| blob | 72e73461c0643bbf128504c93a9c47dd377784aa |
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 */
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 <linux/btrfs.h>
34 #include <linux/uio.h>
46 /*
47 * when auto defrag is enabled we
48 * queue up these defrag structs to remember which
49 * inodes need defragging passes
50 */
53 /* objectid */
54 u64 ino;
55 /*
56 * transid where the defrag was added, we search for
57 * extents newer than this
58 */
59 u64 transid;
61 /* root objectid */
62 u64 root;
64 /* last offset we were able to defrag */
65 u64 last_offset;
67 /* if we've wrapped around back to zero once already */
69 };
73 {
82 else
84 }
86 /* pop a record for an inode into the defrag tree. The lock
87 * must be held already
88 *
89 * If you're inserting a record for an older transid than an
90 * existing record, the transid already in the tree is lowered
91 *
92 * If an existing record is found the defrag item you
93 * pass in is freed
94 */
97 {
115 /* if we're reinserting an entry for
116 * an old defrag run, make sure to
117 * lower the transid of our existing record
118 */
124 }
125 }
130 }
133 {
141 }
143 /*
144 * insert a defrag record for this inode if auto defrag is
145 * enabled
146 */
149 {
152 u64 transid;
163 else
176 /*
177 * If we set IN_DEFRAG flag and evict the inode from memory,
178 * and then re-read this inode, this new inode doesn't have
179 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 */
186 }
189 }
191 /*
192 * Requeue the defrag object. If there is a defrag object that points to
193 * the same inode in the tree, we will merge them together (by
194 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195 */
198 {
205 /*
206 * Here we don't check the IN_DEFRAG flag, because we need merge
207 * them together.
208 */
215 out:
217 }
219 /*
220 * pick the defragable inode that we want, if it doesn't exist, we will get
221 * the next one.
222 */
225 {
246 else
248 }
254 else
256 }
257 out:
262 }
265 {
279 }
281 }
283 #define BTRFS_DEFRAG_BATCH 1024
287 {
296 /* get the inode */
307 }
316 }
319 /* do a chunk of defrag */
327 BTRFS_DEFRAG_BATCH);
329 /*
330 * if we filled the whole defrag batch, there
331 * must be more work to do. Queue this defrag
332 * again
333 */
338 /*
339 * we didn't fill our defrag batch, but
340 * we didn't start at zero. Make sure we loop
341 * around to the start of the file.
342 */
348 }
352 cleanup:
356 }
358 /*
359 * run through the list of inodes in the FS that need
360 * defragging
361 */
363 {
370 /* Pause the auto defragger. */
378 /* find an inode to defrag */
380 first_ino);
388 }
389 }
395 }
398 /*
399 * during unmount, we use the transaction_wait queue to
400 * wait for the defragger to stop
401 */
404 }
406 /* simple helper to fault in pages and copy. This should go away
407 * and be replaced with calls into generic code.
408 */
413 {
423 /*
424 * Copy data from userspace to the current page
425 */
428 /* Flush processor's dcache for this page */
431 /*
432 * if we get a partial write, we can end up with
433 * partially up to date pages. These add
434 * a lot of complexity, so make sure they don't
435 * happen by forcing this copy to be retried.
436 *
437 * The rest of the btrfs_file_write code will fall
438 * back to page at a time copies after we return 0.
439 */
447 /* Return to btrfs_file_write_iter to fault page */
454 pg++;
456 }
457 }
459 }
461 /*
462 * unlocks pages after btrfs_file_write is done with them
463 */
465 {
468 /* page checked is some magic around finding pages that
469 * have been modified without going through btrfs_set_page_dirty
470 * clear it here. There should be no need to mark the pages
471 * accessed as prepare_pages should have marked them accessed
472 * in prepare_pages via find_or_create_page()
473 */
477 }
478 }
480 /*
481 * after copy_from_user, pages need to be dirtied and we need to make
482 * sure holes are created between the current EOF and the start of
483 * any next extents (if required).
484 *
485 * this also makes the decision about creating an inline extent vs
486 * doing real data extents, marking pages dirty and delalloc as required.
487 */
492 {
495 u64 num_bytes;
496 u64 start_pos;
497 u64 end_of_last_block;
506 cached);
515 }
517 /*
518 * we've only changed i_size in ram, and we haven't updated
519 * the disk i_size. There is no need to log the inode
520 * at this time.
521 */
525 }
527 /*
528 * this drops all the extents in the cache that intersect the range
529 * [start, end]. Existing extents are split as required.
530 */
533 {
539 u64 gen;
550 }
567 }
575 }
582 }
600 else
611 }
621 }
646 }
653 }
657 modified);
660 modified);
662 }
665 }
666 next:
671 /* once for us */
673 /* once for the tree*/
675 }
680 }
682 /*
683 * this is very complex, but the basic idea is to drop all extents
684 * in the range start - end. hint_block is filled in with a block number
685 * that would be a good hint to the block allocator for this file.
686 *
687 * If an extent intersects the range but is not entirely inside the range
688 * it is either truncated or split. Anything entirely inside the range
689 * is deleted from the tree.
690 */
696 u32 extent_item_size,
698 {
739 }
741 leafs_visited++;
742 next_slot:
752 }
753 leafs_visited++;
756 }
767 }
787 /* can't happen */
789 }
791 /*
792 * Don't skip extent items representing 0 byte lengths. They
793 * used to be created (bug) if while punching holes we hit
794 * -ENOSPC condition. So if we find one here, just ensure we
795 * delete it, otherwise we would insert a new file extent item
796 * with the same key (offset) as that 0 bytes length file
797 * extent item in the call to setup_items_for_insert() later
798 * in this function.
799 */
806 }
814 }
816 /*
817 * | - range to drop - |
818 * | -------- extent -------- |
819 */
825 }
834 }
860 }
862 }
863 /*
864 * | ---- range to drop ----- |
865 * | -------- extent -------- |
866 */
871 }
885 }
888 /*
889 * | ---- range to drop ----- |
890 * | -------- extent -------- |
891 */
897 }
909 }
911 /*
912 * | ---- range to drop ----- |
913 * | ------ extent ------ |
914 */
916 delete_extent_item:
922 del_nr++;
923 }
936 extent_offset);
940 }
948 }
951 del_nr);
955 }
962 }
965 }
968 /*
969 * Set path->slots[0] to first slot, so that after the delete
970 * if items are move off from our leaf to its immediate left or
971 * right neighbor leafs, we end up with a correct and adjusted
972 * path->slots[0] for our insertion (if replace_extent != 0).
973 */
978 }
981 /*
982 * If btrfs_del_items() was called, it might have deleted a leaf, in
983 * which case it unlocked our path, so check path->locks[0] matches a
984 * write lock.
985 */
1001 }
1004 extent_item_size,
1008 }
1015 }
1020 {
1031 }
1036 {
1039 u64 extent_end;
1064 }
1066 /*
1067 * Mark extent in the range start - end as written.
1068 *
1069 * This changes extent type from 'pre-allocated' to 'regular'. If only
1070 * part of extent is marked as written, the extent will be split into
1071 * two or three.
1072 */
1075 {
1082 u64 bytenr;
1083 u64 num_bytes;
1084 u64 extent_end;
1085 u64 orig_offset;
1086 u64 other_start;
1087 u64 other_end;
1088 u64 split;
1098 again:
1117 BTRFS_FILE_EXTENT_PREALLOC);
1150 }
1151 }
1179 }
1180 }
1191 }
1195 }
1224 }
1226 }
1236 }
1239 del_nr++;
1244 }
1253 }
1256 del_nr++;
1261 }
1266 BTRFS_FILE_EXTENT_REG);
1273 BTRFS_FILE_EXTENT_REG);
1283 }
1284 }
1285 out:
1288 }
1290 /*
1291 * on error we return an unlocked page and the error value
1292 * on success we return a locked page and 0
1293 */
1296 {
1308 }
1309 }
1311 }
1313 /*
1314 * this just gets pages into the page cache and locks them down.
1315 */
1319 {
1333 }
1337 force_uptodate);
1345 }
1347 }
1350 fail:
1354 faili--;
1355 }
1358 }
1360 /*
1361 * This function locks the extent and properly waits for data=ordered extents
1362 * to finish before allowing the pages to be modified if need.
1363 *
1364 * The return value:
1365 * 1 - the extent is locked
1366 * 0 - the extent is not locked, and everything is OK
1367 * -EAGAIN - need re-prepare the pages
1368 * the other < 0 number - Something wrong happens
1369 */
1375 {
1376 u64 start_pos;
1377 u64 last_pos;
1399 }
1403 }
1414 }
1421 }
1424 }
1428 {
1432 u64 num_bytes;
1448 }
1452 }
1462 }
1467 }
1471 loff_t pos)
1472 {
1478 u64 lockstart;
1479 u64 lockend;
1499 offset);
1501 PAGE_CACHE_SIZE);
1508 /*
1509 * Fault pages before locking them in prepare_pages
1510 * to avoid recursive lock
1511 */
1515 }
1520 BTRFS_INODE_PREALLOC)) {
1525 /*
1526 * For nodata cow case, no need to reserve
1527 * data space.
1528 */
1530 /*
1531 * our prealloc extent may be smaller than
1532 * write_bytes, so scale down.
1533 */
1535 PAGE_CACHE_SIZE);
1538 }
1539 }
1544 reserve_metadata:
1549 write_bytes);
1550 else
1553 }
1557 again:
1558 /*
1559 * This is going to setup the pages array with the number of
1560 * pages we want, so we don't really need to worry about the
1561 * contents of pages from loop to loop
1562 */
1565 force_page_uptodate);
1579 }
1584 /*
1585 * if we have trouble faulting in the pages, fall
1586 * back to one page at a time
1587 */
1597 PAGE_CACHE_SIZE);
1598 }
1600 /*
1601 * If we had a short copy we need to release the excess delaloc
1602 * bytes we reserved. We need to increment outstanding_extents
1603 * because btrfs_delalloc_release_space will decrement it, but
1604 * we still have an outstanding extent for the chunk we actually
1605 * managed to copy.
1606 */
1609 PAGE_CACHE_SHIFT;
1614 }
1617 release_bytes);
1619 u64 __pos;
1624 release_bytes);
1625 }
1626 }
1633 NULL);
1637 GFP_NOFS);
1641 }
1656 }
1668 }
1678 }
1679 }
1682 }
1686 loff_t pos)
1687 {
1690 ssize_t written;
1691 ssize_t written_buffered;
1692 loff_t endbyte;
1705 }
1706 /*
1707 * Ensure all data is persisted. We want the next direct IO read to be
1708 * able to read what was just written.
1709 */
1721 out:
1723 }
1726 {
1741 }
1745 {
1749 u64 start_pos;
1750 u64 end_pos;
1753 ssize_t err;
1754 loff_t pos;
1762 }
1769 }
1771 /*
1772 * If BTRFS flips readonly due to some impossible error
1773 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1774 * although we have opened a file as writable, we have
1775 * to stop this write operation to ensure FS consistency.
1776 */
1781 }
1783 /*
1784 * We reserve space for updating the inode when we reserve space for the
1785 * extent we are going to write, so we will enospc out there. We don't
1786 * need to start yet another transaction to update the inode as we will
1787 * update the inode when we finish writing whatever data we write.
1788 */
1795 /* Expand hole size to cover write data, preventing empty gap */
1801 }
1802 }
1813 }
1817 /*
1818 * We also have to set last_sub_trans to the current log transid,
1819 * otherwise subsequent syncs to a file that's been synced in this
1820 * transaction will appear to have already occured.
1821 */
1829 }
1833 out:
1836 }
1839 {
1842 /*
1843 * ordered_data_close is set by settattr when we are about to truncate
1844 * a file from a non-zero size to a zero size. This tries to
1845 * flush down new bytes that may have been written if the
1846 * application were using truncate to replace a file in place.
1847 */
1852 }
1855 {
1863 }
1865 /*
1866 * fsync call for both files and directories. This logs the inode into
1867 * the tree log instead of forcing full commits whenever possible.
1868 *
1869 * It needs to call filemap_fdatawait so that all ordered extent updates are
1870 * in the metadata btree are up to date for copying to the log.
1871 *
1872 * It drops the inode mutex before doing the tree log commit. This is an
1873 * important optimization for directories because holding the mutex prevents
1874 * new operations on the dir while we write to disk.
1875 */
1877 {
1885 u64 len;
1887 /*
1888 * The range length can be represented by u64, we have to do the typecasts
1889 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1890 */
1894 /*
1895 * We write the dirty pages in the range and wait until they complete
1896 * out of the ->i_mutex. If so, we can flush the dirty pages by
1897 * multi-task, and make the performance up. See
1898 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1899 */
1908 /*
1909 * We might have have had more pages made dirty after calling
1910 * start_ordered_ops and before acquiring the inode's i_mutex.
1911 */
1913 /*
1914 * For a full sync, we need to make sure any ordered operations
1915 * start and finish before we start logging the inode, so that
1916 * all extents are persisted and the respective file extent
1917 * items are in the fs/subvol btree.
1918 */
1921 /*
1922 * Start any new ordered operations before starting to log the
1923 * inode. We will wait for them to finish in btrfs_sync_log().
1924 *
1925 * Right before acquiring the inode's mutex, we might have new
1926 * writes dirtying pages, which won't immediately start the
1927 * respective ordered operations - that is done through the
1928 * fill_delalloc callbacks invoked from the writepage and
1929 * writepages address space operations. So make sure we start
1930 * all ordered operations before starting to log our inode. Not
1931 * doing this means that while logging the inode, writeback
1932 * could start and invoke writepage/writepages, which would call
1933 * the fill_delalloc callbacks (cow_file_range,
1934 * submit_compressed_extents). These callbacks add first an
1935 * extent map to the modified list of extents and then create
1936 * the respective ordered operation, which means in
1937 * tree-log.c:btrfs_log_inode() we might capture all existing
1938 * ordered operations (with btrfs_get_logged_extents()) before
1939 * the fill_delalloc callback adds its ordered operation, and by
1940 * the time we visit the modified list of extent maps (with
1941 * btrfs_log_changed_extents()), we see and process the extent
1942 * map they created. We then use the extent map to construct a
1943 * file extent item for logging without waiting for the
1944 * respective ordered operation to finish - this file extent
1945 * item points to a disk location that might not have yet been
1946 * written to, containing random data - so after a crash a log
1947 * replay will make our inode have file extent items that point
1948 * to disk locations containing invalid data, as we returned
1949 * success to userspace without waiting for the respective
1950 * ordered operation to finish, because it wasn't captured by
1951 * btrfs_get_logged_extents().
1952 */
1954 }
1958 }
1961 /*
1962 * If the last transaction that changed this file was before the current
1963 * transaction and we have the full sync flag set in our inode, we can
1964 * bail out now without any syncing.
1965 *
1966 * Note that we can't bail out if the full sync flag isn't set. This is
1967 * because when the full sync flag is set we start all ordered extents
1968 * and wait for them to fully complete - when they complete they update
1969 * the inode's last_trans field through:
1970 *
1971 * btrfs_finish_ordered_io() ->
1972 * btrfs_update_inode_fallback() ->
1973 * btrfs_update_inode() ->
1974 * btrfs_set_inode_last_trans()
1975 *
1976 * So we are sure that last_trans is up to date and can do this check to
1977 * bail out safely. For the fast path, when the full sync flag is not
1978 * set in our inode, we can not do it because we start only our ordered
1979 * extents and don't wait for them to complete (that is when
1980 * btrfs_finish_ordered_io runs), so here at this point their last_trans
1981 * value might be less than or equals to fs_info->last_trans_committed,
1982 * and setting a speculative last_trans for an inode when a buffered
1983 * write is made (such as fs_info->generation + 1 for example) would not
1984 * be reliable since after setting the value and before fsync is called
1985 * any number of transactions can start and commit (transaction kthread
1986 * commits the current transaction periodically), and a transaction
1987 * commit does not start nor waits for ordered extents to complete.
1988 */
1995 /*
1996 * We'v had everything committed since the last time we were
1997 * modified so clear this flag in case it was set for whatever
1998 * reason, it's no longer relevant.
1999 */
2004 }
2006 /*
2007 * ok we haven't committed the transaction yet, lets do a commit
2008 */
2012 /*
2013 * We use start here because we will need to wait on the IO to complete
2014 * in btrfs_sync_log, which could require joining a transaction (for
2015 * example checking cross references in the nocow path). If we use join
2016 * here we could get into a situation where we're waiting on IO to
2017 * happen that is blocked on a transaction trying to commit. With start
2018 * we inc the extwriter counter, so we wait for all extwriters to exit
2019 * before we start blocking join'ers. This comment is to keep somebody
2020 * from thinking they are super smart and changing this to
2021 * btrfs_join_transaction *cough*Josef*cough*.
2022 */
2028 }
2035 /* Fallthrough and commit/free transaction. */
2037 }
2039 /* we've logged all the items and now have a consistent
2040 * version of the file in the log. It is possible that
2041 * someone will come in and modify the file, but that's
2042 * fine because the log is consistent on disk, and we
2043 * have references to all of the file's extents
2044 *
2045 * It is possible that someone will come in and log the
2046 * file again, but that will end up using the synchronization
2047 * inside btrfs_sync_log to keep things safe.
2048 */
2051 /*
2052 * If any of the ordered extents had an error, just return it to user
2053 * space, so that the application knows some writes didn't succeed and
2054 * can take proper action (retry for e.g.). Blindly committing the
2055 * transaction in this case, would fool userspace that everything was
2056 * successful. And we also want to make sure our log doesn't contain
2057 * file extent items pointing to extents that weren't fully written to -
2058 * just like in the non fast fsync path, where we check for the ordered
2059 * operation's error flag before writing to the log tree and return -EIO
2060 * if any of them had this flag set (btrfs_wait_ordered_range) -
2061 * therefore we need to check for errors in the ordered operations,
2062 * which are indicated by ctx.io_err.
2063 */
2068 }
2076 }
2077 }
2083 }
2084 }
2088 }
2089 out:
2091 }
2097 };
2100 {
2110 }
2114 {
2139 }
2143 {
2166 u64 num_bytes;
2178 }
2181 u64 num_bytes;
2188 offset;
2194 }
2203 out:
2234 }
2237 }
2239 /*
2240 * Find a hole extent on given inode and change start/len to the end of hole
2241 * extent.(hole/vacuum extent whose em->start <= start &&
2242 * em->start + em->len > start)
2243 * When a hole extent is found, return 1 and modify start/len.
2244 */
2246 {
2254 else
2257 }
2259 /* Hole or vacuum extent(only exists in no-hole mode) */
2265 }
2268 }
2271 {
2277 u64 lockstart;
2278 u64 lockend;
2279 u64 tail_start;
2280 u64 tail_len;
2282 u64 cur_offset;
2284 u64 drop_end;
2290 u64 ino_size;
2304 /* Already in a large hole */
2307 }
2315 /*
2316 * We needn't truncate any page which is beyond the end of the file
2317 * because we are sure there is no data there.
2318 */
2319 /*
2320 * Only do this if we are in the same page and we aren't doing the
2321 * entire page.
2322 */
2329 }
2331 }
2333 /* zero back part of the first page */
2340 }
2341 }
2343 /* Check the aligned pages after the first unaligned page,
2344 * if offset != orig_start, which means the first unaligned page
2345 * including serveral following pages are already in holes,
2346 * the extra check can be skipped */
2348 /* after truncate page, check hole again */
2357 }
2359 }
2361 /* Check the tail unaligned part is in a hole */
2369 /* zero the front end of the last page */
2376 }
2377 }
2378 }
2383 }
2394 /*
2395 * We need to make sure we have no ordered extents in this range
2396 * and nobody raced in and read a page in this range, if we did
2397 * we need to try again.
2398 */
2406 }
2416 }
2417 }
2423 }
2429 }
2433 /*
2434 * 1 - update the inode
2435 * 1 - removing the extents in the range
2436 * 1 - adding the hole extent if no_holes isn't set
2437 */
2443 }
2446 min_size);
2463 drop_end);
2467 }
2468 }
2476 }
2486 }
2499 }
2500 }
2505 }
2508 /*
2509 * If we are using the NO_HOLES feature we might have had already an
2510 * hole that overlaps a part of the region [lockstart, lockend] and
2511 * ends at (or beyond) lockend. Since we have no file extent items to
2512 * represent holes, drop_end can be less than lockend and so we must
2513 * make sure we have an extent map representing the existing hole (the
2514 * call to __btrfs_drop_extents() might have dropped the existing extent
2515 * map representing the existing hole), otherwise the fast fsync path
2516 * will not record the existence of the hole region
2517 * [existing_hole_start, lockend].
2518 */
2521 /*
2522 * Don't insert file hole extent item if it's for a range beyond eof
2523 * (because it's useless) or if it represents a 0 bytes range (when
2524 * cur_offset == drop_end).
2525 */
2531 }
2532 }
2534 out_trans:
2546 out_free:
2549 out:
2552 out_only_mutex:
2554 /*
2555 * If we only end up zeroing part of a page, we still need to
2556 * update the inode item, so that all the time fields are
2557 * updated as well as the necessary btrfs inode in memory fields
2558 * for detecting, at fsync time, if the inode isn't yet in the
2559 * log tree or it's there but not up to date.
2560 */
2567 }
2568 }
2573 }
2575 /* Helper structure to record which range is already reserved */
2578 u64 start;
2579 u64 len;
2580 };
2582 /*
2583 * Helper function to add falloc range
2584 *
2585 * Caller should have locked the larger range of extent containing
2586 * [start, len)
2587 */
2589 {
2596 /*
2597 * As fallocate iterate by bytenr order, we only need to check
2598 * the last range.
2599 */
2604 }
2605 insert:
2613 }
2617 {
2623 u64 cur_offset;
2624 u64 last_byte;
2625 u64 alloc_start;
2626 u64 alloc_end;
2628 u64 locked_end;
2637 /* Make sure we aren't being give some crap mode */
2644 /*
2645 * Only trigger disk allocation, don't trigger qgroup reserve
2646 *
2647 * For qgroup space, it will be checked later.
2648 */
2658 /*
2659 * TODO: Move these two operations after we have checked
2660 * accurate reserved space, or fallocate can still fail but
2661 * with page truncated or size expanded.
2662 *
2663 * But that's a minor problem and won't do much harm BTW.
2664 */
2667 alloc_start);
2671 /*
2672 * If we are fallocating from the end of the file onward we
2673 * need to zero out the end of the page if i_size lands in the
2674 * middle of a page.
2675 */
2679 }
2681 /*
2682 * wait for ordered IO before we have any locks. We'll loop again
2683 * below with the locks held.
2684 */
2694 /* the extent lock is ordered inside the running
2695 * transaction
2696 */
2708 /*
2709 * we can't wait on the range with the transaction
2710 * running or with the extent lock held
2711 */
2720 }
2721 }
2723 /* First, check if we exceed the qgroup limit */
2732 else
2735 }
2747 }
2752 }
2757 }
2759 /*
2760 * If ret is still 0, means we're OK to fallocate.
2761 * Or just cleanup the list and exit.
2762 */
2771 }
2780 /*
2781 * We didn't need to allocate any more space, but we
2782 * still extended the size of the file so we need to
2783 * update i_size and the inode item.
2784 */
2795 else
2797 }
2798 }
2799 out_unlock:
2802 out:
2803 /*
2804 * As we waited the extent range, the data_rsv_map must be empty
2805 * in the range, as written data range will be released from it.
2806 * And for prealloacted extent, it will also be released when
2807 * its metadata is written.
2808 * So this is completely used as cleanup.
2809 */
2812 /* Let go of our reservation. */
2816 }
2819 {
2823 u64 lockstart;
2824 u64 lockend;
2825 u64 start;
2826 u64 len;
2832 /*
2833 * *offset can be negative, in this case we start finding DATA/HOLE from
2834 * the very start of the file.
2835 */
2842 lockend--;
2854 }
2869 }
2874 else
2876 }
2880 }
2883 {
2898 }
2904 }
2905 }
2908 out:
2911 }
2924 #ifdef CONFIG_COMPAT
2926 #endif
2927 };
2930 {
2933 }
2936 {
2940 NULL);
2945 }
2948 {
2951 /*
2952 * So with compression we will find and lock a dirty page and clear the
2953 * first one as dirty, setup an async extent, and immediately return
2954 * with the entire range locked but with nobody actually marked with
2955 * writeback. So we can't just filemap_write_and_wait_range() and
2956 * expect it to work since it will just kick off a thread to do the
2957 * actual work. So we need to call filemap_fdatawrite_range _again_
2958 * since it will wait on the page lock, which won't be unlocked until
2959 * after the pages have been marked as writeback and so we're good to go
2960 * from there. We have to do this otherwise we'll miss the ordered
2961 * extents and that results in badness. Please Josef, do not think you
2962 * know better and pull this out at some point in the future, it is
2963 * right and you are wrong.
2964 */
2971 }