| blob | 73b547f88bfcaac5ec8e550e5a7d0d27e13014cf |
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 */
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 */
1385 {
1386 u64 start_pos;
1387 u64 last_pos;
1409 }
1413 }
1424 }
1431 }
1434 }
1438 {
1442 u64 num_bytes;
1458 }
1462 }
1472 }
1477 }
1481 loff_t pos)
1482 {
1488 u64 lockstart;
1489 u64 lockend;
1509 offset);
1511 PAGE_CACHE_SIZE);
1518 /*
1519 * Fault pages before locking them in prepare_pages
1520 * to avoid recursive lock
1521 */
1525 }
1531 BTRFS_INODE_PREALLOC)) &&
1533 /*
1534 * For nodata cow case, no need to reserve
1535 * data space.
1536 */
1538 /*
1539 * our prealloc extent may be smaller than
1540 * write_bytes, so scale down.
1541 */
1543 PAGE_CACHE_SIZE);
1546 }
1552 reserve_metadata:
1557 write_bytes);
1558 else
1561 }
1565 again:
1566 /*
1567 * This is going to setup the pages array with the number of
1568 * pages we want, so we don't really need to worry about the
1569 * contents of pages from loop to loop
1570 */
1573 force_page_uptodate);
1587 }
1592 /*
1593 * if we have trouble faulting in the pages, fall
1594 * back to one page at a time
1595 */
1605 PAGE_CACHE_SIZE);
1606 }
1608 /*
1609 * If we had a short copy we need to release the excess delaloc
1610 * bytes we reserved. We need to increment outstanding_extents
1611 * because btrfs_delalloc_release_space will decrement it, but
1612 * we still have an outstanding extent for the chunk we actually
1613 * managed to copy.
1614 */
1617 PAGE_CACHE_SHIFT;
1622 }
1625 release_bytes);
1627 u64 __pos;
1632 release_bytes);
1633 }
1634 }
1641 NULL);
1645 GFP_NOFS);
1649 }
1663 }
1675 }
1685 }
1686 }
1689 }
1693 loff_t pos)
1694 {
1697 ssize_t written;
1698 ssize_t written_buffered;
1699 loff_t endbyte;
1712 }
1713 /*
1714 * Ensure all data is persisted. We want the next direct IO read to be
1715 * able to read what was just written.
1716 */
1728 out:
1730 }
1733 {
1748 }
1752 {
1756 u64 start_pos;
1757 u64 end_pos;
1760 ssize_t err;
1761 loff_t pos;
1769 }
1776 }
1778 /*
1779 * If BTRFS flips readonly due to some impossible error
1780 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1781 * although we have opened a file as writable, we have
1782 * to stop this write operation to ensure FS consistency.
1783 */
1788 }
1790 /*
1791 * We reserve space for updating the inode when we reserve space for the
1792 * extent we are going to write, so we will enospc out there. We don't
1793 * need to start yet another transaction to update the inode as we will
1794 * update the inode when we finish writing whatever data we write.
1795 */
1802 /* Expand hole size to cover write data, preventing empty gap */
1808 }
1809 }
1820 }
1824 /*
1825 * We also have to set last_sub_trans to the current log transid,
1826 * otherwise subsequent syncs to a file that's been synced in this
1827 * transaction will appear to have already occured.
1828 */
1836 }
1840 out:
1843 }
1846 {
1849 /*
1850 * ordered_data_close is set by settattr when we are about to truncate
1851 * a file from a non-zero size to a zero size. This tries to
1852 * flush down new bytes that may have been written if the
1853 * application were using truncate to replace a file in place.
1854 */
1859 }
1862 {
1866 /*
1867 * This is only called in fsync, which would do synchronous writes, so
1868 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1869 * multiple disks using raid profile, a large IO can be split to
1870 * several segments of stripe length (currently 64K).
1871 */
1879 }
1881 /*
1882 * fsync call for both files and directories. This logs the inode into
1883 * the tree log instead of forcing full commits whenever possible.
1884 *
1885 * It needs to call filemap_fdatawait so that all ordered extent updates are
1886 * in the metadata btree are up to date for copying to the log.
1887 *
1888 * It drops the inode mutex before doing the tree log commit. This is an
1889 * important optimization for directories because holding the mutex prevents
1890 * new operations on the dir while we write to disk.
1891 */
1893 {
1901 u64 len;
1903 /*
1904 * If the inode needs a full sync, make sure we use a full range to
1905 * avoid log tree corruption, due to hole detection racing with ordered
1906 * extent completion for adjacent ranges, and assertion failures during
1907 * hole detection.
1908 */
1913 }
1915 /*
1916 * The range length can be represented by u64, we have to do the typecasts
1917 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1918 */
1922 /*
1923 * We write the dirty pages in the range and wait until they complete
1924 * out of the ->i_mutex. If so, we can flush the dirty pages by
1925 * multi-task, and make the performance up. See
1926 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1927 */
1936 /*
1937 * We might have have had more pages made dirty after calling
1938 * start_ordered_ops and before acquiring the inode's i_mutex.
1939 */
1941 /*
1942 * For a full sync, we need to make sure any ordered operations
1943 * start and finish before we start logging the inode, so that
1944 * all extents are persisted and the respective file extent
1945 * items are in the fs/subvol btree.
1946 */
1949 /*
1950 * Start any new ordered operations before starting to log the
1951 * inode. We will wait for them to finish in btrfs_sync_log().
1952 *
1953 * Right before acquiring the inode's mutex, we might have new
1954 * writes dirtying pages, which won't immediately start the
1955 * respective ordered operations - that is done through the
1956 * fill_delalloc callbacks invoked from the writepage and
1957 * writepages address space operations. So make sure we start
1958 * all ordered operations before starting to log our inode. Not
1959 * doing this means that while logging the inode, writeback
1960 * could start and invoke writepage/writepages, which would call
1961 * the fill_delalloc callbacks (cow_file_range,
1962 * submit_compressed_extents). These callbacks add first an
1963 * extent map to the modified list of extents and then create
1964 * the respective ordered operation, which means in
1965 * tree-log.c:btrfs_log_inode() we might capture all existing
1966 * ordered operations (with btrfs_get_logged_extents()) before
1967 * the fill_delalloc callback adds its ordered operation, and by
1968 * the time we visit the modified list of extent maps (with
1969 * btrfs_log_changed_extents()), we see and process the extent
1970 * map they created. We then use the extent map to construct a
1971 * file extent item for logging without waiting for the
1972 * respective ordered operation to finish - this file extent
1973 * item points to a disk location that might not have yet been
1974 * written to, containing random data - so after a crash a log
1975 * replay will make our inode have file extent items that point
1976 * to disk locations containing invalid data, as we returned
1977 * success to userspace without waiting for the respective
1978 * ordered operation to finish, because it wasn't captured by
1979 * btrfs_get_logged_extents().
1980 */
1982 }
1986 }
1989 /*
1990 * If the last transaction that changed this file was before the current
1991 * transaction and we have the full sync flag set in our inode, we can
1992 * bail out now without any syncing.
1993 *
1994 * Note that we can't bail out if the full sync flag isn't set. This is
1995 * because when the full sync flag is set we start all ordered extents
1996 * and wait for them to fully complete - when they complete they update
1997 * the inode's last_trans field through:
1998 *
1999 * btrfs_finish_ordered_io() ->
2000 * btrfs_update_inode_fallback() ->
2001 * btrfs_update_inode() ->
2002 * btrfs_set_inode_last_trans()
2003 *
2004 * So we are sure that last_trans is up to date and can do this check to
2005 * bail out safely. For the fast path, when the full sync flag is not
2006 * set in our inode, we can not do it because we start only our ordered
2007 * extents and don't wait for them to complete (that is when
2008 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2009 * value might be less than or equals to fs_info->last_trans_committed,
2010 * and setting a speculative last_trans for an inode when a buffered
2011 * write is made (such as fs_info->generation + 1 for example) would not
2012 * be reliable since after setting the value and before fsync is called
2013 * any number of transactions can start and commit (transaction kthread
2014 * commits the current transaction periodically), and a transaction
2015 * commit does not start nor waits for ordered extents to complete.
2016 */
2023 /*
2024 * We'v had everything committed since the last time we were
2025 * modified so clear this flag in case it was set for whatever
2026 * reason, it's no longer relevant.
2027 */
2032 }
2034 /*
2035 * ok we haven't committed the transaction yet, lets do a commit
2036 */
2040 /*
2041 * We use start here because we will need to wait on the IO to complete
2042 * in btrfs_sync_log, which could require joining a transaction (for
2043 * example checking cross references in the nocow path). If we use join
2044 * here we could get into a situation where we're waiting on IO to
2045 * happen that is blocked on a transaction trying to commit. With start
2046 * we inc the extwriter counter, so we wait for all extwriters to exit
2047 * before we start blocking join'ers. This comment is to keep somebody
2048 * from thinking they are super smart and changing this to
2049 * btrfs_join_transaction *cough*Josef*cough*.
2050 */
2056 }
2063 /* Fallthrough and commit/free transaction. */
2065 }
2067 /* we've logged all the items and now have a consistent
2068 * version of the file in the log. It is possible that
2069 * someone will come in and modify the file, but that's
2070 * fine because the log is consistent on disk, and we
2071 * have references to all of the file's extents
2072 *
2073 * It is possible that someone will come in and log the
2074 * file again, but that will end up using the synchronization
2075 * inside btrfs_sync_log to keep things safe.
2076 */
2079 /*
2080 * If any of the ordered extents had an error, just return it to user
2081 * space, so that the application knows some writes didn't succeed and
2082 * can take proper action (retry for e.g.). Blindly committing the
2083 * transaction in this case, would fool userspace that everything was
2084 * successful. And we also want to make sure our log doesn't contain
2085 * file extent items pointing to extents that weren't fully written to -
2086 * just like in the non fast fsync path, where we check for the ordered
2087 * operation's error flag before writing to the log tree and return -EIO
2088 * if any of them had this flag set (btrfs_wait_ordered_range) -
2089 * therefore we need to check for errors in the ordered operations,
2090 * which are indicated by ctx.io_err.
2091 */
2096 }
2104 }
2105 }
2111 }
2112 }
2116 }
2117 out:
2119 }
2125 };
2128 {
2138 }
2142 {
2167 }
2171 {
2194 u64 num_bytes;
2206 }
2209 u64 num_bytes;
2216 offset;
2222 }
2231 out:
2262 }
2265 }
2267 /*
2268 * Find a hole extent on given inode and change start/len to the end of hole
2269 * extent.(hole/vacuum extent whose em->start <= start &&
2270 * em->start + em->len > start)
2271 * When a hole extent is found, return 1 and modify start/len.
2272 */
2274 {
2282 else
2285 }
2287 /* Hole or vacuum extent(only exists in no-hole mode) */
2293 }
2296 }
2299 {
2305 u64 lockstart;
2306 u64 lockend;
2307 u64 tail_start;
2308 u64 tail_len;
2310 u64 cur_offset;
2312 u64 drop_end;
2318 u64 ino_size;
2332 /* Already in a large hole */
2335 }
2343 /*
2344 * We needn't truncate any page which is beyond the end of the file
2345 * because we are sure there is no data there.
2346 */
2347 /*
2348 * Only do this if we are in the same page and we aren't doing the
2349 * entire page.
2350 */
2357 }
2359 }
2361 /* zero back part of the first page */
2368 }
2369 }
2371 /* Check the aligned pages after the first unaligned page,
2372 * if offset != orig_start, which means the first unaligned page
2373 * including serveral following pages are already in holes,
2374 * the extra check can be skipped */
2376 /* after truncate page, check hole again */
2385 }
2387 }
2389 /* Check the tail unaligned part is in a hole */
2397 /* zero the front end of the last page */
2404 }
2405 }
2406 }
2411 }
2422 /*
2423 * We need to make sure we have no ordered extents in this range
2424 * and nobody raced in and read a page in this range, if we did
2425 * we need to try again.
2426 */
2434 }
2444 }
2445 }
2451 }
2457 }
2461 /*
2462 * 1 - update the inode
2463 * 1 - removing the extents in the range
2464 * 1 - adding the hole extent if no_holes isn't set
2465 */
2471 }
2474 min_size);
2491 drop_end);
2495 }
2496 }
2504 }
2514 }
2527 }
2528 }
2533 }
2536 /*
2537 * If we are using the NO_HOLES feature we might have had already an
2538 * hole that overlaps a part of the region [lockstart, lockend] and
2539 * ends at (or beyond) lockend. Since we have no file extent items to
2540 * represent holes, drop_end can be less than lockend and so we must
2541 * make sure we have an extent map representing the existing hole (the
2542 * call to __btrfs_drop_extents() might have dropped the existing extent
2543 * map representing the existing hole), otherwise the fast fsync path
2544 * will not record the existence of the hole region
2545 * [existing_hole_start, lockend].
2546 */
2549 /*
2550 * Don't insert file hole extent item if it's for a range beyond eof
2551 * (because it's useless) or if it represents a 0 bytes range (when
2552 * cur_offset == drop_end).
2553 */
2559 }
2560 }
2562 out_trans:
2574 out_free:
2577 out:
2580 out_only_mutex:
2582 /*
2583 * If we only end up zeroing part of a page, we still need to
2584 * update the inode item, so that all the time fields are
2585 * updated as well as the necessary btrfs inode in memory fields
2586 * for detecting, at fsync time, if the inode isn't yet in the
2587 * log tree or it's there but not up to date.
2588 */
2595 }
2596 }
2601 }
2603 /* Helper structure to record which range is already reserved */
2606 u64 start;
2607 u64 len;
2608 };
2610 /*
2611 * Helper function to add falloc range
2612 *
2613 * Caller should have locked the larger range of extent containing
2614 * [start, len)
2615 */
2617 {
2624 /*
2625 * As fallocate iterate by bytenr order, we only need to check
2626 * the last range.
2627 */
2632 }
2633 insert:
2641 }
2645 {
2651 u64 cur_offset;
2652 u64 last_byte;
2653 u64 alloc_start;
2654 u64 alloc_end;
2656 u64 locked_end;
2665 /* Make sure we aren't being give some crap mode */
2672 /*
2673 * Only trigger disk allocation, don't trigger qgroup reserve
2674 *
2675 * For qgroup space, it will be checked later.
2676 */
2686 /*
2687 * TODO: Move these two operations after we have checked
2688 * accurate reserved space, or fallocate can still fail but
2689 * with page truncated or size expanded.
2690 *
2691 * But that's a minor problem and won't do much harm BTW.
2692 */
2695 alloc_start);
2699 /*
2700 * If we are fallocating from the end of the file onward we
2701 * need to zero out the end of the page if i_size lands in the
2702 * middle of a page.
2703 */
2707 }
2709 /*
2710 * wait for ordered IO before we have any locks. We'll loop again
2711 * below with the locks held.
2712 */
2722 /* the extent lock is ordered inside the running
2723 * transaction
2724 */
2736 /*
2737 * we can't wait on the range with the transaction
2738 * running or with the extent lock held
2739 */
2748 }
2749 }
2751 /* First, check if we exceed the qgroup limit */
2760 else
2763 }
2775 }
2780 }
2785 }
2787 /*
2788 * If ret is still 0, means we're OK to fallocate.
2789 * Or just cleanup the list and exit.
2790 */
2799 }
2808 /*
2809 * We didn't need to allocate any more space, but we
2810 * still extended the size of the file so we need to
2811 * update i_size and the inode item.
2812 */
2823 else
2825 }
2826 }
2827 out_unlock:
2830 out:
2831 /*
2832 * As we waited the extent range, the data_rsv_map must be empty
2833 * in the range, as written data range will be released from it.
2834 * And for prealloacted extent, it will also be released when
2835 * its metadata is written.
2836 * So this is completely used as cleanup.
2837 */
2840 /* Let go of our reservation. */
2844 }
2847 {
2851 u64 lockstart;
2852 u64 lockend;
2853 u64 start;
2854 u64 len;
2860 /*
2861 * *offset can be negative, in this case we start finding DATA/HOLE from
2862 * the very start of the file.
2863 */
2870 lockend--;
2882 }
2897 }
2902 else
2904 }
2908 }
2911 {
2926 }
2932 }
2933 }
2936 out:
2939 }
2952 #ifdef CONFIG_COMPAT
2954 #endif
2955 };
2958 {
2961 }
2964 {
2968 NULL);
2973 }
2976 {
2979 /*
2980 * So with compression we will find and lock a dirty page and clear the
2981 * first one as dirty, setup an async extent, and immediately return
2982 * with the entire range locked but with nobody actually marked with
2983 * writeback. So we can't just filemap_write_and_wait_range() and
2984 * expect it to work since it will just kick off a thread to do the
2985 * actual work. So we need to call filemap_fdatawrite_range _again_
2986 * since it will wait on the page lock, which won't be unlocked until
2987 * after the pages have been marked as writeback and so we're good to go
2988 * from there. We have to do this otherwise we'll miss the ordered
2989 * extents and that results in badness. Please Josef, do not think you
2990 * know better and pull this out at some point in the future, it is
2991 * right and you are wrong.
2992 */
2999 }