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         u64 num_bytes;
 117         u64 start_pos;
 118         u64 end_of_last_block;
 119         u64 end_pos = pos + write_bytes;
 120         loff_t isize = i_size_read(inode);
 121
 122         start_pos = pos & ~((u64)root->sectorsize - 1);
 123         num_bytes = (write_bytes + pos - start_pos +
 124                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 125
 126         end_of_last_block = start_pos + num_bytes - 1;
 127         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 128         for (i = 0; i < num_pages; i++) {
 129                 struct page *p = pages[i];
 130                 SetPageUptodate(p);
 131                 ClearPageChecked(p);
 132                 set_page_dirty(p);
 133         }
 134         if (end_pos > isize) {
 135                 i_size_write(inode, end_pos);
 136                 /* we've only changed i_size in ram, and we haven't updated
 137                  * the disk i_size.  There is no need to log the inode
 138                  * at this time.
 139                  */
 140         }
 141         return err;
 142 }
 143
 144 /*
 145  * this drops all the extents in the cache that intersect the range
 146  * [start, end].  Existing extents are split as required.
 147  */
 148 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 149                             int skip_pinned)
 150 {
 151         struct extent_map *em;
 152         struct extent_map *split = NULL;
 153         struct extent_map *split2 = NULL;
 154         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 155         u64 len = end - start + 1;
 156         int ret;
 157         int testend = 1;
 158         unsigned long flags;
 159         int compressed = 0;
 160
 161         WARN_ON(end < start);
 162         if (end == (u64)-1) {
 163                 len = (u64)-1;
 164                 testend = 0;
 165         }
 166         while (1) {
 167                 if (!split)
 168                         split = alloc_extent_map(GFP_NOFS);
 169                 if (!split2)
 170                         split2 = alloc_extent_map(GFP_NOFS);
 171
 172                 write_lock(&em_tree->lock);
 173                 em = lookup_extent_mapping(em_tree, start, len);
 174                 if (!em) {
 175                         write_unlock(&em_tree->lock);
 176                         break;
 177                 }
 178                 flags = em->flags;
 179                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 180                         if (em->start <= start &&
 181                             (!testend || em->start + em->len >= start + len)) {
 182                                 free_extent_map(em);
 183                                 write_unlock(&em_tree->lock);
 184                                 break;
 185                         }
 186                         if (start < em->start) {
 187                                 len = em->start - start;
 188                         } else {
 189                                 len = start + len - (em->start + em->len);
 190                                 start = em->start + em->len;
 191                         }
 192                         free_extent_map(em);
 193                         write_unlock(&em_tree->lock);
 194                         continue;
 195                 }
 196                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 197                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 198                 remove_extent_mapping(em_tree, em);
 199
 200                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 201                     em->start < start) {
 202                         split->start = em->start;
 203                         split->len = start - em->start;
 204                         split->orig_start = em->orig_start;
 205                         split->block_start = em->block_start;
 206
 207                         if (compressed)
 208                                 split->block_len = em->block_len;
 209                         else
 210                                 split->block_len = split->len;
 211
 212                         split->bdev = em->bdev;
 213                         split->flags = flags;
 214                         ret = add_extent_mapping(em_tree, split);
 215                         BUG_ON(ret);
 216                         free_extent_map(split);
 217                         split = split2;
 218                         split2 = NULL;
 219                 }
 220                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 221                     testend && em->start + em->len > start + len) {
 222                         u64 diff = start + len - em->start;
 223
 224                         split->start = start + len;
 225                         split->len = em->start + em->len - (start + len);
 226                         split->bdev = em->bdev;
 227                         split->flags = flags;
 228
 229                         if (compressed) {
 230                                 split->block_len = em->block_len;
 231                                 split->block_start = em->block_start;
 232                                 split->orig_start = em->orig_start;
 233                         } else {
 234                                 split->block_len = split->len;
 235                                 split->block_start = em->block_start + diff;
 236                                 split->orig_start = split->start;
 237                         }
 238
 239                         ret = add_extent_mapping(em_tree, split);
 240                         BUG_ON(ret);
 241                         free_extent_map(split);
 242                         split = NULL;
 243                 }
 244                 write_unlock(&em_tree->lock);
 245
 246                 /* once for us */
 247                 free_extent_map(em);
 248                 /* once for the tree*/
 249                 free_extent_map(em);
 250         }
 251         if (split)
 252                 free_extent_map(split);
 253         if (split2)
 254                 free_extent_map(split2);
 255         return 0;
 256 }
 257
 258 /*
 259  * this is very complex, but the basic idea is to drop all extents
 260  * in the range start - end.  hint_block is filled in with a block number
 261  * that would be a good hint to the block allocator for this file.
 262  *
 263  * If an extent intersects the range but is not entirely inside the range
 264  * it is either truncated or split.  Anything entirely inside the range
 265  * is deleted from the tree.
 266  *
 267  * inline_limit is used to tell this code which offsets in the file to keep
 268  * if they contain inline extents.
 269  */
 270 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 271                        struct btrfs_root *root, struct inode *inode,
 272                        u64 start, u64 end, u64 locked_end,
 273                        u64 inline_limit, u64 *hint_byte, int drop_cache)
 274 {
 275         u64 extent_end = 0;
 276         u64 search_start = start;
 277         u64 ram_bytes = 0;
 278         u64 disk_bytenr = 0;
 279         u64 orig_locked_end = locked_end;
 280         u8 compression;
 281         u8 encryption;
 282         u16 other_encoding = 0;
 283         struct extent_buffer *leaf;
 284         struct btrfs_file_extent_item *extent;
 285         struct btrfs_path *path;
 286         struct btrfs_key key;
 287         struct btrfs_file_extent_item old;
 288         int keep;
 289         int slot;
 290         int bookend;
 291         int found_type = 0;
 292         int found_extent;
 293         int found_inline;
 294         int recow;
 295         int ret;
 296
 297         inline_limit = 0;
 298         if (drop_cache)
 299                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 300
 301         path = btrfs_alloc_path();
 302         if (!path)
 303                 return -ENOMEM;
 304         while (1) {
 305                 recow = 0;
 306                 btrfs_release_path(root, path);
 307                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 308                                                search_start, -1);
 309                 if (ret < 0)
 310                         goto out;
 311                 if (ret > 0) {
 312                         if (path->slots[0] == 0) {
 313                                 ret = 0;
 314                                 goto out;
 315                         }
 316                         path->slots[0]--;
 317                 }
 318 next_slot:
 319                 keep = 0;
 320                 bookend = 0;
 321                 found_extent = 0;
 322                 found_inline = 0;
 323                 compression = 0;
 324                 encryption = 0;
 325                 extent = NULL;
 326                 leaf = path->nodes[0];
 327                 slot = path->slots[0];
 328                 ret = 0;
 329                 btrfs_item_key_to_cpu(leaf, &key, slot);
 330                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 331                     key.offset >= end) {
 332                         goto out;
 333                 }
 334                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 335                     key.objectid != inode->i_ino) {
 336                         goto out;
 337                 }
 338                 if (recow) {
 339                         search_start = max(key.offset, start);
 340                         continue;
 341                 }
 342                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 343                         extent = btrfs_item_ptr(leaf, slot,
 344                                                 struct btrfs_file_extent_item);
 345                         found_type = btrfs_file_extent_type(leaf, extent);
 346                         compression = btrfs_file_extent_compression(leaf,
 347                                                                     extent);
 348                         encryption = btrfs_file_extent_encryption(leaf,
 349                                                                   extent);
 350                         other_encoding = btrfs_file_extent_other_encoding(leaf,
 351                                                                   extent);
 352                         if (found_type == BTRFS_FILE_EXTENT_REG ||
 353                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 354                                 extent_end =
 355                                      btrfs_file_extent_disk_bytenr(leaf,
 356                                                                    extent);
 357                                 if (extent_end)
 358                                         *hint_byte = extent_end;
 359
 360                                 extent_end = key.offset +
 361                                      btrfs_file_extent_num_bytes(leaf, extent);
 362                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
 363                                                                 extent);
 364                                 found_extent = 1;
 365                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 366                                 found_inline = 1;
 367                                 extent_end = key.offset +
 368                                      btrfs_file_extent_inline_len(leaf, extent);
 369                         }
 370                 } else {
 371                         extent_end = search_start;
 372                 }
 373
 374                 /* we found nothing we can drop */
 375                 if ((!found_extent && !found_inline) ||
 376                     search_start >= extent_end) {
 377                         int nextret;
 378                         u32 nritems;
 379                         nritems = btrfs_header_nritems(leaf);
 380                         if (slot >= nritems - 1) {
 381                                 nextret = btrfs_next_leaf(root, path);
 382                                 if (nextret)
 383                                         goto out;
 384                                 recow = 1;
 385                         } else {
 386                                 path->slots[0]++;
 387                         }
 388                         goto next_slot;
 389                 }
 390
 391                 if (end <= extent_end && start >= key.offset && found_inline)
 392                         *hint_byte = EXTENT_MAP_INLINE;
 393
 394                 if (found_extent) {
 395                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 396                                            sizeof(old));
 397                 }
 398
 399                 if (end < extent_end && end >= key.offset) {
 400                         bookend = 1;
 401                         if (found_inline && start <= key.offset)
 402                                 keep = 1;
 403                 }
 404
 405                 if (bookend && found_extent) {
 406                         if (locked_end < extent_end) {
 407                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 408                                                 locked_end, extent_end - 1,
 409                                                 GFP_NOFS);
 410                                 if (!ret) {
 411                                         btrfs_release_path(root, path);
 412                                         lock_extent(&BTRFS_I(inode)->io_tree,
 413                                                 locked_end, extent_end - 1,
 414                                                 GFP_NOFS);
 415                                         locked_end = extent_end;
 416                                         continue;
 417                                 }
 418                                 locked_end = extent_end;
 419                         }
 420                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 421                         if (disk_bytenr != 0) {
 422                                 ret = btrfs_inc_extent_ref(trans, root,
 423                                            disk_bytenr,
 424                                            le64_to_cpu(old.disk_num_bytes), 0,
 425                                            root->root_key.objectid,
 426                                            key.objectid, key.offset -
 427                                            le64_to_cpu(old.offset));
 428                                 BUG_ON(ret);
 429                         }
 430                 }
 431
 432                 if (found_inline) {
 433                         u64 mask = root->sectorsize - 1;
 434                         search_start = (extent_end + mask) & ~mask;
 435                 } else
 436                         search_start = extent_end;
 437
 438                 /* truncate existing extent */
 439                 if (start > key.offset) {
 440                         u64 new_num;
 441                         u64 old_num;
 442                         keep = 1;
 443                         WARN_ON(start & (root->sectorsize - 1));
 444                         if (found_extent) {
 445                                 new_num = start - key.offset;
 446                                 old_num = btrfs_file_extent_num_bytes(leaf,
 447                                                                       extent);
 448                                 *hint_byte =
 449                                         btrfs_file_extent_disk_bytenr(leaf,
 450                                                                       extent);
 451                                 if (btrfs_file_extent_disk_bytenr(leaf,
 452                                                                   extent)) {
 453                                         inode_sub_bytes(inode, old_num -
 454                                                         new_num);
 455                                 }
 456                                 btrfs_set_file_extent_num_bytes(leaf,
 457                                                         extent, new_num);
 458                                 btrfs_mark_buffer_dirty(leaf);
 459                         } else if (key.offset < inline_limit &&
 460                                    (end > extent_end) &&
 461                                    (inline_limit < extent_end)) {
 462                                 u32 new_size;
 463                                 new_size = btrfs_file_extent_calc_inline_size(
 464                                                    inline_limit - key.offset);
 465                                 inode_sub_bytes(inode, extent_end -
 466                                                 inline_limit);
 467                                 btrfs_set_file_extent_ram_bytes(leaf, extent,
 468                                                         new_size);
 469                                 if (!compression && !encryption) {
 470                                         btrfs_truncate_item(trans, root, path,
 471                                                             new_size, 1);
 472                                 }
 473                         }
 474                 }
 475                 /* delete the entire extent */
 476                 if (!keep) {
 477                         if (found_inline)
 478                                 inode_sub_bytes(inode, extent_end -
 479                                                 key.offset);
 480                         ret = btrfs_del_item(trans, root, path);
 481                         /* TODO update progress marker and return */
 482                         BUG_ON(ret);
 483                         extent = NULL;
 484                         btrfs_release_path(root, path);
 485                         /* the extent will be freed later */
 486                 }
 487                 if (bookend && found_inline && start <= key.offset) {
 488                         u32 new_size;
 489                         new_size = btrfs_file_extent_calc_inline_size(
 490                                                    extent_end - end);
 491                         inode_sub_bytes(inode, end - key.offset);
 492                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 493                                                         new_size);
 494                         if (!compression && !encryption)
 495                                 ret = btrfs_truncate_item(trans, root, path,
 496                                                           new_size, 0);
 497                         BUG_ON(ret);
 498                 }
 499                 /* create bookend, splitting the extent in two */
 500                 if (bookend && found_extent) {
 501                         struct btrfs_key ins;
 502                         ins.objectid = inode->i_ino;
 503                         ins.offset = end;
 504                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 505
 506                         btrfs_release_path(root, path);
 507                         path->leave_spinning = 1;
 508                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 509                                                       sizeof(*extent));
 510                         BUG_ON(ret);
 511
 512                         leaf = path->nodes[0];
 513                         extent = btrfs_item_ptr(leaf, path->slots[0],
 514                                                 struct btrfs_file_extent_item);
 515                         write_extent_buffer(leaf, &old,
 516                                             (unsigned long)extent, sizeof(old));
 517
 518                         btrfs_set_file_extent_compression(leaf, extent,
 519                                                           compression);
 520                         btrfs_set_file_extent_encryption(leaf, extent,
 521                                                          encryption);
 522                         btrfs_set_file_extent_other_encoding(leaf, extent,
 523                                                              other_encoding);
 524                         btrfs_set_file_extent_offset(leaf, extent,
 525                                     le64_to_cpu(old.offset) + end - key.offset);
 526                         WARN_ON(le64_to_cpu(old.num_bytes) <
 527                                 (extent_end - end));
 528                         btrfs_set_file_extent_num_bytes(leaf, extent,
 529                                                         extent_end - end);
 530
 531                         /*
 532                          * set the ram bytes to the size of the full extent
 533                          * before splitting.  This is a worst case flag,
 534                          * but its the best we can do because we don't know
 535                          * how splitting affects compression
 536                          */
 537                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 538                                                         ram_bytes);
 539                         btrfs_set_file_extent_type(leaf, extent, found_type);
 540
 541                         btrfs_unlock_up_safe(path, 1);
 542                         btrfs_mark_buffer_dirty(path->nodes[0]);
 543                         btrfs_set_lock_blocking(path->nodes[0]);
 544
 545                         path->leave_spinning = 0;
 546                         btrfs_release_path(root, path);
 547                         if (disk_bytenr != 0)
 548                                 inode_add_bytes(inode, extent_end - end);
 549                 }
 550
 551                 if (found_extent && !keep) {
 552                         u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
 553
 554                         if (old_disk_bytenr != 0) {
 555                                 inode_sub_bytes(inode,
 556                                                 le64_to_cpu(old.num_bytes));
 557                                 ret = btrfs_free_extent(trans, root,
 558                                                 old_disk_bytenr,
 559                                                 le64_to_cpu(old.disk_num_bytes),
 560                                                 0, root->root_key.objectid,
 561                                                 key.objectid, key.offset -
 562                                                 le64_to_cpu(old.offset));
 563                                 BUG_ON(ret);
 564                                 *hint_byte = old_disk_bytenr;
 565                         }
 566                 }
 567
 568                 if (search_start >= end) {
 569                         ret = 0;
 570                         goto out;
 571                 }
 572         }
 573 out:
 574         btrfs_free_path(path);
 575         if (locked_end > orig_locked_end) {
 576                 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
 577                               locked_end - 1, GFP_NOFS);
 578         }
 579         return ret;
 580 }
 581
 582 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 583                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
 584 {
 585         struct btrfs_file_extent_item *fi;
 586         struct btrfs_key key;
 587         u64 extent_end;
 588
 589         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 590                 return 0;
 591
 592         btrfs_item_key_to_cpu(leaf, &key, slot);
 593         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 594                 return 0;
 595
 596         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 597         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 598             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 599             btrfs_file_extent_compression(leaf, fi) ||
 600             btrfs_file_extent_encryption(leaf, fi) ||
 601             btrfs_file_extent_other_encoding(leaf, fi))
 602                 return 0;
 603
 604         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 605         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 606                 return 0;
 607
 608         *start = key.offset;
 609         *end = extent_end;
 610         return 1;
 611 }
 612
 613 /*
 614  * Mark extent in the range start - end as written.
 615  *
 616  * This changes extent type from 'pre-allocated' to 'regular'. If only
 617  * part of extent is marked as written, the extent will be split into
 618  * two or three.
 619  */
 620 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 621                               struct btrfs_root *root,
 622                               struct inode *inode, u64 start, u64 end)
 623 {
 624         struct extent_buffer *leaf;
 625         struct btrfs_path *path;
 626         struct btrfs_file_extent_item *fi;
 627         struct btrfs_key key;
 628         u64 bytenr;
 629         u64 num_bytes;
 630         u64 extent_end;
 631         u64 orig_offset;
 632         u64 other_start;
 633         u64 other_end;
 634         u64 split = start;
 635         u64 locked_end = end;
 636         int extent_type;
 637         int split_end = 1;
 638         int ret;
 639
 640         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 641
 642         path = btrfs_alloc_path();
 643         BUG_ON(!path);
 644 again:
 645         key.objectid = inode->i_ino;
 646         key.type = BTRFS_EXTENT_DATA_KEY;
 647         if (split == start)
 648                 key.offset = split;
 649         else
 650                 key.offset = split - 1;
 651
 652         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 653         if (ret > 0 && path->slots[0] > 0)
 654                 path->slots[0]--;
 655
 656         leaf = path->nodes[0];
 657         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 658         BUG_ON(key.objectid != inode->i_ino ||
 659                key.type != BTRFS_EXTENT_DATA_KEY);
 660         fi = btrfs_item_ptr(leaf, path->slots[0],
 661                             struct btrfs_file_extent_item);
 662         extent_type = btrfs_file_extent_type(leaf, fi);
 663         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
 664         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 665         BUG_ON(key.offset > start || extent_end < end);
 666
 667         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 668         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 669         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 670
 671         if (key.offset == start)
 672                 split = end;
 673
 674         if (key.offset == start && extent_end == end) {
 675                 int del_nr = 0;
 676                 int del_slot = 0;
 677                 other_start = end;
 678                 other_end = 0;
 679                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 680                                      bytenr, &other_start, &other_end)) {
 681                         extent_end = other_end;
 682                         del_slot = path->slots[0] + 1;
 683                         del_nr++;
 684                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 685                                                 0, root->root_key.objectid,
 686                                                 inode->i_ino, orig_offset);
 687                         BUG_ON(ret);
 688                 }
 689                 other_start = 0;
 690                 other_end = start;
 691                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
 692                                      bytenr, &other_start, &other_end)) {
 693                         key.offset = other_start;
 694                         del_slot = path->slots[0];
 695                         del_nr++;
 696                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 697                                                 0, root->root_key.objectid,
 698                                                 inode->i_ino, orig_offset);
 699                         BUG_ON(ret);
 700                 }
 701                 split_end = 0;
 702                 if (del_nr == 0) {
 703                         btrfs_set_file_extent_type(leaf, fi,
 704                                                    BTRFS_FILE_EXTENT_REG);
 705                         goto done;
 706                 }
 707
 708                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 709                                     struct btrfs_file_extent_item);
 710                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 711                 btrfs_set_file_extent_num_bytes(leaf, fi,
 712                                                 extent_end - key.offset);
 713                 btrfs_mark_buffer_dirty(leaf);
 714
 715                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 716                 BUG_ON(ret);
 717                 goto release;
 718         } else if (split == start) {
 719                 if (locked_end < extent_end) {
 720                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 721                                         locked_end, extent_end - 1, GFP_NOFS);
 722                         if (!ret) {
 723                                 btrfs_release_path(root, path);
 724                                 lock_extent(&BTRFS_I(inode)->io_tree,
 725                                         locked_end, extent_end - 1, GFP_NOFS);
 726                                 locked_end = extent_end;
 727                                 goto again;
 728                         }
 729                         locked_end = extent_end;
 730                 }
 731                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
 732         } else  {
 733                 BUG_ON(key.offset != start);
 734                 key.offset = split;
 735                 btrfs_set_file_extent_offset(leaf, fi, key.offset -
 736                                              orig_offset);
 737                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
 738                 btrfs_set_item_key_safe(trans, root, path, &key);
 739                 extent_end = split;
 740         }
 741
 742         if (extent_end == end) {
 743                 split_end = 0;
 744                 extent_type = BTRFS_FILE_EXTENT_REG;
 745         }
 746         if (extent_end == end && split == start) {
 747                 other_start = end;
 748                 other_end = 0;
 749                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 750                                      bytenr, &other_start, &other_end)) {
 751                         path->slots[0]++;
 752                         fi = btrfs_item_ptr(leaf, path->slots[0],
 753                                             struct btrfs_file_extent_item);
 754                         key.offset = split;
 755                         btrfs_set_item_key_safe(trans, root, path, &key);
 756                         btrfs_set_file_extent_offset(leaf, fi, key.offset -
 757                                                      orig_offset);
 758                         btrfs_set_file_extent_num_bytes(leaf, fi,
 759                                                         other_end - split);
 760                         goto done;
 761                 }
 762         }
 763         if (extent_end == end && split == end) {
 764                 other_start = 0;
 765                 other_end = start;
 766                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
 767                                      bytenr, &other_start, &other_end)) {
 768                         path->slots[0]--;
 769                         fi = btrfs_item_ptr(leaf, path->slots[0],
 770                                             struct btrfs_file_extent_item);
 771                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
 772                                                         other_start);
 773                         goto done;
 774                 }
 775         }
 776
 777         btrfs_mark_buffer_dirty(leaf);
 778
 779         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 780                                    root->root_key.objectid,
 781                                    inode->i_ino, orig_offset);
 782         BUG_ON(ret);
 783         btrfs_release_path(root, path);
 784
 785         key.offset = start;
 786         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
 787         BUG_ON(ret);
 788
 789         leaf = path->nodes[0];
 790         fi = btrfs_item_ptr(leaf, path->slots[0],
 791                             struct btrfs_file_extent_item);
 792         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 793         btrfs_set_file_extent_type(leaf, fi, extent_type);
 794         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
 795         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
 796         btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
 797         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
 798         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 799         btrfs_set_file_extent_compression(leaf, fi, 0);
 800         btrfs_set_file_extent_encryption(leaf, fi, 0);
 801         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 802 done:
 803         btrfs_mark_buffer_dirty(leaf);
 804
 805 release:
 806         btrfs_release_path(root, path);
 807         if (split_end && split == start) {
 808                 split = end;
 809                 goto again;
 810         }
 811         if (locked_end > end) {
 812                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
 813                               GFP_NOFS);
 814         }
 815         btrfs_free_path(path);
 816         return 0;
 817 }
 818
 819 /*
 820  * this gets pages into the page cache and locks them down, it also properly
 821  * waits for data=ordered extents to finish before allowing the pages to be
 822  * modified.
 823  */
 824 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 825                          struct page **pages, size_t num_pages,
 826                          loff_t pos, unsigned long first_index,
 827                          unsigned long last_index, size_t write_bytes)
 828 {
 829         int i;
 830         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 831         struct inode *inode = fdentry(file)->d_inode;
 832         int err = 0;
 833         u64 start_pos;
 834         u64 last_pos;
 835
 836         start_pos = pos & ~((u64)root->sectorsize - 1);
 837         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 838
 839         if (start_pos > inode->i_size) {
 840                 err = btrfs_cont_expand(inode, start_pos);
 841                 if (err)
 842                         return err;
 843         }
 844
 845         memset(pages, 0, num_pages * sizeof(struct page *));
 846 again:
 847         for (i = 0; i < num_pages; i++) {
 848                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 849                 if (!pages[i]) {
 850                         err = -ENOMEM;
 851                         BUG_ON(1);
 852                 }
 853                 wait_on_page_writeback(pages[i]);
 854         }
 855         if (start_pos < inode->i_size) {
 856                 struct btrfs_ordered_extent *ordered;
 857                 lock_extent(&BTRFS_I(inode)->io_tree,
 858                             start_pos, last_pos - 1, GFP_NOFS);
 859                 ordered = btrfs_lookup_first_ordered_extent(inode,
 860                                                             last_pos - 1);
 861                 if (ordered &&
 862                     ordered->file_offset + ordered->len > start_pos &&
 863                     ordered->file_offset < last_pos) {
 864                         btrfs_put_ordered_extent(ordered);
 865                         unlock_extent(&BTRFS_I(inode)->io_tree,
 866                                       start_pos, last_pos - 1, GFP_NOFS);
 867                         for (i = 0; i < num_pages; i++) {
 868                                 unlock_page(pages[i]);
 869                                 page_cache_release(pages[i]);
 870                         }
 871                         btrfs_wait_ordered_range(inode, start_pos,
 872                                                  last_pos - start_pos);
 873                         goto again;
 874                 }
 875                 if (ordered)
 876                         btrfs_put_ordered_extent(ordered);
 877
 878                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 879                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 880                                   GFP_NOFS);
 881                 unlock_extent(&BTRFS_I(inode)->io_tree,
 882                               start_pos, last_pos - 1, GFP_NOFS);
 883         }
 884         for (i = 0; i < num_pages; i++) {
 885                 clear_page_dirty_for_io(pages[i]);
 886                 set_page_extent_mapped(pages[i]);
 887                 WARN_ON(!PageLocked(pages[i]));
 888         }
 889         return 0;
 890 }
 891
 892 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 893                                 size_t count, loff_t *ppos)
 894 {
 895         loff_t pos;
 896         loff_t start_pos;
 897         ssize_t num_written = 0;
 898         ssize_t err = 0;
 899         int ret = 0;
 900         struct inode *inode = fdentry(file)->d_inode;
 901         struct btrfs_root *root = BTRFS_I(inode)->root;
 902         struct page **pages = NULL;
 903         int nrptrs;
 904         struct page *pinned[2];
 905         unsigned long first_index;
 906         unsigned long last_index;
 907         int will_write;
 908
 909         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 910                       (file->f_flags & O_DIRECT));
 911
 912         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 913                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 914         pinned[0] = NULL;
 915         pinned[1] = NULL;
 916
 917         pos = *ppos;
 918         start_pos = pos;
 919
 920         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 921         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 922         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 923         if (err)
 924                 goto out_nolock;
 925         if (count == 0)
 926                 goto out_nolock;
 927
 928         err = file_remove_suid(file);
 929         if (err)
 930                 goto out_nolock;
 931         file_update_time(file);
 932
 933         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 934
 935         mutex_lock(&inode->i_mutex);
 936         BTRFS_I(inode)->sequence++;
 937         first_index = pos >> PAGE_CACHE_SHIFT;
 938         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 939
 940         /*
 941          * there are lots of better ways to do this, but this code
 942          * makes sure the first and last page in the file range are
 943          * up to date and ready for cow
 944          */
 945         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 946                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 947                 if (!PageUptodate(pinned[0])) {
 948                         ret = btrfs_readpage(NULL, pinned[0]);
 949                         BUG_ON(ret);
 950                         wait_on_page_locked(pinned[0]);
 951                 } else {
 952                         unlock_page(pinned[0]);
 953                 }
 954         }
 955         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 956                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 957                 if (!PageUptodate(pinned[1])) {
 958                         ret = btrfs_readpage(NULL, pinned[1]);
 959                         BUG_ON(ret);
 960                         wait_on_page_locked(pinned[1]);
 961                 } else {
 962                         unlock_page(pinned[1]);
 963                 }
 964         }
 965
 966         while (count > 0) {
 967                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 968                 size_t write_bytes = min(count, nrptrs *
 969                                         (size_t)PAGE_CACHE_SIZE -
 970                                          offset);
 971                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 972                                         PAGE_CACHE_SHIFT;
 973
 974                 WARN_ON(num_pages > nrptrs);
 975                 memset(pages, 0, sizeof(struct page *) * nrptrs);
 976
 977                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
 978                 if (ret)
 979                         goto out;
 980
 981                 ret = prepare_pages(root, file, pages, num_pages,
 982                                     pos, first_index, last_index,
 983                                     write_bytes);
 984                 if (ret) {
 985                         btrfs_free_reserved_data_space(root, inode,
 986                                                        write_bytes);
 987                         goto out;
 988                 }
 989
 990                 ret = btrfs_copy_from_user(pos, num_pages,
 991                                            write_bytes, pages, buf);
 992                 if (ret) {
 993                         btrfs_free_reserved_data_space(root, inode,
 994                                                        write_bytes);
 995                         btrfs_drop_pages(pages, num_pages);
 996                         goto out;
 997                 }
 998
 999                 ret = dirty_and_release_pages(NULL, root, file, pages,
1000                                               num_pages, pos, write_bytes);
1001                 btrfs_drop_pages(pages, num_pages);
1002                 if (ret) {
1003                         btrfs_free_reserved_data_space(root, inode,
1004                                                        write_bytes);
1005                         goto out;
1006                 }
1007
1008                 if (will_write) {
1009                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1010                                                pos + write_bytes - 1,
1011                                                WB_SYNC_ALL);
1012                 } else {
1013                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1014                                                            num_pages);
1015                         if (num_pages <
1016                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1017                                 btrfs_btree_balance_dirty(root, 1);
1018                         btrfs_throttle(root);
1019                 }
1020
1021                 buf += write_bytes;
1022                 count -= write_bytes;
1023                 pos += write_bytes;
1024                 num_written += write_bytes;
1025
1026                 cond_resched();
1027         }
1028 out:
1029         mutex_unlock(&inode->i_mutex);
1030         if (ret)
1031                 err = ret;
1032
1033 out_nolock:
1034         kfree(pages);
1035         if (pinned[0])
1036                 page_cache_release(pinned[0]);
1037         if (pinned[1])
1038                 page_cache_release(pinned[1]);
1039         *ppos = pos;
1040
1041         /*
1042          * we want to make sure fsync finds this change
1043          * but we haven't joined a transaction running right now.
1044          *
1045          * Later on, someone is sure to update the inode and get the
1046          * real transid recorded.
1047          *
1048          * We set last_trans now to the fs_info generation + 1,
1049          * this will either be one more than the running transaction
1050          * or the generation used for the next transaction if there isn't
1051          * one running right now.
1052          */
1053         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1054
1055         if (num_written > 0 && will_write) {
1056                 struct btrfs_trans_handle *trans;
1057
1058                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1059                 if (err)
1060                         num_written = err;
1061
1062                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1063                         trans = btrfs_start_transaction(root, 1);
1064                         ret = btrfs_log_dentry_safe(trans, root,
1065                                                     file->f_dentry);
1066                         if (ret == 0) {
1067                                 ret = btrfs_sync_log(trans, root);
1068                                 if (ret == 0)
1069                                         btrfs_end_transaction(trans, root);
1070                                 else
1071                                         btrfs_commit_transaction(trans, root);
1072                         } else {
1073                                 btrfs_commit_transaction(trans, root);
1074                         }
1075                 }
1076                 if (file->f_flags & O_DIRECT) {
1077                         invalidate_mapping_pages(inode->i_mapping,
1078                               start_pos >> PAGE_CACHE_SHIFT,
1079                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1080                 }
1081         }
1082         current->backing_dev_info = NULL;
1083         return num_written ? num_written : err;
1084 }
1085
1086 int btrfs_release_file(struct inode *inode, struct file *filp)
1087 {
1088         /*
1089          * ordered_data_close is set by settattr when we are about to truncate
1090          * a file from a non-zero size to a zero size.  This tries to
1091          * flush down new bytes that may have been written if the
1092          * application were using truncate to replace a file in place.
1093          */
1094         if (BTRFS_I(inode)->ordered_data_close) {
1095                 BTRFS_I(inode)->ordered_data_close = 0;
1096                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1097                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1098                         filemap_flush(inode->i_mapping);
1099         }
1100         if (filp->private_data)
1101                 btrfs_ioctl_trans_end(filp);
1102         return 0;
1103 }
1104
1105 /*
1106  * fsync call for both files and directories.  This logs the inode into
1107  * the tree log instead of forcing full commits whenever possible.
1108  *
1109  * It needs to call filemap_fdatawait so that all ordered extent updates are
1110  * in the metadata btree are up to date for copying to the log.
1111  *
1112  * It drops the inode mutex before doing the tree log commit.  This is an
1113  * important optimization for directories because holding the mutex prevents
1114  * new operations on the dir while we write to disk.
1115  */
1116 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1117 {
1118         struct inode *inode = dentry->d_inode;
1119         struct btrfs_root *root = BTRFS_I(inode)->root;
1120         int ret = 0;
1121         struct btrfs_trans_handle *trans;
1122
1123         /*
1124          * check the transaction that last modified this inode
1125          * and see if its already been committed
1126          */
1127         if (!BTRFS_I(inode)->last_trans)
1128                 goto out;
1129
1130         mutex_lock(&root->fs_info->trans_mutex);
1131         if (BTRFS_I(inode)->last_trans <=
1132             root->fs_info->last_trans_committed) {
1133                 BTRFS_I(inode)->last_trans = 0;
1134                 mutex_unlock(&root->fs_info->trans_mutex);
1135                 goto out;
1136         }
1137         mutex_unlock(&root->fs_info->trans_mutex);
1138
1139         root->log_batch++;
1140         filemap_fdatawrite(inode->i_mapping);
1141         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1142         root->log_batch++;
1143
1144         if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1145                 goto out;
1146         /*
1147          * ok we haven't committed the transaction yet, lets do a commit
1148          */
1149         if (file && file->private_data)
1150                 btrfs_ioctl_trans_end(file);
1151
1152         trans = btrfs_start_transaction(root, 1);
1153         if (!trans) {
1154                 ret = -ENOMEM;
1155                 goto out;
1156         }
1157
1158         ret = btrfs_log_dentry_safe(trans, root, dentry);
1159         if (ret < 0)
1160                 goto out;
1161
1162         /* we've logged all the items and now have a consistent
1163          * version of the file in the log.  It is possible that
1164          * someone will come in and modify the file, but that's
1165          * fine because the log is consistent on disk, and we
1166          * have references to all of the file's extents
1167          *
1168          * It is possible that someone will come in and log the
1169          * file again, but that will end up using the synchronization
1170          * inside btrfs_sync_log to keep things safe.
1171          */
1172         mutex_unlock(&dentry->d_inode->i_mutex);
1173
1174         if (ret > 0) {
1175                 ret = btrfs_commit_transaction(trans, root);
1176         } else {
1177                 ret = btrfs_sync_log(trans, root);
1178                 if (ret == 0)
1179                         ret = btrfs_end_transaction(trans, root);
1180                 else
1181                         ret = btrfs_commit_transaction(trans, root);
1182         }
1183         mutex_lock(&dentry->d_inode->i_mutex);
1184 out:
1185         return ret > 0 ? EIO : ret;
1186 }
1187
1188 static struct vm_operations_struct btrfs_file_vm_ops = {
1189         .fault          = filemap_fault,
1190         .page_mkwrite   = btrfs_page_mkwrite,
1191 };
1192
1193 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1194 {
1195         vma->vm_ops = &btrfs_file_vm_ops;
1196         file_accessed(filp);
1197         return 0;
1198 }
1199
1200 struct file_operations btrfs_file_operations = {
1201         .llseek         = generic_file_llseek,
1202         .read           = do_sync_read,
1203         .aio_read       = generic_file_aio_read,
1204         .splice_read    = generic_file_splice_read,
1205         .write          = btrfs_file_write,
1206         .mmap           = btrfs_file_mmap,
1207         .open           = generic_file_open,
1208         .release        = btrfs_release_file,
1209         .fsync          = btrfs_sync_file,
1210         .unlocked_ioctl = btrfs_ioctl,
1211 #ifdef CONFIG_COMPAT
1212         .compat_ioctl   = btrfs_ioctl,
1213 #endif
1214 };