| blob | cecf9aa1081134d255455ee329a4e118b8355216 |
1 /*
2 * linux/fs/ext4/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
17 *
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
19 */
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/buffer_head.h>
28 #include <linux/writeback.h>
29 #include <linux/pagevec.h>
30 #include <linux/mpage.h>
31 #include <linux/namei.h>
32 #include <linux/uio.h>
33 #include <linux/bio.h>
34 #include <linux/workqueue.h>
35 #include <linux/kernel.h>
36 #include <linux/printk.h>
37 #include <linux/slab.h>
38 #include <linux/bitops.h>
45 #include <trace/events/ext4.h>
47 #define MPAGE_DA_EXTENT_TAIL 0x01
51 {
53 __u16 csum_lo;
55 __u32 csum;
63 }
74 }
78 {
91 else
95 }
99 {
100 __u32 csum;
112 }
115 loff_t new_size)
116 {
118 /*
119 * If jinode is zero, then we never opened the file for
120 * writing, so there's no need to call
121 * jbd2_journal_begin_ordered_truncate() since there's no
122 * outstanding writes we need to flush.
123 */
128 new_size);
129 }
138 /*
139 * Test whether an inode is a fast symlink.
140 */
142 {
150 }
152 /*
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
155 * this transaction.
156 */
159 {
162 /*
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
167 */
176 }
178 /*
179 * Called at the last iput() if i_nlink is zero.
180 */
182 {
189 /*
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
203 *
204 * Note that directories do not have this problem because they
205 * don't use page cache.
206 */
215 }
220 }
232 /*
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
235 */
241 /*
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
244 * cleaned up.
245 */
249 }
259 }
263 /*
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
268 */
276 stop_handle:
281 }
282 }
284 /*
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
291 */
295 /*
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
300 * fails.
301 */
303 /* If that failed, just do the required in-core inode clear. */
305 else
310 no_delete:
312 }
314 #ifdef CONFIG_QUOTA
316 {
318 }
319 #endif
321 /*
322 * Called with i_data_sem down, which is important since we can call
323 * ext4_discard_preallocations() from here.
324 */
327 {
340 }
342 /* Update per-inode reservations */
348 /* Update quota subsystem for data blocks */
352 /*
353 * We did fallocate with an offset that is already delayed
354 * allocated. So on delayed allocated writeback we should
355 * not re-claim the quota for fallocated blocks.
356 */
358 }
360 /*
361 * If we have done all the pending block allocations and if
362 * there aren't any writers on the inode, we can discard the
363 * inode's preallocations.
364 */
368 }
373 {
377 "lblock %lu mapped to illegal pblock "
381 }
383 }
385 #define check_block_validity(inode, map) \
386 __check_block_validity((inode), __func__, __LINE__, (map))
388 #ifdef ES_AGGRESSIVE_TEST
394 {
398 /*
399 * There is a race window that the result is not the same.
400 * e.g. xfstests #223 when dioread_nolock enables. The reason
401 * is that we lookup a block mapping in extent status tree with
402 * out taking i_data_sem. So at the time the unwritten extent
403 * could be converted.
404 */
409 EXT4_GET_BLOCKS_KEEP_SIZE);
412 EXT4_GET_BLOCKS_KEEP_SIZE);
413 }
417 /*
418 * We don't check m_len because extent will be collpased in status
419 * tree. So the m_len might not equal.
420 */
425 "es_cached ex [%d/%d/%llu/%x] != "
431 }
432 }
435 /*
436 * The ext4_map_blocks() function tries to look up the requested blocks,
437 * and returns if the blocks are already mapped.
438 *
439 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
440 * and store the allocated blocks in the result buffer head and mark it
441 * mapped.
442 *
443 * If file type is extents based, it will call ext4_ext_map_blocks(),
444 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
445 * based files
446 *
447 * On success, it returns the number of blocks being mapped or allocated.
448 * if create==0 and the blocks are pre-allocated and unwritten block,
449 * the result buffer head is unmapped. If the create ==1, it will make sure
450 * the buffer head is mapped.
451 *
452 * It returns 0 if plain look up failed (blocks have not been allocated), in
453 * that case, buffer head is unmapped
454 *
455 * It returns the error in case of allocation failure.
456 */
459 {
463 #ifdef ES_AGGRESSIVE_TEST
467 #endif
474 /*
475 * ext4_map_blocks returns an int, and m_len is an unsigned int
476 */
480 /* We can handle the block number less than EXT_MAX_BLOCKS */
484 /* Lookup extent status tree firstly */
499 }
500 #ifdef ES_AGGRESSIVE_TEST
503 #endif
505 }
507 /*
508 * Try to see if we can get the block without requesting a new
509 * file system block.
510 */
515 EXT4_GET_BLOCKS_KEEP_SIZE);
518 EXT4_GET_BLOCKS_KEEP_SIZE);
519 }
525 "ES len assertion failed for inode "
529 }
542 }
546 found:
551 }
553 /* If it is only a block(s) look up */
557 /*
558 * Returns if the blocks have already allocated
559 *
560 * Note that if blocks have been preallocated
561 * ext4_ext_get_block() returns the create = 0
562 * with buffer head unmapped.
563 */
565 /*
566 * If we need to convert extent to unwritten
567 * we continue and do the actual work in
568 * ext4_ext_map_blocks()
569 */
573 /*
574 * Here we clear m_flags because after allocating an new extent,
575 * it will be set again.
576 */
579 /*
580 * New blocks allocate and/or writing to unwritten extent
581 * will possibly result in updating i_data, so we take
582 * the write lock of i_data_sem, and call get_block()
583 * with create == 1 flag.
584 */
587 /*
588 * We need to check for EXT4 here because migrate
589 * could have changed the inode type in between
590 */
597 /*
598 * We allocated new blocks which will result in
599 * i_data's format changing. Force the migrate
600 * to fail by clearing migrate flags
601 */
603 }
605 /*
606 * Update reserved blocks/metadata blocks after successful
607 * block allocation which had been deferred till now. We don't
608 * support fallocate for non extent files. So we can update
609 * reserve space here.
610 */
614 }
621 "ES len assertion failed for inode "
625 }
627 /*
628 * If the extent has been zeroed out, we don't need to update
629 * extent status tree.
630 */
635 }
647 }
649 has_zeroout:
655 }
657 }
659 /* Maximum number of blocks we map for direct IO at once. */
660 #define DIO_MAX_BLOCKS 4096
664 {
677 /* Direct IO write... */
682 dio_credits);
686 }
688 }
697 /*
698 * dgc: I suspect unwritten conversion on ext4+DAX is
699 * fundamentally broken here when there are concurrent
700 * read/write in progress on this inode.
701 */
705 }
710 }
714 }
718 {
721 }
723 /*
724 * `handle' can be NULL if create is zero
725 */
728 {
752 /*
753 * Now that we do not always journal data, we should
754 * keep in mind whether this should always journal the
755 * new buffer as metadata. For now, regular file
756 * writes use ext4_get_block instead, so it's not a
757 * problem.
758 */
765 }
769 }
778 errout:
781 }
785 {
799 }
808 {
824 }
828 }
830 }
832 /*
833 * To preserve ordering, it is essential that the hole instantiation and
834 * the data write be encapsulated in a single transaction. We cannot
835 * close off a transaction and start a new one between the ext4_get_block()
836 * and the commit_write(). So doing the jbd2_journal_start at the start of
837 * prepare_write() is the right place.
838 *
839 * Also, this function can nest inside ext4_writepage(). In that case, we
840 * *know* that ext4_writepage() has generated enough buffer credits to do the
841 * whole page. So we won't block on the journal in that case, which is good,
842 * because the caller may be PF_MEMALLOC.
843 *
844 * By accident, ext4 can be reentered when a transaction is open via
845 * quota file writes. If we were to commit the transaction while thus
846 * reentered, there can be a deadlock - we would be holding a quota
847 * lock, and the commit would never complete if another thread had a
848 * transaction open and was blocking on the quota lock - a ranking
849 * violation.
850 *
851 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
852 * will _not_ run commit under these circumstances because handle->h_ref
853 * is elevated. We'll still have enough credits for the tiny quotafile
854 * write.
855 */
858 {
864 /*
865 * __block_write_begin() could have dirtied some buffers. Clean
866 * the dirty bit as jbd2_journal_get_write_access() could complain
867 * otherwise about fs integrity issues. Setting of the dirty bit
868 * by __block_write_begin() isn't a real problem here as we clear
869 * the bit before releasing a page lock and thus writeback cannot
870 * ever write the buffer.
871 */
879 }
884 #ifdef CONFIG_EXT4_FS_ENCRYPTION
887 {
892 sector_t block;
917 }
919 }
935 }
940 }
941 }
946 }
954 }
955 }
956 /*
957 * If we issued read requests, let them complete.
958 */
963 }
969 }
970 #endif
975 {
981 pgoff_t index;
985 /*
986 * Reserve one block more for addition to orphan list in case
987 * we allocate blocks but write fails for some reason
988 */
1001 }
1003 /*
1004 * grab_cache_page_write_begin() can take a long time if the
1005 * system is thrashing due to memory pressure, or if the page
1006 * is being written back. So grab it first before we start
1007 * the transaction handle. This also allows us to allocate
1008 * the page (if needed) without using GFP_NOFS.
1009 */
1010 retry_grab:
1016 retry_journal:
1021 }
1025 /* The page got truncated from under us */
1030 }
1031 /* In case writeback began while the page was unlocked */
1034 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1037 ext4_get_block_write);
1038 else
1040 ext4_get_block);
1041 #else
1044 else
1046 #endif
1050 do_journal_get_write_access);
1051 }
1055 /*
1056 * __block_write_begin may have instantiated a few blocks
1057 * outside i_size. Trim these off again. Don't need
1058 * i_size_read because we hold i_mutex.
1059 *
1060 * Add inode to orphan list in case we crash before
1061 * truncate finishes
1062 */
1069 /*
1070 * If truncate failed early the inode might
1071 * still be on the orphan list; we need to
1072 * make sure the inode is removed from the
1073 * orphan list in that case.
1074 */
1077 }
1084 }
1087 }
1089 /* For write_end() in data=journal mode */
1091 {
1100 }
1102 /*
1103 * We need to pick up the new inode size which generic_commit_write gave us
1104 * `file' can be NULL - eg, when called from page_symlink().
1105 *
1106 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1107 * buffers are managed internally.
1108 */
1113 {
1127 }
1128 }
1139 /*
1140 * it's important to update i_size while still holding page lock:
1141 * page writeout could otherwise come in and zero beyond i_size.
1142 */
1149 /*
1150 * Don't mark the inode dirty under page lock. First, it unnecessarily
1151 * makes the holding time of page lock longer. Second, it forces lock
1152 * ordering of page lock and transaction start for journaling
1153 * filesystems.
1154 */
1159 /* if we have allocated more blocks and copied
1160 * less. We will have blocks allocated outside
1161 * inode->i_size. So truncate them
1162 */
1164 errout:
1171 /*
1172 * If truncate failed early the inode might still be
1173 * on the orphan list; we need to make sure the inode
1174 * is removed from the orphan list in that case.
1175 */
1178 }
1181 }
1187 {
1210 }
1216 }
1230 }
1233 /* if we have allocated more blocks and copied
1234 * less. We will have blocks allocated outside
1235 * inode->i_size. So truncate them
1236 */
1244 /*
1245 * If truncate failed early the inode might still be
1246 * on the orphan list; we need to make sure the inode
1247 * is removed from the orphan list in that case.
1248 */
1251 }
1254 }
1256 /*
1257 * Reserve space for a single cluster
1258 */
1260 {
1265 /*
1266 * We will charge metadata quota at writeout time; this saves
1267 * us from metadata over-estimation, though we may go over by
1268 * a small amount in the end. Here we just reserve for data.
1269 */
1279 }
1285 }
1288 {
1299 /*
1300 * if there aren't enough reserved blocks, then the
1301 * counter is messed up somewhere. Since this
1302 * function is called from invalidate page, it's
1303 * harmless to return without any action.
1304 */
1306 "ino %lu, to_free %d with only %d reserved "
1311 }
1314 /* update fs dirty data blocks counter */
1320 }
1325 {
1333 ext4_fsblk_t lblk;
1346 to_release++;
1347 contiguous_blks++;
1353 contiguous_blks;
1356 }
1364 }
1366 /* If we have released all the blocks belonging to a cluster, then we
1367 * need to release the reserved space for that cluster. */
1376 num_clusters--;
1377 }
1378 }
1380 /*
1381 * Delayed allocation stuff
1382 */
1391 /*
1392 * Extent to map - this can be after first_page because that can be
1393 * fully mapped. We somewhat abuse m_flags to store whether the extent
1394 * is delalloc or unwritten.
1395 */
1398 };
1402 {
1409 /* This is necessary when next_page == 0. */
1420 }
1436 }
1438 }
1441 }
1442 }
1445 {
1464 }
1467 {
1469 }
1471 /*
1472 * This function is grabs code from the very beginning of
1473 * ext4_map_blocks, but assumes that the caller is from delayed write
1474 * time. This function looks up the requested blocks and sets the
1475 * buffer delay bit under the protection of i_data_sem.
1476 */
1480 {
1484 #ifdef ES_AGGRESSIVE_TEST
1488 #endif
1498 /* Lookup extent status tree firstly */
1504 }
1506 /*
1507 * Delayed extent could be allocated by fallocate.
1508 * So we need to check it.
1509 */
1515 }
1526 else
1529 #ifdef ES_AGGRESSIVE_TEST
1531 #endif
1533 }
1535 /*
1536 * Try to see if we can get the block without requesting a new
1537 * file system block.
1538 */
1544 else
1547 add_delayed:
1550 /*
1551 * XXX: __block_prepare_write() unmaps passed block,
1552 * is it OK?
1553 */
1554 /*
1555 * If the block was allocated from previously allocated cluster,
1556 * then we don't need to reserve it again. However we still need
1557 * to reserve metadata for every block we're going to write.
1558 */
1563 /* not enough space to reserve */
1566 }
1567 }
1574 }
1585 "ES len assertion failed for inode "
1589 }
1597 }
1599 out_unlock:
1603 }
1605 /*
1606 * This is a special get_block_t callback which is used by
1607 * ext4_da_write_begin(). It will either return mapped block or
1608 * reserve space for a single block.
1609 *
1610 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1611 * We also have b_blocknr = -1 and b_bdev initialized properly
1612 *
1613 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1614 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1615 * initialized properly.
1616 */
1619 {
1629 /*
1630 * first, we need to know whether the block is allocated already
1631 * preallocated blocks are unmapped but should treated
1632 * the same as allocated blocks.
1633 */
1642 /* A delayed write to unwritten bh should be marked
1643 * new and mapped. Mapped ensures that we don't do
1644 * get_block multiple times when we write to the same
1645 * offset and new ensures that we do proper zero out
1646 * for partial write.
1647 */
1650 }
1652 }
1655 {
1658 }
1661 {
1664 }
1668 {
1690 }
1693 }
1694 /*
1695 * We need to release the page lock before we start the
1696 * journal, so grab a reference so the page won't disappear
1697 * out from under us.
1698 */
1708 }
1714 /* The page got truncated from under us */
1718 }
1728 do_journal_get_write_access);
1731 write_end_fn);
1732 }
1744 out:
1746 out_no_pagelock:
1749 }
1751 /*
1752 * Note that we don't need to start a transaction unless we're journaling data
1753 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1754 * need to file the inode to the transaction's list in ordered mode because if
1755 * we are writing back data added by write(), the inode is already there and if
1756 * we are writing back data modified via mmap(), no one guarantees in which
1757 * transaction the data will hit the disk. In case we are journaling data, we
1758 * cannot start transaction directly because transaction start ranks above page
1759 * lock so we have to do some magic.
1760 *
1761 * This function can get called via...
1762 * - ext4_writepages after taking page lock (have journal handle)
1763 * - journal_submit_inode_data_buffers (no journal handle)
1764 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1765 * - grab_page_cache when doing write_begin (have journal handle)
1766 *
1767 * We don't do any block allocation in this function. If we have page with
1768 * multiple blocks we need to write those buffer_heads that are mapped. This
1769 * is important for mmaped based write. So if we do with blocksize 1K
1770 * truncate(f, 1024);
1771 * a = mmap(f, 0, 4096);
1772 * a[0] = 'a';
1773 * truncate(f, 4096);
1774 * we have in the page first buffer_head mapped via page_mkwrite call back
1775 * but other buffer_heads would be unmapped but dirty (dirty done via the
1776 * do_wp_page). So writepage should write the first block. If we modify
1777 * the mmap area beyond 1024 we will again get a page_fault and the
1778 * page_mkwrite callback will do the block allocation and mark the
1779 * buffer_heads mapped.
1780 *
1781 * We redirty the page if we have any buffer_heads that is either delay or
1782 * unwritten in the page.
1783 *
1784 * We can get recursively called as show below.
1785 *
1786 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1787 * ext4_writepage()
1788 *
1789 * But since we don't do any block allocation we should not deadlock.
1790 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1791 */
1794 {
1796 loff_t size;
1807 else
1811 /*
1812 * We cannot do block allocation or other extent handling in this
1813 * function. If there are buffers needing that, we have to redirty
1814 * the page. But we may reach here when we do a journal commit via
1815 * journal_submit_inode_data_buffers() and in that case we must write
1816 * allocated buffers to achieve data=ordered mode guarantees.
1817 */
1819 ext4_bh_delay_or_unwritten)) {
1822 /*
1823 * For memory cleaning there's no point in writing only
1824 * some buffers. So just bail out. Warn if we came here
1825 * from direct reclaim.
1826 */
1831 }
1833 }
1836 /*
1837 * It's mmapped pagecache. Add buffers and journal it. There
1838 * doesn't seem much point in redirtying the page here.
1839 */
1848 }
1851 /* Drop io_end reference we got from init */
1854 }
1857 {
1865 else
1874 }
1876 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1878 /*
1879 * mballoc gives us at most this number of blocks...
1880 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1881 * The rest of mballoc seems to handle chunks up to full group size.
1882 */
1883 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1885 /*
1886 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1887 *
1888 * @mpd - extent of blocks
1889 * @lblk - logical number of the block in the file
1890 * @bh - buffer head we want to add to the extent
1891 *
1892 * The function is used to collect contig. blocks in the same state. If the
1893 * buffer doesn't require mapping for writeback and we haven't started the
1894 * extent of buffers to map yet, the function returns 'true' immediately - the
1895 * caller can write the buffer right away. Otherwise the function returns true
1896 * if the block has been added to the extent, false if the block couldn't be
1897 * added.
1898 */
1901 {
1904 /* Buffer that doesn't need mapping for writeback? */
1907 /* So far no extent to map => we write the buffer right away */
1911 }
1913 /* First block in the extent? */
1919 }
1921 /* Don't go larger than mballoc is willing to allocate */
1925 /* Can we merge the block to our big extent? */
1930 }
1932 }
1934 /*
1935 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1936 *
1937 * @mpd - extent of blocks for mapping
1938 * @head - the first buffer in the page
1939 * @bh - buffer we should start processing from
1940 * @lblk - logical number of the block in the file corresponding to @bh
1941 *
1942 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1943 * the page for IO if all buffers in this page were mapped and there's no
1944 * accumulated extent of buffers to map or add buffers in the page to the
1945 * extent of buffers to map. The function returns 1 if the caller can continue
1946 * by processing the next page, 0 if it should stop adding buffers to the
1947 * extent to map because we cannot extend it anymore. It can also return value
1948 * < 0 in case of error during IO submission.
1949 */
1953 ext4_lblk_t lblk)
1954 {
1964 /* Found extent to map? */
1967 /* Everything mapped so far and we hit EOF */
1969 }
1971 /* So far everything mapped? Submit the page for IO. */
1976 }
1978 }
1980 /*
1981 * mpage_map_buffers - update buffers corresponding to changed extent and
1982 * submit fully mapped pages for IO
1983 *
1984 * @mpd - description of extent to map, on return next extent to map
1985 *
1986 * Scan buffers corresponding to changed extent (we expect corresponding pages
1987 * to be already locked) and update buffer state according to new extent state.
1988 * We map delalloc buffers to their physical location, clear unwritten bits,
1989 * and mark buffers as uninit when we perform writes to unwritten extents
1990 * and do extent conversion after IO is finished. If the last page is not fully
1991 * mapped, we update @map to the next extent in the last page that needs
1992 * mapping. Otherwise we submit the page for IO.
1993 */
1995 {
2002 ext4_lblk_t lblk;
2003 sector_t pblock;
2014 PAGEVEC_SIZE);
2022 /* Up to 'end' pages must be contiguous */
2029 /*
2030 * Buffer after end of mapped extent.
2031 * Find next buffer in the page to map.
2032 */
2035 /*
2036 * FIXME: If dioread_nolock supports
2037 * blocksize < pagesize, we need to make
2038 * sure we add size mapped so far to
2039 * io_end->size as the following call
2040 * can submit the page for IO.
2041 */
2048 }
2052 }
2056 /*
2057 * FIXME: This is going to break if dioread_nolock
2058 * supports blocksize < pagesize as we will try to
2059 * convert potentially unmapped parts of inode.
2060 */
2062 /* Page fully mapped - let IO run! */
2067 }
2068 start++;
2069 }
2071 }
2072 /* Extent fully mapped and matches with page boundary. We are done. */
2076 }
2079 {
2086 /*
2087 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2088 * to convert an unwritten extent to be initialized (in the case
2089 * where we have written into one or more preallocated blocks). It is
2090 * possible that we're going to need more metadata blocks than
2091 * previously reserved. However we must not fail because we're in
2092 * writeback and there is nothing we can do about it so it might result
2093 * in data loss. So use reserved blocks to allocate metadata if
2094 * possible.
2095 *
2096 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2097 * the blocks in question are delalloc blocks. This indicates
2098 * that the blocks and quotas has already been checked when
2099 * the data was copied into the page cache.
2100 */
2102 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2117 }
2119 }
2128 }
2130 }
2132 /*
2133 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2134 * mpd->len and submit pages underlying it for IO
2135 *
2136 * @handle - handle for journal operations
2137 * @mpd - extent to map
2138 * @give_up_on_write - we set this to true iff there is a fatal error and there
2139 * is no hope of writing the data. The caller should discard
2140 * dirty pages to avoid infinite loops.
2141 *
2142 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2143 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2144 * them to initialized or split the described range from larger unwritten
2145 * extent. Note that we need not map all the described range since allocation
2146 * can return less blocks or the range is covered by more unwritten extents. We
2147 * cannot map more because we are limited by reserved transaction credits. On
2148 * the other hand we always make sure that the last touched page is fully
2149 * mapped so that it can be written out (and thus forward progress is
2150 * guaranteed). After mapping we submit all mapped pages for IO.
2151 */
2155 {
2159 loff_t disksize;
2171 /*
2172 * Let the uper layers retry transient errors.
2173 * In the case of ENOSPC, if ext4_count_free_blocks()
2174 * is non-zero, a commit should free up blocks.
2175 */
2181 }
2183 "Delayed block allocation failed for "
2184 "inode %lu at logical offset %llu with"
2190 "This should not happen!! Data will "
2194 invalidate_dirty_pages:
2197 }
2199 /*
2200 * Update buffer state, submit mapped pages, and get us new
2201 * extent to map
2202 */
2208 update_disksize:
2209 /*
2210 * Update on-disk size after IO is submitted. Races with
2211 * truncate are avoided by checking i_size under i_data_sem.
2212 */
2216 loff_t i_size;
2232 }
2234 }
2236 /*
2237 * Calculate the total number of credits to reserve for one writepages
2238 * iteration. This is called from ext4_writepages(). We map an extent of
2239 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2240 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2241 * bpp - 1 blocks in bpp different extents.
2242 */
2244 {
2249 }
2251 /*
2252 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2253 * and underlying extent to map
2254 *
2255 * @mpd - where to look for pages
2256 *
2257 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2258 * IO immediately. When we find a page which isn't mapped we start accumulating
2259 * extent of buffers underlying these pages that needs mapping (formed by
2260 * either delayed or unwritten buffers). We also lock the pages containing
2261 * these buffers. The extent found is returned in @mpd structure (starting at
2262 * mpd->lblk with length mpd->len blocks).
2263 *
2264 * Note that this function can attach bios to one io_end structure which are
2265 * neither logically nor physically contiguous. Although it may seem as an
2266 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2267 * case as we need to track IO to all buffers underlying a page in one io_end.
2268 */
2270 {
2280 ext4_lblk_t lblk;
2285 else
2300 /*
2301 * At this point, the page may be truncated or
2302 * invalidated (changing page->mapping to NULL), or
2303 * even swizzled back from swapper_space to tmpfs file
2304 * mapping. However, page->index will not change
2305 * because we have a reference on the page.
2306 */
2310 /*
2311 * Accumulated enough dirty pages? This doesn't apply
2312 * to WB_SYNC_ALL mode. For integrity sync we have to
2313 * keep going because someone may be concurrently
2314 * dirtying pages, and we might have synced a lot of
2315 * newly appeared dirty pages, but have not synced all
2316 * of the old dirty pages.
2317 */
2321 /* If we can't merge this page, we are done. */
2326 /*
2327 * If the page is no longer dirty, or its mapping no
2328 * longer corresponds to inode we are writing (which
2329 * means it has been truncated or invalidated), or the
2330 * page is already under writeback and we are not doing
2331 * a data integrity writeback, skip the page
2332 */
2339 }
2347 /* Add all dirty buffers to mpd */
2355 left--;
2356 }
2359 }
2361 out:
2364 }
2368 {
2373 }
2377 {
2393 /*
2394 * No pages to write? This is mainly a kludge to avoid starting
2395 * a transaction for special inodes like journal inode on last iput()
2396 * because that could violate lock ordering on umount
2397 */
2408 }
2410 /*
2411 * If the filesystem has aborted, it is read-only, so return
2412 * right away instead of dumping stack traces later on that
2413 * will obscure the real source of the problem. We test
2414 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2415 * the latter could be true if the filesystem is mounted
2416 * read-only, and in that case, ext4_writepages should
2417 * *never* be called, so if that ever happens, we would want
2418 * the stack trace.
2419 */
2423 }
2426 /*
2427 * We may need to convert up to one extent per block in
2428 * the page and we may dirty the inode.
2429 */
2431 }
2433 /*
2434 * If we have inline data and arrive here, it means that
2435 * we will soon create the block for the 1st page, so
2436 * we'd better clear the inline data here.
2437 */
2439 /* Just inode will be modified... */
2444 }
2446 EXT4_STATE_MAY_INLINE_DATA));
2449 }
2463 }
2468 retry:
2474 /* For each extent of pages we use new io_end */
2479 }
2481 /*
2482 * We have two constraints: We find one extent to map and we
2483 * must always write out whole page (makes a difference when
2484 * blocksize < pagesize) so that we don't block on IO when we
2485 * try to write out the rest of the page. Journalled mode is
2486 * not supported by delalloc.
2487 */
2491 /* start a new transaction */
2499 /* Release allocated io_end */
2502 }
2511 /*
2512 * We scanned the whole range (or exhausted
2513 * nr_to_write), submitted what was mapped and
2514 * didn't find anything needing mapping. We are
2515 * done.
2516 */
2518 }
2519 }
2521 /* Submit prepared bio */
2523 /* Unlock pages we didn't use */
2525 /* Drop our io_end reference we got from init */
2529 /*
2530 * Commit the transaction which would
2531 * free blocks released in the transaction
2532 * and try again
2533 */
2537 }
2538 /* Fatal error - ENOMEM, EIO... */
2541 }
2548 }
2550 /* Update index */
2552 /*
2553 * Set the writeback_index so that range_cyclic
2554 * mode will write it back later
2555 */
2558 out_writepages:
2562 }
2565 {
2569 /*
2570 * switch to non delalloc mode if we are running low
2571 * on free block. The free block accounting via percpu
2572 * counters can get slightly wrong with percpu_counter_batch getting
2573 * accumulated on each CPU without updating global counters
2574 * Delalloc need an accurate free block accounting. So switch
2575 * to non delalloc when we are near to error range.
2576 */
2577 free_clusters =
2579 dirty_clusters =
2581 /*
2582 * Start pushing delalloc when 1/2 of free blocks are dirty.
2583 */
2589 /*
2590 * free block count is less than 150% of dirty blocks
2591 * or free blocks is less than watermark
2592 */
2594 }
2596 }
2598 /* We always reserve for an inode update; the superblock could be there too */
2600 {
2602 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2608 /* We might need to update the superblock to set LARGE_FILE */
2610 }
2615 {
2618 pgoff_t index;
2628 }
2640 }
2642 /*
2643 * grab_cache_page_write_begin() can take a long time if the
2644 * system is thrashing due to memory pressure, or if the page
2645 * is being written back. So grab it first before we start
2646 * the transaction handle. This also allows us to allocate
2647 * the page (if needed) without using GFP_NOFS.
2648 */
2649 retry_grab:
2655 /*
2656 * With delayed allocation, we don't log the i_disksize update
2657 * if there is delayed block allocation. But we still need
2658 * to journalling the i_disksize update if writes to the end
2659 * of file which has an already mapped buffer.
2660 */
2661 retry_journal:
2667 }
2671 /* The page got truncated from under us */
2676 }
2677 /* In case writeback began while the page was unlocked */
2680 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2682 ext4_da_get_block_prep);
2683 #else
2685 #endif
2689 /*
2690 * block_write_begin may have instantiated a few blocks
2691 * outside i_size. Trim these off again. Don't need
2692 * i_size_read because we hold i_mutex.
2693 */
2703 }
2707 }
2709 /*
2710 * Check if we should update i_disksize
2711 * when write to the end of file but not require block allocation
2712 */
2715 {
2730 }
2736 {
2740 loff_t new_i_size;
2752 /*
2753 * generic_write_end() will run mark_inode_dirty() if i_size
2754 * changes. So let's piggyback the i_disksize mark_inode_dirty
2755 * into that.
2756 */
2762 /* We need to mark inode dirty even if
2763 * new_i_size is less that inode->i_size
2764 * bu greater than i_disksize.(hint delalloc)
2765 */
2767 }
2768 }
2774 page);
2775 else
2787 }
2791 {
2792 /*
2793 * Drop reserved blocks
2794 */
2801 out:
2805 }
2807 /*
2808 * Force all delayed allocation blocks to be allocated for a given inode.
2809 */
2811 {
2817 /*
2818 * We do something simple for now. The filemap_flush() will
2819 * also start triggering a write of the data blocks, which is
2820 * not strictly speaking necessary (and for users of
2821 * laptop_mode, not even desirable). However, to do otherwise
2822 * would require replicating code paths in:
2823 *
2824 * ext4_writepages() ->
2825 * write_cache_pages() ---> (via passed in callback function)
2826 * __mpage_da_writepage() -->
2827 * mpage_add_bh_to_extent()
2828 * mpage_da_map_blocks()
2829 *
2830 * The problem is that write_cache_pages(), located in
2831 * mm/page-writeback.c, marks pages clean in preparation for
2832 * doing I/O, which is not desirable if we're not planning on
2833 * doing I/O at all.
2834 *
2835 * We could call write_cache_pages(), and then redirty all of
2836 * the pages by calling redirty_page_for_writepage() but that
2837 * would be ugly in the extreme. So instead we would need to
2838 * replicate parts of the code in the above functions,
2839 * simplifying them because we wouldn't actually intend to
2840 * write out the pages, but rather only collect contiguous
2841 * logical block extents, call the multi-block allocator, and
2842 * then update the buffer heads with the block allocations.
2843 *
2844 * For now, though, we'll cheat by calling filemap_flush(),
2845 * which will map the blocks, and start the I/O, but not
2846 * actually wait for the I/O to complete.
2847 */
2849 }
2851 /*
2852 * bmap() is special. It gets used by applications such as lilo and by
2853 * the swapper to find the on-disk block of a specific piece of data.
2854 *
2855 * Naturally, this is dangerous if the block concerned is still in the
2856 * journal. If somebody makes a swapfile on an ext4 data-journaling
2857 * filesystem and enables swap, then they may get a nasty shock when the
2858 * data getting swapped to that swapfile suddenly gets overwritten by
2859 * the original zero's written out previously to the journal and
2860 * awaiting writeback in the kernel's buffer cache.
2861 *
2862 * So, if we see any bmap calls here on a modified, data-journaled file,
2863 * take extra steps to flush any blocks which might be in the cache.
2864 */
2866 {
2871 /*
2872 * We can get here for an inline file via the FIBMAP ioctl
2873 */
2879 /*
2880 * With delalloc we want to sync the file
2881 * so that we can make sure we allocate
2882 * blocks for file
2883 */
2885 }
2889 /*
2890 * This is a REALLY heavyweight approach, but the use of
2891 * bmap on dirty files is expected to be extremely rare:
2892 * only if we run lilo or swapon on a freshly made file
2893 * do we expect this to happen.
2894 *
2895 * (bmap requires CAP_SYS_RAWIO so this does not
2896 * represent an unprivileged user DOS attack --- we'd be
2897 * in trouble if mortal users could trigger this path at
2898 * will.)
2899 *
2900 * NB. EXT4_STATE_JDATA is not set on files other than
2901 * regular files. If somebody wants to bmap a directory
2902 * or symlink and gets confused because the buffer
2903 * hasn't yet been flushed to disk, they deserve
2904 * everything they get.
2905 */
2915 }
2918 }
2921 {
2934 }
2936 static int
2939 {
2942 /* If the file has inline data, no need to do readpages. */
2947 }
2951 {
2954 /* No journalling happens on data buffers when this function is used */
2958 }
2963 {
2968 /*
2969 * If it's a full truncate we just forget about the pending dirtying
2970 */
2975 }
2977 /* Wrapper for aops... */
2981 {
2983 }
2986 {
2991 /* Page has dirty journalled data -> cannot release */
2996 else
2998 }
3000 /*
3001 * ext4_get_block used when preparing for a DIO write or buffer write.
3002 * We allocate an uinitialized extent if blocks haven't been allocated.
3003 * The extent will be converted to initialized after the IO is complete.
3004 */
3007 {
3011 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3012 }
3016 {
3020 EXT4_GET_BLOCKS_NO_LOCK);
3021 }
3025 {
3028 /* if not async direct IO just return */
3035 size);
3041 }
3043 /*
3044 * For ext4 extent files, ext4 will do direct-io write to holes,
3045 * preallocated extents, and those write extend the file, no need to
3046 * fall back to buffered IO.
3047 *
3048 * For holes, we fallocate those blocks, mark them as unwritten
3049 * If those blocks were preallocated, we mark sure they are split, but
3050 * still keep the range to write as unwritten.
3051 *
3052 * The unwritten extents will be converted to written when DIO is completed.
3053 * For async direct IO, since the IO may still pending when return, we
3054 * set up an end_io call back function, which will do the conversion
3055 * when async direct IO completed.
3056 *
3057 * If the O_DIRECT write will extend the file then add this inode to the
3058 * orphan list. So recovery will truncate it back to the original size
3059 * if the machine crashes during the write.
3060 *
3061 */
3063 loff_t offset)
3064 {
3067 ssize_t ret;
3075 /* Use the old path for reads and writes beyond i_size. */
3081 /*
3082 * Make all waiters for direct IO properly wait also for extent
3083 * conversion. This also disallows race between truncate() and
3084 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3085 */
3089 /* If we do a overwrite dio, i_mutex locking can be released */
3095 }
3097 /*
3098 * We could direct write to holes and fallocate.
3099 *
3100 * Allocated blocks to fill the hole are marked as
3101 * unwritten to prevent parallel buffered read to expose
3102 * the stale data before DIO complete the data IO.
3103 *
3104 * As to previously fallocated extents, ext4 get_block will
3105 * just simply mark the buffer mapped but still keep the
3106 * extents unwritten.
3107 *
3108 * For non AIO case, we will convert those unwritten extents
3109 * to written after return back from blockdev_direct_IO.
3110 *
3111 * For async DIO, the conversion needs to be deferred when the
3112 * IO is completed. The ext4 end_io callback function will be
3113 * called to take care of the conversion work. Here for async
3114 * case, we allocate an io_end structure to hook to the iocb.
3115 */
3123 }
3124 /*
3125 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3126 */
3128 /*
3129 * we save the io structure for current async direct
3130 * IO, so that later ext4_map_blocks() could flag the
3131 * io structure whether there is a unwritten extents
3132 * needs to be converted when IO is completed.
3133 */
3135 }
3142 }
3143 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3145 #endif
3149 else
3152 get_block_func,
3155 /*
3156 * Put our reference to io_end. This can free the io_end structure e.g.
3157 * in sync IO case or in case of error. It can even perform extent
3158 * conversion if all bios we submitted finished before we got here.
3159 * Note that in that case iocb->private can be already set to NULL
3160 * here.
3161 */
3165 /*
3166 * When no IO was submitted ext4_end_io_dio() was not
3167 * called so we have to put iocb's reference.
3168 */
3174 }
3175 }
3177 EXT4_STATE_DIO_UNWRITTEN)) {
3179 /*
3180 * for non AIO case, since the IO is already
3181 * completed, we could do the conversion right here
3182 */
3188 }
3190 retake_lock:
3193 /* take i_mutex locking again if we do a ovewrite dio */
3197 }
3200 }
3203 loff_t offset)
3204 {
3208 ssize_t ret;
3210 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3213 #endif
3215 /*
3216 * If we are doing data journalling we don't support O_DIRECT
3217 */
3221 /* Let buffer I/O handle the inline data case. */
3228 else
3232 }
3234 /*
3235 * Pages can be marked dirty completely asynchronously from ext4's journalling
3236 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3237 * much here because ->set_page_dirty is called under VFS locks. The page is
3238 * not necessarily locked.
3239 *
3240 * We cannot just dirty the page and leave attached buffers clean, because the
3241 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3242 * or jbddirty because all the journalling code will explode.
3243 *
3244 * So what we do is to mark the page "pending dirty" and next time writepage
3245 * is called, propagate that into the buffers appropriately.
3246 */
3248 {
3251 }
3267 };
3283 };
3299 };
3302 {
3315 }
3318 else
3320 }
3324 {
3328 ext4_lblk_t iblock;
3346 /* Find the buffer that contains "offset" */
3351 iblock++;
3353 }
3357 }
3361 /* unmapped? It's a hole - nothing to do */
3365 }
3366 }
3368 /* Ok, it's mapped. Make sure it's up-to-date */
3376 /* Uhhuh. Read error. Complain and punt. */
3381 /* We expect the key to be set. */
3385 }
3386 }
3392 }
3403 }
3405 unlock:
3409 }
3411 /*
3412 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3413 * starting from file offset 'from'. The range to be zero'd must
3414 * be contained with in one block. If the specified range exceeds
3415 * the end of the block it will be shortened to end of the block
3416 * that cooresponds to 'from'
3417 */
3420 {
3426 /*
3427 * correct length if it does not fall between
3428 * 'from' and the end of the block
3429 */
3436 }
3438 /*
3439 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3440 * up to the end of the block which corresponds to `from'.
3441 * This required during truncate. We need to physically zero the tail end
3442 * of that block so it doesn't yield old data if the file is later grown.
3443 */
3446 {
3456 }
3460 {
3474 /* Handle partial zero within the single block */
3480 }
3481 /* Handle partial zero out on the start of the range */
3487 }
3488 /* Handle partial zero out on the end of the range */
3494 }
3497 {
3505 }
3507 /*
3508 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3509 * associated with the given offset and length
3510 *
3511 * @inode: File inode
3512 * @offset: The offset where the hole will begin
3513 * @len: The length of the hole
3514 *
3515 * Returns: 0 on success or negative on failure
3516 */
3519 {
3533 /*
3534 * Write out all dirty pages to avoid race conditions
3535 * Then release them.
3536 */
3542 }
3546 /* No need to punch hole beyond i_size */
3550 /*
3551 * If the hole extends beyond i_size, set the hole
3552 * to end after the page that contains i_size
3553 */
3557 offset;
3558 }
3562 /*
3563 * Attach jinode to inode for jbd2 if we do any zeroing of
3564 * partial block
3565 */
3570 }
3575 /* Now release the pages and zero block aligned part of pages*/
3578 last_block_offset);
3580 /* Wait all existing dio workers, newcomers will block on i_mutex */
3586 else
3593 }
3596 length);
3604 /* If there are no blocks to remove, return now */
3616 }
3621 else
3623 stop_block);
3629 /* Now release the pages again to reduce race window */
3632 last_block_offset);
3636 out_stop:
3638 out_dio:
3640 out_mutex:
3643 }
3646 {
3659 }
3663 }
3668 }
3670 /*
3671 * ext4_truncate()
3672 *
3673 * We block out ext4_get_block() block instantiations across the entire
3674 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3675 * simultaneously on behalf of the same inode.
3676 *
3677 * As we work through the truncate and commit bits of it to the journal there
3678 * is one core, guiding principle: the file's tree must always be consistent on
3679 * disk. We must be able to restart the truncate after a crash.
3680 *
3681 * The file's tree may be transiently inconsistent in memory (although it
3682 * probably isn't), but whenever we close off and commit a journal transaction,
3683 * the contents of (the filesystem + the journal) must be consistent and
3684 * restartable. It's pretty simple, really: bottom up, right to left (although
3685 * left-to-right works OK too).
3686 *
3687 * Note that at recovery time, journal replay occurs *before* the restart of
3688 * truncate against the orphan inode list.
3689 *
3690 * The committed inode has the new, desired i_size (which is the same as
3691 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3692 * that this inode's truncate did not complete and it will again call
3693 * ext4_truncate() to have another go. So there will be instantiated blocks
3694 * to the right of the truncation point in a crashed ext4 filesystem. But
3695 * that's fine - as long as they are linked from the inode, the post-crash
3696 * ext4_truncate() run will find them and release them.
3697 */
3699 {
3705 /*
3706 * There is a possibility that we're either freeing the inode
3707 * or it's a completely new inode. In those cases we might not
3708 * have i_mutex locked because it's not necessary.
3709 */
3728 }
3730 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3734 }
3738 else
3745 }
3750 /*
3751 * We add the inode to the orphan list, so that if this
3752 * truncate spans multiple transactions, and we crash, we will
3753 * resume the truncate when the filesystem recovers. It also
3754 * marks the inode dirty, to catch the new size.
3755 *
3756 * Implication: the file must always be in a sane, consistent
3757 * truncatable state while each transaction commits.
3758 */
3768 else
3776 out_stop:
3777 /*
3778 * If this was a simple ftruncate() and the file will remain alive,
3779 * then we need to clear up the orphan record which we created above.
3780 * However, if this was a real unlink then we were called by
3781 * ext4_evict_inode(), and we allow that function to clean up the
3782 * orphan info for us.
3783 */
3792 }
3794 /*
3795 * ext4_get_inode_loc returns with an extra refcount against the inode's
3796 * underlying buffer_head on success. If 'in_mem' is true, we have all
3797 * data in memory that is needed to recreate the on-disk version of this
3798 * inode.
3799 */
3802 {
3806 ext4_fsblk_t block;
3818 /*
3819 * Figure out the offset within the block group inode table
3820 */
3833 /*
3834 * If the buffer has the write error flag, we have failed
3835 * to write out another inode in the same block. In this
3836 * case, we don't have to read the block because we may
3837 * read the old inode data successfully.
3838 */
3843 /* someone brought it uptodate while we waited */
3846 }
3848 /*
3849 * If we have all information of the inode in memory and this
3850 * is the only valid inode in the block, we need not read the
3851 * block.
3852 */
3859 /* Is the inode bitmap in cache? */
3864 /*
3865 * If the inode bitmap isn't in cache then the
3866 * optimisation may end up performing two reads instead
3867 * of one, so skip it.
3868 */
3872 }
3878 }
3881 /* all other inodes are free, so skip I/O */
3886 }
3887 }
3889 make_io:
3890 /*
3891 * If we need to do any I/O, try to pre-readahead extra
3892 * blocks from the inode table.
3893 */
3900 /* s_inode_readahead_blks is always a power of 2 */
3913 }
3915 /*
3916 * There are other valid inodes in the buffer, this inode
3917 * has in-inode xattrs, or we don't have this inode in memory.
3918 * Read the block from disk.
3919 */
3930 }
3931 }
3932 has_buffer:
3935 }
3938 {
3939 /* We have all inode data except xattrs in memory here. */
3942 }
3945 {
3963 }
3965 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3967 {
3976 EXT4_DIRSYNC_FL);
3988 }
3992 {
3993 blkcnt_t i_blocks ;
3998 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3999 /* we are using combined 48 bit field */
4003 /* i_blocks represent file system block size */
4007 }
4010 }
4011 }
4016 {
4024 }
4027 {
4035 uid_t i_uid;
4036 gid_t i_gid;
4061 }
4065 /* Precompute checksum seed for inode metadata */
4068 __u32 csum;
4075 }
4081 }
4089 }
4098 /* We now have enough fields to check if the inode was active or not.
4099 * This is needed because nfsd might try to access dead inodes
4100 * the test is that same one that e2fsck uses
4101 * NeilBrown 1999oct15
4102 */
4107 /* this inode is deleted */
4110 }
4111 /* The only unlinked inodes we let through here have
4112 * valid i_mode and are being read by the orphan
4113 * recovery code: that's fine, we're about to complete
4114 * the process of deleting those.
4115 * OR it is the EXT4_BOOT_LOADER_INO which is
4116 * not initialized on a new filesystem. */
4117 }
4126 #ifdef CONFIG_QUOTA
4128 #endif
4132 /*
4133 * NOTE! The in-memory inode i_data array is in little-endian order
4134 * even on big-endian machines: we do NOT byteswap the block numbers!
4135 */
4140 /*
4141 * Set transaction id's of transactions that have to be committed
4142 * to finish f[data]sync. We set them to currently running transaction
4143 * as we cannot be sure that the inode or some of its metadata isn't
4144 * part of the transaction - the inode could have been reclaimed and
4145 * now it is reread from disk.
4146 */
4149 tid_t tid;
4154 else
4158 else
4163 }
4167 /* The extra space is currently unused. Use it. */
4169 EXT4_GOOD_OLD_INODE_SIZE;
4172 }
4173 }
4186 }
4187 }
4201 /* Validate extent which is part of inode */
4206 /* Validate block references which are part of inode */
4208 }
4209 }
4232 }
4239 else
4248 }
4254 bad_inode:
4258 }
4261 {
4265 }
4270 {
4276 /*
4277 * i_blocks can be represented in a 32 bit variable
4278 * as multiple of 512 bytes
4279 */
4284 }
4289 /*
4290 * i_blocks can be represented in a 48 bit variable
4291 * as multiple of 512 bytes
4292 */
4298 /* i_block is stored in file system block size */
4302 }
4304 }
4309 };
4313 {
4338 }
4341 }
4343 /*
4344 * Opportunistically update the other time fields for other inodes in
4345 * the same inode table block.
4346 */
4349 {
4356 /*
4357 * Calculate the first inode in the inode table block. Inode
4358 * numbers are one-based. That is, the first inode in a block
4359 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4360 */
4367 }
4368 }
4370 /*
4371 * Post the struct inode info into an on-disk inode location in the
4372 * buffer-cache. This gobbles the caller's reference to the
4373 * buffer_head in the inode location struct.
4374 *
4375 * The caller must have write access to iloc->bh.
4376 */
4380 {
4387 uid_t i_uid;
4388 gid_t i_gid;
4392 /* For fields not tracked in the in-memory inode,
4393 * initialise them to zero for new inodes. */
4404 /*
4405 * Fix up interoperability with old kernels. Otherwise, old inodes get
4406 * re-used with the upper 16 bits of the uid/gid intact
4407 */
4416 }
4422 }
4434 }
4444 }
4447 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4451 }
4463 }
4467 }
4477 }
4478 }
4497 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4500 }
4502 out_brelse:
4506 }
4508 /*
4509 * ext4_write_inode()
4510 *
4511 * We are called from a few places:
4512 *
4513 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4514 * Here, there will be no transaction running. We wait for any running
4515 * transaction to commit.
4516 *
4517 * - Within flush work (sys_sync(), kupdate and such).
4518 * We wait on commit, if told to.
4519 *
4520 * - Within iput_final() -> write_inode_now()
4521 * We wait on commit, if told to.
4522 *
4523 * In all cases it is actually safe for us to return without doing anything,
4524 * because the inode has been copied into a raw inode buffer in
4525 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4526 * writeback.
4527 *
4528 * Note that we are absolutely dependent upon all inode dirtiers doing the
4529 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4530 * which we are interested.
4531 *
4532 * It would be a bug for them to not do this. The code:
4533 *
4534 * mark_inode_dirty(inode)
4535 * stuff();
4536 * inode->i_size = expr;
4537 *
4538 * is in error because write_inode() could occur while `stuff()' is running,
4539 * and the new i_size will be lost. Plus the inode will no longer be on the
4540 * superblock's dirty inode list.
4541 */
4543 {
4554 }
4556 /*
4557 * No need to force transaction in WB_SYNC_NONE mode. Also
4558 * ext4_sync_fs() will force the commit after everything is
4559 * written.
4560 */
4571 /*
4572 * sync(2) will flush the whole buffer cache. No need to do
4573 * it here separately for each inode.
4574 */
4581 }
4583 }
4585 }
4587 /*
4588 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4589 * buffers that are attached to a page stradding i_size and are undergoing
4590 * commit. In that case we have to wait for commit to finish and try again.
4591 */
4593 {
4601 /*
4602 * All buffers in the last page remain valid? Then there's nothing to
4603 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4604 * blocksize case
4605 */
4626 }
4627 }
4629 /*
4630 * ext4_setattr()
4631 *
4632 * Called from notify_change.
4633 *
4634 * We want to trap VFS attempts to truncate the file as soon as
4635 * possible. In particular, we want to make sure that when the VFS
4636 * shrinks i_size, we put the inode on the orphan list and modify
4637 * i_disksize immediately, so that during the subsequent flushing of
4638 * dirty pages and freeing of disk blocks, we can guarantee that any
4639 * commit will leave the blocks being flushed in an unused state on
4640 * disk. (On recovery, the inode will get truncated and the blocks will
4641 * be freed, so we have a strong guarantee that no future commit will
4642 * leave these blocks visible to the user.)
4643 *
4644 * Another thing we have to assure is that if we are in ordered mode
4645 * and inode is still attached to the committing transaction, we must
4646 * we start writeout of all the dirty pages which are being truncated.
4647 * This way we are sure that all the data written in the previous
4648 * transaction are already on disk (truncate waits for pages under
4649 * writeback).
4650 *
4651 * Called with inode->i_mutex down.
4652 */
4654 {
4670 /* (user+group)*(old+new) structure, inode write (sb,
4671 * inode block, ? - but truncate inode update has it) */
4678 }
4683 }
4684 /* Update corresponding info in inode so that everything is in
4685 * one transaction */
4692 }
4704 }
4717 }
4723 }
4727 }
4733 /*
4734 * We have to update i_size under i_data_sem together
4735 * with i_disksize to avoid races with writeback code
4736 * running ext4_wb_update_i_disksize().
4737 */
4746 }
4747 }
4751 /*
4752 * Blocks are going to be removed from the inode. Wait
4753 * for dio in flight. Temporarily disable
4754 * dioread_nolock to prevent livelock.
4755 */
4763 }
4764 /*
4765 * Truncate pagecache after we've waited for commit
4766 * in data=journal mode to make pages freeable.
4767 */
4771 }
4776 }
4778 /*
4779 * If the call to ext4_truncate failed to get a transaction handle at
4780 * all, we need to clean up the in-core orphan list manually.
4781 */
4788 err_out:
4793 }
4797 {
4804 /*
4805 * If there is inline data in the inode, the inode will normally not
4806 * have data blocks allocated (it may have an external xattr block).
4807 * Report at least one sector for such files, so tools like tar, rsync,
4808 * others doen't incorrectly think the file is completely sparse.
4809 */
4813 /*
4814 * We can't update i_blocks if the block allocation is delayed
4815 * otherwise in the case of system crash before the real block
4816 * allocation is done, we will have i_blocks inconsistent with
4817 * on-disk file blocks.
4818 * We always keep i_blocks updated together with real
4819 * allocation. But to not confuse with user, stat
4820 * will return the blocks that include the delayed allocation
4821 * blocks for this file.
4822 */
4827 }
4831 {
4835 }
4837 /*
4838 * Account for index blocks, block groups bitmaps and block group
4839 * descriptor blocks if modify datablocks and index blocks
4840 * worse case, the indexs blocks spread over different block groups
4841 *
4842 * If datablocks are discontiguous, they are possible to spread over
4843 * different block groups too. If they are contiguous, with flexbg,
4844 * they could still across block group boundary.
4845 *
4846 * Also account for superblock, inode, quota and xattr blocks
4847 */
4850 {
4856 /*
4857 * How many index blocks need to touch to map @lblocks logical blocks
4858 * to @pextents physical extents?
4859 */
4864 /*
4865 * Now let's see how many group bitmaps and group descriptors need
4866 * to account
4867 */
4875 /* bitmaps and block group descriptor blocks */
4878 /* Blocks for super block, inode, quota and xattr blocks */
4882 }
4884 /*
4885 * Calculate the total number of credits to reserve to fit
4886 * the modification of a single pages into a single transaction,
4887 * which may include multiple chunks of block allocations.
4888 *
4889 * This could be called via ext4_write_begin()
4890 *
4891 * We need to consider the worse case, when
4892 * one new block per extent.
4893 */
4895 {
4901 /* Account for data blocks for journalled mode */
4905 }
4907 /*
4908 * Calculate the journal credits for a chunk of data modification.
4909 *
4910 * This is called from DIO, fallocate or whoever calling
4911 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4912 *
4913 * journal buffers for data blocks are not included here, as DIO
4914 * and fallocate do no need to journal data buffers.
4915 */
4917 {
4919 }
4921 /*
4922 * The caller must have previously called ext4_reserve_inode_write().
4923 * Give this, we know that the caller already has write access to iloc->bh.
4924 */
4927 {
4933 /* the do_update_inode consumes one bh->b_count */
4936 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4940 }
4942 /*
4943 * On success, We end up with an outstanding reference count against
4944 * iloc->bh. This _must_ be cleaned up later.
4945 */
4947 int
4950 {
4960 }
4961 }
4964 }
4966 /*
4967 * Expand an inode by new_extra_isize bytes.
4968 * Returns 0 on success or negative error number on failure.
4969 */
4974 {
4985 /* No extended attributes present */
4989 new_extra_isize);
4992 }
4994 /* try to expand with EAs present */
4997 }
4999 /*
5000 * What we do here is to mark the in-core inode as clean with respect to inode
5001 * dirtiness (it may still be data-dirty).
5002 * This means that the in-core inode may be reaped by prune_icache
5003 * without having to perform any I/O. This is a very good thing,
5004 * because *any* task may call prune_icache - even ones which
5005 * have a transaction open against a different journal.
5006 *
5007 * Is this cheating? Not really. Sure, we haven't written the
5008 * inode out, but prune_icache isn't a user-visible syncing function.
5009 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5010 * we start and wait on commits.
5011 */
5013 {
5025 /*
5026 * We need extra buffer credits since we may write into EA block
5027 * with this same handle. If journal_extend fails, then it will
5028 * only result in a minor loss of functionality for that inode.
5029 * If this is felt to be critical, then e2fsck should be run to
5030 * force a large enough s_min_extra_isize.
5031 */
5039 EXT4_STATE_NO_EXPAND);
5043 "Unable to expand inode %lu. Delete"
5046 mnt_count =
5048 }
5049 }
5050 }
5051 }
5055 }
5057 /*
5058 * ext4_dirty_inode() is called from __mark_inode_dirty()
5059 *
5060 * We're really interested in the case where a file is being extended.
5061 * i_size has been changed by generic_commit_write() and we thus need
5062 * to include the updated inode in the current transaction.
5063 *
5064 * Also, dquot_alloc_block() will always dirty the inode when blocks
5065 * are allocated to the file.
5066 *
5067 * If the inode is marked synchronous, we don't honour that here - doing
5068 * so would cause a commit on atime updates, which we don't bother doing.
5069 * We handle synchronous inodes at the highest possible level.
5070 *
5071 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5072 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5073 * to copy into the on-disk inode structure are the timestamp files.
5074 */
5076 {
5088 out:
5090 }
5092 #if 0
5093 /*
5094 * Bind an inode's backing buffer_head into this transaction, to prevent
5095 * it from being flushed to disk early. Unlike
5096 * ext4_reserve_inode_write, this leaves behind no bh reference and
5097 * returns no iloc structure, so the caller needs to repeat the iloc
5098 * lookup to mark the inode dirty later.
5099 */
5101 {
5112 NULL,
5115 }
5116 }
5119 }
5120 #endif
5123 {
5128 /*
5129 * We have to be very careful here: changing a data block's
5130 * journaling status dynamically is dangerous. If we write a
5131 * data block to the journal, change the status and then delete
5132 * that block, we risk forgetting to revoke the old log record
5133 * from the journal and so a subsequent replay can corrupt data.
5134 * So, first we make sure that the journal is empty and that
5135 * nobody is changing anything.
5136 */
5143 /* We have to allocate physical blocks for delalloc blocks
5144 * before flushing journal. otherwise delalloc blocks can not
5145 * be allocated any more. even more truncate on delalloc blocks
5146 * could trigger BUG by flushing delalloc blocks in journal.
5147 * There is no delalloc block in non-journal data mode.
5148 */
5153 }
5155 /* Wait for all existing dio workers */
5161 /*
5162 * OK, there are no updates running now, and all cached data is
5163 * synced to disk. We are now in a completely consistent state
5164 * which doesn't have anything in the journal, and we know that
5165 * no filesystem updates are running, so it is safe to modify
5166 * the inode's in-core data-journaling state flag now.
5167 */
5177 }
5179 }
5185 /* Finally we can mark the inode as dirty. */
5197 }
5200 {
5202 }
5205 {
5207 loff_t size;
5219 /* Delalloc case is easy... */
5225 ext4_da_get_block_prep);
5229 }
5233 /* Page got truncated from under us? */
5238 }
5242 else
5244 /*
5245 * Return if we have all the buffers mapped. This avoids the need to do
5246 * journal_start/journal_stop which can block and take a long time
5247 */
5251 ext4_bh_unmapped)) {
5252 /* Wait so that we don't change page under IO */
5256 }
5257 }
5259 /* OK, we need to fill the hole... */
5262 else
5264 retry_alloc:
5270 }
5279 }
5281 }
5285 out_ret:
5287 out:
5290 }