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/falloc.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/slab.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39 #include "tree-log.h"
  40 #include "locking.h"
  41 #include "compat.h"
  42
  43
  44 /* simple helper to fault in pages and copy.  This should go away
  45  * and be replaced with calls into generic code.
  46  */
  47 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  48                                          int write_bytes,
  49                                          struct page **prepared_pages,
  50                                          struct iov_iter *i)
  51 {
  52         size_t copied = 0;
  53         int pg = 0;
  54         int offset = pos & (PAGE_CACHE_SIZE - 1);
  55         int total_copied = 0;
  56
  57         while (write_bytes > 0) {
  58                 size_t count = min_t(size_t,
  59                                      PAGE_CACHE_SIZE - offset, write_bytes);
  60                 struct page *page = prepared_pages[pg];
  61                 /*
  62                  * Copy data from userspace to the current page
  63                  *
  64                  * Disable pagefault to avoid recursive lock since
  65                  * the pages are already locked
  66                  */
  67                 pagefault_disable();
  68                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
  69                 pagefault_enable();
  70
  71                 /* Flush processor's dcache for this page */
  72                 flush_dcache_page(page);
  73                 iov_iter_advance(i, copied);
  74                 write_bytes -= copied;
  75                 total_copied += copied;
  76
  77                 /* Return to btrfs_file_aio_write to fault page */
  78                 if (unlikely(copied == 0)) {
  79                         break;
  80                 }
  81
  82                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
  83                         offset += copied;
  84                 } else {
  85                         pg++;
  86                         offset = 0;
  87                 }
  88         }
  89         return total_copied;
  90 }
  91
  92 /*
  93  * unlocks pages after btrfs_file_write is done with them
  94  */
  95 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
  96 {
  97         size_t i;
  98         for (i = 0; i < num_pages; i++) {
  99                 if (!pages[i])
 100                         break;
 101                 /* page checked is some magic around finding pages that
 102                  * have been modified without going through btrfs_set_page_dirty
 103                  * clear it here
 104                  */
 105                 ClearPageChecked(pages[i]);
 106                 unlock_page(pages[i]);
 107                 mark_page_accessed(pages[i]);
 108                 page_cache_release(pages[i]);
 109         }
 110 }
 111
 112 /*
 113  * after copy_from_user, pages need to be dirtied and we need to make
 114  * sure holes are created between the current EOF and the start of
 115  * any next extents (if required).
 116  *
 117  * this also makes the decision about creating an inline extent vs
 118  * doing real data extents, marking pages dirty and delalloc as required.
 119  */
 120 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 121                                    struct btrfs_root *root,
 122                                    struct file *file,
 123                                    struct page **pages,
 124                                    size_t num_pages,
 125                                    loff_t pos,
 126                                    size_t write_bytes)
 127 {
 128         int err = 0;
 129         int i;
 130         struct inode *inode = fdentry(file)->d_inode;
 131         u64 num_bytes;
 132         u64 start_pos;
 133         u64 end_of_last_block;
 134         u64 end_pos = pos + write_bytes;
 135         loff_t isize = i_size_read(inode);
 136
 137         start_pos = pos & ~((u64)root->sectorsize - 1);
 138         num_bytes = (write_bytes + pos - start_pos +
 139                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 140
 141         end_of_last_block = start_pos + num_bytes - 1;
 142         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 143                                         NULL);
 144         BUG_ON(err);
 145
 146         for (i = 0; i < num_pages; i++) {
 147                 struct page *p = pages[i];
 148                 SetPageUptodate(p);
 149                 ClearPageChecked(p);
 150                 set_page_dirty(p);
 151         }
 152         if (end_pos > isize) {
 153                 i_size_write(inode, end_pos);
 154                 /* we've only changed i_size in ram, and we haven't updated
 155                  * the disk i_size.  There is no need to log the inode
 156                  * at this time.
 157                  */
 158         }
 159         return 0;
 160 }
 161
 162 /*
 163  * this drops all the extents in the cache that intersect the range
 164  * [start, end].  Existing extents are split as required.
 165  */
 166 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 167                             int skip_pinned)
 168 {
 169         struct extent_map *em;
 170         struct extent_map *split = NULL;
 171         struct extent_map *split2 = NULL;
 172         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 173         u64 len = end - start + 1;
 174         int ret;
 175         int testend = 1;
 176         unsigned long flags;
 177         int compressed = 0;
 178
 179         WARN_ON(end < start);
 180         if (end == (u64)-1) {
 181                 len = (u64)-1;
 182                 testend = 0;
 183         }
 184         while (1) {
 185                 if (!split)
 186                         split = alloc_extent_map(GFP_NOFS);
 187                 if (!split2)
 188                         split2 = alloc_extent_map(GFP_NOFS);
 189                 BUG_ON(!split || !split2);
 190
 191                 write_lock(&em_tree->lock);
 192                 em = lookup_extent_mapping(em_tree, start, len);
 193                 if (!em) {
 194                         write_unlock(&em_tree->lock);
 195                         break;
 196                 }
 197                 flags = em->flags;
 198                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 199                         if (testend && em->start + em->len >= start + len) {
 200                                 free_extent_map(em);
 201                                 write_unlock(&em_tree->lock);
 202                                 break;
 203                         }
 204                         start = em->start + em->len;
 205                         if (testend)
 206                                 len = start + len - (em->start + em->len);
 207                         free_extent_map(em);
 208                         write_unlock(&em_tree->lock);
 209                         continue;
 210                 }
 211                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 212                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 213                 remove_extent_mapping(em_tree, em);
 214
 215                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 216                     em->start < start) {
 217                         split->start = em->start;
 218                         split->len = start - em->start;
 219                         split->orig_start = em->orig_start;
 220                         split->block_start = em->block_start;
 221
 222                         if (compressed)
 223                                 split->block_len = em->block_len;
 224                         else
 225                                 split->block_len = split->len;
 226
 227                         split->bdev = em->bdev;
 228                         split->flags = flags;
 229                         split->compress_type = em->compress_type;
 230                         ret = add_extent_mapping(em_tree, split);
 231                         BUG_ON(ret);
 232                         free_extent_map(split);
 233                         split = split2;
 234                         split2 = NULL;
 235                 }
 236                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 237                     testend && em->start + em->len > start + len) {
 238                         u64 diff = start + len - em->start;
 239
 240                         split->start = start + len;
 241                         split->len = em->start + em->len - (start + len);
 242                         split->bdev = em->bdev;
 243                         split->flags = flags;
 244                         split->compress_type = em->compress_type;
 245
 246                         if (compressed) {
 247                                 split->block_len = em->block_len;
 248                                 split->block_start = em->block_start;
 249                                 split->orig_start = em->orig_start;
 250                         } else {
 251                                 split->block_len = split->len;
 252                                 split->block_start = em->block_start + diff;
 253                                 split->orig_start = split->start;
 254                         }
 255
 256                         ret = add_extent_mapping(em_tree, split);
 257                         BUG_ON(ret);
 258                         free_extent_map(split);
 259                         split = NULL;
 260                 }
 261                 write_unlock(&em_tree->lock);
 262
 263                 /* once for us */
 264                 free_extent_map(em);
 265                 /* once for the tree*/
 266                 free_extent_map(em);
 267         }
 268         if (split)
 269                 free_extent_map(split);
 270         if (split2)
 271                 free_extent_map(split2);
 272         return 0;
 273 }
 274
 275 /*
 276  * this is very complex, but the basic idea is to drop all extents
 277  * in the range start - end.  hint_block is filled in with a block number
 278  * that would be a good hint to the block allocator for this file.
 279  *
 280  * If an extent intersects the range but is not entirely inside the range
 281  * it is either truncated or split.  Anything entirely inside the range
 282  * is deleted from the tree.
 283  */
 284 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
 285                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
 286 {
 287         struct btrfs_root *root = BTRFS_I(inode)->root;
 288         struct extent_buffer *leaf;
 289         struct btrfs_file_extent_item *fi;
 290         struct btrfs_path *path;
 291         struct btrfs_key key;
 292         struct btrfs_key new_key;
 293         u64 search_start = start;
 294         u64 disk_bytenr = 0;
 295         u64 num_bytes = 0;
 296         u64 extent_offset = 0;
 297         u64 extent_end = 0;
 298         int del_nr = 0;
 299         int del_slot = 0;
 300         int extent_type;
 301         int recow;
 302         int ret;
 303
 304         if (drop_cache)
 305                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 306
 307         path = btrfs_alloc_path();
 308         if (!path)
 309                 return -ENOMEM;
 310
 311         while (1) {
 312                 recow = 0;
 313                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 314                                                search_start, -1);
 315                 if (ret < 0)
 316                         break;
 317                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 318                         leaf = path->nodes[0];
 319                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 320                         if (key.objectid == inode->i_ino &&
 321                             key.type == BTRFS_EXTENT_DATA_KEY)
 322                                 path->slots[0]--;
 323                 }
 324                 ret = 0;
 325 next_slot:
 326                 leaf = path->nodes[0];
 327                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 328                         BUG_ON(del_nr > 0);
 329                         ret = btrfs_next_leaf(root, path);
 330                         if (ret < 0)
 331                                 break;
 332                         if (ret > 0) {
 333                                 ret = 0;
 334                                 break;
 335                         }
 336                         leaf = path->nodes[0];
 337                         recow = 1;
 338                 }
 339
 340                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 341                 if (key.objectid > inode->i_ino ||
 342                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 343                         break;
 344
 345                 fi = btrfs_item_ptr(leaf, path->slots[0],
 346                                     struct btrfs_file_extent_item);
 347                 extent_type = btrfs_file_extent_type(leaf, fi);
 348
 349                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 350                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 351                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 352                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 353                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 354                         extent_end = key.offset +
 355                                 btrfs_file_extent_num_bytes(leaf, fi);
 356                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 357                         extent_end = key.offset +
 358                                 btrfs_file_extent_inline_len(leaf, fi);
 359                 } else {
 360                         WARN_ON(1);
 361                         extent_end = search_start;
 362                 }
 363
 364                 if (extent_end <= search_start) {
 365                         path->slots[0]++;
 366                         goto next_slot;
 367                 }
 368
 369                 search_start = max(key.offset, start);
 370                 if (recow) {
 371                         btrfs_release_path(root, path);
 372                         continue;
 373                 }
 374
 375                 /*
 376                  *     | - range to drop - |
 377                  *  | -------- extent -------- |
 378                  */
 379                 if (start > key.offset && end < extent_end) {
 380                         BUG_ON(del_nr > 0);
 381                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 382
 383                         memcpy(&new_key, &key, sizeof(new_key));
 384                         new_key.offset = start;
 385                         ret = btrfs_duplicate_item(trans, root, path,
 386                                                    &new_key);
 387                         if (ret == -EAGAIN) {
 388                                 btrfs_release_path(root, path);
 389                                 continue;
 390                         }
 391                         if (ret < 0)
 392                                 break;
 393
 394                         leaf = path->nodes[0];
 395                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 396                                             struct btrfs_file_extent_item);
 397                         btrfs_set_file_extent_num_bytes(leaf, fi,
 398                                                         start - key.offset);
 399
 400                         fi = btrfs_item_ptr(leaf, path->slots[0],
 401                                             struct btrfs_file_extent_item);
 402
 403                         extent_offset += start - key.offset;
 404                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 405                         btrfs_set_file_extent_num_bytes(leaf, fi,
 406                                                         extent_end - start);
 407                         btrfs_mark_buffer_dirty(leaf);
 408
 409                         if (disk_bytenr > 0) {
 410                                 ret = btrfs_inc_extent_ref(trans, root,
 411                                                 disk_bytenr, num_bytes, 0,
 412                                                 root->root_key.objectid,
 413                                                 new_key.objectid,
 414                                                 start - extent_offset);
 415                                 BUG_ON(ret);
 416                                 *hint_byte = disk_bytenr;
 417                         }
 418                         key.offset = start;
 419                 }
 420                 /*
 421                  *  | ---- range to drop ----- |
 422                  *      | -------- extent -------- |
 423                  */
 424                 if (start <= key.offset && end < extent_end) {
 425                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 426
 427                         memcpy(&new_key, &key, sizeof(new_key));
 428                         new_key.offset = end;
 429                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 430
 431                         extent_offset += end - key.offset;
 432                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 433                         btrfs_set_file_extent_num_bytes(leaf, fi,
 434                                                         extent_end - end);
 435                         btrfs_mark_buffer_dirty(leaf);
 436                         if (disk_bytenr > 0) {
 437                                 inode_sub_bytes(inode, end - key.offset);
 438                                 *hint_byte = disk_bytenr;
 439                         }
 440                         break;
 441                 }
 442
 443                 search_start = extent_end;
 444                 /*
 445                  *       | ---- range to drop ----- |
 446                  *  | -------- extent -------- |
 447                  */
 448                 if (start > key.offset && end >= extent_end) {
 449                         BUG_ON(del_nr > 0);
 450                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 451
 452                         btrfs_set_file_extent_num_bytes(leaf, fi,
 453                                                         start - key.offset);
 454                         btrfs_mark_buffer_dirty(leaf);
 455                         if (disk_bytenr > 0) {
 456                                 inode_sub_bytes(inode, extent_end - start);
 457                                 *hint_byte = disk_bytenr;
 458                         }
 459                         if (end == extent_end)
 460                                 break;
 461
 462                         path->slots[0]++;
 463                         goto next_slot;
 464                 }
 465
 466                 /*
 467                  *  | ---- range to drop ----- |
 468                  *    | ------ extent ------ |
 469                  */
 470                 if (start <= key.offset && end >= extent_end) {
 471                         if (del_nr == 0) {
 472                                 del_slot = path->slots[0];
 473                                 del_nr = 1;
 474                         } else {
 475                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 476                                 del_nr++;
 477                         }
 478
 479                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 480                                 inode_sub_bytes(inode,
 481                                                 extent_end - key.offset);
 482                                 extent_end = ALIGN(extent_end,
 483                                                    root->sectorsize);
 484                         } else if (disk_bytenr > 0) {
 485                                 ret = btrfs_free_extent(trans, root,
 486                                                 disk_bytenr, num_bytes, 0,
 487                                                 root->root_key.objectid,
 488                                                 key.objectid, key.offset -
 489                                                 extent_offset);
 490                                 BUG_ON(ret);
 491                                 inode_sub_bytes(inode,
 492                                                 extent_end - key.offset);
 493                                 *hint_byte = disk_bytenr;
 494                         }
 495
 496                         if (end == extent_end)
 497                                 break;
 498
 499                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 500                                 path->slots[0]++;
 501                                 goto next_slot;
 502                         }
 503
 504                         ret = btrfs_del_items(trans, root, path, del_slot,
 505                                               del_nr);
 506                         BUG_ON(ret);
 507
 508                         del_nr = 0;
 509                         del_slot = 0;
 510
 511                         btrfs_release_path(root, path);
 512                         continue;
 513                 }
 514
 515                 BUG_ON(1);
 516         }
 517
 518         if (del_nr > 0) {
 519                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 520                 BUG_ON(ret);
 521         }
 522
 523         btrfs_free_path(path);
 524         return ret;
 525 }
 526
 527 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 528                             u64 objectid, u64 bytenr, u64 orig_offset,
 529                             u64 *start, u64 *end)
 530 {
 531         struct btrfs_file_extent_item *fi;
 532         struct btrfs_key key;
 533         u64 extent_end;
 534
 535         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 536                 return 0;
 537
 538         btrfs_item_key_to_cpu(leaf, &key, slot);
 539         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 540                 return 0;
 541
 542         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 543         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 544             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 545             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 546             btrfs_file_extent_compression(leaf, fi) ||
 547             btrfs_file_extent_encryption(leaf, fi) ||
 548             btrfs_file_extent_other_encoding(leaf, fi))
 549                 return 0;
 550
 551         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 552         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 553                 return 0;
 554
 555         *start = key.offset;
 556         *end = extent_end;
 557         return 1;
 558 }
 559
 560 /*
 561  * Mark extent in the range start - end as written.
 562  *
 563  * This changes extent type from 'pre-allocated' to 'regular'. If only
 564  * part of extent is marked as written, the extent will be split into
 565  * two or three.
 566  */
 567 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 568                               struct inode *inode, u64 start, u64 end)
 569 {
 570         struct btrfs_root *root = BTRFS_I(inode)->root;
 571         struct extent_buffer *leaf;
 572         struct btrfs_path *path;
 573         struct btrfs_file_extent_item *fi;
 574         struct btrfs_key key;
 575         struct btrfs_key new_key;
 576         u64 bytenr;
 577         u64 num_bytes;
 578         u64 extent_end;
 579         u64 orig_offset;
 580         u64 other_start;
 581         u64 other_end;
 582         u64 split;
 583         int del_nr = 0;
 584         int del_slot = 0;
 585         int recow;
 586         int ret;
 587
 588         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 589
 590         path = btrfs_alloc_path();
 591         BUG_ON(!path);
 592 again:
 593         recow = 0;
 594         split = start;
 595         key.objectid = inode->i_ino;
 596         key.type = BTRFS_EXTENT_DATA_KEY;
 597         key.offset = split;
 598
 599         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 600         if (ret > 0 && path->slots[0] > 0)
 601                 path->slots[0]--;
 602
 603         leaf = path->nodes[0];
 604         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 605         BUG_ON(key.objectid != inode->i_ino ||
 606                key.type != BTRFS_EXTENT_DATA_KEY);
 607         fi = btrfs_item_ptr(leaf, path->slots[0],
 608                             struct btrfs_file_extent_item);
 609         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 610                BTRFS_FILE_EXTENT_PREALLOC);
 611         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 612         BUG_ON(key.offset > start || extent_end < end);
 613
 614         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 615         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 616         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 617         memcpy(&new_key, &key, sizeof(new_key));
 618
 619         if (start == key.offset && end < extent_end) {
 620                 other_start = 0;
 621                 other_end = start;
 622                 if (extent_mergeable(leaf, path->slots[0] - 1,
 623                                      inode->i_ino, bytenr, orig_offset,
 624                                      &other_start, &other_end)) {
 625                         new_key.offset = end;
 626                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 627                         fi = btrfs_item_ptr(leaf, path->slots[0],
 628                                             struct btrfs_file_extent_item);
 629                         btrfs_set_file_extent_num_bytes(leaf, fi,
 630                                                         extent_end - end);
 631                         btrfs_set_file_extent_offset(leaf, fi,
 632                                                      end - orig_offset);
 633                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 634                                             struct btrfs_file_extent_item);
 635                         btrfs_set_file_extent_num_bytes(leaf, fi,
 636                                                         end - other_start);
 637                         btrfs_mark_buffer_dirty(leaf);
 638                         goto out;
 639                 }
 640         }
 641
 642         if (start > key.offset && end == extent_end) {
 643                 other_start = end;
 644                 other_end = 0;
 645                 if (extent_mergeable(leaf, path->slots[0] + 1,
 646                                      inode->i_ino, bytenr, orig_offset,
 647                                      &other_start, &other_end)) {
 648                         fi = btrfs_item_ptr(leaf, path->slots[0],
 649                                             struct btrfs_file_extent_item);
 650                         btrfs_set_file_extent_num_bytes(leaf, fi,
 651                                                         start - key.offset);
 652                         path->slots[0]++;
 653                         new_key.offset = start;
 654                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 655
 656                         fi = btrfs_item_ptr(leaf, path->slots[0],
 657                                             struct btrfs_file_extent_item);
 658                         btrfs_set_file_extent_num_bytes(leaf, fi,
 659                                                         other_end - start);
 660                         btrfs_set_file_extent_offset(leaf, fi,
 661                                                      start - orig_offset);
 662                         btrfs_mark_buffer_dirty(leaf);
 663                         goto out;
 664                 }
 665         }
 666
 667         while (start > key.offset || end < extent_end) {
 668                 if (key.offset == start)
 669                         split = end;
 670
 671                 new_key.offset = split;
 672                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
 673                 if (ret == -EAGAIN) {
 674                         btrfs_release_path(root, path);
 675                         goto again;
 676                 }
 677                 BUG_ON(ret < 0);
 678
 679                 leaf = path->nodes[0];
 680                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 681                                     struct btrfs_file_extent_item);
 682                 btrfs_set_file_extent_num_bytes(leaf, fi,
 683                                                 split - key.offset);
 684
 685                 fi = btrfs_item_ptr(leaf, path->slots[0],
 686                                     struct btrfs_file_extent_item);
 687
 688                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 689                 btrfs_set_file_extent_num_bytes(leaf, fi,
 690                                                 extent_end - split);
 691                 btrfs_mark_buffer_dirty(leaf);
 692
 693                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 694                                            root->root_key.objectid,
 695                                            inode->i_ino, orig_offset);
 696                 BUG_ON(ret);
 697
 698                 if (split == start) {
 699                         key.offset = start;
 700                 } else {
 701                         BUG_ON(start != key.offset);
 702                         path->slots[0]--;
 703                         extent_end = end;
 704                 }
 705                 recow = 1;
 706         }
 707
 708         other_start = end;
 709         other_end = 0;
 710         if (extent_mergeable(leaf, path->slots[0] + 1,
 711                              inode->i_ino, bytenr, orig_offset,
 712                              &other_start, &other_end)) {
 713                 if (recow) {
 714                         btrfs_release_path(root, path);
 715                         goto again;
 716                 }
 717                 extent_end = other_end;
 718                 del_slot = path->slots[0] + 1;
 719                 del_nr++;
 720                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 721                                         0, root->root_key.objectid,
 722                                         inode->i_ino, orig_offset);
 723                 BUG_ON(ret);
 724         }
 725         other_start = 0;
 726         other_end = start;
 727         if (extent_mergeable(leaf, path->slots[0] - 1,
 728                              inode->i_ino, bytenr, orig_offset,
 729                              &other_start, &other_end)) {
 730                 if (recow) {
 731                         btrfs_release_path(root, path);
 732                         goto again;
 733                 }
 734                 key.offset = other_start;
 735                 del_slot = path->slots[0];
 736                 del_nr++;
 737                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 738                                         0, root->root_key.objectid,
 739                                         inode->i_ino, orig_offset);
 740                 BUG_ON(ret);
 741         }
 742         if (del_nr == 0) {
 743                 fi = btrfs_item_ptr(leaf, path->slots[0],
 744                            struct btrfs_file_extent_item);
 745                 btrfs_set_file_extent_type(leaf, fi,
 746                                            BTRFS_FILE_EXTENT_REG);
 747                 btrfs_mark_buffer_dirty(leaf);
 748         } else {
 749                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 750                            struct btrfs_file_extent_item);
 751                 btrfs_set_file_extent_type(leaf, fi,
 752                                            BTRFS_FILE_EXTENT_REG);
 753                 btrfs_set_file_extent_num_bytes(leaf, fi,
 754                                                 extent_end - key.offset);
 755                 btrfs_mark_buffer_dirty(leaf);
 756
 757                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 758                 BUG_ON(ret);
 759         }
 760 out:
 761         btrfs_free_path(path);
 762         return 0;
 763 }
 764
 765 /*
 766  * this gets pages into the page cache and locks them down, it also properly
 767  * waits for data=ordered extents to finish before allowing the pages to be
 768  * modified.
 769  */
 770 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 771                          struct page **pages, size_t num_pages,
 772                          loff_t pos, unsigned long first_index,
 773                          unsigned long last_index, size_t write_bytes)
 774 {
 775         struct extent_state *cached_state = NULL;
 776         int i;
 777         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 778         struct inode *inode = fdentry(file)->d_inode;
 779         int err = 0;
 780         u64 start_pos;
 781         u64 last_pos;
 782
 783         start_pos = pos & ~((u64)root->sectorsize - 1);
 784         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 785
 786         if (start_pos > inode->i_size) {
 787                 err = btrfs_cont_expand(inode, start_pos);
 788                 if (err)
 789                         return err;
 790         }
 791
 792         memset(pages, 0, num_pages * sizeof(struct page *));
 793 again:
 794         for (i = 0; i < num_pages; i++) {
 795                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 796                 if (!pages[i]) {
 797                         int c;
 798                         for (c = i - 1; c >= 0; c--) {
 799                                 unlock_page(pages[c]);
 800                                 page_cache_release(pages[c]);
 801                         }
 802                         return -ENOMEM;
 803                 }
 804                 wait_on_page_writeback(pages[i]);
 805         }
 806         if (start_pos < inode->i_size) {
 807                 struct btrfs_ordered_extent *ordered;
 808                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
 809                                  start_pos, last_pos - 1, 0, &cached_state,
 810                                  GFP_NOFS);
 811                 ordered = btrfs_lookup_first_ordered_extent(inode,
 812                                                             last_pos - 1);
 813                 if (ordered &&
 814                     ordered->file_offset + ordered->len > start_pos &&
 815                     ordered->file_offset < last_pos) {
 816                         btrfs_put_ordered_extent(ordered);
 817                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 818                                              start_pos, last_pos - 1,
 819                                              &cached_state, GFP_NOFS);
 820                         for (i = 0; i < num_pages; i++) {
 821                                 unlock_page(pages[i]);
 822                                 page_cache_release(pages[i]);
 823                         }
 824                         btrfs_wait_ordered_range(inode, start_pos,
 825                                                  last_pos - start_pos);
 826                         goto again;
 827                 }
 828                 if (ordered)
 829                         btrfs_put_ordered_extent(ordered);
 830
 831                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
 832                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 833                                   EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
 834                                   GFP_NOFS);
 835                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
 836                                      start_pos, last_pos - 1, &cached_state,
 837                                      GFP_NOFS);
 838         }
 839         for (i = 0; i < num_pages; i++) {
 840                 clear_page_dirty_for_io(pages[i]);
 841                 set_page_extent_mapped(pages[i]);
 842                 WARN_ON(!PageLocked(pages[i]));
 843         }
 844         return 0;
 845 }
 846
 847 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
 848                                     const struct iovec *iov,
 849                                     unsigned long nr_segs, loff_t pos)
 850 {
 851         struct file *file = iocb->ki_filp;
 852         struct inode *inode = fdentry(file)->d_inode;
 853         struct btrfs_root *root = BTRFS_I(inode)->root;
 854         struct page *pinned[2];
 855         struct page **pages = NULL;
 856         struct iov_iter i;
 857         loff_t *ppos = &iocb->ki_pos;
 858         loff_t start_pos;
 859         ssize_t num_written = 0;
 860         ssize_t err = 0;
 861         size_t count;
 862         size_t ocount;
 863         int ret = 0;
 864         int nrptrs;
 865         unsigned long first_index;
 866         unsigned long last_index;
 867         int will_write;
 868         int buffered = 0;
 869         int copied = 0;
 870         int dirty_pages = 0;
 871
 872         will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
 873                       (file->f_flags & O_DIRECT));
 874
 875         pinned[0] = NULL;
 876         pinned[1] = NULL;
 877
 878         start_pos = pos;
 879
 880         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 881
 882         mutex_lock(&inode->i_mutex);
 883
 884         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
 885         if (err)
 886                 goto out;
 887         count = ocount;
 888
 889         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 890         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 891         if (err)
 892                 goto out;
 893
 894         if (count == 0)
 895                 goto out;
 896
 897         err = file_remove_suid(file);
 898         if (err)
 899                 goto out;
 900
 901         /*
 902          * If BTRFS flips readonly due to some impossible error
 903          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
 904          * although we have opened a file as writable, we have
 905          * to stop this write operation to ensure FS consistency.
 906          */
 907         if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 908                 err = -EROFS;
 909                 goto out;
 910         }
 911
 912         file_update_time(file);
 913         BTRFS_I(inode)->sequence++;
 914
 915         if (unlikely(file->f_flags & O_DIRECT)) {
 916                 num_written = generic_file_direct_write(iocb, iov, &nr_segs,
 917                                                         pos, ppos, count,
 918                                                         ocount);
 919                 /*
 920                  * the generic O_DIRECT will update in-memory i_size after the
 921                  * DIOs are done.  But our endio handlers that update the on
 922                  * disk i_size never update past the in memory i_size.  So we
 923                  * need one more update here to catch any additions to the
 924                  * file
 925                  */
 926                 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
 927                         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
 928                         mark_inode_dirty(inode);
 929                 }
 930
 931                 if (num_written < 0) {
 932                         ret = num_written;
 933                         num_written = 0;
 934                         goto out;
 935                 } else if (num_written == count) {
 936                         /* pick up pos changes done by the generic code */
 937                         pos = *ppos;
 938                         goto out;
 939                 }
 940                 /*
 941                  * We are going to do buffered for the rest of the range, so we
 942                  * need to make sure to invalidate the buffered pages when we're
 943                  * done.
 944                  */
 945                 buffered = 1;
 946                 pos += num_written;
 947         }
 948
 949         iov_iter_init(&i, iov, nr_segs, count, num_written);
 950         nrptrs = min((iov_iter_count(&i) + PAGE_CACHE_SIZE - 1) /
 951                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
 952                      (sizeof(struct page *)));
 953         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 954         if (!pages) {
 955                 ret = -ENOMEM;
 956                 goto out;
 957         }
 958
 959         /* generic_write_checks can change our pos */
 960         start_pos = pos;
 961
 962         first_index = pos >> PAGE_CACHE_SHIFT;
 963         last_index = (pos + iov_iter_count(&i)) >> PAGE_CACHE_SHIFT;
 964
 965         /*
 966          * there are lots of better ways to do this, but this code
 967          * makes sure the first and last page in the file range are
 968          * up to date and ready for cow
 969          */
 970         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 971                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 972                 if (!PageUptodate(pinned[0])) {
 973                         ret = btrfs_readpage(NULL, pinned[0]);
 974                         BUG_ON(ret);
 975                         wait_on_page_locked(pinned[0]);
 976                 } else {
 977                         unlock_page(pinned[0]);
 978                 }
 979         }
 980         if ((pos + iov_iter_count(&i)) & (PAGE_CACHE_SIZE - 1)) {
 981                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 982                 if (!PageUptodate(pinned[1])) {
 983                         ret = btrfs_readpage(NULL, pinned[1]);
 984                         BUG_ON(ret);
 985                         wait_on_page_locked(pinned[1]);
 986                 } else {
 987                         unlock_page(pinned[1]);
 988                 }
 989         }
 990
 991         while (iov_iter_count(&i) > 0) {
 992                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 993                 size_t write_bytes = min(iov_iter_count(&i),
 994                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
 995                                          offset);
 996                 size_t num_pages = (write_bytes + offset +
 997                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 998
 999                 WARN_ON(num_pages > nrptrs);
1000                 memset(pages, 0, sizeof(struct page *) * nrptrs);
1001
1002                 /*
1003                  * Fault pages before locking them in prepare_pages
1004                  * to avoid recursive lock
1005                  */
1006                 if (unlikely(iov_iter_fault_in_readable(&i, write_bytes))) {
1007                         ret = -EFAULT;
1008                         goto out;
1009                 }
1010
1011                 ret = btrfs_delalloc_reserve_space(inode,
1012                                         num_pages << PAGE_CACHE_SHIFT);
1013                 if (ret)
1014                         goto out;
1015
1016                 ret = prepare_pages(root, file, pages, num_pages,
1017                                     pos, first_index, last_index,
1018                                     write_bytes);
1019                 if (ret) {
1020                         btrfs_delalloc_release_space(inode,
1021                                         num_pages << PAGE_CACHE_SHIFT);
1022                         goto out;
1023                 }
1024
1025                 copied = btrfs_copy_from_user(pos, num_pages,
1026                                            write_bytes, pages, &i);
1027                 dirty_pages = (copied + offset + PAGE_CACHE_SIZE - 1) >>
1028                                 PAGE_CACHE_SHIFT;
1029
1030                 if (num_pages > dirty_pages) {
1031                         if (copied > 0)
1032                                 atomic_inc(
1033                                         &BTRFS_I(inode)->outstanding_extents);
1034                         btrfs_delalloc_release_space(inode,
1035                                         (num_pages - dirty_pages) <<
1036                                         PAGE_CACHE_SHIFT);
1037                 }
1038
1039                 if (copied > 0) {
1040                         dirty_and_release_pages(NULL, root, file, pages,
1041                                                 dirty_pages, pos, copied);
1042                 }
1043
1044                 btrfs_drop_pages(pages, num_pages);
1045
1046                 if (copied > 0) {
1047                         if (will_write) {
1048                                 filemap_fdatawrite_range(inode->i_mapping, pos,
1049                                                          pos + copied - 1);
1050                         } else {
1051                                 balance_dirty_pages_ratelimited_nr(
1052                                                         inode->i_mapping,
1053                                                         dirty_pages);
1054                                 if (dirty_pages <
1055                                 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1056                                         btrfs_btree_balance_dirty(root, 1);
1057                                 btrfs_throttle(root);
1058                         }
1059                 }
1060
1061                 pos += copied;
1062                 num_written += copied;
1063
1064                 cond_resched();
1065         }
1066 out:
1067         mutex_unlock(&inode->i_mutex);
1068         if (ret)
1069                 err = ret;
1070
1071         kfree(pages);
1072         if (pinned[0])
1073                 page_cache_release(pinned[0]);
1074         if (pinned[1])
1075                 page_cache_release(pinned[1]);
1076         *ppos = pos;
1077
1078         /*
1079          * we want to make sure fsync finds this change
1080          * but we haven't joined a transaction running right now.
1081          *
1082          * Later on, someone is sure to update the inode and get the
1083          * real transid recorded.
1084          *
1085          * We set last_trans now to the fs_info generation + 1,
1086          * this will either be one more than the running transaction
1087          * or the generation used for the next transaction if there isn't
1088          * one running right now.
1089          */
1090         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1091
1092         if (num_written > 0 && will_write) {
1093                 struct btrfs_trans_handle *trans;
1094
1095                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1096                 if (err)
1097                         num_written = err;
1098
1099                 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1100                         trans = btrfs_start_transaction(root, 0);
1101                         if (IS_ERR(trans)) {
1102                                 num_written = PTR_ERR(trans);
1103                                 goto done;
1104                         }
1105                         mutex_lock(&inode->i_mutex);
1106                         ret = btrfs_log_dentry_safe(trans, root,
1107                                                     file->f_dentry);
1108                         mutex_unlock(&inode->i_mutex);
1109                         if (ret == 0) {
1110                                 ret = btrfs_sync_log(trans, root);
1111                                 if (ret == 0)
1112                                         btrfs_end_transaction(trans, root);
1113                                 else
1114                                         btrfs_commit_transaction(trans, root);
1115                         } else if (ret != BTRFS_NO_LOG_SYNC) {
1116                                 btrfs_commit_transaction(trans, root);
1117                         } else {
1118                                 btrfs_end_transaction(trans, root);
1119                         }
1120                 }
1121                 if (file->f_flags & O_DIRECT && buffered) {
1122                         invalidate_mapping_pages(inode->i_mapping,
1123                               start_pos >> PAGE_CACHE_SHIFT,
1124                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1125                 }
1126         }
1127 done:
1128         current->backing_dev_info = NULL;
1129         return num_written ? num_written : err;
1130 }
1131
1132 int btrfs_release_file(struct inode *inode, struct file *filp)
1133 {
1134         /*
1135          * ordered_data_close is set by settattr when we are about to truncate
1136          * a file from a non-zero size to a zero size.  This tries to
1137          * flush down new bytes that may have been written if the
1138          * application were using truncate to replace a file in place.
1139          */
1140         if (BTRFS_I(inode)->ordered_data_close) {
1141                 BTRFS_I(inode)->ordered_data_close = 0;
1142                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1143                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1144                         filemap_flush(inode->i_mapping);
1145         }
1146         if (filp->private_data)
1147                 btrfs_ioctl_trans_end(filp);
1148         return 0;
1149 }
1150
1151 /*
1152  * fsync call for both files and directories.  This logs the inode into
1153  * the tree log instead of forcing full commits whenever possible.
1154  *
1155  * It needs to call filemap_fdatawait so that all ordered extent updates are
1156  * in the metadata btree are up to date for copying to the log.
1157  *
1158  * It drops the inode mutex before doing the tree log commit.  This is an
1159  * important optimization for directories because holding the mutex prevents
1160  * new operations on the dir while we write to disk.
1161  */
1162 int btrfs_sync_file(struct file *file, int datasync)
1163 {
1164         struct dentry *dentry = file->f_path.dentry;
1165         struct inode *inode = dentry->d_inode;
1166         struct btrfs_root *root = BTRFS_I(inode)->root;
1167         int ret = 0;
1168         struct btrfs_trans_handle *trans;
1169
1170
1171         /* we wait first, since the writeback may change the inode */
1172         root->log_batch++;
1173         /* the VFS called filemap_fdatawrite for us */
1174         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1175         root->log_batch++;
1176
1177         /*
1178          * check the transaction that last modified this inode
1179          * and see if its already been committed
1180          */
1181         if (!BTRFS_I(inode)->last_trans)
1182                 goto out;
1183
1184         /*
1185          * if the last transaction that changed this file was before
1186          * the current transaction, we can bail out now without any
1187          * syncing
1188          */
1189         mutex_lock(&root->fs_info->trans_mutex);
1190         if (BTRFS_I(inode)->last_trans <=
1191             root->fs_info->last_trans_committed) {
1192                 BTRFS_I(inode)->last_trans = 0;
1193                 mutex_unlock(&root->fs_info->trans_mutex);
1194                 goto out;
1195         }
1196         mutex_unlock(&root->fs_info->trans_mutex);
1197
1198         /*
1199          * ok we haven't committed the transaction yet, lets do a commit
1200          */
1201         if (file->private_data)
1202                 btrfs_ioctl_trans_end(file);
1203
1204         trans = btrfs_start_transaction(root, 0);
1205         if (IS_ERR(trans)) {
1206                 ret = PTR_ERR(trans);
1207                 goto out;
1208         }
1209
1210         ret = btrfs_log_dentry_safe(trans, root, dentry);
1211         if (ret < 0)
1212                 goto out;
1213
1214         /* we've logged all the items and now have a consistent
1215          * version of the file in the log.  It is possible that
1216          * someone will come in and modify the file, but that's
1217          * fine because the log is consistent on disk, and we
1218          * have references to all of the file's extents
1219          *
1220          * It is possible that someone will come in and log the
1221          * file again, but that will end up using the synchronization
1222          * inside btrfs_sync_log to keep things safe.
1223          */
1224         mutex_unlock(&dentry->d_inode->i_mutex);
1225
1226         if (ret != BTRFS_NO_LOG_SYNC) {
1227                 if (ret > 0) {
1228                         ret = btrfs_commit_transaction(trans, root);
1229                 } else {
1230                         ret = btrfs_sync_log(trans, root);
1231                         if (ret == 0)
1232                                 ret = btrfs_end_transaction(trans, root);
1233                         else
1234                                 ret = btrfs_commit_transaction(trans, root);
1235                 }
1236         } else {
1237                 ret = btrfs_end_transaction(trans, root);
1238         }
1239         mutex_lock(&dentry->d_inode->i_mutex);
1240 out:
1241         return ret > 0 ? -EIO : ret;
1242 }
1243
1244 static const struct vm_operations_struct btrfs_file_vm_ops = {
1245         .fault          = filemap_fault,
1246         .page_mkwrite   = btrfs_page_mkwrite,
1247 };
1248
1249 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1250 {
1251         struct address_space *mapping = filp->f_mapping;
1252
1253         if (!mapping->a_ops->readpage)
1254                 return -ENOEXEC;
1255
1256         file_accessed(filp);
1257         vma->vm_ops = &btrfs_file_vm_ops;
1258         vma->vm_flags |= VM_CAN_NONLINEAR;
1259
1260         return 0;
1261 }
1262
1263 static long btrfs_fallocate(struct file *file, int mode,
1264                             loff_t offset, loff_t len)
1265 {
1266         struct inode *inode = file->f_path.dentry->d_inode;
1267         struct extent_state *cached_state = NULL;
1268         u64 cur_offset;
1269         u64 last_byte;
1270         u64 alloc_start;
1271         u64 alloc_end;
1272         u64 alloc_hint = 0;
1273         u64 locked_end;
1274         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1275         struct extent_map *em;
1276         int ret;
1277
1278         alloc_start = offset & ~mask;
1279         alloc_end =  (offset + len + mask) & ~mask;
1280
1281         /* We only support the FALLOC_FL_KEEP_SIZE mode */
1282         if (mode & ~FALLOC_FL_KEEP_SIZE)
1283                 return -EOPNOTSUPP;
1284
1285         /*
1286          * wait for ordered IO before we have any locks.  We'll loop again
1287          * below with the locks held.
1288          */
1289         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1290
1291         mutex_lock(&inode->i_mutex);
1292         ret = inode_newsize_ok(inode, alloc_end);
1293         if (ret)
1294                 goto out;
1295
1296         if (alloc_start > inode->i_size) {
1297                 ret = btrfs_cont_expand(inode, alloc_start);
1298                 if (ret)
1299                         goto out;
1300         }
1301
1302         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1303         if (ret)
1304                 goto out;
1305
1306         locked_end = alloc_end - 1;
1307         while (1) {
1308                 struct btrfs_ordered_extent *ordered;
1309
1310                 /* the extent lock is ordered inside the running
1311                  * transaction
1312                  */
1313                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1314                                  locked_end, 0, &cached_state, GFP_NOFS);
1315                 ordered = btrfs_lookup_first_ordered_extent(inode,
1316                                                             alloc_end - 1);
1317                 if (ordered &&
1318                     ordered->file_offset + ordered->len > alloc_start &&
1319                     ordered->file_offset < alloc_end) {
1320                         btrfs_put_ordered_extent(ordered);
1321                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1322                                              alloc_start, locked_end,
1323                                              &cached_state, GFP_NOFS);
1324                         /*
1325                          * we can't wait on the range with the transaction
1326                          * running or with the extent lock held
1327                          */
1328                         btrfs_wait_ordered_range(inode, alloc_start,
1329                                                  alloc_end - alloc_start);
1330                 } else {
1331                         if (ordered)
1332                                 btrfs_put_ordered_extent(ordered);
1333                         break;
1334                 }
1335         }
1336
1337         cur_offset = alloc_start;
1338         while (1) {
1339                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1340                                       alloc_end - cur_offset, 0);
1341                 BUG_ON(IS_ERR(em) || !em);
1342                 last_byte = min(extent_map_end(em), alloc_end);
1343                 last_byte = (last_byte + mask) & ~mask;
1344                 if (em->block_start == EXTENT_MAP_HOLE ||
1345                     (cur_offset >= inode->i_size &&
1346                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1347                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1348                                                         last_byte - cur_offset,
1349                                                         1 << inode->i_blkbits,
1350                                                         offset + len,
1351                                                         &alloc_hint);
1352                         if (ret < 0) {
1353                                 free_extent_map(em);
1354                                 break;
1355                         }
1356                 }
1357                 free_extent_map(em);
1358
1359                 cur_offset = last_byte;
1360                 if (cur_offset >= alloc_end) {
1361                         ret = 0;
1362                         break;
1363                 }
1364         }
1365         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1366                              &cached_state, GFP_NOFS);
1367
1368         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1369 out:
1370         mutex_unlock(&inode->i_mutex);
1371         return ret;
1372 }
1373
1374 const struct file_operations btrfs_file_operations = {
1375         .llseek         = generic_file_llseek,
1376         .read           = do_sync_read,
1377         .write          = do_sync_write,
1378         .aio_read       = generic_file_aio_read,
1379         .splice_read    = generic_file_splice_read,
1380         .aio_write      = btrfs_file_aio_write,
1381         .mmap           = btrfs_file_mmap,
1382         .open           = generic_file_open,
1383         .release        = btrfs_release_file,
1384         .fsync          = btrfs_sync_file,
1385         .fallocate      = btrfs_fallocate,
1386         .unlocked_ioctl = btrfs_ioctl,
1387 #ifdef CONFIG_COMPAT
1388         .compat_ioctl   = btrfs_ioctl,
1389 #endif
1390 };