         mutex_unlock(&dentry->d_inode->i_mutex);
1192
1193         if (ret > 0) {
1194                 ret = btrfs_commit_transaction(trans, root);
1195         } else {
1196                 ret = btrfs_sync_log(trans, root);
1197                 if (ret == 0)
1198                         ret = btrfs_end_transaction(trans, root);
1199                 else
1200                         ret = btrfs_commit_transaction(trans, root);
1201         }
1202         mutex_lock(&dentry->d_inode->i_mutex);
1203 out:
1204         return ret > 0 ? EIO : ret;
1205 }
1206
1207 static struct vm_operations_struct btrfs_file_vm_ops = {
1208         .fault          = filemap_fault,
1209         .page_mkwrite   = btrfs_page_mkwrite,
1210 };
1211
1212 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1213 {
1214         vma->vm_ops = &btrfs_file_vm_ops;
1215         file_accessed(filp);
1216         return 0;
1217 }
1218
1219 struct file_operations btrfs_file_operations = {
1220         .llseek         = generic_file_llseek,
1221         .read           = do_sync_read,
1222         .aio_read       = generic_file_aio_read,
1223         .splice_read    = generic_file_splice_read,
1224         .write          = btrfs_file_write,
1225         .mmap           = btrfs_file_mmap,
1226         .open           = generic_file_open,
1227         .release        = btrfs_release_file,
1228         .fsync          = btrfs_sync_file,
1229         .unlocked_ioctl = btrfs_ioctl,
1230 #ifdef CONFIG_COMPAT
1231         .compat_ioctl   = btrfs_ioctl,
1232 #endif
1233 };