| blob | 612fbcf76b5c4820ad6a18d7723f7e65db803d18 |
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 }
1188 {
1211 }
1217 }
1231 }
1234 /* if we have allocated more blocks and copied
1235 * less. We will have blocks allocated outside
1236 * inode->i_size. So truncate them
1237 */
1245 /*
1246 * If truncate failed early the inode might still be
1247 * on the orphan list; we need to make sure the inode
1248 * is removed from the orphan list in that case.
1249 */
1252 }
1255 }
1257 /*
1258 * Reserve space for a single cluster
1259 */
1261 {
1266 /*
1267 * We will charge metadata quota at writeout time; this saves
1268 * us from metadata over-estimation, though we may go over by
1269 * a small amount in the end. Here we just reserve for data.
1270 */
1280 }
1286 }
1289 {
1300 /*
1301 * if there aren't enough reserved blocks, then the
1302 * counter is messed up somewhere. Since this
1303 * function is called from invalidate page, it's
1304 * harmless to return without any action.
1305 */
1307 "ino %lu, to_free %d with only %d reserved "
1312 }
1315 /* update fs dirty data blocks counter */
1321 }
1326 {
1334 ext4_fsblk_t lblk;
1347 to_release++;
1348 contiguous_blks++;
1354 contiguous_blks;
1357 }
1365 }
1367 /* If we have released all the blocks belonging to a cluster, then we
1368 * need to release the reserved space for that cluster. */
1377 num_clusters--;
1378 }
1379 }
1381 /*
1382 * Delayed allocation stuff
1383 */
1392 /*
1393 * Extent to map - this can be after first_page because that can be
1394 * fully mapped. We somewhat abuse m_flags to store whether the extent
1395 * is delalloc or unwritten.
1396 */
1399 };
1403 {
1410 /* This is necessary when next_page == 0. */
1421 }
1437 }
1439 }
1442 }
1443 }
1446 {
1465 }
1468 {
1470 }
1472 /*
1473 * This function is grabs code from the very beginning of
1474 * ext4_map_blocks, but assumes that the caller is from delayed write
1475 * time. This function looks up the requested blocks and sets the
1476 * buffer delay bit under the protection of i_data_sem.
1477 */
1481 {
1485 #ifdef ES_AGGRESSIVE_TEST
1489 #endif
1499 /* Lookup extent status tree firstly */
1505 }
1507 /*
1508 * Delayed extent could be allocated by fallocate.
1509 * So we need to check it.
1510 */
1516 }
1527 else
1530 #ifdef ES_AGGRESSIVE_TEST
1532 #endif
1534 }
1536 /*
1537 * Try to see if we can get the block without requesting a new
1538 * file system block.
1539 */
1545 else
1548 add_delayed:
1551 /*
1552 * XXX: __block_prepare_write() unmaps passed block,
1553 * is it OK?
1554 */
1555 /*
1556 * If the block was allocated from previously allocated cluster,
1557 * then we don't need to reserve it again. However we still need
1558 * to reserve metadata for every block we're going to write.
1559 */
1564 /* not enough space to reserve */
1567 }
1568 }
1575 }
1586 "ES len assertion failed for inode "
1590 }
1598 }
1600 out_unlock:
1604 }
1606 /*
1607 * This is a special get_block_t callback which is used by
1608 * ext4_da_write_begin(). It will either return mapped block or
1609 * reserve space for a single block.
1610 *
1611 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1612 * We also have b_blocknr = -1 and b_bdev initialized properly
1613 *
1614 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1615 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1616 * initialized properly.
1617 */
1620 {
1630 /*
1631 * first, we need to know whether the block is allocated already
1632 * preallocated blocks are unmapped but should treated
1633 * the same as allocated blocks.
1634 */
1643 /* A delayed write to unwritten bh should be marked
1644 * new and mapped. Mapped ensures that we don't do
1645 * get_block multiple times when we write to the same
1646 * offset and new ensures that we do proper zero out
1647 * for partial write.
1648 */
1651 }
1653 }
1656 {
1659 }
1662 {
1665 }
1669 {
1691 }
1694 }
1695 /*
1696 * We need to release the page lock before we start the
1697 * journal, so grab a reference so the page won't disappear
1698 * out from under us.
1699 */
1709 }
1715 /* The page got truncated from under us */
1719 }
1729 do_journal_get_write_access);
1732 write_end_fn);
1733 }
1745 out:
1747 out_no_pagelock:
1750 }
1752 /*
1753 * Note that we don't need to start a transaction unless we're journaling data
1754 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1755 * need to file the inode to the transaction's list in ordered mode because if
1756 * we are writing back data added by write(), the inode is already there and if
1757 * we are writing back data modified via mmap(), no one guarantees in which
1758 * transaction the data will hit the disk. In case we are journaling data, we
1759 * cannot start transaction directly because transaction start ranks above page
1760 * lock so we have to do some magic.
1761 *
1762 * This function can get called via...
1763 * - ext4_writepages after taking page lock (have journal handle)
1764 * - journal_submit_inode_data_buffers (no journal handle)
1765 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1766 * - grab_page_cache when doing write_begin (have journal handle)
1767 *
1768 * We don't do any block allocation in this function. If we have page with
1769 * multiple blocks we need to write those buffer_heads that are mapped. This
1770 * is important for mmaped based write. So if we do with blocksize 1K
1771 * truncate(f, 1024);
1772 * a = mmap(f, 0, 4096);
1773 * a[0] = 'a';
1774 * truncate(f, 4096);
1775 * we have in the page first buffer_head mapped via page_mkwrite call back
1776 * but other buffer_heads would be unmapped but dirty (dirty done via the
1777 * do_wp_page). So writepage should write the first block. If we modify
1778 * the mmap area beyond 1024 we will again get a page_fault and the
1779 * page_mkwrite callback will do the block allocation and mark the
1780 * buffer_heads mapped.
1781 *
1782 * We redirty the page if we have any buffer_heads that is either delay or
1783 * unwritten in the page.
1784 *
1785 * We can get recursively called as show below.
1786 *
1787 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1788 * ext4_writepage()
1789 *
1790 * But since we don't do any block allocation we should not deadlock.
1791 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1792 */
1795 {
1797 loff_t size;
1808 else
1812 /*
1813 * We cannot do block allocation or other extent handling in this
1814 * function. If there are buffers needing that, we have to redirty
1815 * the page. But we may reach here when we do a journal commit via
1816 * journal_submit_inode_data_buffers() and in that case we must write
1817 * allocated buffers to achieve data=ordered mode guarantees.
1818 */
1820 ext4_bh_delay_or_unwritten)) {
1823 /*
1824 * For memory cleaning there's no point in writing only
1825 * some buffers. So just bail out. Warn if we came here
1826 * from direct reclaim.
1827 */
1832 }
1834 }
1837 /*
1838 * It's mmapped pagecache. Add buffers and journal it. There
1839 * doesn't seem much point in redirtying the page here.
1840 */
1849 }
1852 /* Drop io_end reference we got from init */
1855 }
1858 {
1866 else
1875 }
1877 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1879 /*
1880 * mballoc gives us at most this number of blocks...
1881 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1882 * The rest of mballoc seems to handle chunks up to full group size.
1883 */
1884 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1886 /*
1887 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1888 *
1889 * @mpd - extent of blocks
1890 * @lblk - logical number of the block in the file
1891 * @bh - buffer head we want to add to the extent
1892 *
1893 * The function is used to collect contig. blocks in the same state. If the
1894 * buffer doesn't require mapping for writeback and we haven't started the
1895 * extent of buffers to map yet, the function returns 'true' immediately - the
1896 * caller can write the buffer right away. Otherwise the function returns true
1897 * if the block has been added to the extent, false if the block couldn't be
1898 * added.
1899 */
1902 {
1905 /* Buffer that doesn't need mapping for writeback? */
1908 /* So far no extent to map => we write the buffer right away */
1912 }
1914 /* First block in the extent? */
1920 }
1922 /* Don't go larger than mballoc is willing to allocate */
1926 /* Can we merge the block to our big extent? */
1931 }
1933 }
1935 /*
1936 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1937 *
1938 * @mpd - extent of blocks for mapping
1939 * @head - the first buffer in the page
1940 * @bh - buffer we should start processing from
1941 * @lblk - logical number of the block in the file corresponding to @bh
1942 *
1943 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1944 * the page for IO if all buffers in this page were mapped and there's no
1945 * accumulated extent of buffers to map or add buffers in the page to the
1946 * extent of buffers to map. The function returns 1 if the caller can continue
1947 * by processing the next page, 0 if it should stop adding buffers to the
1948 * extent to map because we cannot extend it anymore. It can also return value
1949 * < 0 in case of error during IO submission.
1950 */
1954 ext4_lblk_t lblk)
1955 {
1965 /* Found extent to map? */
1968 /* Everything mapped so far and we hit EOF */
1970 }
1972 /* So far everything mapped? Submit the page for IO. */
1977 }
1979 }
1981 /*
1982 * mpage_map_buffers - update buffers corresponding to changed extent and
1983 * submit fully mapped pages for IO
1984 *
1985 * @mpd - description of extent to map, on return next extent to map
1986 *
1987 * Scan buffers corresponding to changed extent (we expect corresponding pages
1988 * to be already locked) and update buffer state according to new extent state.
1989 * We map delalloc buffers to their physical location, clear unwritten bits,
1990 * and mark buffers as uninit when we perform writes to unwritten extents
1991 * and do extent conversion after IO is finished. If the last page is not fully
1992 * mapped, we update @map to the next extent in the last page that needs
1993 * mapping. Otherwise we submit the page for IO.
1994 */
1996 {
2003 ext4_lblk_t lblk;
2004 sector_t pblock;
2015 PAGEVEC_SIZE);
2023 /* Up to 'end' pages must be contiguous */
2030 /*
2031 * Buffer after end of mapped extent.
2032 * Find next buffer in the page to map.
2033 */
2036 /*
2037 * FIXME: If dioread_nolock supports
2038 * blocksize < pagesize, we need to make
2039 * sure we add size mapped so far to
2040 * io_end->size as the following call
2041 * can submit the page for IO.
2042 */
2049 }
2053 }
2057 /*
2058 * FIXME: This is going to break if dioread_nolock
2059 * supports blocksize < pagesize as we will try to
2060 * convert potentially unmapped parts of inode.
2061 */
2063 /* Page fully mapped - let IO run! */
2068 }
2069 start++;
2070 }
2072 }
2073 /* Extent fully mapped and matches with page boundary. We are done. */
2077 }
2080 {
2087 /*
2088 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2089 * to convert an unwritten extent to be initialized (in the case
2090 * where we have written into one or more preallocated blocks). It is
2091 * possible that we're going to need more metadata blocks than
2092 * previously reserved. However we must not fail because we're in
2093 * writeback and there is nothing we can do about it so it might result
2094 * in data loss. So use reserved blocks to allocate metadata if
2095 * possible.
2096 *
2097 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2098 * the blocks in question are delalloc blocks. This indicates
2099 * that the blocks and quotas has already been checked when
2100 * the data was copied into the page cache.
2101 */
2103 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2118 }
2120 }
2129 }
2131 }
2133 /*
2134 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2135 * mpd->len and submit pages underlying it for IO
2136 *
2137 * @handle - handle for journal operations
2138 * @mpd - extent to map
2139 * @give_up_on_write - we set this to true iff there is a fatal error and there
2140 * is no hope of writing the data. The caller should discard
2141 * dirty pages to avoid infinite loops.
2142 *
2143 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2144 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2145 * them to initialized or split the described range from larger unwritten
2146 * extent. Note that we need not map all the described range since allocation
2147 * can return less blocks or the range is covered by more unwritten extents. We
2148 * cannot map more because we are limited by reserved transaction credits. On
2149 * the other hand we always make sure that the last touched page is fully
2150 * mapped so that it can be written out (and thus forward progress is
2151 * guaranteed). After mapping we submit all mapped pages for IO.
2152 */
2156 {
2160 loff_t disksize;
2172 /*
2173 * Let the uper layers retry transient errors.
2174 * In the case of ENOSPC, if ext4_count_free_blocks()
2175 * is non-zero, a commit should free up blocks.
2176 */
2182 }
2184 "Delayed block allocation failed for "
2185 "inode %lu at logical offset %llu with"
2191 "This should not happen!! Data will "
2195 invalidate_dirty_pages:
2198 }
2200 /*
2201 * Update buffer state, submit mapped pages, and get us new
2202 * extent to map
2203 */
2209 update_disksize:
2210 /*
2211 * Update on-disk size after IO is submitted. Races with
2212 * truncate are avoided by checking i_size under i_data_sem.
2213 */
2217 loff_t i_size;
2233 }
2235 }
2237 /*
2238 * Calculate the total number of credits to reserve for one writepages
2239 * iteration. This is called from ext4_writepages(). We map an extent of
2240 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2241 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2242 * bpp - 1 blocks in bpp different extents.
2243 */
2245 {
2250 }
2252 /*
2253 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2254 * and underlying extent to map
2255 *
2256 * @mpd - where to look for pages
2257 *
2258 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2259 * IO immediately. When we find a page which isn't mapped we start accumulating
2260 * extent of buffers underlying these pages that needs mapping (formed by
2261 * either delayed or unwritten buffers). We also lock the pages containing
2262 * these buffers. The extent found is returned in @mpd structure (starting at
2263 * mpd->lblk with length mpd->len blocks).
2264 *
2265 * Note that this function can attach bios to one io_end structure which are
2266 * neither logically nor physically contiguous. Although it may seem as an
2267 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2268 * case as we need to track IO to all buffers underlying a page in one io_end.
2269 */
2271 {
2281 ext4_lblk_t lblk;
2286 else
2301 /*
2302 * At this point, the page may be truncated or
2303 * invalidated (changing page->mapping to NULL), or
2304 * even swizzled back from swapper_space to tmpfs file
2305 * mapping. However, page->index will not change
2306 * because we have a reference on the page.
2307 */
2311 /*
2312 * Accumulated enough dirty pages? This doesn't apply
2313 * to WB_SYNC_ALL mode. For integrity sync we have to
2314 * keep going because someone may be concurrently
2315 * dirtying pages, and we might have synced a lot of
2316 * newly appeared dirty pages, but have not synced all
2317 * of the old dirty pages.
2318 */
2322 /* If we can't merge this page, we are done. */
2327 /*
2328 * If the page is no longer dirty, or its mapping no
2329 * longer corresponds to inode we are writing (which
2330 * means it has been truncated or invalidated), or the
2331 * page is already under writeback and we are not doing
2332 * a data integrity writeback, skip the page
2333 */
2340 }
2348 /* Add all dirty buffers to mpd */
2356 left--;
2357 }
2360 }
2362 out:
2365 }
2369 {
2374 }
2378 {
2394 /*
2395 * No pages to write? This is mainly a kludge to avoid starting
2396 * a transaction for special inodes like journal inode on last iput()
2397 * because that could violate lock ordering on umount
2398 */
2409 }
2411 /*
2412 * If the filesystem has aborted, it is read-only, so return
2413 * right away instead of dumping stack traces later on that
2414 * will obscure the real source of the problem. We test
2415 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2416 * the latter could be true if the filesystem is mounted
2417 * read-only, and in that case, ext4_writepages should
2418 * *never* be called, so if that ever happens, we would want
2419 * the stack trace.
2420 */
2424 }
2427 /*
2428 * We may need to convert up to one extent per block in
2429 * the page and we may dirty the inode.
2430 */
2432 }
2434 /*
2435 * If we have inline data and arrive here, it means that
2436 * we will soon create the block for the 1st page, so
2437 * we'd better clear the inline data here.
2438 */
2440 /* Just inode will be modified... */
2445 }
2447 EXT4_STATE_MAY_INLINE_DATA));
2450 }
2464 }
2469 retry:
2475 /* For each extent of pages we use new io_end */
2480 }
2482 /*
2483 * We have two constraints: We find one extent to map and we
2484 * must always write out whole page (makes a difference when
2485 * blocksize < pagesize) so that we don't block on IO when we
2486 * try to write out the rest of the page. Journalled mode is
2487 * not supported by delalloc.
2488 */
2492 /* start a new transaction */
2500 /* Release allocated io_end */
2503 }
2512 /*
2513 * We scanned the whole range (or exhausted
2514 * nr_to_write), submitted what was mapped and
2515 * didn't find anything needing mapping. We are
2516 * done.
2517 */
2519 }
2520 }
2522 /* Submit prepared bio */
2524 /* Unlock pages we didn't use */
2526 /* Drop our io_end reference we got from init */
2530 /*
2531 * Commit the transaction which would
2532 * free blocks released in the transaction
2533 * and try again
2534 */
2538 }
2539 /* Fatal error - ENOMEM, EIO... */
2542 }
2549 }
2551 /* Update index */
2553 /*
2554 * Set the writeback_index so that range_cyclic
2555 * mode will write it back later
2556 */
2559 out_writepages:
2563 }
2566 {
2570 /*
2571 * switch to non delalloc mode if we are running low
2572 * on free block. The free block accounting via percpu
2573 * counters can get slightly wrong with percpu_counter_batch getting
2574 * accumulated on each CPU without updating global counters
2575 * Delalloc need an accurate free block accounting. So switch
2576 * to non delalloc when we are near to error range.
2577 */
2578 free_clusters =
2580 dirty_clusters =
2582 /*
2583 * Start pushing delalloc when 1/2 of free blocks are dirty.
2584 */
2590 /*
2591 * free block count is less than 150% of dirty blocks
2592 * or free blocks is less than watermark
2593 */
2595 }
2597 }
2599 /* We always reserve for an inode update; the superblock could be there too */
2601 {
2603 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2609 /* We might need to update the superblock to set LARGE_FILE */
2611 }
2616 {
2619 pgoff_t index;
2629 }
2641 }
2643 /*
2644 * grab_cache_page_write_begin() can take a long time if the
2645 * system is thrashing due to memory pressure, or if the page
2646 * is being written back. So grab it first before we start
2647 * the transaction handle. This also allows us to allocate
2648 * the page (if needed) without using GFP_NOFS.
2649 */
2650 retry_grab:
2656 /*
2657 * With delayed allocation, we don't log the i_disksize update
2658 * if there is delayed block allocation. But we still need
2659 * to journalling the i_disksize update if writes to the end
2660 * of file which has an already mapped buffer.
2661 */
2662 retry_journal:
2668 }
2672 /* The page got truncated from under us */
2677 }
2678 /* In case writeback began while the page was unlocked */
2681 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2683 ext4_da_get_block_prep);
2684 #else
2686 #endif
2690 /*
2691 * block_write_begin may have instantiated a few blocks
2692 * outside i_size. Trim these off again. Don't need
2693 * i_size_read because we hold i_mutex.
2694 */
2704 }
2708 }
2710 /*
2711 * Check if we should update i_disksize
2712 * when write to the end of file but not require block allocation
2713 */
2716 {
2731 }
2737 {
2741 loff_t new_i_size;
2753 /*
2754 * generic_write_end() will run mark_inode_dirty() if i_size
2755 * changes. So let's piggyback the i_disksize mark_inode_dirty
2756 * into that.
2757 */
2763 /* We need to mark inode dirty even if
2764 * new_i_size is less that inode->i_size
2765 * bu greater than i_disksize.(hint delalloc)
2766 */
2768 }
2769 }
2775 page);
2776 else
2788 }
2792 {
2793 /*
2794 * Drop reserved blocks
2795 */
2802 out:
2806 }
2808 /*
2809 * Force all delayed allocation blocks to be allocated for a given inode.
2810 */
2812 {
2818 /*
2819 * We do something simple for now. The filemap_flush() will
2820 * also start triggering a write of the data blocks, which is
2821 * not strictly speaking necessary (and for users of
2822 * laptop_mode, not even desirable). However, to do otherwise
2823 * would require replicating code paths in:
2824 *
2825 * ext4_writepages() ->
2826 * write_cache_pages() ---> (via passed in callback function)
2827 * __mpage_da_writepage() -->
2828 * mpage_add_bh_to_extent()
2829 * mpage_da_map_blocks()
2830 *
2831 * The problem is that write_cache_pages(), located in
2832 * mm/page-writeback.c, marks pages clean in preparation for
2833 * doing I/O, which is not desirable if we're not planning on
2834 * doing I/O at all.
2835 *
2836 * We could call write_cache_pages(), and then redirty all of
2837 * the pages by calling redirty_page_for_writepage() but that
2838 * would be ugly in the extreme. So instead we would need to
2839 * replicate parts of the code in the above functions,
2840 * simplifying them because we wouldn't actually intend to
2841 * write out the pages, but rather only collect contiguous
2842 * logical block extents, call the multi-block allocator, and
2843 * then update the buffer heads with the block allocations.
2844 *
2845 * For now, though, we'll cheat by calling filemap_flush(),
2846 * which will map the blocks, and start the I/O, but not
2847 * actually wait for the I/O to complete.
2848 */
2850 }
2852 /*
2853 * bmap() is special. It gets used by applications such as lilo and by
2854 * the swapper to find the on-disk block of a specific piece of data.
2855 *
2856 * Naturally, this is dangerous if the block concerned is still in the
2857 * journal. If somebody makes a swapfile on an ext4 data-journaling
2858 * filesystem and enables swap, then they may get a nasty shock when the
2859 * data getting swapped to that swapfile suddenly gets overwritten by
2860 * the original zero's written out previously to the journal and
2861 * awaiting writeback in the kernel's buffer cache.
2862 *
2863 * So, if we see any bmap calls here on a modified, data-journaled file,
2864 * take extra steps to flush any blocks which might be in the cache.
2865 */
2867 {
2872 /*
2873 * We can get here for an inline file via the FIBMAP ioctl
2874 */
2880 /*
2881 * With delalloc we want to sync the file
2882 * so that we can make sure we allocate
2883 * blocks for file
2884 */
2886 }
2890 /*
2891 * This is a REALLY heavyweight approach, but the use of
2892 * bmap on dirty files is expected to be extremely rare:
2893 * only if we run lilo or swapon on a freshly made file
2894 * do we expect this to happen.
2895 *
2896 * (bmap requires CAP_SYS_RAWIO so this does not
2897 * represent an unprivileged user DOS attack --- we'd be
2898 * in trouble if mortal users could trigger this path at
2899 * will.)
2900 *
2901 * NB. EXT4_STATE_JDATA is not set on files other than
2902 * regular files. If somebody wants to bmap a directory
2903 * or symlink and gets confused because the buffer
2904 * hasn't yet been flushed to disk, they deserve
2905 * everything they get.
2906 */
2916 }
2919 }
2922 {
2935 }
2937 static int
2940 {
2943 /* If the file has inline data, no need to do readpages. */
2948 }
2952 {
2955 /* No journalling happens on data buffers when this function is used */
2959 }
2964 {
2969 /*
2970 * If it's a full truncate we just forget about the pending dirtying
2971 */
2976 }
2978 /* Wrapper for aops... */
2982 {
2984 }
2987 {
2992 /* Page has dirty journalled data -> cannot release */
2997 else
2999 }
3001 /*
3002 * ext4_get_block used when preparing for a DIO write or buffer write.
3003 * We allocate an uinitialized extent if blocks haven't been allocated.
3004 * The extent will be converted to initialized after the IO is complete.
3005 */
3008 {
3012 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3013 }
3017 {
3021 EXT4_GET_BLOCKS_NO_LOCK);
3022 }
3026 {
3033 }
3037 {
3040 /* if not async direct IO just return */
3047 size);
3053 }
3055 /*
3056 * For ext4 extent files, ext4 will do direct-io write to holes,
3057 * preallocated extents, and those write extend the file, no need to
3058 * fall back to buffered IO.
3059 *
3060 * For holes, we fallocate those blocks, mark them as unwritten
3061 * If those blocks were preallocated, we mark sure they are split, but
3062 * still keep the range to write as unwritten.
3063 *
3064 * The unwritten extents will be converted to written when DIO is completed.
3065 * For async direct IO, since the IO may still pending when return, we
3066 * set up an end_io call back function, which will do the conversion
3067 * when async direct IO completed.
3068 *
3069 * If the O_DIRECT write will extend the file then add this inode to the
3070 * orphan list. So recovery will truncate it back to the original size
3071 * if the machine crashes during the write.
3072 *
3073 */
3075 loff_t offset)
3076 {
3079 ssize_t ret;
3087 /* Use the old path for reads and writes beyond i_size. */
3093 /*
3094 * Make all waiters for direct IO properly wait also for extent
3095 * conversion. This also disallows race between truncate() and
3096 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3097 */
3101 /* If we do a overwrite dio, i_mutex locking can be released */
3107 }
3109 /*
3110 * We could direct write to holes and fallocate.
3111 *
3112 * Allocated blocks to fill the hole are marked as
3113 * unwritten to prevent parallel buffered read to expose
3114 * the stale data before DIO complete the data IO.
3115 *
3116 * As to previously fallocated extents, ext4 get_block will
3117 * just simply mark the buffer mapped but still keep the
3118 * extents unwritten.
3119 *
3120 * For non AIO case, we will convert those unwritten extents
3121 * to written after return back from blockdev_direct_IO.
3122 *
3123 * For async DIO, the conversion needs to be deferred when the
3124 * IO is completed. The ext4 end_io callback function will be
3125 * called to take care of the conversion work. Here for async
3126 * case, we allocate an io_end structure to hook to the iocb.
3127 */
3135 }
3136 /*
3137 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3138 */
3140 /*
3141 * we save the io structure for current async direct
3142 * IO, so that later ext4_map_blocks() could flag the
3143 * io structure whether there is a unwritten extents
3144 * needs to be converted when IO is completed.
3145 */
3147 }
3154 }
3155 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3157 #endif
3161 else
3164 get_block_func,
3167 /*
3168 * Put our reference to io_end. This can free the io_end structure e.g.
3169 * in sync IO case or in case of error. It can even perform extent
3170 * conversion if all bios we submitted finished before we got here.
3171 * Note that in that case iocb->private can be already set to NULL
3172 * here.
3173 */
3177 /*
3178 * When no IO was submitted ext4_end_io_dio() was not
3179 * called so we have to put iocb's reference.
3180 */
3186 }
3187 }
3189 EXT4_STATE_DIO_UNWRITTEN)) {
3191 /*
3192 * for non AIO case, since the IO is already
3193 * completed, we could do the conversion right here
3194 */
3200 }
3202 retake_lock:
3205 /* take i_mutex locking again if we do a ovewrite dio */
3209 }
3212 }
3215 loff_t offset)
3216 {
3220 ssize_t ret;
3222 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3225 #endif
3227 /*
3228 * If we are doing data journalling we don't support O_DIRECT
3229 */
3233 /* Let buffer I/O handle the inline data case. */
3240 else
3244 }
3246 /*
3247 * Pages can be marked dirty completely asynchronously from ext4's journalling
3248 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3249 * much here because ->set_page_dirty is called under VFS locks. The page is
3250 * not necessarily locked.
3251 *
3252 * We cannot just dirty the page and leave attached buffers clean, because the
3253 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3254 * or jbddirty because all the journalling code will explode.
3255 *
3256 * So what we do is to mark the page "pending dirty" and next time writepage
3257 * is called, propagate that into the buffers appropriately.
3258 */
3260 {
3263 }
3279 };
3295 };
3311 };
3314 {
3327 }
3330 else
3332 }
3336 {
3340 ext4_lblk_t iblock;
3358 /* Find the buffer that contains "offset" */
3363 iblock++;
3365 }
3369 }
3373 /* unmapped? It's a hole - nothing to do */
3377 }
3378 }
3380 /* Ok, it's mapped. Make sure it's up-to-date */
3388 /* Uhhuh. Read error. Complain and punt. */
3393 /* We expect the key to be set. */
3397 }
3398 }
3404 }
3415 }
3417 unlock:
3421 }
3423 /*
3424 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3425 * starting from file offset 'from'. The range to be zero'd must
3426 * be contained with in one block. If the specified range exceeds
3427 * the end of the block it will be shortened to end of the block
3428 * that cooresponds to 'from'
3429 */
3432 {
3438 /*
3439 * correct length if it does not fall between
3440 * 'from' and the end of the block
3441 */
3448 }
3450 /*
3451 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3452 * up to the end of the block which corresponds to `from'.
3453 * This required during truncate. We need to physically zero the tail end
3454 * of that block so it doesn't yield old data if the file is later grown.
3455 */
3458 {
3468 }
3472 {
3486 /* Handle partial zero within the single block */
3492 }
3493 /* Handle partial zero out on the start of the range */
3499 }
3500 /* Handle partial zero out on the end of the range */
3506 }
3509 {
3517 }
3519 /*
3520 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3521 * associated with the given offset and length
3522 *
3523 * @inode: File inode
3524 * @offset: The offset where the hole will begin
3525 * @len: The length of the hole
3526 *
3527 * Returns: 0 on success or negative on failure
3528 */
3531 {
3545 /*
3546 * Write out all dirty pages to avoid race conditions
3547 * Then release them.
3548 */
3554 }
3558 /* No need to punch hole beyond i_size */
3562 /*
3563 * If the hole extends beyond i_size, set the hole
3564 * to end after the page that contains i_size
3565 */
3569 offset;
3570 }
3574 /*
3575 * Attach jinode to inode for jbd2 if we do any zeroing of
3576 * partial block
3577 */
3582 }
3587 /* Now release the pages and zero block aligned part of pages*/
3590 last_block_offset);
3592 /* Wait all existing dio workers, newcomers will block on i_mutex */
3598 else
3605 }
3608 length);
3616 /* If there are no blocks to remove, return now */
3628 }
3633 else
3635 stop_block);
3641 /* Now release the pages again to reduce race window */
3644 last_block_offset);
3648 out_stop:
3650 out_dio:
3652 out_mutex:
3655 }
3658 {
3671 }
3675 }
3680 }
3682 /*
3683 * ext4_truncate()
3684 *
3685 * We block out ext4_get_block() block instantiations across the entire
3686 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3687 * simultaneously on behalf of the same inode.
3688 *
3689 * As we work through the truncate and commit bits of it to the journal there
3690 * is one core, guiding principle: the file's tree must always be consistent on
3691 * disk. We must be able to restart the truncate after a crash.
3692 *
3693 * The file's tree may be transiently inconsistent in memory (although it
3694 * probably isn't), but whenever we close off and commit a journal transaction,
3695 * the contents of (the filesystem + the journal) must be consistent and
3696 * restartable. It's pretty simple, really: bottom up, right to left (although
3697 * left-to-right works OK too).
3698 *
3699 * Note that at recovery time, journal replay occurs *before* the restart of
3700 * truncate against the orphan inode list.
3701 *
3702 * The committed inode has the new, desired i_size (which is the same as
3703 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3704 * that this inode's truncate did not complete and it will again call
3705 * ext4_truncate() to have another go. So there will be instantiated blocks
3706 * to the right of the truncation point in a crashed ext4 filesystem. But
3707 * that's fine - as long as they are linked from the inode, the post-crash
3708 * ext4_truncate() run will find them and release them.
3709 */
3711 {
3717 /*
3718 * There is a possibility that we're either freeing the inode
3719 * or it's a completely new inode. In those cases we might not
3720 * have i_mutex locked because it's not necessary.
3721 */
3740 }
3742 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3746 }
3750 else
3757 }
3762 /*
3763 * We add the inode to the orphan list, so that if this
3764 * truncate spans multiple transactions, and we crash, we will
3765 * resume the truncate when the filesystem recovers. It also
3766 * marks the inode dirty, to catch the new size.
3767 *
3768 * Implication: the file must always be in a sane, consistent
3769 * truncatable state while each transaction commits.
3770 */
3780 else
3788 out_stop:
3789 /*
3790 * If this was a simple ftruncate() and the file will remain alive,
3791 * then we need to clear up the orphan record which we created above.
3792 * However, if this was a real unlink then we were called by
3793 * ext4_evict_inode(), and we allow that function to clean up the
3794 * orphan info for us.
3795 */
3804 }
3806 /*
3807 * ext4_get_inode_loc returns with an extra refcount against the inode's
3808 * underlying buffer_head on success. If 'in_mem' is true, we have all
3809 * data in memory that is needed to recreate the on-disk version of this
3810 * inode.
3811 */
3814 {
3818 ext4_fsblk_t block;
3830 /*
3831 * Figure out the offset within the block group inode table
3832 */
3845 /*
3846 * If the buffer has the write error flag, we have failed
3847 * to write out another inode in the same block. In this
3848 * case, we don't have to read the block because we may
3849 * read the old inode data successfully.
3850 */
3855 /* someone brought it uptodate while we waited */
3858 }
3860 /*
3861 * If we have all information of the inode in memory and this
3862 * is the only valid inode in the block, we need not read the
3863 * block.
3864 */
3871 /* Is the inode bitmap in cache? */
3876 /*
3877 * If the inode bitmap isn't in cache then the
3878 * optimisation may end up performing two reads instead
3879 * of one, so skip it.
3880 */
3884 }
3890 }
3893 /* all other inodes are free, so skip I/O */
3898 }
3899 }
3901 make_io:
3902 /*
3903 * If we need to do any I/O, try to pre-readahead extra
3904 * blocks from the inode table.
3905 */
3912 /* s_inode_readahead_blks is always a power of 2 */
3925 }
3927 /*
3928 * There are other valid inodes in the buffer, this inode
3929 * has in-inode xattrs, or we don't have this inode in memory.
3930 * Read the block from disk.
3931 */
3942 }
3943 }
3944 has_buffer:
3947 }
3950 {
3951 /* We have all inode data except xattrs in memory here. */
3954 }
3957 {
3975 }
3977 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3979 {
3988 EXT4_DIRSYNC_FL);
4000 }
4004 {
4005 blkcnt_t i_blocks ;
4010 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4011 /* we are using combined 48 bit field */
4015 /* i_blocks represent file system block size */
4019 }
4022 }
4023 }
4028 {
4036 }
4039 {
4047 uid_t i_uid;
4048 gid_t i_gid;
4073 }
4077 /* Precompute checksum seed for inode metadata */
4080 __u32 csum;
4087 }
4093 }
4101 }
4110 /* We now have enough fields to check if the inode was active or not.
4111 * This is needed because nfsd might try to access dead inodes
4112 * the test is that same one that e2fsck uses
4113 * NeilBrown 1999oct15
4114 */
4119 /* this inode is deleted */
4122 }
4123 /* The only unlinked inodes we let through here have
4124 * valid i_mode and are being read by the orphan
4125 * recovery code: that's fine, we're about to complete
4126 * the process of deleting those.
4127 * OR it is the EXT4_BOOT_LOADER_INO which is
4128 * not initialized on a new filesystem. */
4129 }
4138 #ifdef CONFIG_QUOTA
4140 #endif
4144 /*
4145 * NOTE! The in-memory inode i_data array is in little-endian order
4146 * even on big-endian machines: we do NOT byteswap the block numbers!
4147 */
4152 /*
4153 * Set transaction id's of transactions that have to be committed
4154 * to finish f[data]sync. We set them to currently running transaction
4155 * as we cannot be sure that the inode or some of its metadata isn't
4156 * part of the transaction - the inode could have been reclaimed and
4157 * now it is reread from disk.
4158 */
4161 tid_t tid;
4166 else
4170 else
4175 }
4179 /* The extra space is currently unused. Use it. */
4181 EXT4_GOOD_OLD_INODE_SIZE;
4184 }
4185 }
4198 }
4199 }
4213 /* Validate extent which is part of inode */
4218 /* Validate block references which are part of inode */
4220 }
4221 }
4244 }
4251 else
4260 }
4266 bad_inode:
4270 }
4273 {
4277 }
4282 {
4288 /*
4289 * i_blocks can be represented in a 32 bit variable
4290 * as multiple of 512 bytes
4291 */
4296 }
4301 /*
4302 * i_blocks can be represented in a 48 bit variable
4303 * as multiple of 512 bytes
4304 */
4310 /* i_block is stored in file system block size */
4314 }
4316 }
4321 };
4325 {
4350 }
4353 }
4355 /*
4356 * Opportunistically update the other time fields for other inodes in
4357 * the same inode table block.
4358 */
4361 {
4368 /*
4369 * Calculate the first inode in the inode table block. Inode
4370 * numbers are one-based. That is, the first inode in a block
4371 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4372 */
4379 }
4380 }
4382 /*
4383 * Post the struct inode info into an on-disk inode location in the
4384 * buffer-cache. This gobbles the caller's reference to the
4385 * buffer_head in the inode location struct.
4386 *
4387 * The caller must have write access to iloc->bh.
4388 */
4392 {
4399 uid_t i_uid;
4400 gid_t i_gid;
4404 /* For fields not tracked in the in-memory inode,
4405 * initialise them to zero for new inodes. */
4416 /*
4417 * Fix up interoperability with old kernels. Otherwise, old inodes get
4418 * re-used with the upper 16 bits of the uid/gid intact
4419 */
4428 }
4434 }
4446 }
4456 }
4459 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4463 }
4475 }
4479 }
4489 }
4490 }
4509 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4512 }
4514 out_brelse:
4518 }
4520 /*
4521 * ext4_write_inode()
4522 *
4523 * We are called from a few places:
4524 *
4525 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4526 * Here, there will be no transaction running. We wait for any running
4527 * transaction to commit.
4528 *
4529 * - Within flush work (sys_sync(), kupdate and such).
4530 * We wait on commit, if told to.
4531 *
4532 * - Within iput_final() -> write_inode_now()
4533 * We wait on commit, if told to.
4534 *
4535 * In all cases it is actually safe for us to return without doing anything,
4536 * because the inode has been copied into a raw inode buffer in
4537 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4538 * writeback.
4539 *
4540 * Note that we are absolutely dependent upon all inode dirtiers doing the
4541 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4542 * which we are interested.
4543 *
4544 * It would be a bug for them to not do this. The code:
4545 *
4546 * mark_inode_dirty(inode)
4547 * stuff();
4548 * inode->i_size = expr;
4549 *
4550 * is in error because write_inode() could occur while `stuff()' is running,
4551 * and the new i_size will be lost. Plus the inode will no longer be on the
4552 * superblock's dirty inode list.
4553 */
4555 {
4566 }
4568 /*
4569 * No need to force transaction in WB_SYNC_NONE mode. Also
4570 * ext4_sync_fs() will force the commit after everything is
4571 * written.
4572 */
4583 /*
4584 * sync(2) will flush the whole buffer cache. No need to do
4585 * it here separately for each inode.
4586 */
4593 }
4595 }
4597 }
4599 /*
4600 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4601 * buffers that are attached to a page stradding i_size and are undergoing
4602 * commit. In that case we have to wait for commit to finish and try again.
4603 */
4605 {
4613 /*
4614 * All buffers in the last page remain valid? Then there's nothing to
4615 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4616 * blocksize case
4617 */
4638 }
4639 }
4641 /*
4642 * ext4_setattr()
4643 *
4644 * Called from notify_change.
4645 *
4646 * We want to trap VFS attempts to truncate the file as soon as
4647 * possible. In particular, we want to make sure that when the VFS
4648 * shrinks i_size, we put the inode on the orphan list and modify
4649 * i_disksize immediately, so that during the subsequent flushing of
4650 * dirty pages and freeing of disk blocks, we can guarantee that any
4651 * commit will leave the blocks being flushed in an unused state on
4652 * disk. (On recovery, the inode will get truncated and the blocks will
4653 * be freed, so we have a strong guarantee that no future commit will
4654 * leave these blocks visible to the user.)
4655 *
4656 * Another thing we have to assure is that if we are in ordered mode
4657 * and inode is still attached to the committing transaction, we must
4658 * we start writeout of all the dirty pages which are being truncated.
4659 * This way we are sure that all the data written in the previous
4660 * transaction are already on disk (truncate waits for pages under
4661 * writeback).
4662 *
4663 * Called with inode->i_mutex down.
4664 */
4666 {
4680 }
4685 /* (user+group)*(old+new) structure, inode write (sb,
4686 * inode block, ? - but truncate inode update has it) */
4693 }
4698 }
4699 /* Update corresponding info in inode so that everything is in
4700 * one transaction */
4707 }
4719 }
4732 }
4738 }
4742 }
4743 /*
4744 * Update c/mtime on truncate up, ext4_truncate() will
4745 * update c/mtime in shrink case below
4746 */
4750 }
4756 /*
4757 * We have to update i_size under i_data_sem together
4758 * with i_disksize to avoid races with writeback code
4759 * running ext4_wb_update_i_disksize().
4760 */
4769 }
4770 }
4774 /*
4775 * Blocks are going to be removed from the inode. Wait
4776 * for dio in flight. Temporarily disable
4777 * dioread_nolock to prevent livelock.
4778 */
4786 }
4787 /*
4788 * Truncate pagecache after we've waited for commit
4789 * in data=journal mode to make pages freeable.
4790 */
4794 }
4799 }
4801 /*
4802 * If the call to ext4_truncate failed to get a transaction handle at
4803 * all, we need to clean up the in-core orphan list manually.
4804 */
4811 err_out:
4816 }
4820 {
4827 /*
4828 * If there is inline data in the inode, the inode will normally not
4829 * have data blocks allocated (it may have an external xattr block).
4830 * Report at least one sector for such files, so tools like tar, rsync,
4831 * others doen't incorrectly think the file is completely sparse.
4832 */
4836 /*
4837 * We can't update i_blocks if the block allocation is delayed
4838 * otherwise in the case of system crash before the real block
4839 * allocation is done, we will have i_blocks inconsistent with
4840 * on-disk file blocks.
4841 * We always keep i_blocks updated together with real
4842 * allocation. But to not confuse with user, stat
4843 * will return the blocks that include the delayed allocation
4844 * blocks for this file.
4845 */
4850 }
4854 {
4858 }
4860 /*
4861 * Account for index blocks, block groups bitmaps and block group
4862 * descriptor blocks if modify datablocks and index blocks
4863 * worse case, the indexs blocks spread over different block groups
4864 *
4865 * If datablocks are discontiguous, they are possible to spread over
4866 * different block groups too. If they are contiguous, with flexbg,
4867 * they could still across block group boundary.
4868 *
4869 * Also account for superblock, inode, quota and xattr blocks
4870 */
4873 {
4879 /*
4880 * How many index blocks need to touch to map @lblocks logical blocks
4881 * to @pextents physical extents?
4882 */
4887 /*
4888 * Now let's see how many group bitmaps and group descriptors need
4889 * to account
4890 */
4898 /* bitmaps and block group descriptor blocks */
4901 /* Blocks for super block, inode, quota and xattr blocks */
4905 }
4907 /*
4908 * Calculate the total number of credits to reserve to fit
4909 * the modification of a single pages into a single transaction,
4910 * which may include multiple chunks of block allocations.
4911 *
4912 * This could be called via ext4_write_begin()
4913 *
4914 * We need to consider the worse case, when
4915 * one new block per extent.
4916 */
4918 {
4924 /* Account for data blocks for journalled mode */
4928 }
4930 /*
4931 * Calculate the journal credits for a chunk of data modification.
4932 *
4933 * This is called from DIO, fallocate or whoever calling
4934 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4935 *
4936 * journal buffers for data blocks are not included here, as DIO
4937 * and fallocate do no need to journal data buffers.
4938 */
4940 {
4942 }
4944 /*
4945 * The caller must have previously called ext4_reserve_inode_write().
4946 * Give this, we know that the caller already has write access to iloc->bh.
4947 */
4950 {
4956 /* the do_update_inode consumes one bh->b_count */
4959 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4963 }
4965 /*
4966 * On success, We end up with an outstanding reference count against
4967 * iloc->bh. This _must_ be cleaned up later.
4968 */
4970 int
4973 {
4983 }
4984 }
4987 }
4989 /*
4990 * Expand an inode by new_extra_isize bytes.
4991 * Returns 0 on success or negative error number on failure.
4992 */
4997 {
5008 /* No extended attributes present */
5012 new_extra_isize);
5015 }
5017 /* try to expand with EAs present */
5020 }
5022 /*
5023 * What we do here is to mark the in-core inode as clean with respect to inode
5024 * dirtiness (it may still be data-dirty).
5025 * This means that the in-core inode may be reaped by prune_icache
5026 * without having to perform any I/O. This is a very good thing,
5027 * because *any* task may call prune_icache - even ones which
5028 * have a transaction open against a different journal.
5029 *
5030 * Is this cheating? Not really. Sure, we haven't written the
5031 * inode out, but prune_icache isn't a user-visible syncing function.
5032 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5033 * we start and wait on commits.
5034 */
5036 {
5048 /*
5049 * We need extra buffer credits since we may write into EA block
5050 * with this same handle. If journal_extend fails, then it will
5051 * only result in a minor loss of functionality for that inode.
5052 * If this is felt to be critical, then e2fsck should be run to
5053 * force a large enough s_min_extra_isize.
5054 */
5062 EXT4_STATE_NO_EXPAND);
5066 "Unable to expand inode %lu. Delete"
5069 mnt_count =
5071 }
5072 }
5073 }
5074 }
5078 }
5080 /*
5081 * ext4_dirty_inode() is called from __mark_inode_dirty()
5082 *
5083 * We're really interested in the case where a file is being extended.
5084 * i_size has been changed by generic_commit_write() and we thus need
5085 * to include the updated inode in the current transaction.
5086 *
5087 * Also, dquot_alloc_block() will always dirty the inode when blocks
5088 * are allocated to the file.
5089 *
5090 * If the inode is marked synchronous, we don't honour that here - doing
5091 * so would cause a commit on atime updates, which we don't bother doing.
5092 * We handle synchronous inodes at the highest possible level.
5093 *
5094 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5095 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5096 * to copy into the on-disk inode structure are the timestamp files.
5097 */
5099 {
5111 out:
5113 }
5115 #if 0
5116 /*
5117 * Bind an inode's backing buffer_head into this transaction, to prevent
5118 * it from being flushed to disk early. Unlike
5119 * ext4_reserve_inode_write, this leaves behind no bh reference and
5120 * returns no iloc structure, so the caller needs to repeat the iloc
5121 * lookup to mark the inode dirty later.
5122 */
5124 {
5135 NULL,
5138 }
5139 }
5142 }
5143 #endif
5146 {
5151 /*
5152 * We have to be very careful here: changing a data block's
5153 * journaling status dynamically is dangerous. If we write a
5154 * data block to the journal, change the status and then delete
5155 * that block, we risk forgetting to revoke the old log record
5156 * from the journal and so a subsequent replay can corrupt data.
5157 * So, first we make sure that the journal is empty and that
5158 * nobody is changing anything.
5159 */
5166 /* We have to allocate physical blocks for delalloc blocks
5167 * before flushing journal. otherwise delalloc blocks can not
5168 * be allocated any more. even more truncate on delalloc blocks
5169 * could trigger BUG by flushing delalloc blocks in journal.
5170 * There is no delalloc block in non-journal data mode.
5171 */
5176 }
5178 /* Wait for all existing dio workers */
5184 /*
5185 * OK, there are no updates running now, and all cached data is
5186 * synced to disk. We are now in a completely consistent state
5187 * which doesn't have anything in the journal, and we know that
5188 * no filesystem updates are running, so it is safe to modify
5189 * the inode's in-core data-journaling state flag now.
5190 */
5200 }
5202 }
5208 /* Finally we can mark the inode as dirty. */
5220 }
5223 {
5225 }
5228 {
5230 loff_t size;
5242 /* Delalloc case is easy... */
5248 ext4_da_get_block_prep);
5252 }
5256 /* Page got truncated from under us? */
5261 }
5265 else
5267 /*
5268 * Return if we have all the buffers mapped. This avoids the need to do
5269 * journal_start/journal_stop which can block and take a long time
5270 */
5274 ext4_bh_unmapped)) {
5275 /* Wait so that we don't change page under IO */
5279 }
5280 }
5282 /* OK, we need to fill the hole... */
5285 else
5287 retry_alloc:
5293 }
5302 }
5304 }
5308 out_ret:
5310 out:
5313 }