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/backing-dev.h>
  26 #include <linux/mpage.h>
  27 #include <linux/swap.h>
  28 #include <linux/writeback.h>
  29 #include <linux/statfs.h>
  30 #include <linux/compat.h>
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "ioctl.h"
  36 #include "print-tree.h"
  37 #include "tree-log.h"
  38 #include "locking.h"
  39 #include "compat.h"
  40
  41
  42 /* simple helper to fault in pages and copy.  This should go away
  43  * and be replaced with calls into generic code.
  44  */
  45 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  46                                          int write_bytes,
  47                                          struct page **prepared_pages,
  48                                          const char __user *buf)
  49 {
  50         long page_fault = 0;
  51         int i;
  52         int offset = pos & (PAGE_CACHE_SIZE - 1);
  53
  54         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  55                 size_t count = min_t(size_t,
  56                                      PAGE_CACHE_SIZE - offset, write_bytes);
  57                 struct page *page = prepared_pages[i];
  58                 fault_in_pages_readable(buf, count);
  59
  60                 /* Copy data from userspace to the current page */
  61                 kmap(page);
  62                 page_fault = __copy_from_user(page_address(page) + offset,
  63                                               buf, count);
  64                 /* Flush processor's dcache for this page */
  65                 flush_dcache_page(page);
  66                 kunmap(page);
  67                 buf += count;
  68                 write_bytes -= count;
  69
  70                 if (page_fault)
  71                         break;
  72         }
  73         return page_fault ? -EFAULT : 0;
  74 }
  75
  76 /*
  77  * unlocks pages after btrfs_file_write is done with them
  78  */
  79 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
  80 {
  81         size_t i;
  82         for (i = 0; i < num_pages; i++) {
  83                 if (!pages[i])
  84                         break;
  85                 /* page checked is some magic around finding pages that
  86                  * have been modified without going through btrfs_set_page_dirty
  87                  * clear it here
  88                  */
  89                 ClearPageChecked(pages[i]);
  90                 unlock_page(pages[i]);
  91                 mark_page_accessed(pages[i]);
  92                 page_cache_release(pages[i]);
  93         }
  94 }
  95
  96 /*
  97  * after copy_from_user, pages need to be dirtied and we need to make
  98  * sure holes are created between the current EOF and the start of
  99  * any next extents (if required).
 100  *
 101  * this also makes the decision about creating an inline extent vs
 102  * doing real data extents, marking pages dirty and delalloc as required.
 103  */
 104 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 105                                    struct btrfs_root *root,
 106                                    struct file *file,
 107                                    struct page **pages,
 108                                    size_t num_pages,
 109                                    loff_t pos,
 110                                    size_t write_bytes)
 111 {
 112         int err = 0;
 113         int i;
 114         struct inode *inode = fdentry(file)->d_inode;
 115         u64 num_bytes;
 116         u64 start_pos;
 117         u64 end_of_last_block;
 118         u64 end_pos = pos + write_bytes;
 119         loff_t isize = i_size_read(inode);
 120
 121         start_pos = pos & ~((u64)root->sectorsize - 1);
 122         num_bytes = (write_bytes + pos - start_pos +
 123                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 124
 125         end_of_last_block = start_pos + num_bytes - 1;
 126         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 127         for (i = 0; i < num_pages; i++) {
 128                 struct page *p = pages[i];
 129                 SetPageUptodate(p);
 130                 ClearPageChecked(p);
 131                 set_page_dirty(p);
 132         }
 133         if (end_pos > isize) {
 134                 i_size_write(inode, end_pos);
 135                 /* we've only changed i_size in ram, and we haven't updated
 136                  * the disk i_size.  There is no need to log the inode
 137                  * at this time.
 138                  */
 139         }
 140         return err;
 141 }
 142
 143 /*
 144  * this drops all the extents in the cache that intersect the range
 145  * [start, end].  Existing extents are split as required.
 146  */
 147 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 148                             int skip_pinned)
 149 {
 150         struct extent_map *em;
 151         struct extent_map *split = NULL;
 152         struct extent_map *split2 = NULL;
 153         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 154         u64 len = end - start + 1;
 155         int ret;
 156         int testend = 1;
 157         unsigned long flags;
 158         int compressed = 0;
 159
 160         WARN_ON(end < start);
 161         if (end == (u64)-1) {
 162                 len = (u64)-1;
 163                 testend = 0;
 164         }
 165         while (1) {
 166                 if (!split)
 167                         split = alloc_extent_map(GFP_NOFS);
 168                 if (!split2)
 169                         split2 = alloc_extent_map(GFP_NOFS);
 170
 171                 write_lock(&em_tree->lock);
 172                 em = lookup_extent_mapping(em_tree, start, len);
 173                 if (!em) {
 174                         write_unlock(&em_tree->lock);
 175                         break;
 176                 }
 177                 flags = em->flags;
 178                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 179                         if (em->start <= start &&
 180                             (!testend || em->start + em->len >= start + len)) {
 181                                 free_extent_map(em);
 182                                 write_unlock(&em_tree->lock);
 183                                 break;
 184                         }
 185                         if (start < em->start) {
 186                                 len = em->start - start;
 187                         } else {
 188                                 len = start + len - (em->start + em->len);
 189                                 start = em->start + em->len;
 190                         }
 191                         free_extent_map(em);
 192                         write_unlock(&em_tree->lock);
 193                         continue;
 194                 }
 195                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 196                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 197                 remove_extent_mapping(em_tree, em);
 198
 199                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 200                     em->start < start) {
 201                         split->start = em->start;
 202                         split->len = start - em->start;
 203                         split->orig_start = em->orig_start;
 204                         split->block_start = em->block_start;
 205
 206                         if (compressed)
 207                                 split->block_len = em->block_len;
 208                         else
 209                                 split->block_len = split->len;
 210
 211                         split->bdev = em->bdev;
 212                         split->flags = flags;
 213                         ret = add_extent_mapping(em_tree, split);
 214                         BUG_ON(ret);
 215                         free_extent_map(split);
 216                         split = split2;
 217                         split2 = NULL;
 218                 }
 219                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 220                     testend && em->start + em->len > start + len) {
 221                         u64 diff = start + len - em->start;
 222
 223                         split->start = start + len;
 224                         split->len = em->start + em->len - (start + len);
 225                         split->bdev = em->bdev;
 226                         split->flags = flags;
 227
 228                         if (compressed) {
 229                                 split->block_len = em->block_len;
 230                                 split->block_start = em->block_start;
 231                                 split->orig_start = em->orig_start;
 232                         } else {
 233                                 split->block_len = split->len;
 234                                 split->block_start = em->block_start + diff;
 235                                 split->orig_start = split->start;
 236                         }
 237
 238                         ret = add_extent_mapping(em_tree, split);
 239                         BUG_ON(ret);
 240                         free_extent_map(split);
 241                         split = NULL;
 242                 }
 243                 write_unlock(&em_tree->lock);
 244
 245                 /* once for us */
 246                 free_extent_map(em);
 247                 /* once for the tree*/
 248                 free_extent_map(em);
 249         }
 250         if (split)
 251                 free_extent_map(split);
 252         if (split2)
 253                 free_extent_map(split2);
 254         return 0;
 255 }
 256
 257 /*
 258  * this is very complex, but the basic idea is to drop all extents
 259  * in the range start - end.  hint_block is filled in with a block number
 260  * that would be a good hint to the block allocator for this file.
 261  *
 262  * If an extent intersects the range but is not entirely inside the range
 263  * it is either truncated or split.  Anything entirely inside the range
 264  * is deleted from the tree.
 265  *
 266  * inline_limit is used to tell this code which offsets in the file to keep
 267  * if they contain inline extents.
 268  */
 269 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 270                        struct btrfs_root *root, struct inode *inode,
 271                        u64 start, u64 end, u64 locked_end,
 272                        u64 inline_limit, u64 *hint_byte, int drop_cache)
 273 {
 274         u64 extent_end = 0;
 275         u64 search_start = start;
 276         u64 ram_bytes = 0;
 277         u64 disk_bytenr = 0;
 278         u64 orig_locked_end = locked_end;
 279         u8 compression;
 280         u8 encryption;
 281         u16 other_encoding = 0;
 282         struct extent_buffer *leaf;
 283         struct btrfs_file_extent_item *extent;
 284         struct btrfs_path *path;
 285         struct btrfs_key key;
 286         struct btrfs_file_extent_item old;
 287         int keep;
 288         int slot;
 289         int bookend;
 290         int found_type = 0;
 291         int found_extent;
 292         int found_inline;
 293         int recow;
 294         int ret;
 295
 296         inline_limit = 0;
 297         if (drop_cache)
 298                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 299
 300         path = btrfs_alloc_path();
 301         if (!path)
 302                 return -ENOMEM;
 303         while (1) {
 304                 recow = 0;
 305                 btrfs_release_path(root, path);
 306                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 307                                                search_start, -1);
 308                 if (ret < 0)
 309                         goto out;
 310                 if (ret > 0) {
 311                         if (path->slots[0] == 0) {
 312                                 ret = 0;
 313                                 goto out;
 314                         }
 315                         path->slots[0]--;
 316                 }
 317 next_slot:
 318                 keep = 0;
 319                 bookend = 0;
 320                 found_extent = 0;
 321                 found_inline = 0;
 322                 compression = 0;
 323                 encryption = 0;
 324                 extent = NULL;
 325                 leaf = path->nodes[0];
 326                 slot = path->slots[0];
 327                 ret = 0;
 328                 btrfs_item_key_to_cpu(leaf, &key, slot);
 329                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 330                     key.offset >= end) {
 331                         goto out;
 332                 }
 333                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 334                     key.objectid != inode->i_ino) {
 335                         goto out;
 336                 }
 337                 if (recow) {
 338                         search_start = max(key.offset, start);
 339                         continue;
 340                 }
 341                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 342                         extent = btrfs_item_ptr(leaf, slot,
 343                                                 struct btrfs_file_extent_item);
 344                         found_type = btrfs_file_extent_type(leaf, extent);
 345                         compression = btrfs_file_extent_compression(leaf,
 346                                                                     extent);
 347                         encryption = btrfs_file_extent_encryption(leaf,
 348                                                                   extent);
 349                         other_encoding = btrfs_file_extent_other_encoding(leaf,
 350                                                                   extent);
 351                         if (found_type == BTRFS_FILE_EXTENT_REG ||
 352                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 353                                 extent_end =
 354                                      btrfs_file_extent_disk_bytenr(leaf,
 355                                                                    extent);
 356                                 if (extent_end)
 357                                         *hint_byte = extent_end;
 358
 359                                 extent_end = key.offset +
 360                                      btrfs_file_extent_num_bytes(leaf, extent);
 361                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
 362                                                                 extent);
 363                                 found_extent = 1;
 364                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 365                                 found_inline = 1;
 366                                 extent_end = key.offset +
 367                                      btrfs_file_extent_inline_len(leaf, extent);
 368                         }
 369                 } else {
 370                         extent_end = search_start;
 371                 }
 372
 373                 /* we found nothing we can drop */
 374                 if ((!found_extent && !found_inline) ||
 375                     search_start >= extent_end) {
 376                         int nextret;
 377                         u32 nritems;
 378                         nritems = btrfs_header_nritems(leaf);
 379                         if (slot >= nritems - 1) {
 380                                 nextret = btrfs_next_leaf(root, path);
 381                                 if (nextret)
 382                                         goto out;
 383                                 recow = 1;
 384                         } else {
 385                                 path->slots[0]++;
 386                         }
 387                         goto next_slot;
 388                 }
 389
 390                 if (end <= extent_end && start >= key.offset && found_inline)
 391                         *hint_byte = EXTENT_MAP_INLINE;
 392
 393                 if (found_extent) {
 394                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 395                                            sizeof(old));
 396                 }
 397
 398                 if (end < extent_end && end >= key.offset) {
 399                         bookend = 1;
 400                         if (found_inline && start <= key.offset)
 401                                 keep = 1;
 402                 }
 403
 404                 if (bookend && found_extent) {
 405                         if (locked_end < extent_end) {
 406                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 407                                                 locked_end, extent_end - 1,
 408                                                 GFP_NOFS);
 409                                 if (!ret) {
 410                                         btrfs_release_path(root, path);
 411                                         lock_extent(&BTRFS_I(inode)->io_tree,
 412                                                 locked_end, extent_end - 1,
 413                                                 GFP_NOFS);
 414                                         locked_end = extent_end;
 415                                         continue;
 416                                 }
 417                                 locked_end = extent_end;
 418                         }
 419                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 420                         if (disk_bytenr != 0) {
 421                                 ret = btrfs_inc_extent_ref(trans, root,
 422                                            disk_bytenr,
 423                                            le64_to_cpu(old.disk_num_bytes), 0,
 424                                            root->root_key.objectid,
 425                                            key.objectid, key.offset -
 426                                            le64_to_cpu(old.offset));
 427                                 BUG_ON(ret);
 428                         }
 429                 }
 430
 431                 if (found_inline) {
 432                         u64 mask = root->sectorsize - 1;
 433                         search_start = (extent_end + mask) & ~mask;
 434                 } else
 435                         search_start = extent_end;
 436
 437                 /* truncate existing extent */
 438                 if (start > key.offset) {
 439                         u64 new_num;
 440                         u64 old_num;
 441                         keep = 1;
 442                         WARN_ON(start & (root->sectorsize - 1));
 443                         if (found_extent) {
 444                                 new_num = start - key.offset;
 445                                 old_num = btrfs_file_extent_num_bytes(leaf,
 446                                                                       extent);
 447                                 *hint_byte =
 448                                         btrfs_file_extent_disk_bytenr(leaf,
 449                                                                       extent);
 450                                 if (btrfs_file_extent_disk_bytenr(leaf,
 451                                                                   extent)) {
 452                                         inode_sub_bytes(inode, old_num -
 453                                                         new_num);
 454                                 }
 455                                 btrfs_set_file_extent_num_bytes(leaf,
 456                                                         extent, new_num);
 457                                 btrfs_mark_buffer_dirty(leaf);
 458                         } else if (key.offset < inline_limit &&
 459                                    (end > extent_end) &&
 460                                    (inline_limit < extent_end)) {
 461                                 u32 new_size;
 462                                 new_size = btrfs_file_extent_calc_inline_size(
 463                                                    inline_limit - key.offset);
 464                                 inode_sub_bytes(inode, extent_end -
 465                                                 inline_limit);
 466                                 btrfs_set_file_extent_ram_bytes(leaf, extent,
 467                                                         new_size);
 468                                 if (!compression && !encryption) {
 469                                         btrfs_truncate_item(trans, root, path,
 470                                                             new_size, 1);
 471                                 }
 472                         }
 473                 }
 474                 /* delete the entire extent */
 475                 if (!keep) {
 476                         if (found_inline)
 477                                 inode_sub_bytes(inode, extent_end -
 478                                                 key.offset);
 479                         ret = btrfs_del_item(trans, root, path);
 480                         /* TODO update progress marker and return */
 481                         BUG_ON(ret);
 482                         extent = NULL;
 483                         btrfs_release_path(root, path);
 484                         /* the extent will be freed later */
 485                 }
 486                 if (bookend && found_inline && start <= key.offset) {
 487                         u32 new_size;
 488                         new_size = btrfs_file_extent_calc_inline_size(
 489                                                    extent_end - end);
 490                         inode_sub_bytes(inode, end - key.offset);
 491                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 492                                                         new_size);
 493                         if (!compression && !encryption)
 494                                 ret = btrfs_truncate_item(trans, root, path,
 495                                                           new_size, 0);
 496                         BUG_ON(ret);
 497                 }
 498                 /* create bookend, splitting the extent in two */
 499                 if (bookend && found_extent) {
 500                         struct btrfs_key ins;
 501                         ins.objectid = inode->i_ino;
 502                         ins.offset = end;
 503                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 504
 505                         btrfs_release_path(root, path);
 506                         path->leave_spinning = 1;
 507                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 508                                                       sizeof(*extent));
 509                         BUG_ON(ret);
 510
 511                         leaf = path->nodes[0];
 512                         extent = btrfs_item_ptr(leaf, path->slots[0],
 513                                                 struct btrfs_file_extent_item);
 514                         write_extent_buffer(leaf, &old,
 515                                             (unsigned long)extent, sizeof(old));
 516
 517                         btrfs_set_file_extent_compression(leaf, extent,
 518                                                           compression);
 519                         btrfs_set_file_extent_encryption(leaf, extent,
 520                                                          encryption);
 521                         btrfs_set_file_extent_other_encoding(leaf, extent,
 522                                                              other_encoding);
 523                         btrfs_set_file_extent_offset(leaf, extent,
 524                                     le64_to_cpu(old.offset) + end - key.offset);
 525                         WARN_ON(le64_to_cpu(old.num_bytes) <
 526                                 (extent_end - end));
 527                         btrfs_set_file_extent_num_bytes(leaf, extent,
 528                                                         extent_end - end);
 529
 530                         /*
 531                          * set the ram bytes to the size of the full extent
 532                          * before splitting.  This is a worst case flag,
 533                          * but its the best we can do because we don't know
 534                          * how splitting affects compression
 535                          */
 536                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 537                                                         ram_bytes);
 538                         btrfs_set_file_extent_type(leaf, extent, found_type);
 539
 540                         btrfs_unlock_up_safe(path, 1);
 541                         btrfs_mark_buffer_dirty(path->nodes[0]);
 542                         btrfs_set_lock_blocking(path->nodes[0]);
 543
 544                         path->leave_spinning = 0;
 545                         btrfs_release_path(root, path);
 546                         if (disk_bytenr != 0)
 547                                 inode_add_bytes(inode, extent_end - end);
 548                 }
 549
 550                 if (found_extent && !keep) {
 551                         u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
 552
 553                         if (old_disk_bytenr != 0) {
 554                                 inode_sub_bytes(inode,
 555                                                 le64_to_cpu(old.num_bytes));
 556                                 ret = btrfs_free_extent(trans, root,
 557                                                 old_disk_bytenr,
 558                                                 le64_to_cpu(old.disk_num_bytes),
 559                                                 0, root->root_key.objectid,
 560                                                 key.objectid, key.offset -
 561                                                 le64_to_cpu(old.offset));
 562                                 BUG_ON(ret);
 563                                 *hint_byte = old_disk_bytenr;
 564                         }
 565                 }
 566
 567                 if (search_start >= end) {
 568                         ret = 0;
 569                         goto out;
 570                 }
 571         }
 572 out:
 573         btrfs_free_path(path);
 574         if (locked_end > orig_locked_end) {
 575                 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
 576                               locked_end - 1, GFP_NOFS);
 577         }
 578         return ret;
 579 }
 580
 581 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 582                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
 583 {
 584         struct btrfs_file_extent_item *fi;
 585         struct btrfs_key key;
 586         u64 extent_end;
 587
 588         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 589                 return 0;
 590
 591         btrfs_item_key_to_cpu(leaf, &key, slot);
 592         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 593                 return 0;
 594
 595         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 596         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 597             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 598             btrfs_file_extent_compression(leaf, fi) ||
 599             btrfs_file_extent_encryption(leaf, fi) ||
 600             btrfs_file_extent_other_encoding(leaf, fi))
 601                 return 0;
 602
 603         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 604         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 605                 return 0;
 606
 607         *start = key.offset;
 608         *end = extent_end;
 609         return 1;
 610 }
 611
 612 /*
 613  * Mark extent in the range start - end as written.
 614  *
 615  * This changes extent type from 'pre-allocated' to 'regular'. If only
 616  * part of extent is marked as written, the extent will be split into
 617  * two or three.
 618  */
 619 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 620                               struct btrfs_root *root,
 621                               struct inode *inode, u64 start, u64 end)
 622 {
 623         struct extent_buffer *leaf;
 624         struct btrfs_path *path;
 625         struct btrfs_file_extent_item *fi;
 626         struct btrfs_key key;
 627         u64 bytenr;
 628         u64 num_bytes;
 629         u64 extent_end;
 630         u64 orig_offset;
 631         u64 other_start;
 632         u64 other_end;
 633         u64 split = start;
 634         u64 locked_end = end;
 635         int extent_type;
 636         int split_end = 1;
 637         int ret;
 638
 639         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 640
 641         path = btrfs_alloc_path();
 642         BUG_ON(!path);
 643 again:
 644         key.objectid = inode->i_ino;
 645         key.type = BTRFS_EXTENT_DATA_KEY;
 646         if (split == start)
 647                 key.offset = split;
 648         else
 649                 key.offset = split - 1;
 650
 651         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 652         if (ret > 0 && path->slots[0] > 0)
 653                 path->slots[0]--;
 654
 655         leaf = path->nodes[0];
 656         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 657         BUG_ON(key.objectid != inode->i_ino ||
 658                key.type != BTRFS_EXTENT_DATA_KEY);
 659         fi = btrfs_item_ptr(leaf, path->slots[0],
 660                             struct btrfs_file_extent_item);
 661         extent_type = btrfs_file_extent_type(leaf, fi);
 662         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
 663         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 664         BUG_ON(key.offset > start || extent_end < end);
 665
 666         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 667         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 668         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 669
 670         if (key.offset == start)
 671                 split = end;
 672
 673         if (key.offset == start && extent_end == end) {
 674                 int del_nr = 0;
 675                 int del_slot = 0;
 676                 other_start = end;
 677                 other_end = 0;
 678                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 679                                      bytenr, &other_start, &other_end)) {
 680                         extent_end = other_end;
 681                         del_slot = path->slots[0] + 1;
 682                         del_nr++;
 683                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 684                                                 0, root->root_key.objectid,
 685                                                 inode->i_ino, orig_offset);
 686                         BUG_ON(ret);
 687                 }
 688                 other_start = 0;
 689                 other_end = start;
 690                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
 691                                      bytenr, &other_start, &other_end)) {
 692                         key.offset = other_start;
 693                         del_slot = path->slots[0];
 694                         del_nr++;
 695                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 696                                                 0, root->root_key.objectid,
 697                                                 inode->i_ino, orig_offset);
 698                         BUG_ON(ret);
 699                 }
 700                 split_end = 0;
 701                 if (del_nr == 0) {
 702                         btrfs_set_file_extent_type(leaf, fi,
 703                                                    BTRFS_FILE_EXTENT_REG);
 704                         goto done;
 705                 }
 706
 707                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 708                                     struct btrfs_file_extent_item);
 709                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 710                 btrfs_set_file_extent_num_bytes(leaf, fi,
 711                                                 extent_end - key.offset);
 712                 btrfs_mark_buffer_dirty(leaf);
 713
 714                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 715                 BUG_ON(ret);
 716                 goto release;
 717         } else if (split == start) {
 718                 if (locked_end < extent_end) {
 719                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 720                                         locked_end, extent_end - 1, GFP_NOFS);
 721                         if (!ret) {
 722                                 btrfs_release_path(root, path);
 723                                 lock_extent(&BTRFS_I(inode)->io_tree,
 724                                         locked_end, extent_end - 1, GFP_NOFS);
 725                                 locked_end = extent_end;
 726                                 goto again;
 727                         }
 728                         locked_end = extent_end;
 729                 }
 730                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
 731         } else  {
 732                 BUG_ON(key.offset != start);
 733                 key.offset = split;
 734                 btrfs_set_file_extent_offset(leaf, fi, key.offset -
 735                                              orig_offset);
 736                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
 737                 btrfs_set_item_key_safe(trans, root, path, &key);
 738                 extent_end = split;
 739         }
 740
 741         if (extent_end == end) {
 742                 split_end = 0;
 743                 extent_type = BTRFS_FILE_EXTENT_REG;
 744         }
 745         if (extent_end == end && split == start) {
 746                 other_start = end;
 747                 other_end = 0;
 748                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 749                                      bytenr, &other_start, &other_end)) {
 750                         path->slots[0]++;
 751                         fi = btrfs_item_ptr(leaf, path->slots[0],
 752                                             struct btrfs_file_extent_item);
 753                         key.offset = split;
 754                         btrfs_set_item_key_safe(trans, root, path, &key);
 755                         btrfs_set_file_extent_offset(leaf, fi, key.offset -
 756                                                      orig_offset);
 757                         btrfs_set_file_extent_num_bytes(leaf, fi,
 758                                                         other_end - split);
 759                         goto done;
 760                 }
 761         }
 762         if (extent_end == end && split == end) {
 763                 other_start = 0;
 764                 other_end = start;
 765                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
 766                                      bytenr, &other_start, &other_end)) {
 767                         path->slots[0]--;
 768                         fi = btrfs_item_ptr(leaf, path->slots[0],
 769                                             struct btrfs_file_extent_item);
 770                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
 771                                                         other_start);
 772                         goto done;
 773                 }
 774         }
 775
 776         btrfs_mark_buffer_dirty(leaf);
 777
 778         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 779                                    root->root_key.objectid,
 780                                    inode->i_ino, orig_offset);
 781         BUG_ON(ret);
 782         btrfs_release_path(root, path);
 783
 784         key.offset = start;
 785         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
 786         BUG_ON(ret);
 787
 788         leaf = path->nodes[0];
 789         fi = btrfs_item_ptr(leaf, path->slots[0],
 790                             struct btrfs_file_extent_item);
 791         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 792         btrfs_set_file_extent_type(leaf, fi, extent_type);
 793         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
 794         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
 795         btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
 796         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
 797         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 798         btrfs_set_file_extent_compression(leaf, fi, 0);
 799         btrfs_set_file_extent_encryption(leaf, fi, 0);
 800         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 801 done:
 802         btrfs_mark_buffer_dirty(leaf);
 803
 804 release:
 805         btrfs_release_path(root, path);
 806         if (split_end && split == start) {
 807                 split = end;
 808                 goto again;
 809         }
 810         if (locked_end > end) {
 811                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
 812                               GFP_NOFS);
 813         }
 814         btrfs_free_path(path);
 815         return 0;
 816 }
 817
 818 /*
 819  * this gets pages into the page cache and locks them down, it also properly
 820  * waits for data=ordered extents to finish before allowing the pages to be
 821  * modified.
 822  */
 823 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 824                          struct page **pages, size_t num_pages,
 825                          loff_t pos, unsigned long first_index,
 826                          unsigned long last_index, size_t write_bytes)
 827 {
 828         int i;
 829         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 830         struct inode *inode = fdentry(file)->d_inode;
 831         int err = 0;
 832         u64 start_pos;
 833         u64 last_pos;
 834
 835         start_pos = pos & ~((u64)root->sectorsize - 1);
 836         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 837
 838         if (start_pos > inode->i_size) {
 839                 err = btrfs_cont_expand(inode, start_pos);
 840                 if (err)
 841                         return err;
 842         }
 843
 844         memset(pages, 0, num_pages * sizeof(struct page *));
 845 again:
 846         for (i = 0; i < num_pages; i++) {
 847                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 848                 if (!pages[i]) {
 849                         err = -ENOMEM;
 850                         BUG_ON(1);
 851                 }
 852                 wait_on_page_writeback(pages[i]);
 853         }
 854         if (start_pos < inode->i_size) {
 855                 struct btrfs_ordered_extent *ordered;
 856                 lock_extent(&BTRFS_I(inode)->io_tree,
 857                             start_pos, last_pos - 1, GFP_NOFS);
 858                 ordered = btrfs_lookup_first_ordered_extent(inode,
 859                                                             last_pos - 1);
 860                 if (ordered &&
 861                     ordered->file_offset + ordered->len > start_pos &&
 862                     ordered->file_offset < last_pos) {
 863                         btrfs_put_ordered_extent(ordered);
 864                         unlock_extent(&BTRFS_I(inode)->io_tree,
 865                                       start_pos, last_pos - 1, GFP_NOFS);
 866                         for (i = 0; i < num_pages; i++) {
 867                                 unlock_page(pages[i]);
 868                                 page_cache_release(pages[i]);
 869                         }
 870                         btrfs_wait_ordered_range(inode, start_pos,
 871                                                  last_pos - start_pos);
 872                         goto again;
 873                 }
 874                 if (ordered)
 875                         btrfs_put_ordered_extent(ordered);
 876
 877                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 878                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 879                                   GFP_NOFS);
 880                 unlock_extent(&BTRFS_I(inode)->io_tree,
 881                               start_pos, last_pos - 1, GFP_NOFS);
 882         }
 883         for (i = 0; i < num_pages; i++) {
 884                 clear_page_dirty_for_io(pages[i]);
 885                 set_page_extent_mapped(pages[i]);
 886                 WARN_ON(!PageLocked(pages[i]));
 887         }
 888         return 0;
 889 }
 890
 891 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 892                                 size_t count, loff_t *ppos)
 893 {
 894         loff_t pos;
 895         loff_t start_pos;
 896         ssize_t num_written = 0;
 897         ssize_t err = 0;
 898         int ret = 0;
 899         struct inode *inode = fdentry(file)->d_inode;
 900         struct btrfs_root *root = BTRFS_I(inode)->root;
 901         struct page **pages = NULL;
 902         int nrptrs;
 903         struct page *pinned[2];
 904         unsigned long first_index;
 905         unsigned long last_index;
 906         int will_write;
 907
 908         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 909                       (file->f_flags & O_DIRECT));
 910
 911         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 912                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 913         pinned[0] = NULL;
 914         pinned[1] = NULL;
 915
 916         pos = *ppos;
 917         start_pos = pos;
 918
 919         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 920         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 921         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 922         if (err)
 923                 goto out_nolock;
 924         if (count == 0)
 925                 goto out_nolock;
 926
 927         err = file_remove_suid(file);
 928         if (err)
 929                 goto out_nolock;
 930         file_update_time(file);
 931
 932         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 933
 934         mutex_lock(&inode->i_mutex);
 935         BTRFS_I(inode)->sequence++;
 936         first_index = pos >> PAGE_CACHE_SHIFT;
 937         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 938
 939         /*
 940          * there are lots of better ways to do this, but this code
 941          * makes sure the first and last page in the file range are
 942          * up to date and ready for cow
 943          */
 944         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 945                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 946                 if (!PageUptodate(pinned[0])) {
 947                         ret = btrfs_readpage(NULL, pinned[0]);
 948                         BUG_ON(ret);
 949                         wait_on_page_locked(pinned[0]);
 950                 } else {
 951                         unlock_page(pinned[0]);
 952                 }
 953         }
 954         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 955                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 956                 if (!PageUptodate(pinned[1])) {
 957                         ret = btrfs_readpage(NULL, pinned[1]);
 958                         BUG_ON(ret);
 959                         wait_on_page_locked(pinned[1]);
 960                 } else {
 961                         unlock_page(pinned[1]);
 962                 }
 963         }
 964
 965         while (count > 0) {
 966                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 967                 size_t write_bytes = min(count, nrptrs *
 968                                         (size_t)PAGE_CACHE_SIZE -
 969                                          offset);
 970                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 971                                         PAGE_CACHE_SHIFT;
 972
 973                 WARN_ON(num_pages > nrptrs);
 974                 memset(pages, 0, sizeof(struct page *) * nrptrs);
 975
 976                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
 977                 if (ret)
 978                         goto out;
 979
 980                 ret = prepare_pages(root, file, pages, num_pages,
 981                                     pos, first_index, last_index,
 982                                     write_bytes);
 983                 if (ret) {
 984                         btrfs_free_reserved_data_space(root, inode,
 985                                                        write_bytes);
 986                         goto out;
 987                 }
 988
 989                 ret = btrfs_copy_from_user(pos, num_pages,
 990                                            write_bytes, pages, buf);
 991                 if (ret) {
 992                         btrfs_free_reserved_data_space(root, inode,
 993                                                        write_bytes);
 994                         btrfs_drop_pages(pages, num_pages);
 995                         goto out;
 996                 }
 997
 998                 ret = dirty_and_release_pages(NULL, root, file, pages,
 999                                               num_pages, pos, write_bytes);
1000                 btrfs_drop_pages(pages, num_pages);
1001                 if (ret) {
1002                         btrfs_free_reserved_data_space(root, inode,
1003                                                        write_bytes);
1004                         goto out;
1005                 }
1006
1007                 if (will_write) {
1008                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1009                                                pos + write_bytes - 1,
1010                                                WB_SYNC_ALL);
1011                 } else {
1012                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1013                                                            num_pages);
1014                         if (num_pages <
1015                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1016                                 btrfs_btree_balance_dirty(root, 1);
1017                         btrfs_throttle(root);
1018                 }
1019
1020                 buf += write_bytes;
1021                 count -= write_bytes;
1022                 pos += write_bytes;
1023                 num_written += write_bytes;
1024
1025                 cond_resched();
1026         }
1027 out:
1028         mutex_unlock(&inode->i_mutex);
1029         if (ret)
1030                 err = ret;
1031
1032 out_nolock:
1033         kfree(pages);
1034         if (pinned[0])
1035                 page_cache_release(pinned[0]);
1036         if (pinned[1])
1037                 page_cache_release(pinned[1]);
1038         *ppos = pos;
1039
1040         /*
1041          * we want to make sure fsync finds this change
1042          * but we haven't joined a transaction running right now.
1043          *
1044          * Later on, someone is sure to update the inode and get the
1045          * real transid recorded.
1046          *
1047          * We set last_trans now to the fs_info generation + 1,
1048          * this will either be one more than the running transaction
1049          * or the generation used for the next transaction if there isn't
1050          * one running right now.
1051          */
1052         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1053
1054         if (num_written > 0 && will_write) {
1055                 struct btrfs_trans_handle *trans;
1056
1057                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1058                 if (err)
1059                         num_written = err;
1060
1061                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1062                         trans = btrfs_start_transaction(root, 1);
1063                         ret = btrfs_log_dentry_safe(trans, root,
1064                                                     file->f_dentry);
1065                         if (ret == 0) {
1066                                 ret = btrfs_sync_log(trans, root);
1067                                 if (ret == 0)
1068                                         btrfs_end_transaction(trans, root);
1069                                 else
1070                                         btrfs_commit_transaction(trans, root);
1071                         } else {
1072                                 btrfs_commit_transaction(trans, root);
1073                         }
1074                 }
1075                 if (file->f_flags & O_DIRECT) {
1076                         invalidate_mapping_pages(inode->i_mapping,
1077                               start_pos >> PAGE_CACHE_SHIFT,
1078                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1079                 }
1080         }
1081         current->backing_dev_info = NULL;
1082         return num_written ? num_written : err;
1083 }
1084
1085 int btrfs_release_file(struct inode *inode, struct file *filp)
1086 {
1087         /*
1088          * ordered_data_close is set by settattr when we are about to truncate
1089          * a file from a non-zero size to a zero size.  This tries to
1090          * flush down new bytes that may have been written if the
1091          * application were using truncate to replace a file in place.
1092          */
1093         if (BTRFS_I(inode)->ordered_data_close) {
1094                 BTRFS_I(inode)->ordered_data_close = 0;
1095                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1096                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1097                         filemap_flush(inode->i_mapping);
1098         }
1099         if (filp->private_data)
1100                 btrfs_ioctl_trans_end(filp);
1101         return 0;
1102 }
1103
1104 /*
1105  * fsync call for both files and directories.  This logs the inode into
1106  * the tree log instead of forcing full commits whenever possible.
1107  *
1108  * It needs to call filemap_fdatawait so that all ordered extent updates are
1109  * in the metadata btree are up to date for copying to the log.
1110  *
1111  * It drops the inode mutex before doing the tree log commit.  This is an
1112  * important optimization for directories because holding the mutex prevents
1113  * new operations on the dir while we write to disk.
1114  */
1115 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1116 {
1117         struct inode *inode = dentry->d_inode;
1118         struct btrfs_root *root = BTRFS_I(inode)->root;
1119         int ret = 0;
1120         struct btrfs_trans_handle *trans;
1121
1122         /*
1123          * check the transaction that last modified this inode
1124          * and see if its already been committed
1125          */
1126         if (!BTRFS_I(inode)->last_trans)
1127                 goto out;
1128
1129         mutex_lock(&root->fs_info->trans_mutex);
1130         if (BTRFS_I(inode)->last_trans <=
1131             root->fs_info->last_trans_committed) {
1132                 BTRFS_I(inode)->last_trans = 0;
1133                 mutex_unlock(&root->fs_info->trans_mutex);
1134                 goto out;
1135         }
1136         mutex_unlock(&root->fs_info->trans_mutex);
1137
1138         root->log_batch++;
1139         filemap_fdatawrite(inode->i_mapping);
1140         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1141         root->log_batch++;
1142
1143         if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1144                 goto out;
1145         /*
1146          * ok we haven't committed the transaction yet, lets do a commit
1147          */
1148         if (file && file->private_data)
1149                 btrfs_ioctl_trans_end(file);
1150
1151         trans = btrfs_start_transaction(root, 1);
1152         if (!trans) {
1153                 ret = -ENOMEM;
1154                 goto out;
1155         }
1156
1157         ret = btrfs_log_dentry_safe(trans, root, dentry);
1158         if (ret < 0)
1159                 goto out;
1160
1161         /* we've logged all the items and now have a consistent
1162          * version of the file in the log.  It is possible that
1163          * someone will come in and modify the file, but that's
1164          * fine because the log is consistent on disk, and we
1165          * have references to all of the file's extents
1166          *
1167          * It is possible that someone will come in and log the
1168          * file again, but that will end up using the synchronization
1169          * inside btrfs_sync_log to keep things safe.
1170          */
1171         mutex_unlock(&dentry->d_inode->i_mutex);
1172
1173         if (ret > 0) {
1174                 ret = btrfs_commit_transaction(trans, root);
1175         } else {
1176                 ret = btrfs_sync_log(trans, root);
1177                 if (ret == 0)
1178                         ret = btrfs_end_transaction(trans, root);
1179                 else
1180                         ret = btrfs_commit_transaction(trans, root);
1181         }
1182         mutex_lock(&dentry->d_inode->i_mutex);
1183 out:
1184         return ret > 0 ? EIO : ret;
1185 }
1186
1187 static const struct vm_operations_struct btrfs_file_vm_ops = {
1188         .fault          = filemap_fault,
1189         .page_mkwrite   = btrfs_page_mkwrite,
1190 };
1191
1192 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1193 {
1194         vma->vm_ops = &btrfs_file_vm_ops;
1195         file_accessed(filp);
1196         return 0;
1197 }
1198
1199 struct file_operations btrfs_file_operations = {
1200         .llseek         = generic_file_llseek,
1201         .read           = do_sync_read,
1202         .aio_read       = generic_file_aio_read,
1203         .splice_read    = generic_file_splice_read,
1204         .write          = btrfs_file_write,
1205         .mmap           = btrfs_file_mmap,
1206         .open           = generic_file_open,
1207         .release        = btrfs_release_file,
1208         .fsync          = btrfs_sync_file,
1209         .unlocked_ioctl = btrfs_ioctl,
1210 #ifdef CONFIG_COMPAT
1211         .compat_ioctl   = btrfs_ioctl,
1212 #endif
1213 };