2 * linux/fs/ext4/inode.c
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)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
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>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = le16_to_cpu(raw->i_checksum_lo);
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = le16_to_cpu(raw->i_checksum_hi);
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = cpu_to_le16(csum_lo);
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = cpu_to_le16(csum_hi);
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !ext4_has_metadata_csum(inode->i_sb))
87 provided = le16_to_cpu(raw->i_checksum_lo);
88 calculated = ext4_inode_csum(inode, raw, ei);
89 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
90 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
91 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
95 return provided == calculated;
98 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
99 struct ext4_inode_info *ei)
103 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
104 cpu_to_le32(EXT4_OS_LINUX) ||
105 !ext4_has_metadata_csum(inode->i_sb))
108 csum = ext4_inode_csum(inode, raw, ei);
109 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
110 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
111 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
112 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode *inode,
118 trace_ext4_begin_ordered_truncate(inode, new_size);
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.
125 if (!EXT4_I(inode)->jinode)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
128 EXT4_I(inode)->jinode,
132 static void ext4_invalidatepage(struct page *page, unsigned int offset,
133 unsigned int length);
134 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
135 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
136 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode *inode)
144 int ea_blocks = EXT4_I(inode)->i_file_acl ?
145 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
147 if (ext4_has_inline_data(inode))
150 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
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.
169 BUG_ON(EXT4_JOURNAL(inode) == NULL);
170 jbd_debug(2, "restarting handle %p\n", handle);
171 up_write(&EXT4_I(inode)->i_data_sem);
172 ret = ext4_journal_restart(handle, nblocks);
173 down_write(&EXT4_I(inode)->i_data_sem);
174 ext4_discard_preallocations(inode);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode *inode)
187 trace_ext4_evict_inode(inode);
189 if (inode->i_nlink) {
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.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode) &&
209 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
210 inode->i_ino != EXT4_JOURNAL_INO) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_complete_transaction(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages_final(&inode->i_data);
219 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
223 if (is_bad_inode(inode))
225 dquot_initialize(inode);
227 if (ext4_should_order_data(inode))
228 ext4_begin_ordered_truncate(inode, 0);
229 truncate_inode_pages_final(&inode->i_data);
231 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
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
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
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.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
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)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
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
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
326 void ext4_da_update_reserve_space(struct inode *inode,
327 int used, int quota_claim)
329 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
330 struct ext4_inode_info *ei = EXT4_I(inode);
332 spin_lock(&ei->i_block_reservation_lock);
333 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
334 if (unlikely(used > ei->i_reserved_data_blocks)) {
335 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
336 "with only %d reserved data blocks",
337 __func__, inode->i_ino, used,
338 ei->i_reserved_data_blocks);
340 used = ei->i_reserved_data_blocks;
343 /* Update per-inode reservations */
344 ei->i_reserved_data_blocks -= used;
345 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
347 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
349 /* Update quota subsystem for data blocks */
351 dquot_claim_block(inode, EXT4_C2B(sbi, used));
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.
358 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
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.
366 if ((ei->i_reserved_data_blocks == 0) &&
367 (atomic_read(&inode->i_writecount) == 0))
368 ext4_discard_preallocations(inode);
371 static int __check_block_validity(struct inode *inode, const char *func,
373 struct ext4_map_blocks *map)
375 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
377 ext4_error_inode(inode, func, line, map->m_pblk,
378 "lblock %lu mapped to illegal pblock "
379 "(length %d)", (unsigned long) map->m_lblk,
381 return -EFSCORRUPTED;
386 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
391 if (ext4_encrypted_inode(inode))
392 return ext4_encrypted_zeroout(inode, lblk, pblk, len);
394 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
401 #define check_block_validity(inode, map) \
402 __check_block_validity((inode), __func__, __LINE__, (map))
404 #ifdef ES_AGGRESSIVE_TEST
405 static void ext4_map_blocks_es_recheck(handle_t *handle,
407 struct ext4_map_blocks *es_map,
408 struct ext4_map_blocks *map,
415 * There is a race window that the result is not the same.
416 * e.g. xfstests #223 when dioread_nolock enables. The reason
417 * is that we lookup a block mapping in extent status tree with
418 * out taking i_data_sem. So at the time the unwritten extent
419 * could be converted.
421 down_read(&EXT4_I(inode)->i_data_sem);
422 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
423 retval = ext4_ext_map_blocks(handle, inode, map, flags &
424 EXT4_GET_BLOCKS_KEEP_SIZE);
426 retval = ext4_ind_map_blocks(handle, inode, map, flags &
427 EXT4_GET_BLOCKS_KEEP_SIZE);
429 up_read((&EXT4_I(inode)->i_data_sem));
432 * We don't check m_len because extent will be collpased in status
433 * tree. So the m_len might not equal.
435 if (es_map->m_lblk != map->m_lblk ||
436 es_map->m_flags != map->m_flags ||
437 es_map->m_pblk != map->m_pblk) {
438 printk("ES cache assertion failed for inode: %lu "
439 "es_cached ex [%d/%d/%llu/%x] != "
440 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
441 inode->i_ino, es_map->m_lblk, es_map->m_len,
442 es_map->m_pblk, es_map->m_flags, map->m_lblk,
443 map->m_len, map->m_pblk, map->m_flags,
447 #endif /* ES_AGGRESSIVE_TEST */
450 * The ext4_map_blocks() function tries to look up the requested blocks,
451 * and returns if the blocks are already mapped.
453 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
454 * and store the allocated blocks in the result buffer head and mark it
457 * If file type is extents based, it will call ext4_ext_map_blocks(),
458 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
461 * On success, it returns the number of blocks being mapped or allocated.
462 * if create==0 and the blocks are pre-allocated and unwritten block,
463 * the result buffer head is unmapped. If the create ==1, it will make sure
464 * the buffer head is mapped.
466 * It returns 0 if plain look up failed (blocks have not been allocated), in
467 * that case, buffer head is unmapped
469 * It returns the error in case of allocation failure.
471 int ext4_map_blocks(handle_t *handle, struct inode *inode,
472 struct ext4_map_blocks *map, int flags)
474 struct extent_status es;
477 #ifdef ES_AGGRESSIVE_TEST
478 struct ext4_map_blocks orig_map;
480 memcpy(&orig_map, map, sizeof(*map));
484 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
485 "logical block %lu\n", inode->i_ino, flags, map->m_len,
486 (unsigned long) map->m_lblk);
489 * ext4_map_blocks returns an int, and m_len is an unsigned int
491 if (unlikely(map->m_len > INT_MAX))
492 map->m_len = INT_MAX;
494 /* We can handle the block number less than EXT_MAX_BLOCKS */
495 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
496 return -EFSCORRUPTED;
498 /* Lookup extent status tree firstly */
499 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
500 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
501 map->m_pblk = ext4_es_pblock(&es) +
502 map->m_lblk - es.es_lblk;
503 map->m_flags |= ext4_es_is_written(&es) ?
504 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
505 retval = es.es_len - (map->m_lblk - es.es_lblk);
506 if (retval > map->m_len)
509 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
514 #ifdef ES_AGGRESSIVE_TEST
515 ext4_map_blocks_es_recheck(handle, inode, map,
522 * Try to see if we can get the block without requesting a new
525 down_read(&EXT4_I(inode)->i_data_sem);
526 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
527 retval = ext4_ext_map_blocks(handle, inode, map, flags &
528 EXT4_GET_BLOCKS_KEEP_SIZE);
530 retval = ext4_ind_map_blocks(handle, inode, map, flags &
531 EXT4_GET_BLOCKS_KEEP_SIZE);
536 if (unlikely(retval != map->m_len)) {
537 ext4_warning(inode->i_sb,
538 "ES len assertion failed for inode "
539 "%lu: retval %d != map->m_len %d",
540 inode->i_ino, retval, map->m_len);
544 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
545 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
546 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
547 !(status & EXTENT_STATUS_WRITTEN) &&
548 ext4_find_delalloc_range(inode, map->m_lblk,
549 map->m_lblk + map->m_len - 1))
550 status |= EXTENT_STATUS_DELAYED;
551 ret = ext4_es_insert_extent(inode, map->m_lblk,
552 map->m_len, map->m_pblk, status);
556 up_read((&EXT4_I(inode)->i_data_sem));
559 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
560 ret = check_block_validity(inode, map);
565 /* If it is only a block(s) look up */
566 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
570 * Returns if the blocks have already allocated
572 * Note that if blocks have been preallocated
573 * ext4_ext_get_block() returns the create = 0
574 * with buffer head unmapped.
576 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
578 * If we need to convert extent to unwritten
579 * we continue and do the actual work in
580 * ext4_ext_map_blocks()
582 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
586 * Here we clear m_flags because after allocating an new extent,
587 * it will be set again.
589 map->m_flags &= ~EXT4_MAP_FLAGS;
592 * New blocks allocate and/or writing to unwritten extent
593 * will possibly result in updating i_data, so we take
594 * the write lock of i_data_sem, and call get_block()
595 * with create == 1 flag.
597 down_write(&EXT4_I(inode)->i_data_sem);
600 * We need to check for EXT4 here because migrate
601 * could have changed the inode type in between
603 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
604 retval = ext4_ext_map_blocks(handle, inode, map, flags);
606 retval = ext4_ind_map_blocks(handle, inode, map, flags);
608 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
610 * We allocated new blocks which will result in
611 * i_data's format changing. Force the migrate
612 * to fail by clearing migrate flags
614 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
618 * Update reserved blocks/metadata blocks after successful
619 * block allocation which had been deferred till now. We don't
620 * support fallocate for non extent files. So we can update
621 * reserve space here.
624 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
625 ext4_da_update_reserve_space(inode, retval, 1);
631 if (unlikely(retval != map->m_len)) {
632 ext4_warning(inode->i_sb,
633 "ES len assertion failed for inode "
634 "%lu: retval %d != map->m_len %d",
635 inode->i_ino, retval, map->m_len);
640 * We have to zeroout blocks before inserting them into extent
641 * status tree. Otherwise someone could look them up there and
642 * use them before they are really zeroed.
644 if (flags & EXT4_GET_BLOCKS_ZERO &&
645 map->m_flags & EXT4_MAP_MAPPED &&
646 map->m_flags & EXT4_MAP_NEW) {
647 ret = ext4_issue_zeroout(inode, map->m_lblk,
648 map->m_pblk, map->m_len);
656 * If the extent has been zeroed out, we don't need to update
657 * extent status tree.
659 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
660 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
661 if (ext4_es_is_written(&es))
664 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
665 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
666 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
667 !(status & EXTENT_STATUS_WRITTEN) &&
668 ext4_find_delalloc_range(inode, map->m_lblk,
669 map->m_lblk + map->m_len - 1))
670 status |= EXTENT_STATUS_DELAYED;
671 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
672 map->m_pblk, status);
680 up_write((&EXT4_I(inode)->i_data_sem));
681 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
682 ret = check_block_validity(inode, map);
690 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
691 * we have to be careful as someone else may be manipulating b_state as well.
693 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
695 unsigned long old_state;
696 unsigned long new_state;
698 flags &= EXT4_MAP_FLAGS;
700 /* Dummy buffer_head? Set non-atomically. */
702 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
706 * Someone else may be modifying b_state. Be careful! This is ugly but
707 * once we get rid of using bh as a container for mapping information
708 * to pass to / from get_block functions, this can go away.
711 old_state = READ_ONCE(bh->b_state);
712 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
714 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
717 /* Maximum number of blocks we map for direct IO at once. */
718 #define DIO_MAX_BLOCKS 4096
720 static int _ext4_get_block(struct inode *inode, sector_t iblock,
721 struct buffer_head *bh, int flags)
723 handle_t *handle = ext4_journal_current_handle();
724 struct ext4_map_blocks map;
725 int ret = 0, started = 0;
728 if (ext4_has_inline_data(inode))
732 map.m_len = bh->b_size >> inode->i_blkbits;
734 if (flags && !handle) {
735 /* Direct IO write... */
736 if (map.m_len > DIO_MAX_BLOCKS)
737 map.m_len = DIO_MAX_BLOCKS;
738 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
739 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
741 if (IS_ERR(handle)) {
742 ret = PTR_ERR(handle);
748 ret = ext4_map_blocks(handle, inode, &map, flags);
750 ext4_io_end_t *io_end = ext4_inode_aio(inode);
752 map_bh(bh, inode->i_sb, map.m_pblk);
753 ext4_update_bh_state(bh, map.m_flags);
754 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
755 set_buffer_defer_completion(bh);
756 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
760 ext4_journal_stop(handle);
764 int ext4_get_block(struct inode *inode, sector_t iblock,
765 struct buffer_head *bh, int create)
767 return _ext4_get_block(inode, iblock, bh,
768 create ? EXT4_GET_BLOCKS_CREATE : 0);
772 * `handle' can be NULL if create is zero
774 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
775 ext4_lblk_t block, int map_flags)
777 struct ext4_map_blocks map;
778 struct buffer_head *bh;
779 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
782 J_ASSERT(handle != NULL || create == 0);
786 err = ext4_map_blocks(handle, inode, &map, map_flags);
789 return create ? ERR_PTR(-ENOSPC) : NULL;
793 bh = sb_getblk(inode->i_sb, map.m_pblk);
795 return ERR_PTR(-ENOMEM);
796 if (map.m_flags & EXT4_MAP_NEW) {
797 J_ASSERT(create != 0);
798 J_ASSERT(handle != NULL);
801 * Now that we do not always journal data, we should
802 * keep in mind whether this should always journal the
803 * new buffer as metadata. For now, regular file
804 * writes use ext4_get_block instead, so it's not a
808 BUFFER_TRACE(bh, "call get_create_access");
809 err = ext4_journal_get_create_access(handle, bh);
814 if (!buffer_uptodate(bh)) {
815 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
816 set_buffer_uptodate(bh);
819 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
820 err = ext4_handle_dirty_metadata(handle, inode, bh);
824 BUFFER_TRACE(bh, "not a new buffer");
831 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
832 ext4_lblk_t block, int map_flags)
834 struct buffer_head *bh;
836 bh = ext4_getblk(handle, inode, block, map_flags);
839 if (!bh || buffer_uptodate(bh))
841 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
843 if (buffer_uptodate(bh))
846 return ERR_PTR(-EIO);
849 int ext4_walk_page_buffers(handle_t *handle,
850 struct buffer_head *head,
854 int (*fn)(handle_t *handle,
855 struct buffer_head *bh))
857 struct buffer_head *bh;
858 unsigned block_start, block_end;
859 unsigned blocksize = head->b_size;
861 struct buffer_head *next;
863 for (bh = head, block_start = 0;
864 ret == 0 && (bh != head || !block_start);
865 block_start = block_end, bh = next) {
866 next = bh->b_this_page;
867 block_end = block_start + blocksize;
868 if (block_end <= from || block_start >= to) {
869 if (partial && !buffer_uptodate(bh))
873 err = (*fn)(handle, bh);
881 * To preserve ordering, it is essential that the hole instantiation and
882 * the data write be encapsulated in a single transaction. We cannot
883 * close off a transaction and start a new one between the ext4_get_block()
884 * and the commit_write(). So doing the jbd2_journal_start at the start of
885 * prepare_write() is the right place.
887 * Also, this function can nest inside ext4_writepage(). In that case, we
888 * *know* that ext4_writepage() has generated enough buffer credits to do the
889 * whole page. So we won't block on the journal in that case, which is good,
890 * because the caller may be PF_MEMALLOC.
892 * By accident, ext4 can be reentered when a transaction is open via
893 * quota file writes. If we were to commit the transaction while thus
894 * reentered, there can be a deadlock - we would be holding a quota
895 * lock, and the commit would never complete if another thread had a
896 * transaction open and was blocking on the quota lock - a ranking
899 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
900 * will _not_ run commit under these circumstances because handle->h_ref
901 * is elevated. We'll still have enough credits for the tiny quotafile
904 int do_journal_get_write_access(handle_t *handle,
905 struct buffer_head *bh)
907 int dirty = buffer_dirty(bh);
910 if (!buffer_mapped(bh) || buffer_freed(bh))
913 * __block_write_begin() could have dirtied some buffers. Clean
914 * the dirty bit as jbd2_journal_get_write_access() could complain
915 * otherwise about fs integrity issues. Setting of the dirty bit
916 * by __block_write_begin() isn't a real problem here as we clear
917 * the bit before releasing a page lock and thus writeback cannot
918 * ever write the buffer.
921 clear_buffer_dirty(bh);
922 BUFFER_TRACE(bh, "get write access");
923 ret = ext4_journal_get_write_access(handle, bh);
925 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
929 #ifdef CONFIG_EXT4_FS_ENCRYPTION
930 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
931 get_block_t *get_block)
933 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
934 unsigned to = from + len;
935 struct inode *inode = page->mapping->host;
936 unsigned block_start, block_end;
939 unsigned blocksize = inode->i_sb->s_blocksize;
941 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
942 bool decrypt = false;
944 BUG_ON(!PageLocked(page));
945 BUG_ON(from > PAGE_CACHE_SIZE);
946 BUG_ON(to > PAGE_CACHE_SIZE);
949 if (!page_has_buffers(page))
950 create_empty_buffers(page, blocksize, 0);
951 head = page_buffers(page);
952 bbits = ilog2(blocksize);
953 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
955 for (bh = head, block_start = 0; bh != head || !block_start;
956 block++, block_start = block_end, bh = bh->b_this_page) {
957 block_end = block_start + blocksize;
958 if (block_end <= from || block_start >= to) {
959 if (PageUptodate(page)) {
960 if (!buffer_uptodate(bh))
961 set_buffer_uptodate(bh);
966 clear_buffer_new(bh);
967 if (!buffer_mapped(bh)) {
968 WARN_ON(bh->b_size != blocksize);
969 err = get_block(inode, block, bh, 1);
972 if (buffer_new(bh)) {
973 unmap_underlying_metadata(bh->b_bdev,
975 if (PageUptodate(page)) {
976 clear_buffer_new(bh);
977 set_buffer_uptodate(bh);
978 mark_buffer_dirty(bh);
981 if (block_end > to || block_start < from)
982 zero_user_segments(page, to, block_end,
987 if (PageUptodate(page)) {
988 if (!buffer_uptodate(bh))
989 set_buffer_uptodate(bh);
992 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
993 !buffer_unwritten(bh) &&
994 (block_start < from || block_end > to)) {
995 ll_rw_block(READ, 1, &bh);
997 decrypt = ext4_encrypted_inode(inode) &&
998 S_ISREG(inode->i_mode);
1002 * If we issued read requests, let them complete.
1004 while (wait_bh > wait) {
1005 wait_on_buffer(*--wait_bh);
1006 if (!buffer_uptodate(*wait_bh))
1010 page_zero_new_buffers(page, from, to);
1012 err = ext4_decrypt(page);
1017 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1018 loff_t pos, unsigned len, unsigned flags,
1019 struct page **pagep, void **fsdata)
1021 struct inode *inode = mapping->host;
1022 int ret, needed_blocks;
1029 trace_ext4_write_begin(inode, pos, len, flags);
1031 * Reserve one block more for addition to orphan list in case
1032 * we allocate blocks but write fails for some reason
1034 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1035 index = pos >> PAGE_CACHE_SHIFT;
1036 from = pos & (PAGE_CACHE_SIZE - 1);
1039 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1040 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1049 * grab_cache_page_write_begin() can take a long time if the
1050 * system is thrashing due to memory pressure, or if the page
1051 * is being written back. So grab it first before we start
1052 * the transaction handle. This also allows us to allocate
1053 * the page (if needed) without using GFP_NOFS.
1056 page = grab_cache_page_write_begin(mapping, index, flags);
1062 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1063 if (IS_ERR(handle)) {
1064 page_cache_release(page);
1065 return PTR_ERR(handle);
1069 if (page->mapping != mapping) {
1070 /* The page got truncated from under us */
1072 page_cache_release(page);
1073 ext4_journal_stop(handle);
1076 /* In case writeback began while the page was unlocked */
1077 wait_for_stable_page(page);
1079 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1080 if (ext4_should_dioread_nolock(inode))
1081 ret = ext4_block_write_begin(page, pos, len,
1082 ext4_get_block_write);
1084 ret = ext4_block_write_begin(page, pos, len,
1087 if (ext4_should_dioread_nolock(inode))
1088 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1090 ret = __block_write_begin(page, pos, len, ext4_get_block);
1092 if (!ret && ext4_should_journal_data(inode)) {
1093 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1095 do_journal_get_write_access);
1101 * __block_write_begin may have instantiated a few blocks
1102 * outside i_size. Trim these off again. Don't need
1103 * i_size_read because we hold i_mutex.
1105 * Add inode to orphan list in case we crash before
1108 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1109 ext4_orphan_add(handle, inode);
1111 ext4_journal_stop(handle);
1112 if (pos + len > inode->i_size) {
1113 ext4_truncate_failed_write(inode);
1115 * If truncate failed early the inode might
1116 * still be on the orphan list; we need to
1117 * make sure the inode is removed from the
1118 * orphan list in that case.
1121 ext4_orphan_del(NULL, inode);
1124 if (ret == -ENOSPC &&
1125 ext4_should_retry_alloc(inode->i_sb, &retries))
1127 page_cache_release(page);
1134 /* For write_end() in data=journal mode */
1135 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1138 if (!buffer_mapped(bh) || buffer_freed(bh))
1140 set_buffer_uptodate(bh);
1141 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1142 clear_buffer_meta(bh);
1143 clear_buffer_prio(bh);
1148 * We need to pick up the new inode size which generic_commit_write gave us
1149 * `file' can be NULL - eg, when called from page_symlink().
1151 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1152 * buffers are managed internally.
1154 static int ext4_write_end(struct file *file,
1155 struct address_space *mapping,
1156 loff_t pos, unsigned len, unsigned copied,
1157 struct page *page, void *fsdata)
1159 handle_t *handle = ext4_journal_current_handle();
1160 struct inode *inode = mapping->host;
1161 loff_t old_size = inode->i_size;
1163 int i_size_changed = 0;
1165 trace_ext4_write_end(inode, pos, len, copied);
1166 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1167 ret = ext4_jbd2_file_inode(handle, inode);
1170 page_cache_release(page);
1175 if (ext4_has_inline_data(inode)) {
1176 ret = ext4_write_inline_data_end(inode, pos, len,
1182 copied = block_write_end(file, mapping, pos,
1183 len, copied, page, fsdata);
1185 * it's important to update i_size while still holding page lock:
1186 * page writeout could otherwise come in and zero beyond i_size.
1188 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1190 page_cache_release(page);
1193 pagecache_isize_extended(inode, old_size, pos);
1195 * Don't mark the inode dirty under page lock. First, it unnecessarily
1196 * makes the holding time of page lock longer. Second, it forces lock
1197 * ordering of page lock and transaction start for journaling
1201 ext4_mark_inode_dirty(handle, inode);
1203 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1204 /* if we have allocated more blocks and copied
1205 * less. We will have blocks allocated outside
1206 * inode->i_size. So truncate them
1208 ext4_orphan_add(handle, inode);
1210 ret2 = ext4_journal_stop(handle);
1214 if (pos + len > inode->i_size) {
1215 ext4_truncate_failed_write(inode);
1217 * If truncate failed early the inode might still be
1218 * on the orphan list; we need to make sure the inode
1219 * is removed from the orphan list in that case.
1222 ext4_orphan_del(NULL, inode);
1225 return ret ? ret : copied;
1229 * This is a private version of page_zero_new_buffers() which doesn't
1230 * set the buffer to be dirty, since in data=journalled mode we need
1231 * to call ext4_handle_dirty_metadata() instead.
1233 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1235 unsigned int block_start = 0, block_end;
1236 struct buffer_head *head, *bh;
1238 bh = head = page_buffers(page);
1240 block_end = block_start + bh->b_size;
1241 if (buffer_new(bh)) {
1242 if (block_end > from && block_start < to) {
1243 if (!PageUptodate(page)) {
1244 unsigned start, size;
1246 start = max(from, block_start);
1247 size = min(to, block_end) - start;
1249 zero_user(page, start, size);
1250 set_buffer_uptodate(bh);
1252 clear_buffer_new(bh);
1255 block_start = block_end;
1256 bh = bh->b_this_page;
1257 } while (bh != head);
1260 static int ext4_journalled_write_end(struct file *file,
1261 struct address_space *mapping,
1262 loff_t pos, unsigned len, unsigned copied,
1263 struct page *page, void *fsdata)
1265 handle_t *handle = ext4_journal_current_handle();
1266 struct inode *inode = mapping->host;
1267 loff_t old_size = inode->i_size;
1271 int size_changed = 0;
1273 trace_ext4_journalled_write_end(inode, pos, len, copied);
1274 from = pos & (PAGE_CACHE_SIZE - 1);
1277 BUG_ON(!ext4_handle_valid(handle));
1279 if (ext4_has_inline_data(inode))
1280 copied = ext4_write_inline_data_end(inode, pos, len,
1284 if (!PageUptodate(page))
1286 zero_new_buffers(page, from+copied, to);
1289 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1290 to, &partial, write_end_fn);
1292 SetPageUptodate(page);
1294 size_changed = ext4_update_inode_size(inode, pos + copied);
1295 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1296 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1298 page_cache_release(page);
1301 pagecache_isize_extended(inode, old_size, pos);
1304 ret2 = ext4_mark_inode_dirty(handle, inode);
1309 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1310 /* if we have allocated more blocks and copied
1311 * less. We will have blocks allocated outside
1312 * inode->i_size. So truncate them
1314 ext4_orphan_add(handle, inode);
1316 ret2 = ext4_journal_stop(handle);
1319 if (pos + len > inode->i_size) {
1320 ext4_truncate_failed_write(inode);
1322 * If truncate failed early the inode might still be
1323 * on the orphan list; we need to make sure the inode
1324 * is removed from the orphan list in that case.
1327 ext4_orphan_del(NULL, inode);
1330 return ret ? ret : copied;
1334 * Reserve space for a single cluster
1336 static int ext4_da_reserve_space(struct inode *inode)
1338 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1339 struct ext4_inode_info *ei = EXT4_I(inode);
1343 * We will charge metadata quota at writeout time; this saves
1344 * us from metadata over-estimation, though we may go over by
1345 * a small amount in the end. Here we just reserve for data.
1347 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1351 spin_lock(&ei->i_block_reservation_lock);
1352 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1353 spin_unlock(&ei->i_block_reservation_lock);
1354 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1357 ei->i_reserved_data_blocks++;
1358 trace_ext4_da_reserve_space(inode);
1359 spin_unlock(&ei->i_block_reservation_lock);
1361 return 0; /* success */
1364 static void ext4_da_release_space(struct inode *inode, int to_free)
1366 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1367 struct ext4_inode_info *ei = EXT4_I(inode);
1370 return; /* Nothing to release, exit */
1372 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1374 trace_ext4_da_release_space(inode, to_free);
1375 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1377 * if there aren't enough reserved blocks, then the
1378 * counter is messed up somewhere. Since this
1379 * function is called from invalidate page, it's
1380 * harmless to return without any action.
1382 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1383 "ino %lu, to_free %d with only %d reserved "
1384 "data blocks", inode->i_ino, to_free,
1385 ei->i_reserved_data_blocks);
1387 to_free = ei->i_reserved_data_blocks;
1389 ei->i_reserved_data_blocks -= to_free;
1391 /* update fs dirty data blocks counter */
1392 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1394 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1396 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1399 static void ext4_da_page_release_reservation(struct page *page,
1400 unsigned int offset,
1401 unsigned int length)
1403 int to_release = 0, contiguous_blks = 0;
1404 struct buffer_head *head, *bh;
1405 unsigned int curr_off = 0;
1406 struct inode *inode = page->mapping->host;
1407 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1408 unsigned int stop = offset + length;
1412 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1414 head = page_buffers(page);
1417 unsigned int next_off = curr_off + bh->b_size;
1419 if (next_off > stop)
1422 if ((offset <= curr_off) && (buffer_delay(bh))) {
1425 clear_buffer_delay(bh);
1426 } else if (contiguous_blks) {
1427 lblk = page->index <<
1428 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1429 lblk += (curr_off >> inode->i_blkbits) -
1431 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1432 contiguous_blks = 0;
1434 curr_off = next_off;
1435 } while ((bh = bh->b_this_page) != head);
1437 if (contiguous_blks) {
1438 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1439 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1440 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1443 /* If we have released all the blocks belonging to a cluster, then we
1444 * need to release the reserved space for that cluster. */
1445 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1446 while (num_clusters > 0) {
1447 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1448 ((num_clusters - 1) << sbi->s_cluster_bits);
1449 if (sbi->s_cluster_ratio == 1 ||
1450 !ext4_find_delalloc_cluster(inode, lblk))
1451 ext4_da_release_space(inode, 1);
1458 * Delayed allocation stuff
1461 struct mpage_da_data {
1462 struct inode *inode;
1463 struct writeback_control *wbc;
1465 pgoff_t first_page; /* The first page to write */
1466 pgoff_t next_page; /* Current page to examine */
1467 pgoff_t last_page; /* Last page to examine */
1469 * Extent to map - this can be after first_page because that can be
1470 * fully mapped. We somewhat abuse m_flags to store whether the extent
1471 * is delalloc or unwritten.
1473 struct ext4_map_blocks map;
1474 struct ext4_io_submit io_submit; /* IO submission data */
1477 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1482 struct pagevec pvec;
1483 struct inode *inode = mpd->inode;
1484 struct address_space *mapping = inode->i_mapping;
1486 /* This is necessary when next_page == 0. */
1487 if (mpd->first_page >= mpd->next_page)
1490 index = mpd->first_page;
1491 end = mpd->next_page - 1;
1493 ext4_lblk_t start, last;
1494 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1495 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1496 ext4_es_remove_extent(inode, start, last - start + 1);
1499 pagevec_init(&pvec, 0);
1500 while (index <= end) {
1501 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1504 for (i = 0; i < nr_pages; i++) {
1505 struct page *page = pvec.pages[i];
1506 if (page->index > end)
1508 BUG_ON(!PageLocked(page));
1509 BUG_ON(PageWriteback(page));
1511 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1512 ClearPageUptodate(page);
1516 index = pvec.pages[nr_pages - 1]->index + 1;
1517 pagevec_release(&pvec);
1521 static void ext4_print_free_blocks(struct inode *inode)
1523 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1524 struct super_block *sb = inode->i_sb;
1525 struct ext4_inode_info *ei = EXT4_I(inode);
1527 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1528 EXT4_C2B(EXT4_SB(inode->i_sb),
1529 ext4_count_free_clusters(sb)));
1530 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1531 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1532 (long long) EXT4_C2B(EXT4_SB(sb),
1533 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1534 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1535 (long long) EXT4_C2B(EXT4_SB(sb),
1536 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1537 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1538 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1539 ei->i_reserved_data_blocks);
1543 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1545 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1549 * This function is grabs code from the very beginning of
1550 * ext4_map_blocks, but assumes that the caller is from delayed write
1551 * time. This function looks up the requested blocks and sets the
1552 * buffer delay bit under the protection of i_data_sem.
1554 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1555 struct ext4_map_blocks *map,
1556 struct buffer_head *bh)
1558 struct extent_status es;
1560 sector_t invalid_block = ~((sector_t) 0xffff);
1561 #ifdef ES_AGGRESSIVE_TEST
1562 struct ext4_map_blocks orig_map;
1564 memcpy(&orig_map, map, sizeof(*map));
1567 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1571 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1572 "logical block %lu\n", inode->i_ino, map->m_len,
1573 (unsigned long) map->m_lblk);
1575 /* Lookup extent status tree firstly */
1576 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1577 if (ext4_es_is_hole(&es)) {
1579 down_read(&EXT4_I(inode)->i_data_sem);
1584 * Delayed extent could be allocated by fallocate.
1585 * So we need to check it.
1587 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1588 map_bh(bh, inode->i_sb, invalid_block);
1590 set_buffer_delay(bh);
1594 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1595 retval = es.es_len - (iblock - es.es_lblk);
1596 if (retval > map->m_len)
1597 retval = map->m_len;
1598 map->m_len = retval;
1599 if (ext4_es_is_written(&es))
1600 map->m_flags |= EXT4_MAP_MAPPED;
1601 else if (ext4_es_is_unwritten(&es))
1602 map->m_flags |= EXT4_MAP_UNWRITTEN;
1606 #ifdef ES_AGGRESSIVE_TEST
1607 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1613 * Try to see if we can get the block without requesting a new
1614 * file system block.
1616 down_read(&EXT4_I(inode)->i_data_sem);
1617 if (ext4_has_inline_data(inode))
1619 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1620 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1622 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1628 * XXX: __block_prepare_write() unmaps passed block,
1632 * If the block was allocated from previously allocated cluster,
1633 * then we don't need to reserve it again. However we still need
1634 * to reserve metadata for every block we're going to write.
1636 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1637 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1638 ret = ext4_da_reserve_space(inode);
1640 /* not enough space to reserve */
1646 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1647 ~0, EXTENT_STATUS_DELAYED);
1653 map_bh(bh, inode->i_sb, invalid_block);
1655 set_buffer_delay(bh);
1656 } else if (retval > 0) {
1658 unsigned int status;
1660 if (unlikely(retval != map->m_len)) {
1661 ext4_warning(inode->i_sb,
1662 "ES len assertion failed for inode "
1663 "%lu: retval %d != map->m_len %d",
1664 inode->i_ino, retval, map->m_len);
1668 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1669 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1670 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1671 map->m_pblk, status);
1677 up_read((&EXT4_I(inode)->i_data_sem));
1683 * This is a special get_block_t callback which is used by
1684 * ext4_da_write_begin(). It will either return mapped block or
1685 * reserve space for a single block.
1687 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1688 * We also have b_blocknr = -1 and b_bdev initialized properly
1690 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1691 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1692 * initialized properly.
1694 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1695 struct buffer_head *bh, int create)
1697 struct ext4_map_blocks map;
1700 BUG_ON(create == 0);
1701 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1703 map.m_lblk = iblock;
1707 * first, we need to know whether the block is allocated already
1708 * preallocated blocks are unmapped but should treated
1709 * the same as allocated blocks.
1711 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1715 map_bh(bh, inode->i_sb, map.m_pblk);
1716 ext4_update_bh_state(bh, map.m_flags);
1718 if (buffer_unwritten(bh)) {
1719 /* A delayed write to unwritten bh should be marked
1720 * new and mapped. Mapped ensures that we don't do
1721 * get_block multiple times when we write to the same
1722 * offset and new ensures that we do proper zero out
1723 * for partial write.
1726 set_buffer_mapped(bh);
1731 static int bget_one(handle_t *handle, struct buffer_head *bh)
1737 static int bput_one(handle_t *handle, struct buffer_head *bh)
1743 static int __ext4_journalled_writepage(struct page *page,
1746 struct address_space *mapping = page->mapping;
1747 struct inode *inode = mapping->host;
1748 struct buffer_head *page_bufs = NULL;
1749 handle_t *handle = NULL;
1750 int ret = 0, err = 0;
1751 int inline_data = ext4_has_inline_data(inode);
1752 struct buffer_head *inode_bh = NULL;
1754 ClearPageChecked(page);
1757 BUG_ON(page->index != 0);
1758 BUG_ON(len > ext4_get_max_inline_size(inode));
1759 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1760 if (inode_bh == NULL)
1763 page_bufs = page_buffers(page);
1768 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1772 * We need to release the page lock before we start the
1773 * journal, so grab a reference so the page won't disappear
1774 * out from under us.
1779 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1780 ext4_writepage_trans_blocks(inode));
1781 if (IS_ERR(handle)) {
1782 ret = PTR_ERR(handle);
1784 goto out_no_pagelock;
1786 BUG_ON(!ext4_handle_valid(handle));
1790 if (page->mapping != mapping) {
1791 /* The page got truncated from under us */
1792 ext4_journal_stop(handle);
1798 BUFFER_TRACE(inode_bh, "get write access");
1799 ret = ext4_journal_get_write_access(handle, inode_bh);
1801 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1804 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1805 do_journal_get_write_access);
1807 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1812 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1813 err = ext4_journal_stop(handle);
1817 if (!ext4_has_inline_data(inode))
1818 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1820 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1829 * Note that we don't need to start a transaction unless we're journaling data
1830 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1831 * need to file the inode to the transaction's list in ordered mode because if
1832 * we are writing back data added by write(), the inode is already there and if
1833 * we are writing back data modified via mmap(), no one guarantees in which
1834 * transaction the data will hit the disk. In case we are journaling data, we
1835 * cannot start transaction directly because transaction start ranks above page
1836 * lock so we have to do some magic.
1838 * This function can get called via...
1839 * - ext4_writepages after taking page lock (have journal handle)
1840 * - journal_submit_inode_data_buffers (no journal handle)
1841 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1842 * - grab_page_cache when doing write_begin (have journal handle)
1844 * We don't do any block allocation in this function. If we have page with
1845 * multiple blocks we need to write those buffer_heads that are mapped. This
1846 * is important for mmaped based write. So if we do with blocksize 1K
1847 * truncate(f, 1024);
1848 * a = mmap(f, 0, 4096);
1850 * truncate(f, 4096);
1851 * we have in the page first buffer_head mapped via page_mkwrite call back
1852 * but other buffer_heads would be unmapped but dirty (dirty done via the
1853 * do_wp_page). So writepage should write the first block. If we modify
1854 * the mmap area beyond 1024 we will again get a page_fault and the
1855 * page_mkwrite callback will do the block allocation and mark the
1856 * buffer_heads mapped.
1858 * We redirty the page if we have any buffer_heads that is either delay or
1859 * unwritten in the page.
1861 * We can get recursively called as show below.
1863 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1866 * But since we don't do any block allocation we should not deadlock.
1867 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1869 static int ext4_writepage(struct page *page,
1870 struct writeback_control *wbc)
1875 struct buffer_head *page_bufs = NULL;
1876 struct inode *inode = page->mapping->host;
1877 struct ext4_io_submit io_submit;
1878 bool keep_towrite = false;
1880 trace_ext4_writepage(page);
1881 size = i_size_read(inode);
1882 if (page->index == size >> PAGE_CACHE_SHIFT)
1883 len = size & ~PAGE_CACHE_MASK;
1885 len = PAGE_CACHE_SIZE;
1887 page_bufs = page_buffers(page);
1889 * We cannot do block allocation or other extent handling in this
1890 * function. If there are buffers needing that, we have to redirty
1891 * the page. But we may reach here when we do a journal commit via
1892 * journal_submit_inode_data_buffers() and in that case we must write
1893 * allocated buffers to achieve data=ordered mode guarantees.
1895 * Also, if there is only one buffer per page (the fs block
1896 * size == the page size), if one buffer needs block
1897 * allocation or needs to modify the extent tree to clear the
1898 * unwritten flag, we know that the page can't be written at
1899 * all, so we might as well refuse the write immediately.
1900 * Unfortunately if the block size != page size, we can't as
1901 * easily detect this case using ext4_walk_page_buffers(), but
1902 * for the extremely common case, this is an optimization that
1903 * skips a useless round trip through ext4_bio_write_page().
1905 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1906 ext4_bh_delay_or_unwritten)) {
1907 redirty_page_for_writepage(wbc, page);
1908 if ((current->flags & PF_MEMALLOC) ||
1909 (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) {
1911 * For memory cleaning there's no point in writing only
1912 * some buffers. So just bail out. Warn if we came here
1913 * from direct reclaim.
1915 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1920 keep_towrite = true;
1923 if (PageChecked(page) && ext4_should_journal_data(inode))
1925 * It's mmapped pagecache. Add buffers and journal it. There
1926 * doesn't seem much point in redirtying the page here.
1928 return __ext4_journalled_writepage(page, len);
1930 ext4_io_submit_init(&io_submit, wbc);
1931 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1932 if (!io_submit.io_end) {
1933 redirty_page_for_writepage(wbc, page);
1937 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1938 ext4_io_submit(&io_submit);
1939 /* Drop io_end reference we got from init */
1940 ext4_put_io_end_defer(io_submit.io_end);
1944 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1947 loff_t size = i_size_read(mpd->inode);
1950 BUG_ON(page->index != mpd->first_page);
1951 if (page->index == size >> PAGE_CACHE_SHIFT)
1952 len = size & ~PAGE_CACHE_MASK;
1954 len = PAGE_CACHE_SIZE;
1955 clear_page_dirty_for_io(page);
1956 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1958 mpd->wbc->nr_to_write--;
1964 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1967 * mballoc gives us at most this number of blocks...
1968 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1969 * The rest of mballoc seems to handle chunks up to full group size.
1971 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1974 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1976 * @mpd - extent of blocks
1977 * @lblk - logical number of the block in the file
1978 * @bh - buffer head we want to add to the extent
1980 * The function is used to collect contig. blocks in the same state. If the
1981 * buffer doesn't require mapping for writeback and we haven't started the
1982 * extent of buffers to map yet, the function returns 'true' immediately - the
1983 * caller can write the buffer right away. Otherwise the function returns true
1984 * if the block has been added to the extent, false if the block couldn't be
1987 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1988 struct buffer_head *bh)
1990 struct ext4_map_blocks *map = &mpd->map;
1992 /* Buffer that doesn't need mapping for writeback? */
1993 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1994 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1995 /* So far no extent to map => we write the buffer right away */
1996 if (map->m_len == 0)
2001 /* First block in the extent? */
2002 if (map->m_len == 0) {
2005 map->m_flags = bh->b_state & BH_FLAGS;
2009 /* Don't go larger than mballoc is willing to allocate */
2010 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2013 /* Can we merge the block to our big extent? */
2014 if (lblk == map->m_lblk + map->m_len &&
2015 (bh->b_state & BH_FLAGS) == map->m_flags) {
2023 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2025 * @mpd - extent of blocks for mapping
2026 * @head - the first buffer in the page
2027 * @bh - buffer we should start processing from
2028 * @lblk - logical number of the block in the file corresponding to @bh
2030 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2031 * the page for IO if all buffers in this page were mapped and there's no
2032 * accumulated extent of buffers to map or add buffers in the page to the
2033 * extent of buffers to map. The function returns 1 if the caller can continue
2034 * by processing the next page, 0 if it should stop adding buffers to the
2035 * extent to map because we cannot extend it anymore. It can also return value
2036 * < 0 in case of error during IO submission.
2038 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2039 struct buffer_head *head,
2040 struct buffer_head *bh,
2043 struct inode *inode = mpd->inode;
2045 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2046 >> inode->i_blkbits;
2049 BUG_ON(buffer_locked(bh));
2051 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2052 /* Found extent to map? */
2055 /* Everything mapped so far and we hit EOF */
2058 } while (lblk++, (bh = bh->b_this_page) != head);
2059 /* So far everything mapped? Submit the page for IO. */
2060 if (mpd->map.m_len == 0) {
2061 err = mpage_submit_page(mpd, head->b_page);
2065 return lblk < blocks;
2069 * mpage_map_buffers - update buffers corresponding to changed extent and
2070 * submit fully mapped pages for IO
2072 * @mpd - description of extent to map, on return next extent to map
2074 * Scan buffers corresponding to changed extent (we expect corresponding pages
2075 * to be already locked) and update buffer state according to new extent state.
2076 * We map delalloc buffers to their physical location, clear unwritten bits,
2077 * and mark buffers as uninit when we perform writes to unwritten extents
2078 * and do extent conversion after IO is finished. If the last page is not fully
2079 * mapped, we update @map to the next extent in the last page that needs
2080 * mapping. Otherwise we submit the page for IO.
2082 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2084 struct pagevec pvec;
2086 struct inode *inode = mpd->inode;
2087 struct buffer_head *head, *bh;
2088 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2094 start = mpd->map.m_lblk >> bpp_bits;
2095 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2096 lblk = start << bpp_bits;
2097 pblock = mpd->map.m_pblk;
2099 pagevec_init(&pvec, 0);
2100 while (start <= end) {
2101 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2105 for (i = 0; i < nr_pages; i++) {
2106 struct page *page = pvec.pages[i];
2108 if (page->index > end)
2110 /* Up to 'end' pages must be contiguous */
2111 BUG_ON(page->index != start);
2112 bh = head = page_buffers(page);
2114 if (lblk < mpd->map.m_lblk)
2116 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2118 * Buffer after end of mapped extent.
2119 * Find next buffer in the page to map.
2122 mpd->map.m_flags = 0;
2124 * FIXME: If dioread_nolock supports
2125 * blocksize < pagesize, we need to make
2126 * sure we add size mapped so far to
2127 * io_end->size as the following call
2128 * can submit the page for IO.
2130 err = mpage_process_page_bufs(mpd, head,
2132 pagevec_release(&pvec);
2137 if (buffer_delay(bh)) {
2138 clear_buffer_delay(bh);
2139 bh->b_blocknr = pblock++;
2141 clear_buffer_unwritten(bh);
2142 } while (lblk++, (bh = bh->b_this_page) != head);
2145 * FIXME: This is going to break if dioread_nolock
2146 * supports blocksize < pagesize as we will try to
2147 * convert potentially unmapped parts of inode.
2149 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2150 /* Page fully mapped - let IO run! */
2151 err = mpage_submit_page(mpd, page);
2153 pagevec_release(&pvec);
2158 pagevec_release(&pvec);
2160 /* Extent fully mapped and matches with page boundary. We are done. */
2162 mpd->map.m_flags = 0;
2166 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2168 struct inode *inode = mpd->inode;
2169 struct ext4_map_blocks *map = &mpd->map;
2170 int get_blocks_flags;
2171 int err, dioread_nolock;
2173 trace_ext4_da_write_pages_extent(inode, map);
2175 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2176 * to convert an unwritten extent to be initialized (in the case
2177 * where we have written into one or more preallocated blocks). It is
2178 * possible that we're going to need more metadata blocks than
2179 * previously reserved. However we must not fail because we're in
2180 * writeback and there is nothing we can do about it so it might result
2181 * in data loss. So use reserved blocks to allocate metadata if
2184 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2185 * the blocks in question are delalloc blocks. This indicates
2186 * that the blocks and quotas has already been checked when
2187 * the data was copied into the page cache.
2189 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2190 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2191 dioread_nolock = ext4_should_dioread_nolock(inode);
2193 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2194 if (map->m_flags & (1 << BH_Delay))
2195 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2197 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2200 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2201 if (!mpd->io_submit.io_end->handle &&
2202 ext4_handle_valid(handle)) {
2203 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2204 handle->h_rsv_handle = NULL;
2206 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2209 BUG_ON(map->m_len == 0);
2210 if (map->m_flags & EXT4_MAP_NEW) {
2211 struct block_device *bdev = inode->i_sb->s_bdev;
2214 for (i = 0; i < map->m_len; i++)
2215 unmap_underlying_metadata(bdev, map->m_pblk + i);
2221 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2222 * mpd->len and submit pages underlying it for IO
2224 * @handle - handle for journal operations
2225 * @mpd - extent to map
2226 * @give_up_on_write - we set this to true iff there is a fatal error and there
2227 * is no hope of writing the data. The caller should discard
2228 * dirty pages to avoid infinite loops.
2230 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2231 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2232 * them to initialized or split the described range from larger unwritten
2233 * extent. Note that we need not map all the described range since allocation
2234 * can return less blocks or the range is covered by more unwritten extents. We
2235 * cannot map more because we are limited by reserved transaction credits. On
2236 * the other hand we always make sure that the last touched page is fully
2237 * mapped so that it can be written out (and thus forward progress is
2238 * guaranteed). After mapping we submit all mapped pages for IO.
2240 static int mpage_map_and_submit_extent(handle_t *handle,
2241 struct mpage_da_data *mpd,
2242 bool *give_up_on_write)
2244 struct inode *inode = mpd->inode;
2245 struct ext4_map_blocks *map = &mpd->map;
2250 mpd->io_submit.io_end->offset =
2251 ((loff_t)map->m_lblk) << inode->i_blkbits;
2253 err = mpage_map_one_extent(handle, mpd);
2255 struct super_block *sb = inode->i_sb;
2257 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2258 goto invalidate_dirty_pages;
2260 * Let the uper layers retry transient errors.
2261 * In the case of ENOSPC, if ext4_count_free_blocks()
2262 * is non-zero, a commit should free up blocks.
2264 if ((err == -ENOMEM) ||
2265 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2267 goto update_disksize;
2270 ext4_msg(sb, KERN_CRIT,
2271 "Delayed block allocation failed for "
2272 "inode %lu at logical offset %llu with"
2273 " max blocks %u with error %d",
2275 (unsigned long long)map->m_lblk,
2276 (unsigned)map->m_len, -err);
2277 ext4_msg(sb, KERN_CRIT,
2278 "This should not happen!! Data will "
2281 ext4_print_free_blocks(inode);
2282 invalidate_dirty_pages:
2283 *give_up_on_write = true;
2288 * Update buffer state, submit mapped pages, and get us new
2291 err = mpage_map_and_submit_buffers(mpd);
2293 goto update_disksize;
2294 } while (map->m_len);
2298 * Update on-disk size after IO is submitted. Races with
2299 * truncate are avoided by checking i_size under i_data_sem.
2301 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2302 if (disksize > EXT4_I(inode)->i_disksize) {
2306 down_write(&EXT4_I(inode)->i_data_sem);
2307 i_size = i_size_read(inode);
2308 if (disksize > i_size)
2310 if (disksize > EXT4_I(inode)->i_disksize)
2311 EXT4_I(inode)->i_disksize = disksize;
2312 err2 = ext4_mark_inode_dirty(handle, inode);
2313 up_write(&EXT4_I(inode)->i_data_sem);
2315 ext4_error(inode->i_sb,
2316 "Failed to mark inode %lu dirty",
2325 * Calculate the total number of credits to reserve for one writepages
2326 * iteration. This is called from ext4_writepages(). We map an extent of
2327 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2328 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2329 * bpp - 1 blocks in bpp different extents.
2331 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2333 int bpp = ext4_journal_blocks_per_page(inode);
2335 return ext4_meta_trans_blocks(inode,
2336 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2340 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2341 * and underlying extent to map
2343 * @mpd - where to look for pages
2345 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2346 * IO immediately. When we find a page which isn't mapped we start accumulating
2347 * extent of buffers underlying these pages that needs mapping (formed by
2348 * either delayed or unwritten buffers). We also lock the pages containing
2349 * these buffers. The extent found is returned in @mpd structure (starting at
2350 * mpd->lblk with length mpd->len blocks).
2352 * Note that this function can attach bios to one io_end structure which are
2353 * neither logically nor physically contiguous. Although it may seem as an
2354 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2355 * case as we need to track IO to all buffers underlying a page in one io_end.
2357 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2359 struct address_space *mapping = mpd->inode->i_mapping;
2360 struct pagevec pvec;
2361 unsigned int nr_pages;
2362 long left = mpd->wbc->nr_to_write;
2363 pgoff_t index = mpd->first_page;
2364 pgoff_t end = mpd->last_page;
2367 int blkbits = mpd->inode->i_blkbits;
2369 struct buffer_head *head;
2371 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2372 tag = PAGECACHE_TAG_TOWRITE;
2374 tag = PAGECACHE_TAG_DIRTY;
2376 pagevec_init(&pvec, 0);
2378 mpd->next_page = index;
2379 while (index <= end) {
2380 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2381 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2385 for (i = 0; i < nr_pages; i++) {
2386 struct page *page = pvec.pages[i];
2389 * At this point, the page may be truncated or
2390 * invalidated (changing page->mapping to NULL), or
2391 * even swizzled back from swapper_space to tmpfs file
2392 * mapping. However, page->index will not change
2393 * because we have a reference on the page.
2395 if (page->index > end)
2399 * Accumulated enough dirty pages? This doesn't apply
2400 * to WB_SYNC_ALL mode. For integrity sync we have to
2401 * keep going because someone may be concurrently
2402 * dirtying pages, and we might have synced a lot of
2403 * newly appeared dirty pages, but have not synced all
2404 * of the old dirty pages.
2406 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2409 /* If we can't merge this page, we are done. */
2410 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2415 * If the page is no longer dirty, or its mapping no
2416 * longer corresponds to inode we are writing (which
2417 * means it has been truncated or invalidated), or the
2418 * page is already under writeback and we are not doing
2419 * a data integrity writeback, skip the page
2421 if (!PageDirty(page) ||
2422 (PageWriteback(page) &&
2423 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2424 unlikely(page->mapping != mapping)) {
2429 wait_on_page_writeback(page);
2430 BUG_ON(PageWriteback(page));
2432 if (mpd->map.m_len == 0)
2433 mpd->first_page = page->index;
2434 mpd->next_page = page->index + 1;
2435 /* Add all dirty buffers to mpd */
2436 lblk = ((ext4_lblk_t)page->index) <<
2437 (PAGE_CACHE_SHIFT - blkbits);
2438 head = page_buffers(page);
2439 err = mpage_process_page_bufs(mpd, head, head, lblk);
2445 pagevec_release(&pvec);
2450 pagevec_release(&pvec);
2454 static int __writepage(struct page *page, struct writeback_control *wbc,
2457 struct address_space *mapping = data;
2458 int ret = ext4_writepage(page, wbc);
2459 mapping_set_error(mapping, ret);
2463 static int ext4_writepages(struct address_space *mapping,
2464 struct writeback_control *wbc)
2466 pgoff_t writeback_index = 0;
2467 long nr_to_write = wbc->nr_to_write;
2468 int range_whole = 0;
2470 handle_t *handle = NULL;
2471 struct mpage_da_data mpd;
2472 struct inode *inode = mapping->host;
2473 int needed_blocks, rsv_blocks = 0, ret = 0;
2474 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2476 struct blk_plug plug;
2477 bool give_up_on_write = false;
2479 trace_ext4_writepages(inode, wbc);
2481 if (dax_mapping(mapping))
2482 return dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2486 * No pages to write? This is mainly a kludge to avoid starting
2487 * a transaction for special inodes like journal inode on last iput()
2488 * because that could violate lock ordering on umount
2490 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2491 goto out_writepages;
2493 if (ext4_should_journal_data(inode)) {
2494 struct blk_plug plug;
2496 blk_start_plug(&plug);
2497 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2498 blk_finish_plug(&plug);
2499 goto out_writepages;
2503 * If the filesystem has aborted, it is read-only, so return
2504 * right away instead of dumping stack traces later on that
2505 * will obscure the real source of the problem. We test
2506 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2507 * the latter could be true if the filesystem is mounted
2508 * read-only, and in that case, ext4_writepages should
2509 * *never* be called, so if that ever happens, we would want
2512 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2514 goto out_writepages;
2517 if (ext4_should_dioread_nolock(inode)) {
2519 * We may need to convert up to one extent per block in
2520 * the page and we may dirty the inode.
2522 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2526 * If we have inline data and arrive here, it means that
2527 * we will soon create the block for the 1st page, so
2528 * we'd better clear the inline data here.
2530 if (ext4_has_inline_data(inode)) {
2531 /* Just inode will be modified... */
2532 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2533 if (IS_ERR(handle)) {
2534 ret = PTR_ERR(handle);
2535 goto out_writepages;
2537 BUG_ON(ext4_test_inode_state(inode,
2538 EXT4_STATE_MAY_INLINE_DATA));
2539 ext4_destroy_inline_data(handle, inode);
2540 ext4_journal_stop(handle);
2543 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2546 if (wbc->range_cyclic) {
2547 writeback_index = mapping->writeback_index;
2548 if (writeback_index)
2550 mpd.first_page = writeback_index;
2553 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2554 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2559 ext4_io_submit_init(&mpd.io_submit, wbc);
2561 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2562 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2564 blk_start_plug(&plug);
2565 while (!done && mpd.first_page <= mpd.last_page) {
2566 /* For each extent of pages we use new io_end */
2567 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2568 if (!mpd.io_submit.io_end) {
2574 * We have two constraints: We find one extent to map and we
2575 * must always write out whole page (makes a difference when
2576 * blocksize < pagesize) so that we don't block on IO when we
2577 * try to write out the rest of the page. Journalled mode is
2578 * not supported by delalloc.
2580 BUG_ON(ext4_should_journal_data(inode));
2581 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2583 /* start a new transaction */
2584 handle = ext4_journal_start_with_reserve(inode,
2585 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2586 if (IS_ERR(handle)) {
2587 ret = PTR_ERR(handle);
2588 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2589 "%ld pages, ino %lu; err %d", __func__,
2590 wbc->nr_to_write, inode->i_ino, ret);
2591 /* Release allocated io_end */
2592 ext4_put_io_end(mpd.io_submit.io_end);
2596 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2597 ret = mpage_prepare_extent_to_map(&mpd);
2600 ret = mpage_map_and_submit_extent(handle, &mpd,
2604 * We scanned the whole range (or exhausted
2605 * nr_to_write), submitted what was mapped and
2606 * didn't find anything needing mapping. We are
2612 ext4_journal_stop(handle);
2613 /* Submit prepared bio */
2614 ext4_io_submit(&mpd.io_submit);
2615 /* Unlock pages we didn't use */
2616 mpage_release_unused_pages(&mpd, give_up_on_write);
2617 /* Drop our io_end reference we got from init */
2618 ext4_put_io_end(mpd.io_submit.io_end);
2620 if (ret == -ENOSPC && sbi->s_journal) {
2622 * Commit the transaction which would
2623 * free blocks released in the transaction
2626 jbd2_journal_force_commit_nested(sbi->s_journal);
2630 /* Fatal error - ENOMEM, EIO... */
2634 blk_finish_plug(&plug);
2635 if (!ret && !cycled && wbc->nr_to_write > 0) {
2637 mpd.last_page = writeback_index - 1;
2643 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2645 * Set the writeback_index so that range_cyclic
2646 * mode will write it back later
2648 mapping->writeback_index = mpd.first_page;
2651 trace_ext4_writepages_result(inode, wbc, ret,
2652 nr_to_write - wbc->nr_to_write);
2656 static int ext4_nonda_switch(struct super_block *sb)
2658 s64 free_clusters, dirty_clusters;
2659 struct ext4_sb_info *sbi = EXT4_SB(sb);
2662 * switch to non delalloc mode if we are running low
2663 * on free block. The free block accounting via percpu
2664 * counters can get slightly wrong with percpu_counter_batch getting
2665 * accumulated on each CPU without updating global counters
2666 * Delalloc need an accurate free block accounting. So switch
2667 * to non delalloc when we are near to error range.
2670 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2672 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2674 * Start pushing delalloc when 1/2 of free blocks are dirty.
2676 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2677 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2679 if (2 * free_clusters < 3 * dirty_clusters ||
2680 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2682 * free block count is less than 150% of dirty blocks
2683 * or free blocks is less than watermark
2690 /* We always reserve for an inode update; the superblock could be there too */
2691 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2693 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2696 if (pos + len <= 0x7fffffffULL)
2699 /* We might need to update the superblock to set LARGE_FILE */
2703 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2704 loff_t pos, unsigned len, unsigned flags,
2705 struct page **pagep, void **fsdata)
2707 int ret, retries = 0;
2710 struct inode *inode = mapping->host;
2713 index = pos >> PAGE_CACHE_SHIFT;
2715 if (ext4_nonda_switch(inode->i_sb)) {
2716 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2717 return ext4_write_begin(file, mapping, pos,
2718 len, flags, pagep, fsdata);
2720 *fsdata = (void *)0;
2721 trace_ext4_da_write_begin(inode, pos, len, flags);
2723 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2724 ret = ext4_da_write_inline_data_begin(mapping, inode,
2734 * grab_cache_page_write_begin() can take a long time if the
2735 * system is thrashing due to memory pressure, or if the page
2736 * is being written back. So grab it first before we start
2737 * the transaction handle. This also allows us to allocate
2738 * the page (if needed) without using GFP_NOFS.
2741 page = grab_cache_page_write_begin(mapping, index, flags);
2747 * With delayed allocation, we don't log the i_disksize update
2748 * if there is delayed block allocation. But we still need
2749 * to journalling the i_disksize update if writes to the end
2750 * of file which has an already mapped buffer.
2753 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2754 ext4_da_write_credits(inode, pos, len));
2755 if (IS_ERR(handle)) {
2756 page_cache_release(page);
2757 return PTR_ERR(handle);
2761 if (page->mapping != mapping) {
2762 /* The page got truncated from under us */
2764 page_cache_release(page);
2765 ext4_journal_stop(handle);
2768 /* In case writeback began while the page was unlocked */
2769 wait_for_stable_page(page);
2771 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2772 ret = ext4_block_write_begin(page, pos, len,
2773 ext4_da_get_block_prep);
2775 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2779 ext4_journal_stop(handle);
2781 * block_write_begin may have instantiated a few blocks
2782 * outside i_size. Trim these off again. Don't need
2783 * i_size_read because we hold i_mutex.
2785 if (pos + len > inode->i_size)
2786 ext4_truncate_failed_write(inode);
2788 if (ret == -ENOSPC &&
2789 ext4_should_retry_alloc(inode->i_sb, &retries))
2792 page_cache_release(page);
2801 * Check if we should update i_disksize
2802 * when write to the end of file but not require block allocation
2804 static int ext4_da_should_update_i_disksize(struct page *page,
2805 unsigned long offset)
2807 struct buffer_head *bh;
2808 struct inode *inode = page->mapping->host;
2812 bh = page_buffers(page);
2813 idx = offset >> inode->i_blkbits;
2815 for (i = 0; i < idx; i++)
2816 bh = bh->b_this_page;
2818 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2823 static int ext4_da_write_end(struct file *file,
2824 struct address_space *mapping,
2825 loff_t pos, unsigned len, unsigned copied,
2826 struct page *page, void *fsdata)
2828 struct inode *inode = mapping->host;
2830 handle_t *handle = ext4_journal_current_handle();
2832 unsigned long start, end;
2833 int write_mode = (int)(unsigned long)fsdata;
2835 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2836 return ext4_write_end(file, mapping, pos,
2837 len, copied, page, fsdata);
2839 trace_ext4_da_write_end(inode, pos, len, copied);
2840 start = pos & (PAGE_CACHE_SIZE - 1);
2841 end = start + copied - 1;
2844 * generic_write_end() will run mark_inode_dirty() if i_size
2845 * changes. So let's piggyback the i_disksize mark_inode_dirty
2848 new_i_size = pos + copied;
2849 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2850 if (ext4_has_inline_data(inode) ||
2851 ext4_da_should_update_i_disksize(page, end)) {
2852 ext4_update_i_disksize(inode, new_i_size);
2853 /* We need to mark inode dirty even if
2854 * new_i_size is less that inode->i_size
2855 * bu greater than i_disksize.(hint delalloc)
2857 ext4_mark_inode_dirty(handle, inode);
2861 if (write_mode != CONVERT_INLINE_DATA &&
2862 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2863 ext4_has_inline_data(inode))
2864 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2867 ret2 = generic_write_end(file, mapping, pos, len, copied,
2873 ret2 = ext4_journal_stop(handle);
2877 return ret ? ret : copied;
2880 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2881 unsigned int length)
2884 * Drop reserved blocks
2886 BUG_ON(!PageLocked(page));
2887 if (!page_has_buffers(page))
2890 ext4_da_page_release_reservation(page, offset, length);
2893 ext4_invalidatepage(page, offset, length);
2899 * Force all delayed allocation blocks to be allocated for a given inode.
2901 int ext4_alloc_da_blocks(struct inode *inode)
2903 trace_ext4_alloc_da_blocks(inode);
2905 if (!EXT4_I(inode)->i_reserved_data_blocks)
2909 * We do something simple for now. The filemap_flush() will
2910 * also start triggering a write of the data blocks, which is
2911 * not strictly speaking necessary (and for users of
2912 * laptop_mode, not even desirable). However, to do otherwise
2913 * would require replicating code paths in:
2915 * ext4_writepages() ->
2916 * write_cache_pages() ---> (via passed in callback function)
2917 * __mpage_da_writepage() -->
2918 * mpage_add_bh_to_extent()
2919 * mpage_da_map_blocks()
2921 * The problem is that write_cache_pages(), located in
2922 * mm/page-writeback.c, marks pages clean in preparation for
2923 * doing I/O, which is not desirable if we're not planning on
2926 * We could call write_cache_pages(), and then redirty all of
2927 * the pages by calling redirty_page_for_writepage() but that
2928 * would be ugly in the extreme. So instead we would need to
2929 * replicate parts of the code in the above functions,
2930 * simplifying them because we wouldn't actually intend to
2931 * write out the pages, but rather only collect contiguous
2932 * logical block extents, call the multi-block allocator, and
2933 * then update the buffer heads with the block allocations.
2935 * For now, though, we'll cheat by calling filemap_flush(),
2936 * which will map the blocks, and start the I/O, but not
2937 * actually wait for the I/O to complete.
2939 return filemap_flush(inode->i_mapping);
2943 * bmap() is special. It gets used by applications such as lilo and by
2944 * the swapper to find the on-disk block of a specific piece of data.
2946 * Naturally, this is dangerous if the block concerned is still in the
2947 * journal. If somebody makes a swapfile on an ext4 data-journaling
2948 * filesystem and enables swap, then they may get a nasty shock when the
2949 * data getting swapped to that swapfile suddenly gets overwritten by
2950 * the original zero's written out previously to the journal and
2951 * awaiting writeback in the kernel's buffer cache.
2953 * So, if we see any bmap calls here on a modified, data-journaled file,
2954 * take extra steps to flush any blocks which might be in the cache.
2956 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2958 struct inode *inode = mapping->host;
2963 * We can get here for an inline file via the FIBMAP ioctl
2965 if (ext4_has_inline_data(inode))
2968 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2969 test_opt(inode->i_sb, DELALLOC)) {
2971 * With delalloc we want to sync the file
2972 * so that we can make sure we allocate
2975 filemap_write_and_wait(mapping);
2978 if (EXT4_JOURNAL(inode) &&
2979 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2981 * This is a REALLY heavyweight approach, but the use of
2982 * bmap on dirty files is expected to be extremely rare:
2983 * only if we run lilo or swapon on a freshly made file
2984 * do we expect this to happen.
2986 * (bmap requires CAP_SYS_RAWIO so this does not
2987 * represent an unprivileged user DOS attack --- we'd be
2988 * in trouble if mortal users could trigger this path at
2991 * NB. EXT4_STATE_JDATA is not set on files other than
2992 * regular files. If somebody wants to bmap a directory
2993 * or symlink and gets confused because the buffer
2994 * hasn't yet been flushed to disk, they deserve
2995 * everything they get.
2998 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2999 journal = EXT4_JOURNAL(inode);
3000 jbd2_journal_lock_updates(journal);
3001 err = jbd2_journal_flush(journal);
3002 jbd2_journal_unlock_updates(journal);
3008 return generic_block_bmap(mapping, block, ext4_get_block);
3011 static int ext4_readpage(struct file *file, struct page *page)
3014 struct inode *inode = page->mapping->host;
3016 trace_ext4_readpage(page);
3018 if (ext4_has_inline_data(inode))
3019 ret = ext4_readpage_inline(inode, page);
3022 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3028 ext4_readpages(struct file *file, struct address_space *mapping,
3029 struct list_head *pages, unsigned nr_pages)
3031 struct inode *inode = mapping->host;
3033 /* If the file has inline data, no need to do readpages. */
3034 if (ext4_has_inline_data(inode))
3037 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3040 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3041 unsigned int length)
3043 trace_ext4_invalidatepage(page, offset, length);
3045 /* No journalling happens on data buffers when this function is used */
3046 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3048 block_invalidatepage(page, offset, length);
3051 static int __ext4_journalled_invalidatepage(struct page *page,
3052 unsigned int offset,
3053 unsigned int length)
3055 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3057 trace_ext4_journalled_invalidatepage(page, offset, length);
3060 * If it's a full truncate we just forget about the pending dirtying
3062 if (offset == 0 && length == PAGE_CACHE_SIZE)
3063 ClearPageChecked(page);
3065 return jbd2_journal_invalidatepage(journal, page, offset, length);
3068 /* Wrapper for aops... */
3069 static void ext4_journalled_invalidatepage(struct page *page,
3070 unsigned int offset,
3071 unsigned int length)
3073 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3076 static int ext4_releasepage(struct page *page, gfp_t wait)
3078 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3080 trace_ext4_releasepage(page);
3082 /* Page has dirty journalled data -> cannot release */
3083 if (PageChecked(page))
3086 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3088 return try_to_free_buffers(page);
3092 * ext4_get_block used when preparing for a DIO write or buffer write.
3093 * We allocate an uinitialized extent if blocks haven't been allocated.
3094 * The extent will be converted to initialized after the IO is complete.
3096 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3097 struct buffer_head *bh_result, int create)
3099 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3100 inode->i_ino, create);
3101 return _ext4_get_block(inode, iblock, bh_result,
3102 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3105 static int ext4_get_block_overwrite(struct inode *inode, sector_t iblock,
3106 struct buffer_head *bh_result, int create)
3110 ext4_debug("ext4_get_block_overwrite: inode %lu, create flag %d\n",
3111 inode->i_ino, create);
3112 ret = _ext4_get_block(inode, iblock, bh_result, 0);
3114 * Blocks should have been preallocated! ext4_file_write_iter() checks
3117 WARN_ON_ONCE(!buffer_mapped(bh_result));
3122 #ifdef CONFIG_FS_DAX
3123 int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
3124 struct buffer_head *bh_result, int create)
3128 struct ext4_map_blocks map;
3129 handle_t *handle = NULL;
3132 ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n",
3133 inode->i_ino, create);
3134 map.m_lblk = iblock;
3135 map.m_len = bh_result->b_size >> inode->i_blkbits;
3136 credits = ext4_chunk_trans_blocks(inode, map.m_len);
3138 flags |= EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_CREATE_ZERO;
3139 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
3140 if (IS_ERR(handle)) {
3141 ret = PTR_ERR(handle);
3146 ret = ext4_map_blocks(handle, inode, &map, flags);
3148 err = ext4_journal_stop(handle);
3149 if (ret >= 0 && err < 0)
3154 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3158 * We are protected by i_mmap_sem so we know block cannot go
3159 * away from under us even though we dropped i_data_sem.
3160 * Convert extent to written and write zeros there.
3162 * Note: We may get here even when create == 0.
3164 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
3165 if (IS_ERR(handle)) {
3166 ret = PTR_ERR(handle);
3170 err = ext4_map_blocks(handle, inode, &map,
3171 EXT4_GET_BLOCKS_CONVERT | EXT4_GET_BLOCKS_CREATE_ZERO);
3174 err2 = ext4_journal_stop(handle);
3175 if (err2 < 0 && ret > 0)
3179 WARN_ON_ONCE(ret == 0 && create);
3181 map_bh(bh_result, inode->i_sb, map.m_pblk);
3182 bh_result->b_state = (bh_result->b_state & ~EXT4_MAP_FLAGS) |
3185 * At least for now we have to clear BH_New so that DAX code
3186 * doesn't attempt to zero blocks again in a racy way.
3188 bh_result->b_state &= ~(1 << BH_New);
3189 bh_result->b_size = map.m_len << inode->i_blkbits;
3196 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3197 ssize_t size, void *private)
3199 ext4_io_end_t *io_end = iocb->private;
3201 /* if not async direct IO just return */
3205 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3206 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3207 iocb->private, io_end->inode->i_ino, iocb, offset,
3210 iocb->private = NULL;
3211 io_end->offset = offset;
3212 io_end->size = size;
3213 ext4_put_io_end(io_end);
3217 * For ext4 extent files, ext4 will do direct-io write to holes,
3218 * preallocated extents, and those write extend the file, no need to
3219 * fall back to buffered IO.
3221 * For holes, we fallocate those blocks, mark them as unwritten
3222 * If those blocks were preallocated, we mark sure they are split, but
3223 * still keep the range to write as unwritten.
3225 * The unwritten extents will be converted to written when DIO is completed.
3226 * For async direct IO, since the IO may still pending when return, we
3227 * set up an end_io call back function, which will do the conversion
3228 * when async direct IO completed.
3230 * If the O_DIRECT write will extend the file then add this inode to the
3231 * orphan list. So recovery will truncate it back to the original size
3232 * if the machine crashes during the write.
3235 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3238 struct file *file = iocb->ki_filp;
3239 struct inode *inode = file->f_mapping->host;
3241 size_t count = iov_iter_count(iter);
3243 get_block_t *get_block_func = NULL;
3245 loff_t final_size = offset + count;
3246 ext4_io_end_t *io_end = NULL;
3248 /* Use the old path for reads and writes beyond i_size. */
3249 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3250 return ext4_ind_direct_IO(iocb, iter, offset);
3252 BUG_ON(iocb->private == NULL);
3255 * Make all waiters for direct IO properly wait also for extent
3256 * conversion. This also disallows race between truncate() and
3257 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3259 if (iov_iter_rw(iter) == WRITE)
3260 inode_dio_begin(inode);
3262 /* If we do a overwrite dio, i_mutex locking can be released */
3263 overwrite = *((int *)iocb->private);
3266 inode_unlock(inode);
3269 * We could direct write to holes and fallocate.
3271 * Allocated blocks to fill the hole are marked as
3272 * unwritten to prevent parallel buffered read to expose
3273 * the stale data before DIO complete the data IO.
3275 * As to previously fallocated extents, ext4 get_block will
3276 * just simply mark the buffer mapped but still keep the
3277 * extents unwritten.
3279 * For non AIO case, we will convert those unwritten extents
3280 * to written after return back from blockdev_direct_IO.
3282 * For async DIO, the conversion needs to be deferred when the
3283 * IO is completed. The ext4 end_io callback function will be
3284 * called to take care of the conversion work. Here for async
3285 * case, we allocate an io_end structure to hook to the iocb.
3287 iocb->private = NULL;
3289 get_block_func = ext4_get_block_overwrite;
3291 ext4_inode_aio_set(inode, NULL);
3292 if (!is_sync_kiocb(iocb)) {
3293 io_end = ext4_init_io_end(inode, GFP_NOFS);
3299 * Grab reference for DIO. Will be dropped in
3302 iocb->private = ext4_get_io_end(io_end);
3304 * we save the io structure for current async direct
3305 * IO, so that later ext4_map_blocks() could flag the
3306 * io structure whether there is a unwritten extents
3307 * needs to be converted when IO is completed.
3309 ext4_inode_aio_set(inode, io_end);
3311 get_block_func = ext4_get_block_write;
3312 dio_flags = DIO_LOCKING;
3314 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3315 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3318 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3319 ext4_end_io_dio, dio_flags);
3321 ret = __blockdev_direct_IO(iocb, inode,
3322 inode->i_sb->s_bdev, iter, offset,
3324 ext4_end_io_dio, NULL, dio_flags);
3327 * Put our reference to io_end. This can free the io_end structure e.g.
3328 * in sync IO case or in case of error. It can even perform extent
3329 * conversion if all bios we submitted finished before we got here.
3330 * Note that in that case iocb->private can be already set to NULL
3334 ext4_inode_aio_set(inode, NULL);
3335 ext4_put_io_end(io_end);
3337 * When no IO was submitted ext4_end_io_dio() was not
3338 * called so we have to put iocb's reference.
3340 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3341 WARN_ON(iocb->private != io_end);
3342 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3343 ext4_put_io_end(io_end);
3344 iocb->private = NULL;
3347 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3348 EXT4_STATE_DIO_UNWRITTEN)) {
3351 * for non AIO case, since the IO is already
3352 * completed, we could do the conversion right here
3354 err = ext4_convert_unwritten_extents(NULL, inode,
3358 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3362 if (iov_iter_rw(iter) == WRITE)
3363 inode_dio_end(inode);
3364 /* take i_mutex locking again if we do a ovewrite dio */
3371 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3374 struct file *file = iocb->ki_filp;
3375 struct inode *inode = file->f_mapping->host;
3376 size_t count = iov_iter_count(iter);
3379 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3380 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3385 * If we are doing data journalling we don't support O_DIRECT
3387 if (ext4_should_journal_data(inode))
3390 /* Let buffer I/O handle the inline data case. */
3391 if (ext4_has_inline_data(inode))
3394 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3395 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3396 ret = ext4_ext_direct_IO(iocb, iter, offset);
3398 ret = ext4_ind_direct_IO(iocb, iter, offset);
3399 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3404 * Pages can be marked dirty completely asynchronously from ext4's journalling
3405 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3406 * much here because ->set_page_dirty is called under VFS locks. The page is
3407 * not necessarily locked.
3409 * We cannot just dirty the page and leave attached buffers clean, because the
3410 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3411 * or jbddirty because all the journalling code will explode.
3413 * So what we do is to mark the page "pending dirty" and next time writepage
3414 * is called, propagate that into the buffers appropriately.
3416 static int ext4_journalled_set_page_dirty(struct page *page)
3418 SetPageChecked(page);
3419 return __set_page_dirty_nobuffers(page);
3422 static const struct address_space_operations ext4_aops = {
3423 .readpage = ext4_readpage,
3424 .readpages = ext4_readpages,
3425 .writepage = ext4_writepage,
3426 .writepages = ext4_writepages,
3427 .write_begin = ext4_write_begin,
3428 .write_end = ext4_write_end,
3430 .invalidatepage = ext4_invalidatepage,
3431 .releasepage = ext4_releasepage,
3432 .direct_IO = ext4_direct_IO,
3433 .migratepage = buffer_migrate_page,
3434 .is_partially_uptodate = block_is_partially_uptodate,
3435 .error_remove_page = generic_error_remove_page,
3438 static const struct address_space_operations ext4_journalled_aops = {
3439 .readpage = ext4_readpage,
3440 .readpages = ext4_readpages,
3441 .writepage = ext4_writepage,
3442 .writepages = ext4_writepages,
3443 .write_begin = ext4_write_begin,
3444 .write_end = ext4_journalled_write_end,
3445 .set_page_dirty = ext4_journalled_set_page_dirty,
3447 .invalidatepage = ext4_journalled_invalidatepage,
3448 .releasepage = ext4_releasepage,
3449 .direct_IO = ext4_direct_IO,
3450 .is_partially_uptodate = block_is_partially_uptodate,
3451 .error_remove_page = generic_error_remove_page,
3454 static const struct address_space_operations ext4_da_aops = {
3455 .readpage = ext4_readpage,
3456 .readpages = ext4_readpages,
3457 .writepage = ext4_writepage,
3458 .writepages = ext4_writepages,
3459 .write_begin = ext4_da_write_begin,
3460 .write_end = ext4_da_write_end,
3462 .invalidatepage = ext4_da_invalidatepage,
3463 .releasepage = ext4_releasepage,
3464 .direct_IO = ext4_direct_IO,
3465 .migratepage = buffer_migrate_page,
3466 .is_partially_uptodate = block_is_partially_uptodate,
3467 .error_remove_page = generic_error_remove_page,
3470 void ext4_set_aops(struct inode *inode)
3472 switch (ext4_inode_journal_mode(inode)) {
3473 case EXT4_INODE_ORDERED_DATA_MODE:
3474 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3476 case EXT4_INODE_WRITEBACK_DATA_MODE:
3477 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3479 case EXT4_INODE_JOURNAL_DATA_MODE:
3480 inode->i_mapping->a_ops = &ext4_journalled_aops;
3485 if (test_opt(inode->i_sb, DELALLOC))
3486 inode->i_mapping->a_ops = &ext4_da_aops;
3488 inode->i_mapping->a_ops = &ext4_aops;
3491 static int __ext4_block_zero_page_range(handle_t *handle,
3492 struct address_space *mapping, loff_t from, loff_t length)
3494 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3495 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3496 unsigned blocksize, pos;
3498 struct inode *inode = mapping->host;
3499 struct buffer_head *bh;
3503 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3504 mapping_gfp_constraint(mapping, ~__GFP_FS));
3508 blocksize = inode->i_sb->s_blocksize;
3510 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3512 if (!page_has_buffers(page))
3513 create_empty_buffers(page, blocksize, 0);
3515 /* Find the buffer that contains "offset" */
3516 bh = page_buffers(page);
3518 while (offset >= pos) {
3519 bh = bh->b_this_page;
3523 if (buffer_freed(bh)) {
3524 BUFFER_TRACE(bh, "freed: skip");
3527 if (!buffer_mapped(bh)) {
3528 BUFFER_TRACE(bh, "unmapped");
3529 ext4_get_block(inode, iblock, bh, 0);
3530 /* unmapped? It's a hole - nothing to do */
3531 if (!buffer_mapped(bh)) {
3532 BUFFER_TRACE(bh, "still unmapped");
3537 /* Ok, it's mapped. Make sure it's up-to-date */
3538 if (PageUptodate(page))
3539 set_buffer_uptodate(bh);
3541 if (!buffer_uptodate(bh)) {
3543 ll_rw_block(READ, 1, &bh);
3545 /* Uhhuh. Read error. Complain and punt. */
3546 if (!buffer_uptodate(bh))
3548 if (S_ISREG(inode->i_mode) &&
3549 ext4_encrypted_inode(inode)) {
3550 /* We expect the key to be set. */
3551 BUG_ON(!ext4_has_encryption_key(inode));
3552 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3553 WARN_ON_ONCE(ext4_decrypt(page));
3556 if (ext4_should_journal_data(inode)) {
3557 BUFFER_TRACE(bh, "get write access");
3558 err = ext4_journal_get_write_access(handle, bh);
3562 zero_user(page, offset, length);
3563 BUFFER_TRACE(bh, "zeroed end of block");
3565 if (ext4_should_journal_data(inode)) {
3566 err = ext4_handle_dirty_metadata(handle, inode, bh);
3569 mark_buffer_dirty(bh);
3570 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3571 err = ext4_jbd2_file_inode(handle, inode);
3576 page_cache_release(page);
3581 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3582 * starting from file offset 'from'. The range to be zero'd must
3583 * be contained with in one block. If the specified range exceeds
3584 * the end of the block it will be shortened to end of the block
3585 * that cooresponds to 'from'
3587 static int ext4_block_zero_page_range(handle_t *handle,
3588 struct address_space *mapping, loff_t from, loff_t length)
3590 struct inode *inode = mapping->host;
3591 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3592 unsigned blocksize = inode->i_sb->s_blocksize;
3593 unsigned max = blocksize - (offset & (blocksize - 1));
3596 * correct length if it does not fall between
3597 * 'from' and the end of the block
3599 if (length > max || length < 0)
3603 return dax_zero_page_range(inode, from, length, ext4_get_block);
3604 return __ext4_block_zero_page_range(handle, mapping, from, length);
3608 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3609 * up to the end of the block which corresponds to `from'.
3610 * This required during truncate. We need to physically zero the tail end
3611 * of that block so it doesn't yield old data if the file is later grown.
3613 static int ext4_block_truncate_page(handle_t *handle,
3614 struct address_space *mapping, loff_t from)
3616 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3619 struct inode *inode = mapping->host;
3621 blocksize = inode->i_sb->s_blocksize;
3622 length = blocksize - (offset & (blocksize - 1));
3624 return ext4_block_zero_page_range(handle, mapping, from, length);
3627 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3628 loff_t lstart, loff_t length)
3630 struct super_block *sb = inode->i_sb;
3631 struct address_space *mapping = inode->i_mapping;
3632 unsigned partial_start, partial_end;
3633 ext4_fsblk_t start, end;
3634 loff_t byte_end = (lstart + length - 1);
3637 partial_start = lstart & (sb->s_blocksize - 1);
3638 partial_end = byte_end & (sb->s_blocksize - 1);
3640 start = lstart >> sb->s_blocksize_bits;
3641 end = byte_end >> sb->s_blocksize_bits;
3643 /* Handle partial zero within the single block */
3645 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3646 err = ext4_block_zero_page_range(handle, mapping,
3650 /* Handle partial zero out on the start of the range */
3651 if (partial_start) {
3652 err = ext4_block_zero_page_range(handle, mapping,
3653 lstart, sb->s_blocksize);
3657 /* Handle partial zero out on the end of the range */
3658 if (partial_end != sb->s_blocksize - 1)
3659 err = ext4_block_zero_page_range(handle, mapping,
3660 byte_end - partial_end,
3665 int ext4_can_truncate(struct inode *inode)
3667 if (S_ISREG(inode->i_mode))
3669 if (S_ISDIR(inode->i_mode))
3671 if (S_ISLNK(inode->i_mode))
3672 return !ext4_inode_is_fast_symlink(inode);
3677 * We have to make sure i_disksize gets properly updated before we truncate
3678 * page cache due to hole punching or zero range. Otherwise i_disksize update
3679 * can get lost as it may have been postponed to submission of writeback but
3680 * that will never happen after we truncate page cache.
3682 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3686 loff_t size = i_size_read(inode);
3688 WARN_ON(!inode_is_locked(inode));
3689 if (offset > size || offset + len < size)
3692 if (EXT4_I(inode)->i_disksize >= size)
3695 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3697 return PTR_ERR(handle);
3698 ext4_update_i_disksize(inode, size);
3699 ext4_mark_inode_dirty(handle, inode);
3700 ext4_journal_stop(handle);
3706 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3707 * associated with the given offset and length
3709 * @inode: File inode
3710 * @offset: The offset where the hole will begin
3711 * @len: The length of the hole
3713 * Returns: 0 on success or negative on failure
3716 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3718 struct super_block *sb = inode->i_sb;
3719 ext4_lblk_t first_block, stop_block;
3720 struct address_space *mapping = inode->i_mapping;
3721 loff_t first_block_offset, last_block_offset;
3723 unsigned int credits;
3726 if (!S_ISREG(inode->i_mode))
3729 trace_ext4_punch_hole(inode, offset, length, 0);
3732 * Write out all dirty pages to avoid race conditions
3733 * Then release them.
3735 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3736 ret = filemap_write_and_wait_range(mapping, offset,
3737 offset + length - 1);
3744 /* No need to punch hole beyond i_size */
3745 if (offset >= inode->i_size)
3749 * If the hole extends beyond i_size, set the hole
3750 * to end after the page that contains i_size
3752 if (offset + length > inode->i_size) {
3753 length = inode->i_size +
3754 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3758 if (offset & (sb->s_blocksize - 1) ||
3759 (offset + length) & (sb->s_blocksize - 1)) {
3761 * Attach jinode to inode for jbd2 if we do any zeroing of
3764 ret = ext4_inode_attach_jinode(inode);
3770 /* Wait all existing dio workers, newcomers will block on i_mutex */
3771 ext4_inode_block_unlocked_dio(inode);
3772 inode_dio_wait(inode);
3775 * Prevent page faults from reinstantiating pages we have released from
3778 down_write(&EXT4_I(inode)->i_mmap_sem);
3779 first_block_offset = round_up(offset, sb->s_blocksize);
3780 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3782 /* Now release the pages and zero block aligned part of pages*/
3783 if (last_block_offset > first_block_offset) {
3784 ret = ext4_update_disksize_before_punch(inode, offset, length);
3787 truncate_pagecache_range(inode, first_block_offset,
3791 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3792 credits = ext4_writepage_trans_blocks(inode);
3794 credits = ext4_blocks_for_truncate(inode);
3795 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3796 if (IS_ERR(handle)) {
3797 ret = PTR_ERR(handle);
3798 ext4_std_error(sb, ret);
3802 ret = ext4_zero_partial_blocks(handle, inode, offset,
3807 first_block = (offset + sb->s_blocksize - 1) >>
3808 EXT4_BLOCK_SIZE_BITS(sb);
3809 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3811 /* If there are no blocks to remove, return now */
3812 if (first_block >= stop_block)
3815 down_write(&EXT4_I(inode)->i_data_sem);
3816 ext4_discard_preallocations(inode);
3818 ret = ext4_es_remove_extent(inode, first_block,
3819 stop_block - first_block);
3821 up_write(&EXT4_I(inode)->i_data_sem);
3825 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3826 ret = ext4_ext_remove_space(inode, first_block,
3829 ret = ext4_ind_remove_space(handle, inode, first_block,
3832 up_write(&EXT4_I(inode)->i_data_sem);
3834 ext4_handle_sync(handle);
3836 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3837 ext4_mark_inode_dirty(handle, inode);
3839 ext4_journal_stop(handle);
3841 up_write(&EXT4_I(inode)->i_mmap_sem);
3842 ext4_inode_resume_unlocked_dio(inode);
3844 inode_unlock(inode);
3848 int ext4_inode_attach_jinode(struct inode *inode)
3850 struct ext4_inode_info *ei = EXT4_I(inode);
3851 struct jbd2_inode *jinode;
3853 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3856 jinode = jbd2_alloc_inode(GFP_KERNEL);
3857 spin_lock(&inode->i_lock);
3860 spin_unlock(&inode->i_lock);
3863 ei->jinode = jinode;
3864 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3867 spin_unlock(&inode->i_lock);
3868 if (unlikely(jinode != NULL))
3869 jbd2_free_inode(jinode);
3876 * We block out ext4_get_block() block instantiations across the entire
3877 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3878 * simultaneously on behalf of the same inode.
3880 * As we work through the truncate and commit bits of it to the journal there
3881 * is one core, guiding principle: the file's tree must always be consistent on
3882 * disk. We must be able to restart the truncate after a crash.
3884 * The file's tree may be transiently inconsistent in memory (although it
3885 * probably isn't), but whenever we close off and commit a journal transaction,
3886 * the contents of (the filesystem + the journal) must be consistent and
3887 * restartable. It's pretty simple, really: bottom up, right to left (although
3888 * left-to-right works OK too).
3890 * Note that at recovery time, journal replay occurs *before* the restart of
3891 * truncate against the orphan inode list.
3893 * The committed inode has the new, desired i_size (which is the same as
3894 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3895 * that this inode's truncate did not complete and it will again call
3896 * ext4_truncate() to have another go. So there will be instantiated blocks
3897 * to the right of the truncation point in a crashed ext4 filesystem. But
3898 * that's fine - as long as they are linked from the inode, the post-crash
3899 * ext4_truncate() run will find them and release them.
3901 void ext4_truncate(struct inode *inode)
3903 struct ext4_inode_info *ei = EXT4_I(inode);
3904 unsigned int credits;
3906 struct address_space *mapping = inode->i_mapping;
3909 * There is a possibility that we're either freeing the inode
3910 * or it's a completely new inode. In those cases we might not
3911 * have i_mutex locked because it's not necessary.
3913 if (!(inode->i_state & (I_NEW|I_FREEING)))
3914 WARN_ON(!inode_is_locked(inode));
3915 trace_ext4_truncate_enter(inode);
3917 if (!ext4_can_truncate(inode))
3920 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3922 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3923 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3925 if (ext4_has_inline_data(inode)) {
3928 ext4_inline_data_truncate(inode, &has_inline);
3933 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3934 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3935 if (ext4_inode_attach_jinode(inode) < 0)
3939 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3940 credits = ext4_writepage_trans_blocks(inode);
3942 credits = ext4_blocks_for_truncate(inode);
3944 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3945 if (IS_ERR(handle)) {
3946 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3950 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3951 ext4_block_truncate_page(handle, mapping, inode->i_size);
3954 * We add the inode to the orphan list, so that if this
3955 * truncate spans multiple transactions, and we crash, we will
3956 * resume the truncate when the filesystem recovers. It also
3957 * marks the inode dirty, to catch the new size.
3959 * Implication: the file must always be in a sane, consistent
3960 * truncatable state while each transaction commits.
3962 if (ext4_orphan_add(handle, inode))
3965 down_write(&EXT4_I(inode)->i_data_sem);
3967 ext4_discard_preallocations(inode);
3969 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3970 ext4_ext_truncate(handle, inode);
3972 ext4_ind_truncate(handle, inode);
3974 up_write(&ei->i_data_sem);
3977 ext4_handle_sync(handle);
3981 * If this was a simple ftruncate() and the file will remain alive,
3982 * then we need to clear up the orphan record which we created above.
3983 * However, if this was a real unlink then we were called by
3984 * ext4_evict_inode(), and we allow that function to clean up the
3985 * orphan info for us.
3988 ext4_orphan_del(handle, inode);
3990 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3991 ext4_mark_inode_dirty(handle, inode);
3992 ext4_journal_stop(handle);
3994 trace_ext4_truncate_exit(inode);
3998 * ext4_get_inode_loc returns with an extra refcount against the inode's
3999 * underlying buffer_head on success. If 'in_mem' is true, we have all
4000 * data in memory that is needed to recreate the on-disk version of this
4003 static int __ext4_get_inode_loc(struct inode *inode,
4004 struct ext4_iloc *iloc, int in_mem)
4006 struct ext4_group_desc *gdp;
4007 struct buffer_head *bh;
4008 struct super_block *sb = inode->i_sb;
4010 int inodes_per_block, inode_offset;
4013 if (!ext4_valid_inum(sb, inode->i_ino))
4014 return -EFSCORRUPTED;
4016 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4017 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4022 * Figure out the offset within the block group inode table
4024 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4025 inode_offset = ((inode->i_ino - 1) %
4026 EXT4_INODES_PER_GROUP(sb));
4027 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4028 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4030 bh = sb_getblk(sb, block);
4033 if (!buffer_uptodate(bh)) {
4037 * If the buffer has the write error flag, we have failed
4038 * to write out another inode in the same block. In this
4039 * case, we don't have to read the block because we may
4040 * read the old inode data successfully.
4042 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4043 set_buffer_uptodate(bh);
4045 if (buffer_uptodate(bh)) {
4046 /* someone brought it uptodate while we waited */
4052 * If we have all information of the inode in memory and this
4053 * is the only valid inode in the block, we need not read the
4057 struct buffer_head *bitmap_bh;
4060 start = inode_offset & ~(inodes_per_block - 1);
4062 /* Is the inode bitmap in cache? */
4063 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4064 if (unlikely(!bitmap_bh))
4068 * If the inode bitmap isn't in cache then the
4069 * optimisation may end up performing two reads instead
4070 * of one, so skip it.
4072 if (!buffer_uptodate(bitmap_bh)) {
4076 for (i = start; i < start + inodes_per_block; i++) {
4077 if (i == inode_offset)
4079 if (ext4_test_bit(i, bitmap_bh->b_data))
4083 if (i == start + inodes_per_block) {
4084 /* all other inodes are free, so skip I/O */
4085 memset(bh->b_data, 0, bh->b_size);
4086 set_buffer_uptodate(bh);
4094 * If we need to do any I/O, try to pre-readahead extra
4095 * blocks from the inode table.
4097 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4098 ext4_fsblk_t b, end, table;
4100 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4102 table = ext4_inode_table(sb, gdp);
4103 /* s_inode_readahead_blks is always a power of 2 */
4104 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4108 num = EXT4_INODES_PER_GROUP(sb);
4109 if (ext4_has_group_desc_csum(sb))
4110 num -= ext4_itable_unused_count(sb, gdp);
4111 table += num / inodes_per_block;
4115 sb_breadahead(sb, b++);
4119 * There are other valid inodes in the buffer, this inode
4120 * has in-inode xattrs, or we don't have this inode in memory.
4121 * Read the block from disk.
4123 trace_ext4_load_inode(inode);
4125 bh->b_end_io = end_buffer_read_sync;
4126 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4128 if (!buffer_uptodate(bh)) {
4129 EXT4_ERROR_INODE_BLOCK(inode, block,
4130 "unable to read itable block");
4140 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4142 /* We have all inode data except xattrs in memory here. */
4143 return __ext4_get_inode_loc(inode, iloc,
4144 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4147 void ext4_set_inode_flags(struct inode *inode)
4149 unsigned int flags = EXT4_I(inode)->i_flags;
4150 unsigned int new_fl = 0;
4152 if (flags & EXT4_SYNC_FL)
4154 if (flags & EXT4_APPEND_FL)
4156 if (flags & EXT4_IMMUTABLE_FL)
4157 new_fl |= S_IMMUTABLE;
4158 if (flags & EXT4_NOATIME_FL)
4159 new_fl |= S_NOATIME;
4160 if (flags & EXT4_DIRSYNC_FL)
4161 new_fl |= S_DIRSYNC;
4162 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
4164 inode_set_flags(inode, new_fl,
4165 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4168 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4169 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4171 unsigned int vfs_fl;
4172 unsigned long old_fl, new_fl;
4175 vfs_fl = ei->vfs_inode.i_flags;
4176 old_fl = ei->i_flags;
4177 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4178 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4180 if (vfs_fl & S_SYNC)
4181 new_fl |= EXT4_SYNC_FL;
4182 if (vfs_fl & S_APPEND)
4183 new_fl |= EXT4_APPEND_FL;
4184 if (vfs_fl & S_IMMUTABLE)
4185 new_fl |= EXT4_IMMUTABLE_FL;
4186 if (vfs_fl & S_NOATIME)
4187 new_fl |= EXT4_NOATIME_FL;
4188 if (vfs_fl & S_DIRSYNC)
4189 new_fl |= EXT4_DIRSYNC_FL;
4190 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4193 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4194 struct ext4_inode_info *ei)
4197 struct inode *inode = &(ei->vfs_inode);
4198 struct super_block *sb = inode->i_sb;
4200 if (ext4_has_feature_huge_file(sb)) {
4201 /* we are using combined 48 bit field */
4202 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4203 le32_to_cpu(raw_inode->i_blocks_lo);
4204 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4205 /* i_blocks represent file system block size */
4206 return i_blocks << (inode->i_blkbits - 9);
4211 return le32_to_cpu(raw_inode->i_blocks_lo);
4215 static inline void ext4_iget_extra_inode(struct inode *inode,
4216 struct ext4_inode *raw_inode,
4217 struct ext4_inode_info *ei)
4219 __le32 *magic = (void *)raw_inode +
4220 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4221 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4222 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4223 ext4_find_inline_data_nolock(inode);
4225 EXT4_I(inode)->i_inline_off = 0;
4228 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4230 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, EXT4_FEATURE_RO_COMPAT_PROJECT))
4232 *projid = EXT4_I(inode)->i_projid;
4236 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4238 struct ext4_iloc iloc;
4239 struct ext4_inode *raw_inode;
4240 struct ext4_inode_info *ei;
4241 struct inode *inode;
4242 journal_t *journal = EXT4_SB(sb)->s_journal;
4249 inode = iget_locked(sb, ino);
4251 return ERR_PTR(-ENOMEM);
4252 if (!(inode->i_state & I_NEW))
4258 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4261 raw_inode = ext4_raw_inode(&iloc);
4263 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4264 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4265 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4266 EXT4_INODE_SIZE(inode->i_sb)) {
4267 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4268 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4269 EXT4_INODE_SIZE(inode->i_sb));
4270 ret = -EFSCORRUPTED;
4274 ei->i_extra_isize = 0;
4276 /* Precompute checksum seed for inode metadata */
4277 if (ext4_has_metadata_csum(sb)) {
4278 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4280 __le32 inum = cpu_to_le32(inode->i_ino);
4281 __le32 gen = raw_inode->i_generation;
4282 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4284 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4288 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4289 EXT4_ERROR_INODE(inode, "checksum invalid");
4294 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4295 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4296 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4297 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4298 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4299 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4300 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4302 i_projid = EXT4_DEF_PROJID;
4304 if (!(test_opt(inode->i_sb, NO_UID32))) {
4305 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4306 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4308 i_uid_write(inode, i_uid);
4309 i_gid_write(inode, i_gid);
4310 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4311 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4313 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4314 ei->i_inline_off = 0;
4315 ei->i_dir_start_lookup = 0;
4316 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4317 /* We now have enough fields to check if the inode was active or not.
4318 * This is needed because nfsd might try to access dead inodes
4319 * the test is that same one that e2fsck uses
4320 * NeilBrown 1999oct15
4322 if (inode->i_nlink == 0) {
4323 if ((inode->i_mode == 0 ||
4324 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4325 ino != EXT4_BOOT_LOADER_INO) {
4326 /* this inode is deleted */
4330 /* The only unlinked inodes we let through here have
4331 * valid i_mode and are being read by the orphan
4332 * recovery code: that's fine, we're about to complete
4333 * the process of deleting those.
4334 * OR it is the EXT4_BOOT_LOADER_INO which is
4335 * not initialized on a new filesystem. */
4337 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4338 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4339 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4340 if (ext4_has_feature_64bit(sb))
4342 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4343 inode->i_size = ext4_isize(raw_inode);
4344 ei->i_disksize = inode->i_size;
4346 ei->i_reserved_quota = 0;
4348 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4349 ei->i_block_group = iloc.block_group;
4350 ei->i_last_alloc_group = ~0;
4352 * NOTE! The in-memory inode i_data array is in little-endian order
4353 * even on big-endian machines: we do NOT byteswap the block numbers!
4355 for (block = 0; block < EXT4_N_BLOCKS; block++)
4356 ei->i_data[block] = raw_inode->i_block[block];
4357 INIT_LIST_HEAD(&ei->i_orphan);
4360 * Set transaction id's of transactions that have to be committed
4361 * to finish f[data]sync. We set them to currently running transaction
4362 * as we cannot be sure that the inode or some of its metadata isn't
4363 * part of the transaction - the inode could have been reclaimed and
4364 * now it is reread from disk.
4367 transaction_t *transaction;
4370 read_lock(&journal->j_state_lock);
4371 if (journal->j_running_transaction)
4372 transaction = journal->j_running_transaction;
4374 transaction = journal->j_committing_transaction;
4376 tid = transaction->t_tid;
4378 tid = journal->j_commit_sequence;
4379 read_unlock(&journal->j_state_lock);
4380 ei->i_sync_tid = tid;
4381 ei->i_datasync_tid = tid;
4384 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4385 if (ei->i_extra_isize == 0) {
4386 /* The extra space is currently unused. Use it. */
4387 ei->i_extra_isize = sizeof(struct ext4_inode) -
4388 EXT4_GOOD_OLD_INODE_SIZE;
4390 ext4_iget_extra_inode(inode, raw_inode, ei);
4394 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4395 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4396 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4397 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4399 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4400 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4401 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4402 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4404 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4409 if (ei->i_file_acl &&
4410 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4411 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4413 ret = -EFSCORRUPTED;
4415 } else if (!ext4_has_inline_data(inode)) {
4416 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4417 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4418 (S_ISLNK(inode->i_mode) &&
4419 !ext4_inode_is_fast_symlink(inode))))
4420 /* Validate extent which is part of inode */
4421 ret = ext4_ext_check_inode(inode);
4422 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4423 (S_ISLNK(inode->i_mode) &&
4424 !ext4_inode_is_fast_symlink(inode))) {
4425 /* Validate block references which are part of inode */
4426 ret = ext4_ind_check_inode(inode);
4432 if (S_ISREG(inode->i_mode)) {
4433 inode->i_op = &ext4_file_inode_operations;
4434 inode->i_fop = &ext4_file_operations;
4435 ext4_set_aops(inode);
4436 } else if (S_ISDIR(inode->i_mode)) {
4437 inode->i_op = &ext4_dir_inode_operations;
4438 inode->i_fop = &ext4_dir_operations;
4439 } else if (S_ISLNK(inode->i_mode)) {
4440 if (ext4_encrypted_inode(inode)) {
4441 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4442 ext4_set_aops(inode);
4443 } else if (ext4_inode_is_fast_symlink(inode)) {
4444 inode->i_link = (char *)ei->i_data;
4445 inode->i_op = &ext4_fast_symlink_inode_operations;
4446 nd_terminate_link(ei->i_data, inode->i_size,
4447 sizeof(ei->i_data) - 1);
4449 inode->i_op = &ext4_symlink_inode_operations;
4450 ext4_set_aops(inode);
4452 inode_nohighmem(inode);
4453 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4454 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4455 inode->i_op = &ext4_special_inode_operations;
4456 if (raw_inode->i_block[0])
4457 init_special_inode(inode, inode->i_mode,
4458 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4460 init_special_inode(inode, inode->i_mode,
4461 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4462 } else if (ino == EXT4_BOOT_LOADER_INO) {
4463 make_bad_inode(inode);
4465 ret = -EFSCORRUPTED;
4466 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4470 ext4_set_inode_flags(inode);
4471 unlock_new_inode(inode);
4477 return ERR_PTR(ret);
4480 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4482 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4483 return ERR_PTR(-EFSCORRUPTED);
4484 return ext4_iget(sb, ino);
4487 static int ext4_inode_blocks_set(handle_t *handle,
4488 struct ext4_inode *raw_inode,
4489 struct ext4_inode_info *ei)
4491 struct inode *inode = &(ei->vfs_inode);
4492 u64 i_blocks = inode->i_blocks;
4493 struct super_block *sb = inode->i_sb;
4495 if (i_blocks <= ~0U) {
4497 * i_blocks can be represented in a 32 bit variable
4498 * as multiple of 512 bytes
4500 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4501 raw_inode->i_blocks_high = 0;
4502 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4505 if (!ext4_has_feature_huge_file(sb))
4508 if (i_blocks <= 0xffffffffffffULL) {
4510 * i_blocks can be represented in a 48 bit variable
4511 * as multiple of 512 bytes
4513 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4514 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4515 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4517 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4518 /* i_block is stored in file system block size */
4519 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4520 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4521 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4526 struct other_inode {
4527 unsigned long orig_ino;
4528 struct ext4_inode *raw_inode;
4531 static int other_inode_match(struct inode * inode, unsigned long ino,
4534 struct other_inode *oi = (struct other_inode *) data;
4536 if ((inode->i_ino != ino) ||
4537 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4538 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4539 ((inode->i_state & I_DIRTY_TIME) == 0))
4541 spin_lock(&inode->i_lock);
4542 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4543 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4544 (inode->i_state & I_DIRTY_TIME)) {
4545 struct ext4_inode_info *ei = EXT4_I(inode);
4547 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4548 spin_unlock(&inode->i_lock);
4550 spin_lock(&ei->i_raw_lock);
4551 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4552 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4553 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4554 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4555 spin_unlock(&ei->i_raw_lock);
4556 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4559 spin_unlock(&inode->i_lock);
4564 * Opportunistically update the other time fields for other inodes in
4565 * the same inode table block.
4567 static void ext4_update_other_inodes_time(struct super_block *sb,
4568 unsigned long orig_ino, char *buf)
4570 struct other_inode oi;
4572 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4573 int inode_size = EXT4_INODE_SIZE(sb);
4575 oi.orig_ino = orig_ino;
4577 * Calculate the first inode in the inode table block. Inode
4578 * numbers are one-based. That is, the first inode in a block
4579 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4581 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4582 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4583 if (ino == orig_ino)
4585 oi.raw_inode = (struct ext4_inode *) buf;
4586 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4591 * Post the struct inode info into an on-disk inode location in the
4592 * buffer-cache. This gobbles the caller's reference to the
4593 * buffer_head in the inode location struct.
4595 * The caller must have write access to iloc->bh.
4597 static int ext4_do_update_inode(handle_t *handle,
4598 struct inode *inode,
4599 struct ext4_iloc *iloc)
4601 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4602 struct ext4_inode_info *ei = EXT4_I(inode);
4603 struct buffer_head *bh = iloc->bh;
4604 struct super_block *sb = inode->i_sb;
4605 int err = 0, rc, block;
4606 int need_datasync = 0, set_large_file = 0;
4611 spin_lock(&ei->i_raw_lock);
4613 /* For fields not tracked in the in-memory inode,
4614 * initialise them to zero for new inodes. */
4615 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4616 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4618 ext4_get_inode_flags(ei);
4619 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4620 i_uid = i_uid_read(inode);
4621 i_gid = i_gid_read(inode);
4622 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4623 if (!(test_opt(inode->i_sb, NO_UID32))) {
4624 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4625 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4627 * Fix up interoperability with old kernels. Otherwise, old inodes get
4628 * re-used with the upper 16 bits of the uid/gid intact
4631 raw_inode->i_uid_high =
4632 cpu_to_le16(high_16_bits(i_uid));
4633 raw_inode->i_gid_high =
4634 cpu_to_le16(high_16_bits(i_gid));
4636 raw_inode->i_uid_high = 0;
4637 raw_inode->i_gid_high = 0;
4640 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4641 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4642 raw_inode->i_uid_high = 0;
4643 raw_inode->i_gid_high = 0;
4645 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4647 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4648 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4649 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4650 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4652 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4654 spin_unlock(&ei->i_raw_lock);
4657 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4658 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4659 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4660 raw_inode->i_file_acl_high =
4661 cpu_to_le16(ei->i_file_acl >> 32);
4662 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4663 if (ei->i_disksize != ext4_isize(raw_inode)) {
4664 ext4_isize_set(raw_inode, ei->i_disksize);
4667 if (ei->i_disksize > 0x7fffffffULL) {
4668 if (!ext4_has_feature_large_file(sb) ||
4669 EXT4_SB(sb)->s_es->s_rev_level ==
4670 cpu_to_le32(EXT4_GOOD_OLD_REV))
4673 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4674 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4675 if (old_valid_dev(inode->i_rdev)) {
4676 raw_inode->i_block[0] =
4677 cpu_to_le32(old_encode_dev(inode->i_rdev));
4678 raw_inode->i_block[1] = 0;
4680 raw_inode->i_block[0] = 0;
4681 raw_inode->i_block[1] =
4682 cpu_to_le32(new_encode_dev(inode->i_rdev));
4683 raw_inode->i_block[2] = 0;
4685 } else if (!ext4_has_inline_data(inode)) {
4686 for (block = 0; block < EXT4_N_BLOCKS; block++)
4687 raw_inode->i_block[block] = ei->i_data[block];
4690 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4691 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4692 if (ei->i_extra_isize) {
4693 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4694 raw_inode->i_version_hi =
4695 cpu_to_le32(inode->i_version >> 32);
4696 raw_inode->i_extra_isize =
4697 cpu_to_le16(ei->i_extra_isize);
4701 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
4702 EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4703 i_projid != EXT4_DEF_PROJID);
4705 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4706 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4707 raw_inode->i_projid = cpu_to_le32(i_projid);
4709 ext4_inode_csum_set(inode, raw_inode, ei);
4710 spin_unlock(&ei->i_raw_lock);
4711 if (inode->i_sb->s_flags & MS_LAZYTIME)
4712 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4715 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4716 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4719 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4720 if (set_large_file) {
4721 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4722 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4725 ext4_update_dynamic_rev(sb);
4726 ext4_set_feature_large_file(sb);
4727 ext4_handle_sync(handle);
4728 err = ext4_handle_dirty_super(handle, sb);
4730 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4733 ext4_std_error(inode->i_sb, err);
4738 * ext4_write_inode()
4740 * We are called from a few places:
4742 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4743 * Here, there will be no transaction running. We wait for any running
4744 * transaction to commit.
4746 * - Within flush work (sys_sync(), kupdate and such).
4747 * We wait on commit, if told to.
4749 * - Within iput_final() -> write_inode_now()
4750 * We wait on commit, if told to.
4752 * In all cases it is actually safe for us to return without doing anything,
4753 * because the inode has been copied into a raw inode buffer in
4754 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4757 * Note that we are absolutely dependent upon all inode dirtiers doing the
4758 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4759 * which we are interested.
4761 * It would be a bug for them to not do this. The code:
4763 * mark_inode_dirty(inode)
4765 * inode->i_size = expr;
4767 * is in error because write_inode() could occur while `stuff()' is running,
4768 * and the new i_size will be lost. Plus the inode will no longer be on the
4769 * superblock's dirty inode list.
4771 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4775 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4778 if (EXT4_SB(inode->i_sb)->s_journal) {
4779 if (ext4_journal_current_handle()) {
4780 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4786 * No need to force transaction in WB_SYNC_NONE mode. Also
4787 * ext4_sync_fs() will force the commit after everything is
4790 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4793 err = ext4_force_commit(inode->i_sb);
4795 struct ext4_iloc iloc;
4797 err = __ext4_get_inode_loc(inode, &iloc, 0);
4801 * sync(2) will flush the whole buffer cache. No need to do
4802 * it here separately for each inode.
4804 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4805 sync_dirty_buffer(iloc.bh);
4806 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4807 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4808 "IO error syncing inode");
4817 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4818 * buffers that are attached to a page stradding i_size and are undergoing
4819 * commit. In that case we have to wait for commit to finish and try again.
4821 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4825 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4826 tid_t commit_tid = 0;
4829 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4831 * All buffers in the last page remain valid? Then there's nothing to
4832 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4835 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4838 page = find_lock_page(inode->i_mapping,
4839 inode->i_size >> PAGE_CACHE_SHIFT);
4842 ret = __ext4_journalled_invalidatepage(page, offset,
4843 PAGE_CACHE_SIZE - offset);
4845 page_cache_release(page);
4849 read_lock(&journal->j_state_lock);
4850 if (journal->j_committing_transaction)
4851 commit_tid = journal->j_committing_transaction->t_tid;
4852 read_unlock(&journal->j_state_lock);
4854 jbd2_log_wait_commit(journal, commit_tid);
4861 * Called from notify_change.
4863 * We want to trap VFS attempts to truncate the file as soon as
4864 * possible. In particular, we want to make sure that when the VFS
4865 * shrinks i_size, we put the inode on the orphan list and modify
4866 * i_disksize immediately, so that during the subsequent flushing of
4867 * dirty pages and freeing of disk blocks, we can guarantee that any
4868 * commit will leave the blocks being flushed in an unused state on
4869 * disk. (On recovery, the inode will get truncated and the blocks will
4870 * be freed, so we have a strong guarantee that no future commit will
4871 * leave these blocks visible to the user.)
4873 * Another thing we have to assure is that if we are in ordered mode
4874 * and inode is still attached to the committing transaction, we must
4875 * we start writeout of all the dirty pages which are being truncated.
4876 * This way we are sure that all the data written in the previous
4877 * transaction are already on disk (truncate waits for pages under
4880 * Called with inode->i_mutex down.
4882 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4884 struct inode *inode = d_inode(dentry);
4887 const unsigned int ia_valid = attr->ia_valid;
4889 error = inode_change_ok(inode, attr);
4893 if (is_quota_modification(inode, attr)) {
4894 error = dquot_initialize(inode);
4898 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4899 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4902 /* (user+group)*(old+new) structure, inode write (sb,
4903 * inode block, ? - but truncate inode update has it) */
4904 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4905 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4906 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4907 if (IS_ERR(handle)) {
4908 error = PTR_ERR(handle);
4911 error = dquot_transfer(inode, attr);
4913 ext4_journal_stop(handle);
4916 /* Update corresponding info in inode so that everything is in
4917 * one transaction */
4918 if (attr->ia_valid & ATTR_UID)
4919 inode->i_uid = attr->ia_uid;
4920 if (attr->ia_valid & ATTR_GID)
4921 inode->i_gid = attr->ia_gid;
4922 error = ext4_mark_inode_dirty(handle, inode);
4923 ext4_journal_stop(handle);
4926 if (attr->ia_valid & ATTR_SIZE) {
4928 loff_t oldsize = inode->i_size;
4929 int shrink = (attr->ia_size <= inode->i_size);
4931 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4932 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4934 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4937 if (!S_ISREG(inode->i_mode))
4940 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4941 inode_inc_iversion(inode);
4943 if (ext4_should_order_data(inode) &&
4944 (attr->ia_size < inode->i_size)) {
4945 error = ext4_begin_ordered_truncate(inode,
4950 if (attr->ia_size != inode->i_size) {
4951 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4952 if (IS_ERR(handle)) {
4953 error = PTR_ERR(handle);
4956 if (ext4_handle_valid(handle) && shrink) {
4957 error = ext4_orphan_add(handle, inode);
4961 * Update c/mtime on truncate up, ext4_truncate() will
4962 * update c/mtime in shrink case below
4965 inode->i_mtime = ext4_current_time(inode);
4966 inode->i_ctime = inode->i_mtime;
4968 down_write(&EXT4_I(inode)->i_data_sem);
4969 EXT4_I(inode)->i_disksize = attr->ia_size;
4970 rc = ext4_mark_inode_dirty(handle, inode);
4974 * We have to update i_size under i_data_sem together
4975 * with i_disksize to avoid races with writeback code
4976 * running ext4_wb_update_i_disksize().
4979 i_size_write(inode, attr->ia_size);
4980 up_write(&EXT4_I(inode)->i_data_sem);
4981 ext4_journal_stop(handle);
4984 ext4_orphan_del(NULL, inode);
4989 pagecache_isize_extended(inode, oldsize, inode->i_size);
4992 * Blocks are going to be removed from the inode. Wait
4993 * for dio in flight. Temporarily disable
4994 * dioread_nolock to prevent livelock.
4997 if (!ext4_should_journal_data(inode)) {
4998 ext4_inode_block_unlocked_dio(inode);
4999 inode_dio_wait(inode);
5000 ext4_inode_resume_unlocked_dio(inode);
5002 ext4_wait_for_tail_page_commit(inode);
5004 down_write(&EXT4_I(inode)->i_mmap_sem);
5006 * Truncate pagecache after we've waited for commit
5007 * in data=journal mode to make pages freeable.
5009 truncate_pagecache(inode, inode->i_size);
5011 ext4_truncate(inode);
5012 up_write(&EXT4_I(inode)->i_mmap_sem);
5016 setattr_copy(inode, attr);
5017 mark_inode_dirty(inode);
5021 * If the call to ext4_truncate failed to get a transaction handle at
5022 * all, we need to clean up the in-core orphan list manually.
5024 if (orphan && inode->i_nlink)
5025 ext4_orphan_del(NULL, inode);
5027 if (!rc && (ia_valid & ATTR_MODE))
5028 rc = posix_acl_chmod(inode, inode->i_mode);
5031 ext4_std_error(inode->i_sb, error);
5037 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5040 struct inode *inode;
5041 unsigned long long delalloc_blocks;
5043 inode = d_inode(dentry);
5044 generic_fillattr(inode, stat);
5047 * If there is inline data in the inode, the inode will normally not
5048 * have data blocks allocated (it may have an external xattr block).
5049 * Report at least one sector for such files, so tools like tar, rsync,
5050 * others doen't incorrectly think the file is completely sparse.
5052 if (unlikely(ext4_has_inline_data(inode)))
5053 stat->blocks += (stat->size + 511) >> 9;
5056 * We can't update i_blocks if the block allocation is delayed
5057 * otherwise in the case of system crash before the real block
5058 * allocation is done, we will have i_blocks inconsistent with
5059 * on-disk file blocks.
5060 * We always keep i_blocks updated together with real
5061 * allocation. But to not confuse with user, stat
5062 * will return the blocks that include the delayed allocation
5063 * blocks for this file.
5065 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5066 EXT4_I(inode)->i_reserved_data_blocks);
5067 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5071 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5074 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5075 return ext4_ind_trans_blocks(inode, lblocks);
5076 return ext4_ext_index_trans_blocks(inode, pextents);
5080 * Account for index blocks, block groups bitmaps and block group
5081 * descriptor blocks if modify datablocks and index blocks
5082 * worse case, the indexs blocks spread over different block groups
5084 * If datablocks are discontiguous, they are possible to spread over
5085 * different block groups too. If they are contiguous, with flexbg,
5086 * they could still across block group boundary.
5088 * Also account for superblock, inode, quota and xattr blocks
5090 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5093 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5099 * How many index blocks need to touch to map @lblocks logical blocks
5100 * to @pextents physical extents?
5102 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5107 * Now let's see how many group bitmaps and group descriptors need
5110 groups = idxblocks + pextents;
5112 if (groups > ngroups)
5114 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5115 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5117 /* bitmaps and block group descriptor blocks */
5118 ret += groups + gdpblocks;
5120 /* Blocks for super block, inode, quota and xattr blocks */
5121 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5127 * Calculate the total number of credits to reserve to fit
5128 * the modification of a single pages into a single transaction,
5129 * which may include multiple chunks of block allocations.
5131 * This could be called via ext4_write_begin()
5133 * We need to consider the worse case, when
5134 * one new block per extent.
5136 int ext4_writepage_trans_blocks(struct inode *inode)
5138 int bpp = ext4_journal_blocks_per_page(inode);
5141 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5143 /* Account for data blocks for journalled mode */
5144 if (ext4_should_journal_data(inode))
5150 * Calculate the journal credits for a chunk of data modification.
5152 * This is called from DIO, fallocate or whoever calling
5153 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5155 * journal buffers for data blocks are not included here, as DIO
5156 * and fallocate do no need to journal data buffers.
5158 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5160 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5164 * The caller must have previously called ext4_reserve_inode_write().
5165 * Give this, we know that the caller already has write access to iloc->bh.
5167 int ext4_mark_iloc_dirty(handle_t *handle,
5168 struct inode *inode, struct ext4_iloc *iloc)
5172 if (IS_I_VERSION(inode))
5173 inode_inc_iversion(inode);
5175 /* the do_update_inode consumes one bh->b_count */
5178 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5179 err = ext4_do_update_inode(handle, inode, iloc);
5185 * On success, We end up with an outstanding reference count against
5186 * iloc->bh. This _must_ be cleaned up later.
5190 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5191 struct ext4_iloc *iloc)
5195 err = ext4_get_inode_loc(inode, iloc);
5197 BUFFER_TRACE(iloc->bh, "get_write_access");
5198 err = ext4_journal_get_write_access(handle, iloc->bh);
5204 ext4_std_error(inode->i_sb, err);
5209 * Expand an inode by new_extra_isize bytes.
5210 * Returns 0 on success or negative error number on failure.
5212 static int ext4_expand_extra_isize(struct inode *inode,
5213 unsigned int new_extra_isize,
5214 struct ext4_iloc iloc,
5217 struct ext4_inode *raw_inode;
5218 struct ext4_xattr_ibody_header *header;
5220 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5223 raw_inode = ext4_raw_inode(&iloc);
5225 header = IHDR(inode, raw_inode);
5227 /* No extended attributes present */
5228 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5229 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5230 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5232 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5236 /* try to expand with EAs present */
5237 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5242 * What we do here is to mark the in-core inode as clean with respect to inode
5243 * dirtiness (it may still be data-dirty).
5244 * This means that the in-core inode may be reaped by prune_icache
5245 * without having to perform any I/O. This is a very good thing,
5246 * because *any* task may call prune_icache - even ones which
5247 * have a transaction open against a different journal.
5249 * Is this cheating? Not really. Sure, we haven't written the
5250 * inode out, but prune_icache isn't a user-visible syncing function.
5251 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5252 * we start and wait on commits.
5254 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5256 struct ext4_iloc iloc;
5257 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5258 static unsigned int mnt_count;
5262 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5263 err = ext4_reserve_inode_write(handle, inode, &iloc);
5264 if (ext4_handle_valid(handle) &&
5265 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5266 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5268 * We need extra buffer credits since we may write into EA block
5269 * with this same handle. If journal_extend fails, then it will
5270 * only result in a minor loss of functionality for that inode.
5271 * If this is felt to be critical, then e2fsck should be run to
5272 * force a large enough s_min_extra_isize.
5274 if ((jbd2_journal_extend(handle,
5275 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5276 ret = ext4_expand_extra_isize(inode,
5277 sbi->s_want_extra_isize,
5280 ext4_set_inode_state(inode,
5281 EXT4_STATE_NO_EXPAND);
5283 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5284 ext4_warning(inode->i_sb,
5285 "Unable to expand inode %lu. Delete"
5286 " some EAs or run e2fsck.",
5289 le16_to_cpu(sbi->s_es->s_mnt_count);
5295 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5300 * ext4_dirty_inode() is called from __mark_inode_dirty()
5302 * We're really interested in the case where a file is being extended.
5303 * i_size has been changed by generic_commit_write() and we thus need
5304 * to include the updated inode in the current transaction.
5306 * Also, dquot_alloc_block() will always dirty the inode when blocks
5307 * are allocated to the file.
5309 * If the inode is marked synchronous, we don't honour that here - doing
5310 * so would cause a commit on atime updates, which we don't bother doing.
5311 * We handle synchronous inodes at the highest possible level.
5313 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5314 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5315 * to copy into the on-disk inode structure are the timestamp files.
5317 void ext4_dirty_inode(struct inode *inode, int flags)
5321 if (flags == I_DIRTY_TIME)
5323 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5327 ext4_mark_inode_dirty(handle, inode);
5329 ext4_journal_stop(handle);
5336 * Bind an inode's backing buffer_head into this transaction, to prevent
5337 * it from being flushed to disk early. Unlike
5338 * ext4_reserve_inode_write, this leaves behind no bh reference and
5339 * returns no iloc structure, so the caller needs to repeat the iloc
5340 * lookup to mark the inode dirty later.
5342 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5344 struct ext4_iloc iloc;
5348 err = ext4_get_inode_loc(inode, &iloc);
5350 BUFFER_TRACE(iloc.bh, "get_write_access");
5351 err = jbd2_journal_get_write_access(handle, iloc.bh);
5353 err = ext4_handle_dirty_metadata(handle,
5359 ext4_std_error(inode->i_sb, err);
5364 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5371 * We have to be very careful here: changing a data block's
5372 * journaling status dynamically is dangerous. If we write a
5373 * data block to the journal, change the status and then delete
5374 * that block, we risk forgetting to revoke the old log record
5375 * from the journal and so a subsequent replay can corrupt data.
5376 * So, first we make sure that the journal is empty and that
5377 * nobody is changing anything.
5380 journal = EXT4_JOURNAL(inode);
5383 if (is_journal_aborted(journal))
5385 /* We have to allocate physical blocks for delalloc blocks
5386 * before flushing journal. otherwise delalloc blocks can not
5387 * be allocated any more. even more truncate on delalloc blocks
5388 * could trigger BUG by flushing delalloc blocks in journal.
5389 * There is no delalloc block in non-journal data mode.
5391 if (val && test_opt(inode->i_sb, DELALLOC)) {
5392 err = ext4_alloc_da_blocks(inode);
5397 /* Wait for all existing dio workers */
5398 ext4_inode_block_unlocked_dio(inode);
5399 inode_dio_wait(inode);
5401 jbd2_journal_lock_updates(journal);
5404 * OK, there are no updates running now, and all cached data is
5405 * synced to disk. We are now in a completely consistent state
5406 * which doesn't have anything in the journal, and we know that
5407 * no filesystem updates are running, so it is safe to modify
5408 * the inode's in-core data-journaling state flag now.
5412 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5414 err = jbd2_journal_flush(journal);
5416 jbd2_journal_unlock_updates(journal);
5417 ext4_inode_resume_unlocked_dio(inode);
5420 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5422 ext4_set_aops(inode);
5424 jbd2_journal_unlock_updates(journal);
5425 ext4_inode_resume_unlocked_dio(inode);
5427 /* Finally we can mark the inode as dirty. */
5429 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5431 return PTR_ERR(handle);
5433 err = ext4_mark_inode_dirty(handle, inode);
5434 ext4_handle_sync(handle);
5435 ext4_journal_stop(handle);
5436 ext4_std_error(inode->i_sb, err);
5441 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5443 return !buffer_mapped(bh);
5446 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5448 struct page *page = vmf->page;
5452 struct file *file = vma->vm_file;
5453 struct inode *inode = file_inode(file);
5454 struct address_space *mapping = inode->i_mapping;
5456 get_block_t *get_block;
5459 sb_start_pagefault(inode->i_sb);
5460 file_update_time(vma->vm_file);
5462 down_read(&EXT4_I(inode)->i_mmap_sem);
5463 /* Delalloc case is easy... */
5464 if (test_opt(inode->i_sb, DELALLOC) &&
5465 !ext4_should_journal_data(inode) &&
5466 !ext4_nonda_switch(inode->i_sb)) {
5468 ret = block_page_mkwrite(vma, vmf,
5469 ext4_da_get_block_prep);
5470 } while (ret == -ENOSPC &&
5471 ext4_should_retry_alloc(inode->i_sb, &retries));
5476 size = i_size_read(inode);
5477 /* Page got truncated from under us? */
5478 if (page->mapping != mapping || page_offset(page) > size) {
5480 ret = VM_FAULT_NOPAGE;
5484 if (page->index == size >> PAGE_CACHE_SHIFT)
5485 len = size & ~PAGE_CACHE_MASK;
5487 len = PAGE_CACHE_SIZE;
5489 * Return if we have all the buffers mapped. This avoids the need to do
5490 * journal_start/journal_stop which can block and take a long time
5492 if (page_has_buffers(page)) {
5493 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5495 ext4_bh_unmapped)) {
5496 /* Wait so that we don't change page under IO */
5497 wait_for_stable_page(page);
5498 ret = VM_FAULT_LOCKED;
5503 /* OK, we need to fill the hole... */
5504 if (ext4_should_dioread_nolock(inode))
5505 get_block = ext4_get_block_write;
5507 get_block = ext4_get_block;
5509 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5510 ext4_writepage_trans_blocks(inode));
5511 if (IS_ERR(handle)) {
5512 ret = VM_FAULT_SIGBUS;
5515 ret = block_page_mkwrite(vma, vmf, get_block);
5516 if (!ret && ext4_should_journal_data(inode)) {
5517 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5518 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5520 ret = VM_FAULT_SIGBUS;
5521 ext4_journal_stop(handle);
5524 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5526 ext4_journal_stop(handle);
5527 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5530 ret = block_page_mkwrite_return(ret);
5532 up_read(&EXT4_I(inode)->i_mmap_sem);
5533 sb_end_pagefault(inode->i_sb);
5537 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5539 struct inode *inode = file_inode(vma->vm_file);
5542 down_read(&EXT4_I(inode)->i_mmap_sem);
5543 err = filemap_fault(vma, vmf);
5544 up_read(&EXT4_I(inode)->i_mmap_sem);