| blob | b3bd912df6bfaf475f9304eba7fb1b9ce6368e87 |
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/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
52 {
54 __u16 csum_lo;
56 __u32 csum;
64 }
75 }
79 {
92 else
96 }
100 {
101 __u32 csum;
113 }
116 loff_t new_size)
117 {
119 /*
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
124 */
129 new_size);
130 }
139 /*
140 * Test whether an inode is a fast symlink.
141 */
143 {
151 }
153 /*
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
156 * this transaction.
157 */
160 {
163 /*
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
168 */
177 }
179 /*
180 * Called at the last iput() if i_nlink is zero.
181 */
183 {
190 /*
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
204 *
205 * Note that directories do not have this problem because they
206 * don't use page cache.
207 */
216 }
221 }
233 /*
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
236 */
242 /*
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
245 * cleaned up.
246 */
250 }
260 }
264 /*
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
269 */
277 stop_handle:
282 }
283 }
285 /*
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
292 */
296 /*
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
301 * fails.
302 */
304 /* If that failed, just do the required in-core inode clear. */
306 else
311 no_delete:
313 }
315 #ifdef CONFIG_QUOTA
317 {
319 }
320 #endif
322 /*
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
325 */
328 {
341 }
343 /* Update per-inode reservations */
349 /* Update quota subsystem for data blocks */
353 /*
354 * We did fallocate with an offset that is already delayed
355 * allocated. So on delayed allocated writeback we should
356 * not re-claim the quota for fallocated blocks.
357 */
359 }
361 /*
362 * If we have done all the pending block allocations and if
363 * there aren't any writers on the inode, we can discard the
364 * inode's preallocations.
365 */
369 }
374 {
378 "lblock %lu mapped to illegal pblock "
382 }
384 }
386 #define check_block_validity(inode, map) \
387 __check_block_validity((inode), __func__, __LINE__, (map))
389 #ifdef ES_AGGRESSIVE_TEST
395 {
399 /*
400 * There is a race window that the result is not the same.
401 * e.g. xfstests #223 when dioread_nolock enables. The reason
402 * is that we lookup a block mapping in extent status tree with
403 * out taking i_data_sem. So at the time the unwritten extent
404 * could be converted.
405 */
410 EXT4_GET_BLOCKS_KEEP_SIZE);
413 EXT4_GET_BLOCKS_KEEP_SIZE);
414 }
418 /*
419 * We don't check m_len because extent will be collpased in status
420 * tree. So the m_len might not equal.
421 */
426 "es_cached ex [%d/%d/%llu/%x] != "
432 }
433 }
436 /*
437 * The ext4_map_blocks() function tries to look up the requested blocks,
438 * and returns if the blocks are already mapped.
439 *
440 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
441 * and store the allocated blocks in the result buffer head and mark it
442 * mapped.
443 *
444 * If file type is extents based, it will call ext4_ext_map_blocks(),
445 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
446 * based files
447 *
448 * On success, it returns the number of blocks being mapped or allocated.
449 * if create==0 and the blocks are pre-allocated and unwritten block,
450 * the result buffer head is unmapped. If the create ==1, it will make sure
451 * the buffer head is mapped.
452 *
453 * It returns 0 if plain look up failed (blocks have not been allocated), in
454 * that case, buffer head is unmapped
455 *
456 * It returns the error in case of allocation failure.
457 */
460 {
464 #ifdef ES_AGGRESSIVE_TEST
468 #endif
475 /*
476 * ext4_map_blocks returns an int, and m_len is an unsigned int
477 */
481 /* We can handle the block number less than EXT_MAX_BLOCKS */
485 /* Lookup extent status tree firstly */
500 }
501 #ifdef ES_AGGRESSIVE_TEST
504 #endif
506 }
508 /*
509 * Try to see if we can get the block without requesting a new
510 * file system block.
511 */
516 EXT4_GET_BLOCKS_KEEP_SIZE);
519 EXT4_GET_BLOCKS_KEEP_SIZE);
520 }
526 "ES len assertion failed for inode "
530 }
543 }
547 found:
552 }
554 /* If it is only a block(s) look up */
558 /*
559 * Returns if the blocks have already allocated
560 *
561 * Note that if blocks have been preallocated
562 * ext4_ext_get_block() returns the create = 0
563 * with buffer head unmapped.
564 */
566 /*
567 * If we need to convert extent to unwritten
568 * we continue and do the actual work in
569 * ext4_ext_map_blocks()
570 */
574 /*
575 * Here we clear m_flags because after allocating an new extent,
576 * it will be set again.
577 */
580 /*
581 * New blocks allocate and/or writing to unwritten extent
582 * will possibly result in updating i_data, so we take
583 * the write lock of i_data_sem, and call get_block()
584 * with create == 1 flag.
585 */
588 /*
589 * We need to check for EXT4 here because migrate
590 * could have changed the inode type in between
591 */
598 /*
599 * We allocated new blocks which will result in
600 * i_data's format changing. Force the migrate
601 * to fail by clearing migrate flags
602 */
604 }
606 /*
607 * Update reserved blocks/metadata blocks after successful
608 * block allocation which had been deferred till now. We don't
609 * support fallocate for non extent files. So we can update
610 * reserve space here.
611 */
615 }
622 "ES len assertion failed for inode "
626 }
628 /*
629 * If the extent has been zeroed out, we don't need to update
630 * extent status tree.
631 */
636 }
648 }
650 has_zeroout:
656 }
658 }
660 /* Maximum number of blocks we map for direct IO at once. */
661 #define DIO_MAX_BLOCKS 4096
665 {
678 /* Direct IO write... */
683 dio_credits);
687 }
689 }
698 /*
699 * dgc: I suspect unwritten conversion on ext4+DAX is
700 * fundamentally broken here when there are concurrent
701 * read/write in progress on this inode.
702 */
706 }
711 }
715 }
719 {
722 }
724 /*
725 * `handle' can be NULL if create is zero
726 */
729 {
753 /*
754 * Now that we do not always journal data, we should
755 * keep in mind whether this should always journal the
756 * new buffer as metadata. For now, regular file
757 * writes use ext4_get_block instead, so it's not a
758 * problem.
759 */
766 }
770 }
779 errout:
782 }
786 {
800 }
809 {
825 }
829 }
831 }
833 /*
834 * To preserve ordering, it is essential that the hole instantiation and
835 * the data write be encapsulated in a single transaction. We cannot
836 * close off a transaction and start a new one between the ext4_get_block()
837 * and the commit_write(). So doing the jbd2_journal_start at the start of
838 * prepare_write() is the right place.
839 *
840 * Also, this function can nest inside ext4_writepage(). In that case, we
841 * *know* that ext4_writepage() has generated enough buffer credits to do the
842 * whole page. So we won't block on the journal in that case, which is good,
843 * because the caller may be PF_MEMALLOC.
844 *
845 * By accident, ext4 can be reentered when a transaction is open via
846 * quota file writes. If we were to commit the transaction while thus
847 * reentered, there can be a deadlock - we would be holding a quota
848 * lock, and the commit would never complete if another thread had a
849 * transaction open and was blocking on the quota lock - a ranking
850 * violation.
851 *
852 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
853 * will _not_ run commit under these circumstances because handle->h_ref
854 * is elevated. We'll still have enough credits for the tiny quotafile
855 * write.
856 */
859 {
865 /*
866 * __block_write_begin() could have dirtied some buffers. Clean
867 * the dirty bit as jbd2_journal_get_write_access() could complain
868 * otherwise about fs integrity issues. Setting of the dirty bit
869 * by __block_write_begin() isn't a real problem here as we clear
870 * the bit before releasing a page lock and thus writeback cannot
871 * ever write the buffer.
872 */
880 }
885 #ifdef CONFIG_EXT4_FS_ENCRYPTION
888 {
893 sector_t block;
918 }
920 }
936 }
941 }
942 }
947 }
955 }
956 }
957 /*
958 * If we issued read requests, let them complete.
959 */
964 }
970 }
971 #endif
976 {
982 pgoff_t index;
986 /*
987 * Reserve one block more for addition to orphan list in case
988 * we allocate blocks but write fails for some reason
989 */
1002 }
1004 /*
1005 * grab_cache_page_write_begin() can take a long time if the
1006 * system is thrashing due to memory pressure, or if the page
1007 * is being written back. So grab it first before we start
1008 * the transaction handle. This also allows us to allocate
1009 * the page (if needed) without using GFP_NOFS.
1010 */
1011 retry_grab:
1017 retry_journal:
1022 }
1026 /* The page got truncated from under us */
1031 }
1032 /* In case writeback began while the page was unlocked */
1035 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1038 ext4_get_block_write);
1039 else
1041 ext4_get_block);
1042 #else
1045 else
1047 #endif
1051 do_journal_get_write_access);
1052 }
1056 /*
1057 * __block_write_begin may have instantiated a few blocks
1058 * outside i_size. Trim these off again. Don't need
1059 * i_size_read because we hold i_mutex.
1060 *
1061 * Add inode to orphan list in case we crash before
1062 * truncate finishes
1063 */
1070 /*
1071 * If truncate failed early the inode might
1072 * still be on the orphan list; we need to
1073 * make sure the inode is removed from the
1074 * orphan list in that case.
1075 */
1078 }
1085 }
1088 }
1090 /* For write_end() in data=journal mode */
1092 {
1101 }
1103 /*
1104 * We need to pick up the new inode size which generic_commit_write gave us
1105 * `file' can be NULL - eg, when called from page_symlink().
1106 *
1107 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1108 * buffers are managed internally.
1109 */
1114 {
1128 }
1129 }
1140 /*
1141 * it's important to update i_size while still holding page lock:
1142 * page writeout could otherwise come in and zero beyond i_size.
1143 */
1150 /*
1151 * Don't mark the inode dirty under page lock. First, it unnecessarily
1152 * makes the holding time of page lock longer. Second, it forces lock
1153 * ordering of page lock and transaction start for journaling
1154 * filesystems.
1155 */
1160 /* if we have allocated more blocks and copied
1161 * less. We will have blocks allocated outside
1162 * inode->i_size. So truncate them
1163 */
1165 errout:
1172 /*
1173 * If truncate failed early the inode might still be
1174 * on the orphan list; we need to make sure the inode
1175 * is removed from the orphan list in that case.
1176 */
1179 }
1182 }
1184 /*
1185 * This is a private version of page_zero_new_buffers() which doesn't
1186 * set the buffer to be dirty, since in data=journalled mode we need
1187 * to call ext4_handle_dirty_metadata() instead.
1188 */
1190 {
1207 }
1209 }
1210 }
1214 }
1220 {
1243 }
1249 }
1263 }
1266 /* if we have allocated more blocks and copied
1267 * less. We will have blocks allocated outside
1268 * inode->i_size. So truncate them
1269 */
1277 /*
1278 * If truncate failed early the inode might still be
1279 * on the orphan list; we need to make sure the inode
1280 * is removed from the orphan list in that case.
1281 */
1284 }
1287 }
1289 /*
1290 * Reserve space for a single cluster
1291 */
1293 {
1298 /*
1299 * We will charge metadata quota at writeout time; this saves
1300 * us from metadata over-estimation, though we may go over by
1301 * a small amount in the end. Here we just reserve for data.
1302 */
1312 }
1318 }
1321 {
1332 /*
1333 * if there aren't enough reserved blocks, then the
1334 * counter is messed up somewhere. Since this
1335 * function is called from invalidate page, it's
1336 * harmless to return without any action.
1337 */
1339 "ino %lu, to_free %d with only %d reserved "
1344 }
1347 /* update fs dirty data blocks counter */
1353 }
1358 {
1366 ext4_fsblk_t lblk;
1379 to_release++;
1380 contiguous_blks++;
1386 contiguous_blks;
1389 }
1397 }
1399 /* If we have released all the blocks belonging to a cluster, then we
1400 * need to release the reserved space for that cluster. */
1409 num_clusters--;
1410 }
1411 }
1413 /*
1414 * Delayed allocation stuff
1415 */
1424 /*
1425 * Extent to map - this can be after first_page because that can be
1426 * fully mapped. We somewhat abuse m_flags to store whether the extent
1427 * is delalloc or unwritten.
1428 */
1431 };
1435 {
1442 /* This is necessary when next_page == 0. */
1453 }
1469 }
1471 }
1474 }
1475 }
1478 {
1497 }
1500 {
1502 }
1504 /*
1505 * This function is grabs code from the very beginning of
1506 * ext4_map_blocks, but assumes that the caller is from delayed write
1507 * time. This function looks up the requested blocks and sets the
1508 * buffer delay bit under the protection of i_data_sem.
1509 */
1513 {
1517 #ifdef ES_AGGRESSIVE_TEST
1521 #endif
1531 /* Lookup extent status tree firstly */
1537 }
1539 /*
1540 * Delayed extent could be allocated by fallocate.
1541 * So we need to check it.
1542 */
1548 }
1559 else
1562 #ifdef ES_AGGRESSIVE_TEST
1564 #endif
1566 }
1568 /*
1569 * Try to see if we can get the block without requesting a new
1570 * file system block.
1571 */
1577 else
1580 add_delayed:
1583 /*
1584 * XXX: __block_prepare_write() unmaps passed block,
1585 * is it OK?
1586 */
1587 /*
1588 * If the block was allocated from previously allocated cluster,
1589 * then we don't need to reserve it again. However we still need
1590 * to reserve metadata for every block we're going to write.
1591 */
1596 /* not enough space to reserve */
1599 }
1600 }
1607 }
1618 "ES len assertion failed for inode "
1622 }
1630 }
1632 out_unlock:
1636 }
1638 /*
1639 * This is a special get_block_t callback which is used by
1640 * ext4_da_write_begin(). It will either return mapped block or
1641 * reserve space for a single block.
1642 *
1643 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1644 * We also have b_blocknr = -1 and b_bdev initialized properly
1645 *
1646 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1647 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1648 * initialized properly.
1649 */
1652 {
1662 /*
1663 * first, we need to know whether the block is allocated already
1664 * preallocated blocks are unmapped but should treated
1665 * the same as allocated blocks.
1666 */
1675 /* A delayed write to unwritten bh should be marked
1676 * new and mapped. Mapped ensures that we don't do
1677 * get_block multiple times when we write to the same
1678 * offset and new ensures that we do proper zero out
1679 * for partial write.
1680 */
1683 }
1685 }
1688 {
1691 }
1694 {
1697 }
1701 {
1723 }
1726 }
1727 /*
1728 * We need to release the page lock before we start the
1729 * journal, so grab a reference so the page won't disappear
1730 * out from under us.
1731 */
1741 }
1747 /* The page got truncated from under us */
1751 }
1761 do_journal_get_write_access);
1764 write_end_fn);
1765 }
1777 out:
1779 out_no_pagelock:
1782 }
1784 /*
1785 * Note that we don't need to start a transaction unless we're journaling data
1786 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1787 * need to file the inode to the transaction's list in ordered mode because if
1788 * we are writing back data added by write(), the inode is already there and if
1789 * we are writing back data modified via mmap(), no one guarantees in which
1790 * transaction the data will hit the disk. In case we are journaling data, we
1791 * cannot start transaction directly because transaction start ranks above page
1792 * lock so we have to do some magic.
1793 *
1794 * This function can get called via...
1795 * - ext4_writepages after taking page lock (have journal handle)
1796 * - journal_submit_inode_data_buffers (no journal handle)
1797 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1798 * - grab_page_cache when doing write_begin (have journal handle)
1799 *
1800 * We don't do any block allocation in this function. If we have page with
1801 * multiple blocks we need to write those buffer_heads that are mapped. This
1802 * is important for mmaped based write. So if we do with blocksize 1K
1803 * truncate(f, 1024);
1804 * a = mmap(f, 0, 4096);
1805 * a[0] = 'a';
1806 * truncate(f, 4096);
1807 * we have in the page first buffer_head mapped via page_mkwrite call back
1808 * but other buffer_heads would be unmapped but dirty (dirty done via the
1809 * do_wp_page). So writepage should write the first block. If we modify
1810 * the mmap area beyond 1024 we will again get a page_fault and the
1811 * page_mkwrite callback will do the block allocation and mark the
1812 * buffer_heads mapped.
1813 *
1814 * We redirty the page if we have any buffer_heads that is either delay or
1815 * unwritten in the page.
1816 *
1817 * We can get recursively called as show below.
1818 *
1819 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1820 * ext4_writepage()
1821 *
1822 * But since we don't do any block allocation we should not deadlock.
1823 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1824 */
1827 {
1829 loff_t size;
1840 else
1844 /*
1845 * We cannot do block allocation or other extent handling in this
1846 * function. If there are buffers needing that, we have to redirty
1847 * the page. But we may reach here when we do a journal commit via
1848 * journal_submit_inode_data_buffers() and in that case we must write
1849 * allocated buffers to achieve data=ordered mode guarantees.
1850 *
1851 * Also, if there is only one buffer per page (the fs block
1852 * size == the page size), if one buffer needs block
1853 * allocation or needs to modify the extent tree to clear the
1854 * unwritten flag, we know that the page can't be written at
1855 * all, so we might as well refuse the write immediately.
1856 * Unfortunately if the block size != page size, we can't as
1857 * easily detect this case using ext4_walk_page_buffers(), but
1858 * for the extremely common case, this is an optimization that
1859 * skips a useless round trip through ext4_bio_write_page().
1860 */
1862 ext4_bh_delay_or_unwritten)) {
1866 /*
1867 * For memory cleaning there's no point in writing only
1868 * some buffers. So just bail out. Warn if we came here
1869 * from direct reclaim.
1870 */
1875 }
1877 }
1880 /*
1881 * It's mmapped pagecache. Add buffers and journal it. There
1882 * doesn't seem much point in redirtying the page here.
1883 */
1892 }
1895 /* Drop io_end reference we got from init */
1898 }
1901 {
1909 else
1918 }
1920 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1922 /*
1923 * mballoc gives us at most this number of blocks...
1924 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1925 * The rest of mballoc seems to handle chunks up to full group size.
1926 */
1927 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1929 /*
1930 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1931 *
1932 * @mpd - extent of blocks
1933 * @lblk - logical number of the block in the file
1934 * @bh - buffer head we want to add to the extent
1935 *
1936 * The function is used to collect contig. blocks in the same state. If the
1937 * buffer doesn't require mapping for writeback and we haven't started the
1938 * extent of buffers to map yet, the function returns 'true' immediately - the
1939 * caller can write the buffer right away. Otherwise the function returns true
1940 * if the block has been added to the extent, false if the block couldn't be
1941 * added.
1942 */
1945 {
1948 /* Buffer that doesn't need mapping for writeback? */
1951 /* So far no extent to map => we write the buffer right away */
1955 }
1957 /* First block in the extent? */
1963 }
1965 /* Don't go larger than mballoc is willing to allocate */
1969 /* Can we merge the block to our big extent? */
1974 }
1976 }
1978 /*
1979 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1980 *
1981 * @mpd - extent of blocks for mapping
1982 * @head - the first buffer in the page
1983 * @bh - buffer we should start processing from
1984 * @lblk - logical number of the block in the file corresponding to @bh
1985 *
1986 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1987 * the page for IO if all buffers in this page were mapped and there's no
1988 * accumulated extent of buffers to map or add buffers in the page to the
1989 * extent of buffers to map. The function returns 1 if the caller can continue
1990 * by processing the next page, 0 if it should stop adding buffers to the
1991 * extent to map because we cannot extend it anymore. It can also return value
1992 * < 0 in case of error during IO submission.
1993 */
1997 ext4_lblk_t lblk)
1998 {
2008 /* Found extent to map? */
2011 /* Everything mapped so far and we hit EOF */
2013 }
2015 /* So far everything mapped? Submit the page for IO. */
2020 }
2022 }
2024 /*
2025 * mpage_map_buffers - update buffers corresponding to changed extent and
2026 * submit fully mapped pages for IO
2027 *
2028 * @mpd - description of extent to map, on return next extent to map
2029 *
2030 * Scan buffers corresponding to changed extent (we expect corresponding pages
2031 * to be already locked) and update buffer state according to new extent state.
2032 * We map delalloc buffers to their physical location, clear unwritten bits,
2033 * and mark buffers as uninit when we perform writes to unwritten extents
2034 * and do extent conversion after IO is finished. If the last page is not fully
2035 * mapped, we update @map to the next extent in the last page that needs
2036 * mapping. Otherwise we submit the page for IO.
2037 */
2039 {
2046 ext4_lblk_t lblk;
2047 sector_t pblock;
2058 PAGEVEC_SIZE);
2066 /* Up to 'end' pages must be contiguous */
2073 /*
2074 * Buffer after end of mapped extent.
2075 * Find next buffer in the page to map.
2076 */
2079 /*
2080 * FIXME: If dioread_nolock supports
2081 * blocksize < pagesize, we need to make
2082 * sure we add size mapped so far to
2083 * io_end->size as the following call
2084 * can submit the page for IO.
2085 */
2092 }
2096 }
2100 /*
2101 * FIXME: This is going to break if dioread_nolock
2102 * supports blocksize < pagesize as we will try to
2103 * convert potentially unmapped parts of inode.
2104 */
2106 /* Page fully mapped - let IO run! */
2111 }
2112 start++;
2113 }
2115 }
2116 /* Extent fully mapped and matches with page boundary. We are done. */
2120 }
2123 {
2130 /*
2131 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2132 * to convert an unwritten extent to be initialized (in the case
2133 * where we have written into one or more preallocated blocks). It is
2134 * possible that we're going to need more metadata blocks than
2135 * previously reserved. However we must not fail because we're in
2136 * writeback and there is nothing we can do about it so it might result
2137 * in data loss. So use reserved blocks to allocate metadata if
2138 * possible.
2139 *
2140 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2141 * the blocks in question are delalloc blocks. This indicates
2142 * that the blocks and quotas has already been checked when
2143 * the data was copied into the page cache.
2144 */
2146 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2161 }
2163 }
2172 }
2174 }
2176 /*
2177 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2178 * mpd->len and submit pages underlying it for IO
2179 *
2180 * @handle - handle for journal operations
2181 * @mpd - extent to map
2182 * @give_up_on_write - we set this to true iff there is a fatal error and there
2183 * is no hope of writing the data. The caller should discard
2184 * dirty pages to avoid infinite loops.
2185 *
2186 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2187 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2188 * them to initialized or split the described range from larger unwritten
2189 * extent. Note that we need not map all the described range since allocation
2190 * can return less blocks or the range is covered by more unwritten extents. We
2191 * cannot map more because we are limited by reserved transaction credits. On
2192 * the other hand we always make sure that the last touched page is fully
2193 * mapped so that it can be written out (and thus forward progress is
2194 * guaranteed). After mapping we submit all mapped pages for IO.
2195 */
2199 {
2203 loff_t disksize;
2215 /*
2216 * Let the uper layers retry transient errors.
2217 * In the case of ENOSPC, if ext4_count_free_blocks()
2218 * is non-zero, a commit should free up blocks.
2219 */
2225 }
2227 "Delayed block allocation failed for "
2228 "inode %lu at logical offset %llu with"
2234 "This should not happen!! Data will "
2238 invalidate_dirty_pages:
2241 }
2243 /*
2244 * Update buffer state, submit mapped pages, and get us new
2245 * extent to map
2246 */
2252 update_disksize:
2253 /*
2254 * Update on-disk size after IO is submitted. Races with
2255 * truncate are avoided by checking i_size under i_data_sem.
2256 */
2260 loff_t i_size;
2276 }
2278 }
2280 /*
2281 * Calculate the total number of credits to reserve for one writepages
2282 * iteration. This is called from ext4_writepages(). We map an extent of
2283 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2284 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2285 * bpp - 1 blocks in bpp different extents.
2286 */
2288 {
2293 }
2295 /*
2296 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2297 * and underlying extent to map
2298 *
2299 * @mpd - where to look for pages
2300 *
2301 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2302 * IO immediately. When we find a page which isn't mapped we start accumulating
2303 * extent of buffers underlying these pages that needs mapping (formed by
2304 * either delayed or unwritten buffers). We also lock the pages containing
2305 * these buffers. The extent found is returned in @mpd structure (starting at
2306 * mpd->lblk with length mpd->len blocks).
2307 *
2308 * Note that this function can attach bios to one io_end structure which are
2309 * neither logically nor physically contiguous. Although it may seem as an
2310 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2311 * case as we need to track IO to all buffers underlying a page in one io_end.
2312 */
2314 {
2324 ext4_lblk_t lblk;
2329 else
2344 /*
2345 * At this point, the page may be truncated or
2346 * invalidated (changing page->mapping to NULL), or
2347 * even swizzled back from swapper_space to tmpfs file
2348 * mapping. However, page->index will not change
2349 * because we have a reference on the page.
2350 */
2354 /*
2355 * Accumulated enough dirty pages? This doesn't apply
2356 * to WB_SYNC_ALL mode. For integrity sync we have to
2357 * keep going because someone may be concurrently
2358 * dirtying pages, and we might have synced a lot of
2359 * newly appeared dirty pages, but have not synced all
2360 * of the old dirty pages.
2361 */
2365 /* If we can't merge this page, we are done. */
2370 /*
2371 * If the page is no longer dirty, or its mapping no
2372 * longer corresponds to inode we are writing (which
2373 * means it has been truncated or invalidated), or the
2374 * page is already under writeback and we are not doing
2375 * a data integrity writeback, skip the page
2376 */
2383 }
2391 /* Add all dirty buffers to mpd */
2399 left--;
2400 }
2403 }
2405 out:
2408 }
2412 {
2417 }
2421 {
2437 /*
2438 * No pages to write? This is mainly a kludge to avoid starting
2439 * a transaction for special inodes like journal inode on last iput()
2440 * because that could violate lock ordering on umount
2441 */
2452 }
2454 /*
2455 * If the filesystem has aborted, it is read-only, so return
2456 * right away instead of dumping stack traces later on that
2457 * will obscure the real source of the problem. We test
2458 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2459 * the latter could be true if the filesystem is mounted
2460 * read-only, and in that case, ext4_writepages should
2461 * *never* be called, so if that ever happens, we would want
2462 * the stack trace.
2463 */
2467 }
2470 /*
2471 * We may need to convert up to one extent per block in
2472 * the page and we may dirty the inode.
2473 */
2475 }
2477 /*
2478 * If we have inline data and arrive here, it means that
2479 * we will soon create the block for the 1st page, so
2480 * we'd better clear the inline data here.
2481 */
2483 /* Just inode will be modified... */
2488 }
2490 EXT4_STATE_MAY_INLINE_DATA));
2493 }
2507 }
2512 retry:
2518 /* For each extent of pages we use new io_end */
2523 }
2525 /*
2526 * We have two constraints: We find one extent to map and we
2527 * must always write out whole page (makes a difference when
2528 * blocksize < pagesize) so that we don't block on IO when we
2529 * try to write out the rest of the page. Journalled mode is
2530 * not supported by delalloc.
2531 */
2535 /* start a new transaction */
2543 /* Release allocated io_end */
2546 }
2555 /*
2556 * We scanned the whole range (or exhausted
2557 * nr_to_write), submitted what was mapped and
2558 * didn't find anything needing mapping. We are
2559 * done.
2560 */
2562 }
2563 }
2565 /* Submit prepared bio */
2567 /* Unlock pages we didn't use */
2569 /* Drop our io_end reference we got from init */
2573 /*
2574 * Commit the transaction which would
2575 * free blocks released in the transaction
2576 * and try again
2577 */
2581 }
2582 /* Fatal error - ENOMEM, EIO... */
2585 }
2592 }
2594 /* Update index */
2596 /*
2597 * Set the writeback_index so that range_cyclic
2598 * mode will write it back later
2599 */
2602 out_writepages:
2606 }
2609 {
2613 /*
2614 * switch to non delalloc mode if we are running low
2615 * on free block. The free block accounting via percpu
2616 * counters can get slightly wrong with percpu_counter_batch getting
2617 * accumulated on each CPU without updating global counters
2618 * Delalloc need an accurate free block accounting. So switch
2619 * to non delalloc when we are near to error range.
2620 */
2621 free_clusters =
2623 dirty_clusters =
2625 /*
2626 * Start pushing delalloc when 1/2 of free blocks are dirty.
2627 */
2633 /*
2634 * free block count is less than 150% of dirty blocks
2635 * or free blocks is less than watermark
2636 */
2638 }
2640 }
2642 /* We always reserve for an inode update; the superblock could be there too */
2644 {
2651 /* We might need to update the superblock to set LARGE_FILE */
2653 }
2658 {
2661 pgoff_t index;
2671 }
2683 }
2685 /*
2686 * grab_cache_page_write_begin() can take a long time if the
2687 * system is thrashing due to memory pressure, or if the page
2688 * is being written back. So grab it first before we start
2689 * the transaction handle. This also allows us to allocate
2690 * the page (if needed) without using GFP_NOFS.
2691 */
2692 retry_grab:
2698 /*
2699 * With delayed allocation, we don't log the i_disksize update
2700 * if there is delayed block allocation. But we still need
2701 * to journalling the i_disksize update if writes to the end
2702 * of file which has an already mapped buffer.
2703 */
2704 retry_journal:
2710 }
2714 /* The page got truncated from under us */
2719 }
2720 /* In case writeback began while the page was unlocked */
2723 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2725 ext4_da_get_block_prep);
2726 #else
2728 #endif
2732 /*
2733 * block_write_begin may have instantiated a few blocks
2734 * outside i_size. Trim these off again. Don't need
2735 * i_size_read because we hold i_mutex.
2736 */
2746 }
2750 }
2752 /*
2753 * Check if we should update i_disksize
2754 * when write to the end of file but not require block allocation
2755 */
2758 {
2773 }
2779 {
2783 loff_t new_i_size;
2795 /*
2796 * generic_write_end() will run mark_inode_dirty() if i_size
2797 * changes. So let's piggyback the i_disksize mark_inode_dirty
2798 * into that.
2799 */
2805 /* We need to mark inode dirty even if
2806 * new_i_size is less that inode->i_size
2807 * bu greater than i_disksize.(hint delalloc)
2808 */
2810 }
2811 }
2817 page);
2818 else
2830 }
2834 {
2835 /*
2836 * Drop reserved blocks
2837 */
2844 out:
2848 }
2850 /*
2851 * Force all delayed allocation blocks to be allocated for a given inode.
2852 */
2854 {
2860 /*
2861 * We do something simple for now. The filemap_flush() will
2862 * also start triggering a write of the data blocks, which is
2863 * not strictly speaking necessary (and for users of
2864 * laptop_mode, not even desirable). However, to do otherwise
2865 * would require replicating code paths in:
2866 *
2867 * ext4_writepages() ->
2868 * write_cache_pages() ---> (via passed in callback function)
2869 * __mpage_da_writepage() -->
2870 * mpage_add_bh_to_extent()
2871 * mpage_da_map_blocks()
2872 *
2873 * The problem is that write_cache_pages(), located in
2874 * mm/page-writeback.c, marks pages clean in preparation for
2875 * doing I/O, which is not desirable if we're not planning on
2876 * doing I/O at all.
2877 *
2878 * We could call write_cache_pages(), and then redirty all of
2879 * the pages by calling redirty_page_for_writepage() but that
2880 * would be ugly in the extreme. So instead we would need to
2881 * replicate parts of the code in the above functions,
2882 * simplifying them because we wouldn't actually intend to
2883 * write out the pages, but rather only collect contiguous
2884 * logical block extents, call the multi-block allocator, and
2885 * then update the buffer heads with the block allocations.
2886 *
2887 * For now, though, we'll cheat by calling filemap_flush(),
2888 * which will map the blocks, and start the I/O, but not
2889 * actually wait for the I/O to complete.
2890 */
2892 }
2894 /*
2895 * bmap() is special. It gets used by applications such as lilo and by
2896 * the swapper to find the on-disk block of a specific piece of data.
2897 *
2898 * Naturally, this is dangerous if the block concerned is still in the
2899 * journal. If somebody makes a swapfile on an ext4 data-journaling
2900 * filesystem and enables swap, then they may get a nasty shock when the
2901 * data getting swapped to that swapfile suddenly gets overwritten by
2902 * the original zero's written out previously to the journal and
2903 * awaiting writeback in the kernel's buffer cache.
2904 *
2905 * So, if we see any bmap calls here on a modified, data-journaled file,
2906 * take extra steps to flush any blocks which might be in the cache.
2907 */
2909 {
2914 /*
2915 * We can get here for an inline file via the FIBMAP ioctl
2916 */
2922 /*
2923 * With delalloc we want to sync the file
2924 * so that we can make sure we allocate
2925 * blocks for file
2926 */
2928 }
2932 /*
2933 * This is a REALLY heavyweight approach, but the use of
2934 * bmap on dirty files is expected to be extremely rare:
2935 * only if we run lilo or swapon on a freshly made file
2936 * do we expect this to happen.
2937 *
2938 * (bmap requires CAP_SYS_RAWIO so this does not
2939 * represent an unprivileged user DOS attack --- we'd be
2940 * in trouble if mortal users could trigger this path at
2941 * will.)
2942 *
2943 * NB. EXT4_STATE_JDATA is not set on files other than
2944 * regular files. If somebody wants to bmap a directory
2945 * or symlink and gets confused because the buffer
2946 * hasn't yet been flushed to disk, they deserve
2947 * everything they get.
2948 */
2958 }
2961 }
2964 {
2977 }
2979 static int
2982 {
2985 /* If the file has inline data, no need to do readpages. */
2990 }
2994 {
2997 /* No journalling happens on data buffers when this function is used */
3001 }
3006 {
3011 /*
3012 * If it's a full truncate we just forget about the pending dirtying
3013 */
3018 }
3020 /* Wrapper for aops... */
3024 {
3026 }
3029 {
3034 /* Page has dirty journalled data -> cannot release */
3039 else
3041 }
3043 /*
3044 * ext4_get_block used when preparing for a DIO write or buffer write.
3045 * We allocate an uinitialized extent if blocks haven't been allocated.
3046 * The extent will be converted to initialized after the IO is complete.
3047 */
3050 {
3054 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3055 }
3059 {
3063 EXT4_GET_BLOCKS_NO_LOCK);
3064 }
3068 {
3075 }
3079 {
3082 /* if not async direct IO just return */
3089 size);
3095 }
3097 /*
3098 * For ext4 extent files, ext4 will do direct-io write to holes,
3099 * preallocated extents, and those write extend the file, no need to
3100 * fall back to buffered IO.
3101 *
3102 * For holes, we fallocate those blocks, mark them as unwritten
3103 * If those blocks were preallocated, we mark sure they are split, but
3104 * still keep the range to write as unwritten.
3105 *
3106 * The unwritten extents will be converted to written when DIO is completed.
3107 * For async direct IO, since the IO may still pending when return, we
3108 * set up an end_io call back function, which will do the conversion
3109 * when async direct IO completed.
3110 *
3111 * If the O_DIRECT write will extend the file then add this inode to the
3112 * orphan list. So recovery will truncate it back to the original size
3113 * if the machine crashes during the write.
3114 *
3115 */
3117 loff_t offset)
3118 {
3121 ssize_t ret;
3129 /* Use the old path for reads and writes beyond i_size. */
3135 /*
3136 * Make all waiters for direct IO properly wait also for extent
3137 * conversion. This also disallows race between truncate() and
3138 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3139 */
3143 /* If we do a overwrite dio, i_mutex locking can be released */
3149 }
3151 /*
3152 * We could direct write to holes and fallocate.
3153 *
3154 * Allocated blocks to fill the hole are marked as
3155 * unwritten to prevent parallel buffered read to expose
3156 * the stale data before DIO complete the data IO.
3157 *
3158 * As to previously fallocated extents, ext4 get_block will
3159 * just simply mark the buffer mapped but still keep the
3160 * extents unwritten.
3161 *
3162 * For non AIO case, we will convert those unwritten extents
3163 * to written after return back from blockdev_direct_IO.
3164 *
3165 * For async DIO, the conversion needs to be deferred when the
3166 * IO is completed. The ext4 end_io callback function will be
3167 * called to take care of the conversion work. Here for async
3168 * case, we allocate an io_end structure to hook to the iocb.
3169 */
3177 }
3178 /*
3179 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3180 */
3182 /*
3183 * we save the io structure for current async direct
3184 * IO, so that later ext4_map_blocks() could flag the
3185 * io structure whether there is a unwritten extents
3186 * needs to be converted when IO is completed.
3187 */
3189 }
3196 }
3197 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3199 #endif
3203 else
3206 get_block_func,
3209 /*
3210 * Put our reference to io_end. This can free the io_end structure e.g.
3211 * in sync IO case or in case of error. It can even perform extent
3212 * conversion if all bios we submitted finished before we got here.
3213 * Note that in that case iocb->private can be already set to NULL
3214 * here.
3215 */
3219 /*
3220 * When no IO was submitted ext4_end_io_dio() was not
3221 * called so we have to put iocb's reference.
3222 */
3228 }
3229 }
3231 EXT4_STATE_DIO_UNWRITTEN)) {
3233 /*
3234 * for non AIO case, since the IO is already
3235 * completed, we could do the conversion right here
3236 */
3242 }
3244 retake_lock:
3247 /* take i_mutex locking again if we do a ovewrite dio */
3251 }
3254 }
3257 loff_t offset)
3258 {
3262 ssize_t ret;
3264 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3267 #endif
3269 /*
3270 * If we are doing data journalling we don't support O_DIRECT
3271 */
3275 /* Let buffer I/O handle the inline data case. */
3282 else
3286 }
3288 /*
3289 * Pages can be marked dirty completely asynchronously from ext4's journalling
3290 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3291 * much here because ->set_page_dirty is called under VFS locks. The page is
3292 * not necessarily locked.
3293 *
3294 * We cannot just dirty the page and leave attached buffers clean, because the
3295 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3296 * or jbddirty because all the journalling code will explode.
3297 *
3298 * So what we do is to mark the page "pending dirty" and next time writepage
3299 * is called, propagate that into the buffers appropriately.
3300 */
3302 {
3305 }
3321 };
3337 };
3353 };
3356 {
3369 }
3372 else
3374 }
3378 {
3382 ext4_lblk_t iblock;
3400 /* Find the buffer that contains "offset" */
3405 iblock++;
3407 }
3411 }
3415 /* unmapped? It's a hole - nothing to do */
3419 }
3420 }
3422 /* Ok, it's mapped. Make sure it's up-to-date */
3430 /* Uhhuh. Read error. Complain and punt. */
3435 /* We expect the key to be set. */
3439 }
3440 }
3446 }
3457 }
3459 unlock:
3463 }
3465 /*
3466 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3467 * starting from file offset 'from'. The range to be zero'd must
3468 * be contained with in one block. If the specified range exceeds
3469 * the end of the block it will be shortened to end of the block
3470 * that cooresponds to 'from'
3471 */
3474 {
3480 /*
3481 * correct length if it does not fall between
3482 * 'from' and the end of the block
3483 */
3490 }
3492 /*
3493 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3494 * up to the end of the block which corresponds to `from'.
3495 * This required during truncate. We need to physically zero the tail end
3496 * of that block so it doesn't yield old data if the file is later grown.
3497 */
3500 {
3510 }
3514 {
3528 /* Handle partial zero within the single block */
3534 }
3535 /* Handle partial zero out on the start of the range */
3541 }
3542 /* Handle partial zero out on the end of the range */
3548 }
3551 {
3559 }
3561 /*
3562 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3563 * associated with the given offset and length
3564 *
3565 * @inode: File inode
3566 * @offset: The offset where the hole will begin
3567 * @len: The length of the hole
3568 *
3569 * Returns: 0 on success or negative on failure
3570 */
3573 {
3587 /*
3588 * Write out all dirty pages to avoid race conditions
3589 * Then release them.
3590 */
3596 }
3600 /* No need to punch hole beyond i_size */
3604 /*
3605 * If the hole extends beyond i_size, set the hole
3606 * to end after the page that contains i_size
3607 */
3611 offset;
3612 }
3616 /*
3617 * Attach jinode to inode for jbd2 if we do any zeroing of
3618 * partial block
3619 */
3624 }
3629 /* Now release the pages and zero block aligned part of pages*/
3632 last_block_offset);
3634 /* Wait all existing dio workers, newcomers will block on i_mutex */
3640 else
3647 }
3650 length);
3658 /* If there are no blocks to remove, return now */
3670 }
3675 else
3677 stop_block);
3683 /* Now release the pages again to reduce race window */
3686 last_block_offset);
3690 out_stop:
3692 out_dio:
3694 out_mutex:
3697 }
3700 {
3713 }
3717 }
3722 }
3724 /*
3725 * ext4_truncate()
3726 *
3727 * We block out ext4_get_block() block instantiations across the entire
3728 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3729 * simultaneously on behalf of the same inode.
3730 *
3731 * As we work through the truncate and commit bits of it to the journal there
3732 * is one core, guiding principle: the file's tree must always be consistent on
3733 * disk. We must be able to restart the truncate after a crash.
3734 *
3735 * The file's tree may be transiently inconsistent in memory (although it
3736 * probably isn't), but whenever we close off and commit a journal transaction,
3737 * the contents of (the filesystem + the journal) must be consistent and
3738 * restartable. It's pretty simple, really: bottom up, right to left (although
3739 * left-to-right works OK too).
3740 *
3741 * Note that at recovery time, journal replay occurs *before* the restart of
3742 * truncate against the orphan inode list.
3743 *
3744 * The committed inode has the new, desired i_size (which is the same as
3745 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3746 * that this inode's truncate did not complete and it will again call
3747 * ext4_truncate() to have another go. So there will be instantiated blocks
3748 * to the right of the truncation point in a crashed ext4 filesystem. But
3749 * that's fine - as long as they are linked from the inode, the post-crash
3750 * ext4_truncate() run will find them and release them.
3751 */
3753 {
3759 /*
3760 * There is a possibility that we're either freeing the inode
3761 * or it's a completely new inode. In those cases we might not
3762 * have i_mutex locked because it's not necessary.
3763 */
3782 }
3784 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3788 }
3792 else
3799 }
3804 /*
3805 * We add the inode to the orphan list, so that if this
3806 * truncate spans multiple transactions, and we crash, we will
3807 * resume the truncate when the filesystem recovers. It also
3808 * marks the inode dirty, to catch the new size.
3809 *
3810 * Implication: the file must always be in a sane, consistent
3811 * truncatable state while each transaction commits.
3812 */
3822 else
3830 out_stop:
3831 /*
3832 * If this was a simple ftruncate() and the file will remain alive,
3833 * then we need to clear up the orphan record which we created above.
3834 * However, if this was a real unlink then we were called by
3835 * ext4_evict_inode(), and we allow that function to clean up the
3836 * orphan info for us.
3837 */
3846 }
3848 /*
3849 * ext4_get_inode_loc returns with an extra refcount against the inode's
3850 * underlying buffer_head on success. If 'in_mem' is true, we have all
3851 * data in memory that is needed to recreate the on-disk version of this
3852 * inode.
3853 */
3856 {
3860 ext4_fsblk_t block;
3872 /*
3873 * Figure out the offset within the block group inode table
3874 */
3887 /*
3888 * If the buffer has the write error flag, we have failed
3889 * to write out another inode in the same block. In this
3890 * case, we don't have to read the block because we may
3891 * read the old inode data successfully.
3892 */
3897 /* someone brought it uptodate while we waited */
3900 }
3902 /*
3903 * If we have all information of the inode in memory and this
3904 * is the only valid inode in the block, we need not read the
3905 * block.
3906 */
3913 /* Is the inode bitmap in cache? */
3918 /*
3919 * If the inode bitmap isn't in cache then the
3920 * optimisation may end up performing two reads instead
3921 * of one, so skip it.
3922 */
3926 }
3932 }
3935 /* all other inodes are free, so skip I/O */
3940 }
3941 }
3943 make_io:
3944 /*
3945 * If we need to do any I/O, try to pre-readahead extra
3946 * blocks from the inode table.
3947 */
3954 /* s_inode_readahead_blks is always a power of 2 */
3967 }
3969 /*
3970 * There are other valid inodes in the buffer, this inode
3971 * has in-inode xattrs, or we don't have this inode in memory.
3972 * Read the block from disk.
3973 */
3984 }
3985 }
3986 has_buffer:
3989 }
3992 {
3993 /* We have all inode data except xattrs in memory here. */
3996 }
3999 {
4017 }
4019 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4021 {
4030 EXT4_DIRSYNC_FL);
4042 }
4046 {
4047 blkcnt_t i_blocks ;
4052 /* we are using combined 48 bit field */
4056 /* i_blocks represent file system block size */
4060 }
4063 }
4064 }
4069 {
4077 }
4080 {
4088 uid_t i_uid;
4089 gid_t i_gid;
4114 }
4118 /* Precompute checksum seed for inode metadata */
4121 __u32 csum;
4128 }
4134 }
4142 }
4151 /* We now have enough fields to check if the inode was active or not.
4152 * This is needed because nfsd might try to access dead inodes
4153 * the test is that same one that e2fsck uses
4154 * NeilBrown 1999oct15
4155 */
4160 /* this inode is deleted */
4163 }
4164 /* The only unlinked inodes we let through here have
4165 * valid i_mode and are being read by the orphan
4166 * recovery code: that's fine, we're about to complete
4167 * the process of deleting those.
4168 * OR it is the EXT4_BOOT_LOADER_INO which is
4169 * not initialized on a new filesystem. */
4170 }
4179 #ifdef CONFIG_QUOTA
4181 #endif
4185 /*
4186 * NOTE! The in-memory inode i_data array is in little-endian order
4187 * even on big-endian machines: we do NOT byteswap the block numbers!
4188 */
4193 /*
4194 * Set transaction id's of transactions that have to be committed
4195 * to finish f[data]sync. We set them to currently running transaction
4196 * as we cannot be sure that the inode or some of its metadata isn't
4197 * part of the transaction - the inode could have been reclaimed and
4198 * now it is reread from disk.
4199 */
4202 tid_t tid;
4207 else
4211 else
4216 }
4220 /* The extra space is currently unused. Use it. */
4222 EXT4_GOOD_OLD_INODE_SIZE;
4225 }
4226 }
4239 }
4240 }
4254 /* Validate extent which is part of inode */
4259 /* Validate block references which are part of inode */
4261 }
4262 }
4285 }
4293 else
4302 }
4308 bad_inode:
4312 }
4315 {
4319 }
4324 {
4330 /*
4331 * i_blocks can be represented in a 32 bit variable
4332 * as multiple of 512 bytes
4333 */
4338 }
4343 /*
4344 * i_blocks can be represented in a 48 bit variable
4345 * as multiple of 512 bytes
4346 */
4352 /* i_block is stored in file system block size */
4356 }
4358 }
4363 };
4367 {
4392 }
4395 }
4397 /*
4398 * Opportunistically update the other time fields for other inodes in
4399 * the same inode table block.
4400 */
4403 {
4410 /*
4411 * Calculate the first inode in the inode table block. Inode
4412 * numbers are one-based. That is, the first inode in a block
4413 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4414 */
4421 }
4422 }
4424 /*
4425 * Post the struct inode info into an on-disk inode location in the
4426 * buffer-cache. This gobbles the caller's reference to the
4427 * buffer_head in the inode location struct.
4428 *
4429 * The caller must have write access to iloc->bh.
4430 */
4434 {
4441 uid_t i_uid;
4442 gid_t i_gid;
4446 /* For fields not tracked in the in-memory inode,
4447 * initialise them to zero for new inodes. */
4458 /*
4459 * Fix up interoperability with old kernels. Otherwise, old inodes get
4460 * re-used with the upper 16 bits of the uid/gid intact
4461 */
4470 }
4476 }
4488 }
4498 }
4504 }
4516 }
4520 }
4530 }
4531 }
4552 }
4554 out_brelse:
4558 }
4560 /*
4561 * ext4_write_inode()
4562 *
4563 * We are called from a few places:
4564 *
4565 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4566 * Here, there will be no transaction running. We wait for any running
4567 * transaction to commit.
4568 *
4569 * - Within flush work (sys_sync(), kupdate and such).
4570 * We wait on commit, if told to.
4571 *
4572 * - Within iput_final() -> write_inode_now()
4573 * We wait on commit, if told to.
4574 *
4575 * In all cases it is actually safe for us to return without doing anything,
4576 * because the inode has been copied into a raw inode buffer in
4577 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4578 * writeback.
4579 *
4580 * Note that we are absolutely dependent upon all inode dirtiers doing the
4581 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4582 * which we are interested.
4583 *
4584 * It would be a bug for them to not do this. The code:
4585 *
4586 * mark_inode_dirty(inode)
4587 * stuff();
4588 * inode->i_size = expr;
4589 *
4590 * is in error because write_inode() could occur while `stuff()' is running,
4591 * and the new i_size will be lost. Plus the inode will no longer be on the
4592 * superblock's dirty inode list.
4593 */
4595 {
4606 }
4608 /*
4609 * No need to force transaction in WB_SYNC_NONE mode. Also
4610 * ext4_sync_fs() will force the commit after everything is
4611 * written.
4612 */
4623 /*
4624 * sync(2) will flush the whole buffer cache. No need to do
4625 * it here separately for each inode.
4626 */
4633 }
4635 }
4637 }
4639 /*
4640 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4641 * buffers that are attached to a page stradding i_size and are undergoing
4642 * commit. In that case we have to wait for commit to finish and try again.
4643 */
4645 {
4653 /*
4654 * All buffers in the last page remain valid? Then there's nothing to
4655 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4656 * blocksize case
4657 */
4678 }
4679 }
4681 /*
4682 * ext4_setattr()
4683 *
4684 * Called from notify_change.
4685 *
4686 * We want to trap VFS attempts to truncate the file as soon as
4687 * possible. In particular, we want to make sure that when the VFS
4688 * shrinks i_size, we put the inode on the orphan list and modify
4689 * i_disksize immediately, so that during the subsequent flushing of
4690 * dirty pages and freeing of disk blocks, we can guarantee that any
4691 * commit will leave the blocks being flushed in an unused state on
4692 * disk. (On recovery, the inode will get truncated and the blocks will
4693 * be freed, so we have a strong guarantee that no future commit will
4694 * leave these blocks visible to the user.)
4695 *
4696 * Another thing we have to assure is that if we are in ordered mode
4697 * and inode is still attached to the committing transaction, we must
4698 * we start writeout of all the dirty pages which are being truncated.
4699 * This way we are sure that all the data written in the previous
4700 * transaction are already on disk (truncate waits for pages under
4701 * writeback).
4702 *
4703 * Called with inode->i_mutex down.
4704 */
4706 {
4720 }
4725 /* (user+group)*(old+new) structure, inode write (sb,
4726 * inode block, ? - but truncate inode update has it) */
4733 }
4738 }
4739 /* Update corresponding info in inode so that everything is in
4740 * one transaction */
4747 }
4759 }
4772 }
4778 }
4782 }
4783 /*
4784 * Update c/mtime on truncate up, ext4_truncate() will
4785 * update c/mtime in shrink case below
4786 */
4790 }
4796 /*
4797 * We have to update i_size under i_data_sem together
4798 * with i_disksize to avoid races with writeback code
4799 * running ext4_wb_update_i_disksize().
4800 */
4809 }
4810 }
4814 /*
4815 * Blocks are going to be removed from the inode. Wait
4816 * for dio in flight. Temporarily disable
4817 * dioread_nolock to prevent livelock.
4818 */
4826 }
4827 /*
4828 * Truncate pagecache after we've waited for commit
4829 * in data=journal mode to make pages freeable.
4830 */
4834 }
4839 }
4841 /*
4842 * If the call to ext4_truncate failed to get a transaction handle at
4843 * all, we need to clean up the in-core orphan list manually.
4844 */
4851 err_out:
4856 }
4860 {
4867 /*
4868 * If there is inline data in the inode, the inode will normally not
4869 * have data blocks allocated (it may have an external xattr block).
4870 * Report at least one sector for such files, so tools like tar, rsync,
4871 * others doen't incorrectly think the file is completely sparse.
4872 */
4876 /*
4877 * We can't update i_blocks if the block allocation is delayed
4878 * otherwise in the case of system crash before the real block
4879 * allocation is done, we will have i_blocks inconsistent with
4880 * on-disk file blocks.
4881 * We always keep i_blocks updated together with real
4882 * allocation. But to not confuse with user, stat
4883 * will return the blocks that include the delayed allocation
4884 * blocks for this file.
4885 */
4890 }
4894 {
4898 }
4900 /*
4901 * Account for index blocks, block groups bitmaps and block group
4902 * descriptor blocks if modify datablocks and index blocks
4903 * worse case, the indexs blocks spread over different block groups
4904 *
4905 * If datablocks are discontiguous, they are possible to spread over
4906 * different block groups too. If they are contiguous, with flexbg,
4907 * they could still across block group boundary.
4908 *
4909 * Also account for superblock, inode, quota and xattr blocks
4910 */
4913 {
4919 /*
4920 * How many index blocks need to touch to map @lblocks logical blocks
4921 * to @pextents physical extents?
4922 */
4927 /*
4928 * Now let's see how many group bitmaps and group descriptors need
4929 * to account
4930 */
4938 /* bitmaps and block group descriptor blocks */
4941 /* Blocks for super block, inode, quota and xattr blocks */
4945 }
4947 /*
4948 * Calculate the total number of credits to reserve to fit
4949 * the modification of a single pages into a single transaction,
4950 * which may include multiple chunks of block allocations.
4951 *
4952 * This could be called via ext4_write_begin()
4953 *
4954 * We need to consider the worse case, when
4955 * one new block per extent.
4956 */
4958 {
4964 /* Account for data blocks for journalled mode */
4968 }
4970 /*
4971 * Calculate the journal credits for a chunk of data modification.
4972 *
4973 * This is called from DIO, fallocate or whoever calling
4974 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4975 *
4976 * journal buffers for data blocks are not included here, as DIO
4977 * and fallocate do no need to journal data buffers.
4978 */
4980 {
4982 }
4984 /*
4985 * The caller must have previously called ext4_reserve_inode_write().
4986 * Give this, we know that the caller already has write access to iloc->bh.
4987 */
4990 {
4996 /* the do_update_inode consumes one bh->b_count */
4999 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5003 }
5005 /*
5006 * On success, We end up with an outstanding reference count against
5007 * iloc->bh. This _must_ be cleaned up later.
5008 */
5010 int
5013 {
5023 }
5024 }
5027 }
5029 /*
5030 * Expand an inode by new_extra_isize bytes.
5031 * Returns 0 on success or negative error number on failure.
5032 */
5037 {
5048 /* No extended attributes present */
5052 new_extra_isize);
5055 }
5057 /* try to expand with EAs present */
5060 }
5062 /*
5063 * What we do here is to mark the in-core inode as clean with respect to inode
5064 * dirtiness (it may still be data-dirty).
5065 * This means that the in-core inode may be reaped by prune_icache
5066 * without having to perform any I/O. This is a very good thing,
5067 * because *any* task may call prune_icache - even ones which
5068 * have a transaction open against a different journal.
5069 *
5070 * Is this cheating? Not really. Sure, we haven't written the
5071 * inode out, but prune_icache isn't a user-visible syncing function.
5072 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5073 * we start and wait on commits.
5074 */
5076 {
5088 /*
5089 * We need extra buffer credits since we may write into EA block
5090 * with this same handle. If journal_extend fails, then it will
5091 * only result in a minor loss of functionality for that inode.
5092 * If this is felt to be critical, then e2fsck should be run to
5093 * force a large enough s_min_extra_isize.
5094 */
5102 EXT4_STATE_NO_EXPAND);
5106 "Unable to expand inode %lu. Delete"
5109 mnt_count =
5111 }
5112 }
5113 }
5114 }
5118 }
5120 /*
5121 * ext4_dirty_inode() is called from __mark_inode_dirty()
5122 *
5123 * We're really interested in the case where a file is being extended.
5124 * i_size has been changed by generic_commit_write() and we thus need
5125 * to include the updated inode in the current transaction.
5126 *
5127 * Also, dquot_alloc_block() will always dirty the inode when blocks
5128 * are allocated to the file.
5129 *
5130 * If the inode is marked synchronous, we don't honour that here - doing
5131 * so would cause a commit on atime updates, which we don't bother doing.
5132 * We handle synchronous inodes at the highest possible level.
5133 *
5134 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5135 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5136 * to copy into the on-disk inode structure are the timestamp files.
5137 */
5139 {
5151 out:
5153 }
5155 #if 0
5156 /*
5157 * Bind an inode's backing buffer_head into this transaction, to prevent
5158 * it from being flushed to disk early. Unlike
5159 * ext4_reserve_inode_write, this leaves behind no bh reference and
5160 * returns no iloc structure, so the caller needs to repeat the iloc
5161 * lookup to mark the inode dirty later.
5162 */
5164 {
5175 NULL,
5178 }
5179 }
5182 }
5183 #endif
5186 {
5191 /*
5192 * We have to be very careful here: changing a data block's
5193 * journaling status dynamically is dangerous. If we write a
5194 * data block to the journal, change the status and then delete
5195 * that block, we risk forgetting to revoke the old log record
5196 * from the journal and so a subsequent replay can corrupt data.
5197 * So, first we make sure that the journal is empty and that
5198 * nobody is changing anything.
5199 */
5206 /* We have to allocate physical blocks for delalloc blocks
5207 * before flushing journal. otherwise delalloc blocks can not
5208 * be allocated any more. even more truncate on delalloc blocks
5209 * could trigger BUG by flushing delalloc blocks in journal.
5210 * There is no delalloc block in non-journal data mode.
5211 */
5216 }
5218 /* Wait for all existing dio workers */
5224 /*
5225 * OK, there are no updates running now, and all cached data is
5226 * synced to disk. We are now in a completely consistent state
5227 * which doesn't have anything in the journal, and we know that
5228 * no filesystem updates are running, so it is safe to modify
5229 * the inode's in-core data-journaling state flag now.
5230 */
5240 }
5242 }
5248 /* Finally we can mark the inode as dirty. */
5260 }
5263 {
5265 }
5268 {
5270 loff_t size;
5282 /* Delalloc case is easy... */
5288 ext4_da_get_block_prep);
5292 }
5296 /* Page got truncated from under us? */
5301 }
5305 else
5307 /*
5308 * Return if we have all the buffers mapped. This avoids the need to do
5309 * journal_start/journal_stop which can block and take a long time
5310 */
5314 ext4_bh_unmapped)) {
5315 /* Wait so that we don't change page under IO */
5319 }
5320 }
5322 /* OK, we need to fill the hole... */
5325 else
5327 retry_alloc:
5333 }
5342 }
5344 }
5348 out_ret:
5350 out:
5353 }