SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "ioctl.h"
  37 #include "print-tree.h"
  38 #include "tree-log.h"
  39 #include "locking.h"
  40 #include "compat.h"
  41
  42
  43 /* simple helper to fault in pages and copy.  This should go away
  44  * and be replaced with calls into generic code.
  45  */
  46 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  47                                          int write_bytes,
  48                                          struct page **prepared_pages,
  49                                          const char __user *buf)
  50 {
  51         long page_fault = 0;
  52         int i;
  53         int offset = pos & (PAGE_CACHE_SIZE - 1);
  54
  55         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  56                 size_t count = min_t(size_t,
  57                                      PAGE_CACHE_SIZE - offset, write_bytes);
  58                 struct page *page = prepared_pages[i];
  59                 fault_in_pages_readable(buf, count);
  60
  61                 /* Copy data from userspace to the current page */
  62                 kmap(page);
  63                 page_fault = __copy_from_user(page_address(page) + offset,
  64                                               buf, count);
  65                 /* Flush processor's dcache for this page */
  66                 flush_dcache_page(page);
  67                 kunmap(page);
  68                 buf += count;
  69                 write_bytes -= count;
  70
  71                 if (page_fault)
  72                         break;
  73         }
  74         return page_fault ? -EFAULT : 0;
  75 }
  76
  77 /*
  78  * unlocks pages after btrfs_file_write is done with them
  79  */
  80 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
  81 {
  82         size_t i;
  83         for (i = 0; i < num_pages; i++) {
  84                 if (!pages[i])
  85                         break;
  86                 /* page checked is some magic around finding pages that
  87                  * have been modified without going through btrfs_set_page_dirty
  88                  * clear it here
  89                  */
  90                 ClearPageChecked(pages[i]);
  91                 unlock_page(pages[i]);
  92                 mark_page_accessed(pages[i]);
  93                 page_cache_release(pages[i]);
  94         }
  95 }
  96
  97 /*
  98  * after copy_from_user, pages need to be dirtied and we need to make
  99  * sure holes are created between the current EOF and the start of
 100  * any next extents (if required).
 101  *
 102  * this also makes the decision about creating an inline extent vs
 103  * doing real data extents, marking pages dirty and delalloc as required.
 104  */
 105 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 106                                    struct btrfs_root *root,
 107                                    struct file *file,
 108                                    struct page **pages,
 109                                    size_t num_pages,
 110                                    loff_t pos,
 111                                    size_t write_bytes)
 112 {
 113         int err = 0;
 114         int i;
 115         struct inode *inode = fdentry(file)->d_inode;
 116         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 117         u64 hint_byte;
 118         u64 num_bytes;
 119         u64 start_pos;
 120         u64 end_of_last_block;
 121         u64 end_pos = pos + write_bytes;
 122         loff_t isize = i_size_read(inode);
 123
 124         start_pos = pos & ~((u64)root->sectorsize - 1);
 125         num_bytes = (write_bytes + pos - start_pos +
 126                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 127
 128         end_of_last_block = start_pos + num_bytes - 1;
 129
 130         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 131         trans = btrfs_join_transaction(root, 1);
 132         if (!trans) {
 133                 err = -ENOMEM;
 134                 goto out_unlock;
 135         }
 136         btrfs_set_trans_block_group(trans, inode);
 137         hint_byte = 0;
 138
 139         /* check for reserved extents on each page, we don't want
 140          * to reset the delalloc bit on things that already have
 141          * extents reserved.
 142          */
 143         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 144         for (i = 0; i < num_pages; i++) {
 145                 struct page *p = pages[i];
 146                 SetPageUptodate(p);
 147                 ClearPageChecked(p);
 148                 set_page_dirty(p);
 149         }
 150         if (end_pos > isize) {
 151                 i_size_write(inode, end_pos);
 152                 /* we've only changed i_size in ram, and we haven't updated
 153                  * the disk i_size.  There is no need to log the inode
 154                  * at this time.
 155                  */
 156         }
 157         err = btrfs_end_transaction(trans, root);
 158 out_unlock:
 159         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 160         return err;
 161 }
 162
 163 /*
 164  * this drops all the extents in the cache that intersect the range
 165  * [start, end].  Existing extents are split as required.
 166  */
 167 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 168                             int skip_pinned)
 169 {
 170         struct extent_map *em;
 171         struct extent_map *split = NULL;
 172         struct extent_map *split2 = NULL;
 173         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 174         u64 len = end - start + 1;
 175         int ret;
 176         int testend = 1;
 177         unsigned long flags;
 178         int compressed = 0;
 179
 180         WARN_ON(end < start);
 181         if (end == (u64)-1) {
 182                 len = (u64)-1;
 183                 testend = 0;
 184         }
 185         while (1) {
 186                 if (!split)
 187                         split = alloc_extent_map(GFP_NOFS);
 188                 if (!split2)
 189                         split2 = alloc_extent_map(GFP_NOFS);
 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                         write_unlock(&em_tree->lock);
 200                         if (em->start <= start &&
 201                             (!testend || em->start + em->len >= start + len)) {
 202                                 free_extent_map(em);
 203                                 break;
 204                         }
 205                         if (start < em->start) {
 206                                 len = em->start - start;
 207                         } else {
 208                                 len = start + len - (em->start + em->len);
 209                                 start = em->start + em->len;
 210                         }
 211                         free_extent_map(em);
 212                         continue;
 213                 }
 214                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 215                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 216                 remove_extent_mapping(em_tree, em);
 217
 218                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 219                     em->start < start) {
 220                         split->start = em->start;
 221                         split->len = start - em->start;
 222                         split->orig_start = em->orig_start;
 223                         split->block_start = em->block_start;
 224
 225                         if (compressed)
 226                                 split->block_len = em->block_len;
 227                         else
 228                                 split->block_len = split->len;
 229
 230                         split->bdev = em->bdev;
 231                         split->flags = flags;
 232                         ret = add_extent_mapping(em_tree, split);
 233                         BUG_ON(ret);
 234                         free_extent_map(split);
 235                         split = split2;
 236                         split2 = NULL;
 237                 }
 238                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 239                     testend && em->start + em->len > start + len) {
 240                         u64 diff = start + len - em->start;
 241
 242                         split->start = start + len;
 243                         split->len = em->start + em->len - (start + len);
 244                         split->bdev = em->bdev;
 245                         split->flags = flags;
 246
 247                         if (compressed) {
 248                                 split->block_len = em->block_len;
 249                                 split->block_start = em->block_start;
 250                                 split->orig_start = em->orig_start;
 251                         } else {
 252                                 split->block_len = split->len;
 253                                 split->block_start = em->block_start + diff;
 254                                 split->orig_start = split->start;
 255                         }
 256
 257                         ret = add_extent_mapping(em_tree, split);
 258                         BUG_ON(ret);
 259                         free_extent_map(split);
 260                         split = NULL;
 261                 }
 262                 write_unlock(&em_tree->lock);
 263
 264                 /* once for us */
 265                 free_extent_map(em);
 266                 /* once for the tree*/
 267                 free_extent_map(em);
 268         }
 269         if (split)
 270                 free_extent_map(split);
 271         if (split2)
 272                 free_extent_map(split2);
 273         return 0;
 274 }
 275
 276 /*
 277  * this is very complex, but the basic idea is to drop all extents
 278  * in the range start - end.  hint_block is filled in with a block number
 279  * that would be a good hint to the block allocator for this file.
 280  *
 281  * If an extent intersects the range but is not entirely inside the range
 282  * it is either truncated or split.  Anything entirely inside the range
 283  * is deleted from the tree.
 284  *
 285  * inline_limit is used to tell this code which offsets in the file to keep
 286  * if they contain inline extents.
 287  */
 288 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 289                        struct btrfs_root *root, struct inode *inode,
 290                        u64 start, u64 end, u64 locked_end,
 291                        u64 inline_limit, u64 *hint_byte)
 292 {
 293         u64 extent_end = 0;
 294         u64 search_start = start;
 295         u64 ram_bytes = 0;
 296         u64 disk_bytenr = 0;
 297         u64 orig_locked_end = locked_end;
 298         u8 compression;
 299         u8 encryption;
 300         u16 other_encoding = 0;
 301         struct extent_buffer *leaf;
 302         struct btrfs_file_extent_item *extent;
 303         struct btrfs_path *path;
 304         struct btrfs_key key;
 305         struct btrfs_file_extent_item old;
 306         int keep;
 307         int slot;
 308         int bookend;
 309         int found_type = 0;
 310         int found_extent;
 311         int found_inline;
 312         int recow;
 313         int ret;
 314
 315         inline_limit = 0;
 316         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 317
 318         path = btrfs_alloc_path();
 319         if (!path)
 320                 return -ENOMEM;
 321         while (1) {
 322                 recow = 0;
 323                 btrfs_release_path(root, path);
 324                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 325                                                search_start, -1);
 326                 if (ret < 0)
 327                         goto out;
 328                 if (ret > 0) {
 329                         if (path->slots[0] == 0) {
 330                                 ret = 0;
 331                                 goto out;
 332                         }
 333                         path->slots[0]--;
 334                 }
 335 next_slot:
 336                 keep = 0;
 337                 bookend = 0;
 338                 found_extent = 0;
 339                 found_inline = 0;
 340                 compression = 0;
 341                 encryption = 0;
 342                 extent = NULL;
 343                 leaf = path->nodes[0];
 344                 slot = path->slots[0];
 345                 ret = 0;
 346                 btrfs_item_key_to_cpu(leaf, &key, slot);
 347                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 348                     key.offset >= end) {
 349                         goto out;
 350                 }
 351                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 352                     key.objectid != inode->i_ino) {
 353                         goto out;
 354                 }
 355                 if (recow) {
 356                         search_start = max(key.offset, start);
 357                         continue;
 358                 }
 359                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 360                         extent = btrfs_item_ptr(leaf, slot,
 361                                                 struct btrfs_file_extent_item);
 362                         found_type = btrfs_file_extent_type(leaf, extent);
 363                         compression = btrfs_file_extent_compression(leaf,
 364                                                                     extent);
 365                         encryption = btrfs_file_extent_encryption(leaf,
 366                                                                   extent);
 367                         other_encoding = btrfs_file_extent_other_encoding(leaf,
 368                                                                   extent);
 369                         if (found_type == BTRFS_FILE_EXTENT_REG ||
 370                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 371                                 extent_end =
 372                                      btrfs_file_extent_disk_bytenr(leaf,
 373                                                                    extent);
 374                                 if (extent_end)
 375                                         *hint_byte = extent_end;
 376
 377                                 extent_end = key.offset +
 378                                      btrfs_file_extent_num_bytes(leaf, extent);
 379                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
 380                                                                 extent);
 381                                 found_extent = 1;
 382                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 383                                 found_inline = 1;
 384                                 extent_end = key.offset +
 385                                      btrfs_file_extent_inline_len(leaf, extent);
 386                         }
 387                 } else {
 388                         extent_end = search_start;
 389                 }
 390
 391                 /* we found nothing we can drop */
 392                 if ((!found_extent && !found_inline) ||
 393                     search_start >= extent_end) {
 394                         int nextret;
 395                         u32 nritems;
 396                         nritems = btrfs_header_nritems(leaf);
 397                         if (slot >= nritems - 1) {
 398                                 nextret = btrfs_next_leaf(root, path);
 399                                 if (nextret)
 400                                         goto out;
 401                                 recow = 1;
 402                         } else {
 403                                 path->slots[0]++;
 404                         }
 405                         goto next_slot;
 406                 }
 407
 408                 if (end <= extent_end && start >= key.offset && found_inline)
 409                         *hint_byte = EXTENT_MAP_INLINE;
 410
 411                 if (found_extent) {
 412                         read_extent_buffer(leaf, &old, (unsigned long)extent,
 413                                            sizeof(old));
 414                 }
 415
 416                 if (end < extent_end && end >= key.offset) {
 417                         bookend = 1;
 418                         if (found_inline && start <= key.offset)
 419                                 keep = 1;
 420                 }
 421
 422                 if (bookend && found_extent) {
 423                         if (locked_end < extent_end) {
 424                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 425                                                 locked_end, extent_end - 1,
 426                                                 GFP_NOFS);
 427                                 if (!ret) {
 428                                         btrfs_release_path(root, path);
 429                                         lock_extent(&BTRFS_I(inode)->io_tree,
 430                                                 locked_end, extent_end - 1,
 431                                                 GFP_NOFS);
 432                                         locked_end = extent_end;
 433                                         continue;
 434                                 }
 435                                 locked_end = extent_end;
 436                         }
 437                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
 438                         if (disk_bytenr != 0) {
 439                                 ret = btrfs_inc_extent_ref(trans, root,
 440                                            disk_bytenr,
 441                                            le64_to_cpu(old.disk_num_bytes), 0,
 442                                            root->root_key.objectid,
 443                                            key.objectid, key.offset -
 444                                            le64_to_cpu(old.offset));
 445                                 BUG_ON(ret);
 446                         }
 447                 }
 448
 449                 if (found_inline) {
 450                         u64 mask = root->sectorsize - 1;
 451                         search_start = (extent_end + mask) & ~mask;
 452                 } else
 453                         search_start = extent_end;
 454
 455                 /* truncate existing extent */
 456                 if (start > key.offset) {
 457                         u64 new_num;
 458                         u64 old_num;
 459                         keep = 1;
 460                         WARN_ON(start & (root->sectorsize - 1));
 461                         if (found_extent) {
 462                                 new_num = start - key.offset;
 463                                 old_num = btrfs_file_extent_num_bytes(leaf,
 464                                                                       extent);
 465                                 *hint_byte =
 466                                         btrfs_file_extent_disk_bytenr(leaf,
 467                                                                       extent);
 468                                 if (btrfs_file_extent_disk_bytenr(leaf,
 469                                                                   extent)) {
 470                                         inode_sub_bytes(inode, old_num -
 471                                                         new_num);
 472                                 }
 473                                 btrfs_set_file_extent_num_bytes(leaf,
 474                                                         extent, new_num);
 475                                 btrfs_mark_buffer_dirty(leaf);
 476                         } else if (key.offset < inline_limit &&
 477                                    (end > extent_end) &&
 478                                    (inline_limit < extent_end)) {
 479                                 u32 new_size;
 480                                 new_size = btrfs_file_extent_calc_inline_size(
 481                                                    inline_limit - key.offset);
 482                                 inode_sub_bytes(inode, extent_end -
 483                                                 inline_limit);
 484                                 btrfs_set_file_extent_ram_bytes(leaf, extent,
 485                                                         new_size);
 486                                 if (!compression && !encryption) {
 487                                         btrfs_truncate_item(trans, root, path,
 488                                                             new_size, 1);
 489                                 }
 490                         }
 491                 }
 492                 /* delete the entire extent */
 493                 if (!keep) {
 494                         if (found_inline)
 495                                 inode_sub_bytes(inode, extent_end -
 496                                                 key.offset);
 497                         ret = btrfs_del_item(trans, root, path);
 498                         /* TODO update progress marker and return */
 499                         BUG_ON(ret);
 500                         extent = NULL;
 501                         btrfs_release_path(root, path);
 502                         /* the extent will be freed later */
 503                 }
 504                 if (bookend && found_inline && start <= key.offset) {
 505                         u32 new_size;
 506                         new_size = btrfs_file_extent_calc_inline_size(
 507                                                    extent_end - end);
 508                         inode_sub_bytes(inode, end - key.offset);
 509                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 510                                                         new_size);
 511                         if (!compression && !encryption)
 512                                 ret = btrfs_truncate_item(trans, root, path,
 513                                                           new_size, 0);
 514                         BUG_ON(ret);
 515                 }
 516                 /* create bookend, splitting the extent in two */
 517                 if (bookend && found_extent) {
 518                         struct btrfs_key ins;
 519                         ins.objectid = inode->i_ino;
 520                         ins.offset = end;
 521                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 522
 523                         btrfs_release_path(root, path);
 524                         path->leave_spinning = 1;
 525                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 526                                                       sizeof(*extent));
 527                         BUG_ON(ret);
 528
 529                         leaf = path->nodes[0];
 530                         extent = btrfs_item_ptr(leaf, path->slots[0],
 531                                                 struct btrfs_file_extent_item);
 532                         write_extent_buffer(leaf, &old,
 533                                             (unsigned long)extent, sizeof(old));
 534
 535                         btrfs_set_file_extent_compression(leaf, extent,
 536                                                           compression);
 537                         btrfs_set_file_extent_encryption(leaf, extent,
 538                                                          encryption);
 539                         btrfs_set_file_extent_other_encoding(leaf, extent,
 540                                                              other_encoding);
 541                         btrfs_set_file_extent_offset(leaf, extent,
 542                                     le64_to_cpu(old.offset) + end - key.offset);
 543                         WARN_ON(le64_to_cpu(old.num_bytes) <
 544                                 (extent_end - end));
 545                         btrfs_set_file_extent_num_bytes(leaf, extent,
 546                                                         extent_end - end);
 547
 548                         /*
 549                          * set the ram bytes to the size of the full extent
 550                          * before splitting.  This is a worst case flag,
 551                          * but its the best we can do because we don't know
 552                          * how splitting affects compression
 553                          */
 554                         btrfs_set_file_extent_ram_bytes(leaf, extent,
 555                                                         ram_bytes);
 556                         btrfs_set_file_extent_type(leaf, extent, found_type);
 557
 558                         btrfs_unlock_up_safe(path, 1);
 559                         btrfs_mark_buffer_dirty(path->nodes[0]);
 560                         btrfs_set_lock_blocking(path->nodes[0]);
 561
 562                         path->leave_spinning = 0;
 563                         btrfs_release_path(root, path);
 564                         if (disk_bytenr != 0)
 565                                 inode_add_bytes(inode, extent_end - end);
 566                 }
 567
 568                 if (found_extent && !keep) {
 569                         u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
 570
 571                         if (old_disk_bytenr != 0) {
 572                                 inode_sub_bytes(inode,
 573                                                 le64_to_cpu(old.num_bytes));
 574                                 ret = btrfs_free_extent(trans, root,
 575                                                 old_disk_bytenr,
 576                                                 le64_to_cpu(old.disk_num_bytes),
 577                                                 0, root->root_key.objectid,
 578                                                 key.objectid, key.offset -
 579                                                 le64_to_cpu(old.offset));
 580                                 BUG_ON(ret);
 581                                 *hint_byte = old_disk_bytenr;
 582                         }
 583                 }
 584
 585                 if (search_start >= end) {
 586                         ret = 0;
 587                         goto out;
 588                 }
 589         }
 590 out:
 591         btrfs_free_path(path);
 592         if (locked_end > orig_locked_end) {
 593                 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
 594                               locked_end - 1, GFP_NOFS);
 595         }
 596         return ret;
 597 }
 598
 599 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 600                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
 601 {
 602         struct btrfs_file_extent_item *fi;
 603         struct btrfs_key key;
 604         u64 extent_end;
 605
 606         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 607                 return 0;
 608
 609         btrfs_item_key_to_cpu(leaf, &key, slot);
 610         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 611                 return 0;
 612
 613         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 614         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 615             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 616             btrfs_file_extent_compression(leaf, fi) ||
 617             btrfs_file_extent_encryption(leaf, fi) ||
 618             btrfs_file_extent_other_encoding(leaf, fi))
 619                 return 0;
 620
 621         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 622         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 623                 return 0;
 624
 625         *start = key.offset;
 626         *end = extent_end;
 627         return 1;
 628 }
 629
 630 /*
 631  * Mark extent in the range start - end as written.
 632  *
 633  * This changes extent type from 'pre-allocated' to 'regular'. If only
 634  * part of extent is marked as written, the extent will be split into
 635  * two or three.
 636  */
 637 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 638                               struct btrfs_root *root,
 639                               struct inode *inode, u64 start, u64 end)
 640 {
 641         struct extent_buffer *leaf;
 642         struct btrfs_path *path;
 643         struct btrfs_file_extent_item *fi;
 644         struct btrfs_key key;
 645         u64 bytenr;
 646         u64 num_bytes;
 647         u64 extent_end;
 648         u64 orig_offset;
 649         u64 other_start;
 650         u64 other_end;
 651         u64 split = start;
 652         u64 locked_end = end;
 653         int extent_type;
 654         int split_end = 1;
 655         int ret;
 656
 657         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 658
 659         path = btrfs_alloc_path();
 660         BUG_ON(!path);
 661 again:
 662         key.objectid = inode->i_ino;
 663         key.type = BTRFS_EXTENT_DATA_KEY;
 664         if (split == start)
 665                 key.offset = split;
 666         else
 667                 key.offset = split - 1;
 668
 669         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 670         if (ret > 0 && path->slots[0] > 0)
 671                 path->slots[0]--;
 672
 673         leaf = path->nodes[0];
 674         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 675         BUG_ON(key.objectid != inode->i_ino ||
 676                key.type != BTRFS_EXTENT_DATA_KEY);
 677         fi = btrfs_item_ptr(leaf, path->slots[0],
 678                             struct btrfs_file_extent_item);
 679         extent_type = btrfs_file_extent_type(leaf, fi);
 680         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
 681         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 682         BUG_ON(key.offset > start || extent_end < end);
 683
 684         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 685         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 686         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 687
 688         if (key.offset == start)
 689                 split = end;
 690
 691         if (key.offset == start && extent_end == end) {
 692                 int del_nr = 0;
 693                 int del_slot = 0;
 694                 other_start = end;
 695                 other_end = 0;
 696                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 697                                      bytenr, &other_start, &other_end)) {
 698                         extent_end = other_end;
 699                         del_slot = path->slots[0] + 1;
 700                         del_nr++;
 701                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 702                                                 0, root->root_key.objectid,
 703                                                 inode->i_ino, orig_offset);
 704                         BUG_ON(ret);
 705                 }
 706                 other_start = 0;
 707                 other_end = start;
 708                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
 709                                      bytenr, &other_start, &other_end)) {
 710                         key.offset = other_start;
 711                         del_slot = path->slots[0];
 712                         del_nr++;
 713                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 714                                                 0, root->root_key.objectid,
 715                                                 inode->i_ino, orig_offset);
 716                         BUG_ON(ret);
 717                 }
 718                 split_end = 0;
 719                 if (del_nr == 0) {
 720                         btrfs_set_file_extent_type(leaf, fi,
 721                                                    BTRFS_FILE_EXTENT_REG);
 722                         goto done;
 723                 }
 724
 725                 fi = btrfs_item_ptr(leaf, del_slot - 1,
 726                                     struct btrfs_file_extent_item);
 727                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 728                 btrfs_set_file_extent_num_bytes(leaf, fi,
 729                                                 extent_end - key.offset);
 730                 btrfs_mark_buffer_dirty(leaf);
 731
 732                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 733                 BUG_ON(ret);
 734                 goto release;
 735         } else if (split == start) {
 736                 if (locked_end < extent_end) {
 737                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
 738                                         locked_end, extent_end - 1, GFP_NOFS);
 739                         if (!ret) {
 740                                 btrfs_release_path(root, path);
 741                                 lock_extent(&BTRFS_I(inode)->io_tree,
 742                                         locked_end, extent_end - 1, GFP_NOFS);
 743                                 locked_end = extent_end;
 744                                 goto again;
 745                         }
 746                         locked_end = extent_end;
 747                 }
 748                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
 749         } else  {
 750                 BUG_ON(key.offset != start);
 751                 key.offset = split;
 752                 btrfs_set_file_extent_offset(leaf, fi, key.offset -
 753                                              orig_offset);
 754                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
 755                 btrfs_set_item_key_safe(trans, root, path, &key);
 756                 extent_end = split;
 757         }
 758
 759         if (extent_end == end) {
 760                 split_end = 0;
 761                 extent_type = BTRFS_FILE_EXTENT_REG;
 762         }
 763         if (extent_end == end && split == start) {
 764                 other_start = end;
 765                 other_end = 0;
 766                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 767                                      bytenr, &other_start, &other_end)) {
 768                         path->slots[0]++;
 769                         fi = btrfs_item_ptr(leaf, path->slots[0],
 770                                             struct btrfs_file_extent_item);
 771                         key.offset = split;
 772                         btrfs_set_item_key_safe(trans, root, path, &key);
 773                         btrfs_set_file_extent_offset(leaf, fi, key.offset -
 774                                                      orig_offset);
 775                         btrfs_set_file_extent_num_bytes(leaf, fi,
 776                                                         other_end - split);
 777                         goto done;
 778                 }
 779         }
 780         if (extent_end == end && split == end) {
 781                 other_start = 0;
 782                 other_end = start;
 783                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
 784                                      bytenr, &other_start, &other_end)) {
 785                         path->slots[0]--;
 786                         fi = btrfs_item_ptr(leaf, path->slots[0],
 787                                             struct btrfs_file_extent_item);
 788                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
 789                                                         other_start);
 790                         goto done;
 791                 }
 792         }
 793
 794         btrfs_mark_buffer_dirty(leaf);
 795
 796         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 797                                    root->root_key.objectid,
 798                                    inode->i_ino, orig_offset);
 799         BUG_ON(ret);
 800         btrfs_release_path(root, path);
 801
 802         key.offset = start;
 803         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
 804         BUG_ON(ret);
 805
 806         leaf = path->nodes[0];
 807         fi = btrfs_item_ptr(leaf, path->slots[0],
 808                             struct btrfs_file_extent_item);
 809         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 810         btrfs_set_file_extent_type(leaf, fi, extent_type);
 811         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
 812         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
 813         btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
 814         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
 815         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 816         btrfs_set_file_extent_compression(leaf, fi, 0);
 817         btrfs_set_file_extent_encryption(leaf, fi, 0);
 818         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 819 done:
 820         btrfs_mark_buffer_dirty(leaf);
 821
 822 release:
 823         btrfs_release_path(root, path);
 824         if (split_end && split == start) {
 825                 split = end;
 826                 goto again;
 827         }
 828         if (locked_end > end) {
 829                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
 830                               GFP_NOFS);
 831         }
 832         btrfs_free_path(path);
 833         return 0;
 834 }
 835
 836 /*
 837  * this gets pages into the page cache and locks them down, it also properly
 838  * waits for data=ordered extents to finish before allowing the pages to be
 839  * modified.
 840  */
 841 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 842                          struct page **pages, size_t num_pages,
 843                          loff_t pos, unsigned long first_index,
 844                          unsigned long last_index, size_t write_bytes)
 845 {
 846         int i;
 847         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 848         struct inode *inode = fdentry(file)->d_inode;
 849         int err = 0;
 850         u64 start_pos;
 851         u64 last_pos;
 852
 853         start_pos = pos & ~((u64)root->sectorsize - 1);
 854         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 855
 856         if (start_pos > inode->i_size) {
 857                 err = btrfs_cont_expand(inode, start_pos);
 858                 if (err)
 859                         return err;
 860         }
 861
 862         memset(pages, 0, num_pages * sizeof(struct page *));
 863 again:
 864         for (i = 0; i < num_pages; i++) {
 865                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 866                 if (!pages[i]) {
 867                         err = -ENOMEM;
 868                         BUG_ON(1);
 869                 }
 870                 wait_on_page_writeback(pages[i]);
 871         }
 872         if (start_pos < inode->i_size) {
 873                 struct btrfs_ordered_extent *ordered;
 874                 lock_extent(&BTRFS_I(inode)->io_tree,
 875                             start_pos, last_pos - 1, GFP_NOFS);
 876                 ordered = btrfs_lookup_first_ordered_extent(inode,
 877                                                             last_pos - 1);
 878                 if (ordered &&
 879                     ordered->file_offset + ordered->len > start_pos &&
 880                     ordered->file_offset < last_pos) {
 881                         btrfs_put_ordered_extent(ordered);
 882                         unlock_extent(&BTRFS_I(inode)->io_tree,
 883                                       start_pos, last_pos - 1, GFP_NOFS);
 884                         for (i = 0; i < num_pages; i++) {
 885                                 unlock_page(pages[i]);
 886                                 page_cache_release(pages[i]);
 887                         }
 888                         btrfs_wait_ordered_range(inode, start_pos,
 889                                                  last_pos - start_pos);
 890                         goto again;
 891                 }
 892                 if (ordered)
 893                         btrfs_put_ordered_extent(ordered);
 894
 895                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 896                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 897                                   GFP_NOFS);
 898                 unlock_extent(&BTRFS_I(inode)->io_tree,
 899                               start_pos, last_pos - 1, GFP_NOFS);
 900         }
 901         for (i = 0; i < num_pages; i++) {
 902                 clear_page_dirty_for_io(pages[i]);
 903                 set_page_extent_mapped(pages[i]);
 904                 WARN_ON(!PageLocked(pages[i]));
 905         }
 906         return 0;
 907 }
 908
 909 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 910                                 size_t count, loff_t *ppos)
 911 {
 912         loff_t pos;
 913         loff_t start_pos;
 914         ssize_t num_written = 0;
 915         ssize_t err = 0;
 916         int ret = 0;
 917         struct inode *inode = fdentry(file)->d_inode;
 918         struct btrfs_root *root = BTRFS_I(inode)->root;
 919         struct page **pages = NULL;
 920         int nrptrs;
 921         struct page *pinned[2];
 922         unsigned long first_index;
 923         unsigned long last_index;
 924         int will_write;
 925
 926         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 927                       (file->f_flags & O_DIRECT));
 928
 929         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 930                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 931         pinned[0] = NULL;
 932         pinned[1] = NULL;
 933
 934         pos = *ppos;
 935         start_pos = pos;
 936
 937         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 938         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 939         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 940         if (err)
 941                 goto out_nolock;
 942         if (count == 0)
 943                 goto out_nolock;
 944
 945         err = file_remove_suid(file);
 946         if (err)
 947                 goto out_nolock;
 948         file_update_time(file);
 949
 950         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 951
 952         mutex_lock(&inode->i_mutex);
 953         BTRFS_I(inode)->sequence++;
 954         first_index = pos >> PAGE_CACHE_SHIFT;
 955         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 956
 957         /*
 958          * there are lots of better ways to do this, but this code
 959          * makes sure the first and last page in the file range are
 960          * up to date and ready for cow
 961          */
 962         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 963                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 964                 if (!PageUptodate(pinned[0])) {
 965                         ret = btrfs_readpage(NULL, pinned[0]);
 966                         BUG_ON(ret);
 967                         wait_on_page_locked(pinned[0]);
 968                 } else {
 969                         unlock_page(pinned[0]);
 970                 }
 971         }
 972         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 973                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 974                 if (!PageUptodate(pinned[1])) {
 975                         ret = btrfs_readpage(NULL, pinned[1]);
 976                         BUG_ON(ret);
 977                         wait_on_page_locked(pinned[1]);
 978                 } else {
 979                         unlock_page(pinned[1]);
 980                 }
 981         }
 982
 983         while (count > 0) {
 984                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 985                 size_t write_bytes = min(count, nrptrs *
 986                                         (size_t)PAGE_CACHE_SIZE -
 987                                          offset);
 988                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 989                                         PAGE_CACHE_SHIFT;
 990
 991                 WARN_ON(num_pages > nrptrs);
 992                 memset(pages, 0, sizeof(struct page *) * nrptrs);
 993
 994                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
 995                 if (ret)
 996                         goto out;
 997
 998                 ret = prepare_pages(root, file, pages, num_pages,
 999                                     pos, first_index, last_index,
1000                                     write_bytes);
1001                 if (ret) {
1002                         btrfs_free_reserved_data_space(root, inode,
1003                                                        write_bytes);
1004                         goto out;
1005                 }
1006
1007                 ret = btrfs_copy_from_user(pos, num_pages,
1008                                            write_bytes, pages, buf);
1009                 if (ret) {
1010                         btrfs_free_reserved_data_space(root, inode,
1011                                                        write_bytes);
1012                         btrfs_drop_pages(pages, num_pages);
1013                         goto out;
1014                 }
1015
1016                 ret = dirty_and_release_pages(NULL, root, file, pages,
1017                                               num_pages, pos, write_bytes);
1018                 btrfs_drop_pages(pages, num_pages);
1019                 if (ret) {
1020                         btrfs_free_reserved_data_space(root, inode,
1021                                                        write_bytes);
1022                         goto out;
1023                 }
1024
1025                 if (will_write) {
1026                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1027                                                pos + write_bytes - 1,
1028                                                WB_SYNC_ALL);
1029                 } else {
1030                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1031                                                            num_pages);
1032                         if (num_pages <
1033                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1034                                 btrfs_btree_balance_dirty(root, 1);
1035                         btrfs_throttle(root);
1036                 }
1037
1038                 buf += write_bytes;
1039                 count -= write_bytes;
1040                 pos += write_bytes;
1041                 num_written += write_bytes;
1042
1043                 cond_resched();
1044         }
1045 out:
1046         mutex_unlock(&inode->i_mutex);
1047         if (ret)
1048                 err = ret;
1049
1050 out_nolock:
1051         kfree(pages);
1052         if (pinned[0])
1053                 page_cache_release(pinned[0]);
1054         if (pinned[1])
1055                 page_cache_release(pinned[1]);
1056         *ppos = pos;
1057
1058         /*
1059          * we want to make sure fsync finds this change
1060          * but we haven't joined a transaction running right now.
1061          *
1062          * Later on, someone is sure to update the inode and get the
1063          * real transid recorded.
1064          *
1065          * We set last_trans now to the fs_info generation + 1,
1066          * this will either be one more than the running transaction
1067          * or the generation used for the next transaction if there isn't
1068          * one running right now.
1069          */
1070         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1071
1072         if (num_written > 0 && will_write) {
1073                 struct btrfs_trans_handle *trans;
1074
1075                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1076                 if (err)
1077                         num_written = err;
1078
1079                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1080                         trans = btrfs_start_transaction(root, 1);
1081                         ret = btrfs_log_dentry_safe(trans, root,
1082                                                     file->f_dentry);
1083                         if (ret == 0) {
1084                                 ret = btrfs_sync_log(trans, root);
1085                                 if (ret == 0)
1086                                         btrfs_end_transaction(trans, root);
1087                                 else
1088                                         btrfs_commit_transaction(trans, root);
1089                         } else {
1090                                 btrfs_commit_transaction(trans, root);
1091                         }
1092                 }
1093                 if (file->f_flags & O_DIRECT) {
1094                         invalidate_mapping_pages(inode->i_mapping,
1095                               start_pos >> PAGE_CACHE_SHIFT,
1096                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1097                 }
1098         }
1099         current->backing_dev_info = NULL;
1100         return num_written ? num_written : err;
1101 }
1102
1103 int btrfs_release_file(struct inode *inode, struct file *filp)
1104 {
1105         /*
1106          * ordered_data_close is set by settattr when we are about to truncate
1107          * a file from a non-zero size to a zero size.  This tries to
1108          * flush down new bytes that may have been written if the
1109          * application were using truncate to replace a file in place.
1110          */
1111         if (BTRFS_I(inode)->ordered_data_close) {
1112                 BTRFS_I(inode)->ordered_data_close = 0;
1113                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1114                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1115                         filemap_flush(inode->i_mapping);
1116         }
1117         if (filp->private_data)
1118                 btrfs_ioctl_trans_end(filp);
1119         return 0;
1120 }
1121
1122 /*
1123  * fsync call for both files and directories.  This logs the inode into
1124  * the tree log instead of forcing full commits whenever possible.
1125  *
1126  * It needs to call filemap_fdatawait so that all ordered extent updates are
1127  * in the metadata btree are up to date for copying to the log.
1128  *
1129  * It drops the inode mutex before doing the tree log commit.  This is an
1130  * important optimization for directories because holding the mutex prevents
1131  * new operations on the dir while we write to disk.
1132  */
1133 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1134 {
1135         struct inode *inode = dentry->d_inode;
1136         struct btrfs_root *root = BTRFS_I(inode)->root;
1137         int ret = 0;
1138         struct btrfs_trans_handle *trans;
1139
1140         /*
1141          * check the transaction that last modified this inode
1142          * and see if its already been committed
1143          */
1144         if (!BTRFS_I(inode)->last_trans)
1145                 goto out;
1146
1147         mutex_lock(&root->fs_info->trans_mutex);
1148         if (BTRFS_I(inode)->last_trans <=
1149             root->fs_info->last_trans_committed) {
1150                 BTRFS_I(inode)->last_trans = 0;
1151                 mutex_unlock(&root->fs_info->trans_mutex);
1152                 goto out;
1153         }
1154         mutex_unlock(&root->fs_info->trans_mutex);
1155
1156         root->log_batch++;
1157         filemap_fdatawrite(inode->i_mapping);
1158         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1159         root->log_batch++;
1160
1161         if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1162                 goto out;
1163         /*
1164          * ok we haven't committed the transaction yet, lets do a commit
1165          */
1166         if (file && file->private_data)
1167                 btrfs_ioctl_trans_end(file);
1168
1169         trans = btrfs_start_transaction(root, 1);
1170         if (!trans) {
1171                 ret = -ENOMEM;
1172                 goto out;
1173         }
1174
1175         ret = btrfs_log_dentry_safe(trans, root, dentry);
1176         if (ret < 0)
1177                 goto out;
1178
1179         /* we've logged all the items and now have a consistent
1180          * version of the file in the log.  It is possible that
1181          * someone will come in and modify the file, but that's
1182          * fine because the log is consistent on disk, and we
1183          * have references to all of the file's extents
1184          *
1185          * It is possible that someone will come in and log the
1186          * file again, but that will end up using the synchronization
1187          * inside btrfs_sync_log to keep things safe.
1188          */
1189         mutex_unlock(&dentry->d_inode->i_mutex);
1190
1191         if (ret > 0) {
1192                 ret = btrfs_commit_transaction(trans, root);
1193         } else {
1194                 ret = btrfs_sync_log(trans, root);
1195                 if (ret == 0)
1196                         ret = btrfs_end_transaction(trans, root);
1197                 else
1198                         ret = btrfs_commit_transaction(trans, root);
1199         }
1200         mutex_lock(&dentry->d_inode->i_mutex);
1201 out:
1202         return ret > 0 ? EIO : ret;
1203 }
1204
1205 static struct vm_operations_struct btrfs_file_vm_ops = {
1206         .fault          = filemap_fault,
1207         .page_mkwrite   = btrfs_page_mkwrite,
1208 };
1209
1210 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1211 {
1212         vma->vm_ops = &btrfs_file_vm_ops;
1213         file_accessed(filp);
1214         return 0;
1215 }
1216
1217 struct file_operations btrfs_file_operations = {
1218         .llseek         = generic_file_llseek,
1219         .read           = do_sync_read,
1220         .aio_read       = generic_file_aio_read,
1221         .splice_read    = generic_file_splice_read,
1222         .write          = btrfs_file_write,
1223         .mmap           = btrfs_file_mmap,
1224         .open           = generic_file_open,
1225         .release        = btrfs_release_file,
1226         .fsync          = btrfs_sync_file,
1227         .unlocked_ioctl = btrfs_ioctl,
1228 #ifdef CONFIG_COMPAT
1229         .compat_ioctl   = btrfs_ioctl,
1230 #endif
1231 };