| blob | 8d7b5a45c00523f4ca7ee5d58810b9d571ad4148 |
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>
47 /*
48 * when auto defrag is enabled we
49 * queue up these defrag structs to remember which
50 * inodes need defragging passes
51 */
54 /* objectid */
55 u64 ino;
56 /*
57 * transid where the defrag was added, we search for
58 * extents newer than this
59 */
60 u64 transid;
62 /* root objectid */
63 u64 root;
65 /* last offset we were able to defrag */
66 u64 last_offset;
68 /* if we've wrapped around back to zero once already */
70 };
74 {
83 else
85 }
87 /* pop a record for an inode into the defrag tree. The lock
88 * must be held already
89 *
90 * If you're inserting a record for an older transid than an
91 * existing record, the transid already in the tree is lowered
92 *
93 * If an existing record is found the defrag item you
94 * pass in is freed
95 */
98 {
116 /* if we're reinserting an entry for
117 * an old defrag run, make sure to
118 * lower the transid of our existing record
119 */
125 }
126 }
131 }
134 {
142 }
144 /*
145 * insert a defrag record for this inode if auto defrag is
146 * enabled
147 */
150 {
153 u64 transid;
164 else
177 /*
178 * If we set IN_DEFRAG flag and evict the inode from memory,
179 * and then re-read this inode, this new inode doesn't have
180 * IN_DEFRAG flag. At the case, we may find the existed defrag.
181 */
187 }
190 }
192 /*
193 * Requeue the defrag object. If there is a defrag object that points to
194 * the same inode in the tree, we will merge them together (by
195 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
196 */
199 {
206 /*
207 * Here we don't check the IN_DEFRAG flag, because we need merge
208 * them together.
209 */
216 out:
218 }
220 /*
221 * pick the defragable inode that we want, if it doesn't exist, we will get
222 * the next one.
223 */
226 {
247 else
249 }
255 else
257 }
258 out:
263 }
266 {
280 }
282 }
284 #define BTRFS_DEFRAG_BATCH 1024
288 {
297 /* get the inode */
308 }
317 }
320 /* do a chunk of defrag */
328 BTRFS_DEFRAG_BATCH);
330 /*
331 * if we filled the whole defrag batch, there
332 * must be more work to do. Queue this defrag
333 * again
334 */
339 /*
340 * we didn't fill our defrag batch, but
341 * we didn't start at zero. Make sure we loop
342 * around to the start of the file.
343 */
349 }
353 cleanup:
357 }
359 /*
360 * run through the list of inodes in the FS that need
361 * defragging
362 */
364 {
371 /* Pause the auto defragger. */
379 /* find an inode to defrag */
381 first_ino);
389 }
390 }
396 }
399 /*
400 * during unmount, we use the transaction_wait queue to
401 * wait for the defragger to stop
402 */
405 }
407 /* simple helper to fault in pages and copy. This should go away
408 * and be replaced with calls into generic code.
409 */
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 */
1297 {
1309 }
1313 }
1314 }
1316 }
1318 /*
1319 * this just gets pages into the page cache and locks them down.
1320 */
1324 {
1332 again:
1339 }
1343 force_uptodate);
1352 }
1355 }
1357 }
1360 fail:
1364 faili--;
1365 }
1368 }
1370 /*
1371 * This function locks the extent and properly waits for data=ordered extents
1372 * to finish before allowing the pages to be modified if need.
1373 *
1374 * The return value:
1375 * 1 - the extent is locked
1376 * 0 - the extent is not locked, and everything is OK
1377 * -EAGAIN - need re-prepare the pages
1378 * the other < 0 number - Something wrong happens
1379 */
1386 {
1388 u64 start_pos;
1389 u64 last_pos;
1394 last_pos = start_pos
1412 }
1416 }
1427 }
1434 }
1437 }
1441 {
1445 u64 num_bytes;
1461 }
1465 }
1475 }
1480 }
1484 loff_t pos)
1485 {
1491 u64 lockstart;
1492 u64 lockend;
1513 offset);
1515 PAGE_SIZE);
1524 /*
1525 * Fault pages before locking them in prepare_pages
1526 * to avoid recursive lock
1527 */
1531 }
1538 BTRFS_INODE_PREALLOC)) &&
1540 /*
1541 * For nodata cow case, no need to reserve
1542 * data space.
1543 */
1545 /*
1546 * our prealloc extent may be smaller than
1547 * write_bytes, so scale down.
1548 */
1550 PAGE_SIZE);
1554 }
1560 reserve_metadata:
1565 write_bytes);
1566 else
1569 }
1573 again:
1574 /*
1575 * This is going to setup the pages array with the number of
1576 * pages we want, so we don't really need to worry about the
1577 * contents of pages from loop to loop
1578 */
1581 force_page_uptodate);
1595 }
1599 /*
1600 * if we have trouble faulting in the pages, fall
1601 * back to one page at a time
1602 */
1612 PAGE_SIZE);
1613 }
1615 /*
1616 * If we had a short copy we need to release the excess delaloc
1617 * bytes we reserved. We need to increment outstanding_extents
1618 * because btrfs_delalloc_release_space will decrement it, but
1619 * we still have an outstanding extent for the chunk we actually
1620 * managed to copy.
1621 */
1623 reserve_bytes);
1627 dirty_sectors);
1636 }
1639 release_bytes);
1641 u64 __pos;
1646 release_bytes);
1647 }
1648 }
1656 NULL);
1660 GFP_NOFS);
1664 }
1678 }
1690 }
1700 }
1701 }
1704 }
1708 loff_t pos)
1709 {
1712 ssize_t written;
1713 ssize_t written_buffered;
1714 loff_t endbyte;
1727 }
1728 /*
1729 * Ensure all data is persisted. We want the next direct IO read to be
1730 * able to read what was just written.
1731 */
1743 out:
1745 }
1748 {
1763 }
1767 {
1771 u64 start_pos;
1772 u64 end_pos;
1775 ssize_t err;
1776 loff_t pos;
1778 loff_t oldsize;
1786 }
1793 }
1795 /*
1796 * If BTRFS flips readonly due to some impossible error
1797 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1798 * although we have opened a file as writable, we have
1799 * to stop this write operation to ensure FS consistency.
1800 */
1805 }
1807 /*
1808 * We reserve space for updating the inode when we reserve space for the
1809 * extent we are going to write, so we will enospc out there. We don't
1810 * need to start yet another transaction to update the inode as we will
1811 * update the inode when we finish writing whatever data we write.
1812 */
1820 /* Expand hole size to cover write data, preventing empty gap */
1826 }
1829 }
1843 }
1847 /*
1848 * We also have to set last_sub_trans to the current log transid,
1849 * otherwise subsequent syncs to a file that's been synced in this
1850 * transaction will appear to have already occurred.
1851 */
1859 }
1863 out:
1866 }
1869 {
1872 /*
1873 * ordered_data_close is set by settattr when we are about to truncate
1874 * a file from a non-zero size to a zero size. This tries to
1875 * flush down new bytes that may have been written if the
1876 * application were using truncate to replace a file in place.
1877 */
1882 }
1885 {
1893 }
1895 /*
1896 * fsync call for both files and directories. This logs the inode into
1897 * the tree log instead of forcing full commits whenever possible.
1898 *
1899 * It needs to call filemap_fdatawait so that all ordered extent updates are
1900 * in the metadata btree are up to date for copying to the log.
1901 *
1902 * It drops the inode mutex before doing the tree log commit. This is an
1903 * important optimization for directories because holding the mutex prevents
1904 * new operations on the dir while we write to disk.
1905 */
1907 {
1915 u64 len;
1917 /*
1918 * The range length can be represented by u64, we have to do the typecasts
1919 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1920 */
1924 /*
1925 * We write the dirty pages in the range and wait until they complete
1926 * out of the ->i_mutex. If so, we can flush the dirty pages by
1927 * multi-task, and make the performance up. See
1928 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1929 */
1938 /*
1939 * We might have have had more pages made dirty after calling
1940 * start_ordered_ops and before acquiring the inode's i_mutex.
1941 */
1943 /*
1944 * For a full sync, we need to make sure any ordered operations
1945 * start and finish before we start logging the inode, so that
1946 * all extents are persisted and the respective file extent
1947 * items are in the fs/subvol btree.
1948 */
1951 /*
1952 * Start any new ordered operations before starting to log the
1953 * inode. We will wait for them to finish in btrfs_sync_log().
1954 *
1955 * Right before acquiring the inode's mutex, we might have new
1956 * writes dirtying pages, which won't immediately start the
1957 * respective ordered operations - that is done through the
1958 * fill_delalloc callbacks invoked from the writepage and
1959 * writepages address space operations. So make sure we start
1960 * all ordered operations before starting to log our inode. Not
1961 * doing this means that while logging the inode, writeback
1962 * could start and invoke writepage/writepages, which would call
1963 * the fill_delalloc callbacks (cow_file_range,
1964 * submit_compressed_extents). These callbacks add first an
1965 * extent map to the modified list of extents and then create
1966 * the respective ordered operation, which means in
1967 * tree-log.c:btrfs_log_inode() we might capture all existing
1968 * ordered operations (with btrfs_get_logged_extents()) before
1969 * the fill_delalloc callback adds its ordered operation, and by
1970 * the time we visit the modified list of extent maps (with
1971 * btrfs_log_changed_extents()), we see and process the extent
1972 * map they created. We then use the extent map to construct a
1973 * file extent item for logging without waiting for the
1974 * respective ordered operation to finish - this file extent
1975 * item points to a disk location that might not have yet been
1976 * written to, containing random data - so after a crash a log
1977 * replay will make our inode have file extent items that point
1978 * to disk locations containing invalid data, as we returned
1979 * success to userspace without waiting for the respective
1980 * ordered operation to finish, because it wasn't captured by
1981 * btrfs_get_logged_extents().
1982 */
1984 }
1988 }
1991 /*
1992 * If the last transaction that changed this file was before the current
1993 * transaction and we have the full sync flag set in our inode, we can
1994 * bail out now without any syncing.
1995 *
1996 * Note that we can't bail out if the full sync flag isn't set. This is
1997 * because when the full sync flag is set we start all ordered extents
1998 * and wait for them to fully complete - when they complete they update
1999 * the inode's last_trans field through:
2000 *
2001 * btrfs_finish_ordered_io() ->
2002 * btrfs_update_inode_fallback() ->
2003 * btrfs_update_inode() ->
2004 * btrfs_set_inode_last_trans()
2005 *
2006 * So we are sure that last_trans is up to date and can do this check to
2007 * bail out safely. For the fast path, when the full sync flag is not
2008 * set in our inode, we can not do it because we start only our ordered
2009 * extents and don't wait for them to complete (that is when
2010 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2011 * value might be less than or equals to fs_info->last_trans_committed,
2012 * and setting a speculative last_trans for an inode when a buffered
2013 * write is made (such as fs_info->generation + 1 for example) would not
2014 * be reliable since after setting the value and before fsync is called
2015 * any number of transactions can start and commit (transaction kthread
2016 * commits the current transaction periodically), and a transaction
2017 * commit does not start nor waits for ordered extents to complete.
2018 */
2026 /*
2027 * We'v had everything committed since the last time we were
2028 * modified so clear this flag in case it was set for whatever
2029 * reason, it's no longer relevant.
2030 */
2035 }
2037 /*
2038 * ok we haven't committed the transaction yet, lets do a commit
2039 */
2043 /*
2044 * We use start here because we will need to wait on the IO to complete
2045 * in btrfs_sync_log, which could require joining a transaction (for
2046 * example checking cross references in the nocow path). If we use join
2047 * here we could get into a situation where we're waiting on IO to
2048 * happen that is blocked on a transaction trying to commit. With start
2049 * we inc the extwriter counter, so we wait for all extwriters to exit
2050 * before we start blocking join'ers. This comment is to keep somebody
2051 * from thinking they are super smart and changing this to
2052 * btrfs_join_transaction *cough*Josef*cough*.
2053 */
2059 }
2066 /* Fallthrough and commit/free transaction. */
2068 }
2070 /* we've logged all the items and now have a consistent
2071 * version of the file in the log. It is possible that
2072 * someone will come in and modify the file, but that's
2073 * fine because the log is consistent on disk, and we
2074 * have references to all of the file's extents
2075 *
2076 * It is possible that someone will come in and log the
2077 * file again, but that will end up using the synchronization
2078 * inside btrfs_sync_log to keep things safe.
2079 */
2082 /*
2083 * If any of the ordered extents had an error, just return it to user
2084 * space, so that the application knows some writes didn't succeed and
2085 * can take proper action (retry for e.g.). Blindly committing the
2086 * transaction in this case, would fool userspace that everything was
2087 * successful. And we also want to make sure our log doesn't contain
2088 * file extent items pointing to extents that weren't fully written to -
2089 * just like in the non fast fsync path, where we check for the ordered
2090 * operation's error flag before writing to the log tree and return -EIO
2091 * if any of them had this flag set (btrfs_wait_ordered_range) -
2092 * therefore we need to check for errors in the ordered operations,
2093 * which are indicated by ctx.io_err.
2094 */
2099 }
2107 }
2108 }
2114 }
2115 }
2119 }
2120 out:
2122 }
2128 };
2131 {
2141 }
2145 {
2170 }
2174 {
2197 u64 num_bytes;
2209 }
2212 u64 num_bytes;
2219 offset;
2225 }
2234 out:
2265 }
2268 }
2270 /*
2271 * Find a hole extent on given inode and change start/len to the end of hole
2272 * extent.(hole/vacuum extent whose em->start <= start &&
2273 * em->start + em->len > start)
2274 * When a hole extent is found, return 1 and modify start/len.
2275 */
2277 {
2285 else
2288 }
2290 /* Hole or vacuum extent(only exists in no-hole mode) */
2296 }
2299 }
2302 {
2308 u64 lockstart;
2309 u64 lockend;
2310 u64 tail_start;
2311 u64 tail_len;
2313 u64 cur_offset;
2315 u64 drop_end;
2321 u64 ino_size;
2335 /* Already in a large hole */
2338 }
2345 /*
2346 * We needn't truncate any block which is beyond the end of the file
2347 * because we are sure there is no data there.
2348 */
2349 /*
2350 * Only do this if we are in the same block and we aren't doing the
2351 * entire block.
2352 */
2359 }
2361 }
2363 /* zero back part of the first block */
2370 }
2371 }
2373 /* Check the aligned pages after the first unaligned page,
2374 * if offset != orig_start, which means the first unaligned page
2375 * including serveral following pages are already in holes,
2376 * the extra check can be skipped */
2378 /* after truncate page, check hole again */
2387 }
2389 }
2391 /* Check the tail unaligned part is in a hole */
2399 /* zero the front end of the last page */
2407 }
2408 }
2409 }
2414 }
2425 /*
2426 * We need to make sure we have no ordered extents in this range
2427 * and nobody raced in and read a page in this range, if we did
2428 * we need to try again.
2429 */
2437 }
2447 }
2448 }
2454 }
2460 }
2464 /*
2465 * 1 - update the inode
2466 * 1 - removing the extents in the range
2467 * 1 - adding the hole extent if no_holes isn't set
2468 */
2474 }
2477 min_size);
2494 drop_end);
2498 }
2499 }
2507 }
2517 }
2530 }
2531 }
2536 }
2539 /*
2540 * If we are using the NO_HOLES feature we might have had already an
2541 * hole that overlaps a part of the region [lockstart, lockend] and
2542 * ends at (or beyond) lockend. Since we have no file extent items to
2543 * represent holes, drop_end can be less than lockend and so we must
2544 * make sure we have an extent map representing the existing hole (the
2545 * call to __btrfs_drop_extents() might have dropped the existing extent
2546 * map representing the existing hole), otherwise the fast fsync path
2547 * will not record the existence of the hole region
2548 * [existing_hole_start, lockend].
2549 */
2552 /*
2553 * Don't insert file hole extent item if it's for a range beyond eof
2554 * (because it's useless) or if it represents a 0 bytes range (when
2555 * cur_offset == drop_end).
2556 */
2562 }
2563 }
2565 out_trans:
2577 out_free:
2580 out:
2583 out_only_mutex:
2585 /*
2586 * If we only end up zeroing part of a page, we still need to
2587 * update the inode item, so that all the time fields are
2588 * updated as well as the necessary btrfs inode in memory fields
2589 * for detecting, at fsync time, if the inode isn't yet in the
2590 * log tree or it's there but not up to date.
2591 */
2598 }
2599 }
2604 }
2606 /* Helper structure to record which range is already reserved */
2609 u64 start;
2610 u64 len;
2611 };
2613 /*
2614 * Helper function to add falloc range
2615 *
2616 * Caller should have locked the larger range of extent containing
2617 * [start, len)
2618 */
2620 {
2627 /*
2628 * As fallocate iterate by bytenr order, we only need to check
2629 * the last range.
2630 */
2635 }
2636 insert:
2644 }
2648 {
2654 u64 cur_offset;
2655 u64 last_byte;
2656 u64 alloc_start;
2657 u64 alloc_end;
2659 u64 locked_end;
2668 /* Make sure we aren't being give some crap mode */
2675 /*
2676 * Only trigger disk allocation, don't trigger qgroup reserve
2677 *
2678 * For qgroup space, it will be checked later.
2679 */
2690 }
2692 /*
2693 * TODO: Move these two operations after we have checked
2694 * accurate reserved space, or fallocate can still fail but
2695 * with page truncated or size expanded.
2696 *
2697 * But that's a minor problem and won't do much harm BTW.
2698 */
2701 alloc_start);
2705 /*
2706 * If we are fallocating from the end of the file onward we
2707 * need to zero out the end of the block if i_size lands in the
2708 * middle of a block.
2709 */
2713 }
2715 /*
2716 * wait for ordered IO before we have any locks. We'll loop again
2717 * below with the locks held.
2718 */
2728 /* the extent lock is ordered inside the running
2729 * transaction
2730 */
2742 /*
2743 * we can't wait on the range with the transaction
2744 * running or with the extent lock held
2745 */
2754 }
2755 }
2757 /* First, check if we exceed the qgroup limit */
2766 else
2769 }
2781 }
2786 }
2791 }
2793 /*
2794 * If ret is still 0, means we're OK to fallocate.
2795 * Or just cleanup the list and exit.
2796 */
2805 }
2814 /*
2815 * We didn't need to allocate any more space, but we
2816 * still extended the size of the file so we need to
2817 * update i_size and the inode item.
2818 */
2829 else
2831 }
2832 }
2833 out_unlock:
2836 out:
2837 /*
2838 * As we waited the extent range, the data_rsv_map must be empty
2839 * in the range, as written data range will be released from it.
2840 * And for prealloacted extent, it will also be released when
2841 * its metadata is written.
2842 * So this is completely used as cleanup.
2843 */
2846 /* Let go of our reservation. */
2850 }
2853 {
2857 u64 lockstart;
2858 u64 lockend;
2859 u64 start;
2860 u64 len;
2866 /*
2867 * *offset can be negative, in this case we start finding DATA/HOLE from
2868 * the very start of the file.
2869 */
2876 lockend--;
2888 }
2903 }
2908 else
2910 }
2914 }
2917 {
2932 }
2938 }
2939 }
2942 out:
2945 }
2958 #ifdef CONFIG_COMPAT
2960 #endif
2964 };
2967 {
2969 }
2972 {
2976 NULL);
2981 }
2984 {
2987 /*
2988 * So with compression we will find and lock a dirty page and clear the
2989 * first one as dirty, setup an async extent, and immediately return
2990 * with the entire range locked but with nobody actually marked with
2991 * writeback. So we can't just filemap_write_and_wait_range() and
2992 * expect it to work since it will just kick off a thread to do the
2993 * actual work. So we need to call filemap_fdatawrite_range _again_
2994 * since it will wait on the page lock, which won't be unlocked until
2995 * after the pages have been marked as writeback and so we're good to go
2996 * from there. We have to do this otherwise we'll miss the ordered
2997 * extents and that results in badness. Please Josef, do not think you
2998 * know better and pull this out at some point in the future, it is
2999 * right and you are wrong.
3000 */
3007 }