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);
222 if (is_bad_inode(inode))
224 dquot_initialize(inode);
226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
228 truncate_inode_pages_final(&inode->i_data);
231 * Protect us against freezing - iput() caller didn't have to have any
232 * protection against it
234 sb_start_intwrite(inode->i_sb);
235 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
236 ext4_blocks_for_truncate(inode)+3);
237 if (IS_ERR(handle)) {
238 ext4_std_error(inode->i_sb, PTR_ERR(handle));
240 * If we're going to skip the normal cleanup, we still need to
241 * make sure that the in-core orphan linked list is properly
244 ext4_orphan_del(NULL, inode);
245 sb_end_intwrite(inode->i_sb);
250 ext4_handle_sync(handle);
252 err = ext4_mark_inode_dirty(handle, inode);
254 ext4_warning(inode->i_sb,
255 "couldn't mark inode dirty (err %d)", err);
259 ext4_truncate(inode);
262 * ext4_ext_truncate() doesn't reserve any slop when it
263 * restarts journal transactions; therefore there may not be
264 * enough credits left in the handle to remove the inode from
265 * the orphan list and set the dtime field.
267 if (!ext4_handle_has_enough_credits(handle, 3)) {
268 err = ext4_journal_extend(handle, 3);
270 err = ext4_journal_restart(handle, 3);
272 ext4_warning(inode->i_sb,
273 "couldn't extend journal (err %d)", err);
275 ext4_journal_stop(handle);
276 ext4_orphan_del(NULL, inode);
277 sb_end_intwrite(inode->i_sb);
283 * Kill off the orphan record which ext4_truncate created.
284 * AKPM: I think this can be inside the above `if'.
285 * Note that ext4_orphan_del() has to be able to cope with the
286 * deletion of a non-existent orphan - this is because we don't
287 * know if ext4_truncate() actually created an orphan record.
288 * (Well, we could do this if we need to, but heck - it works)
290 ext4_orphan_del(handle, inode);
291 EXT4_I(inode)->i_dtime = get_seconds();
294 * One subtle ordering requirement: if anything has gone wrong
295 * (transaction abort, IO errors, whatever), then we can still
296 * do these next steps (the fs will already have been marked as
297 * having errors), but we can't free the inode if the mark_dirty
300 if (ext4_mark_inode_dirty(handle, inode))
301 /* If that failed, just do the required in-core inode clear. */
302 ext4_clear_inode(inode);
304 ext4_free_inode(handle, inode);
305 ext4_journal_stop(handle);
306 sb_end_intwrite(inode->i_sb);
309 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
313 qsize_t *ext4_get_reserved_space(struct inode *inode)
315 return &EXT4_I(inode)->i_reserved_quota;
320 * Called with i_data_sem down, which is important since we can call
321 * ext4_discard_preallocations() from here.
323 void ext4_da_update_reserve_space(struct inode *inode,
324 int used, int quota_claim)
326 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
327 struct ext4_inode_info *ei = EXT4_I(inode);
329 spin_lock(&ei->i_block_reservation_lock);
330 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
331 if (unlikely(used > ei->i_reserved_data_blocks)) {
332 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
333 "with only %d reserved data blocks",
334 __func__, inode->i_ino, used,
335 ei->i_reserved_data_blocks);
337 used = ei->i_reserved_data_blocks;
340 /* Update per-inode reservations */
341 ei->i_reserved_data_blocks -= used;
342 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
344 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
346 /* Update quota subsystem for data blocks */
348 dquot_claim_block(inode, EXT4_C2B(sbi, used));
351 * We did fallocate with an offset that is already delayed
352 * allocated. So on delayed allocated writeback we should
353 * not re-claim the quota for fallocated blocks.
355 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
359 * If we have done all the pending block allocations and if
360 * there aren't any writers on the inode, we can discard the
361 * inode's preallocations.
363 if ((ei->i_reserved_data_blocks == 0) &&
364 (atomic_read(&inode->i_writecount) == 0))
365 ext4_discard_preallocations(inode);
368 static int __check_block_validity(struct inode *inode, const char *func,
370 struct ext4_map_blocks *map)
372 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
374 ext4_error_inode(inode, func, line, map->m_pblk,
375 "lblock %lu mapped to illegal pblock "
376 "(length %d)", (unsigned long) map->m_lblk,
378 return -EFSCORRUPTED;
383 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
388 if (ext4_encrypted_inode(inode))
389 return ext4_encrypted_zeroout(inode, lblk, pblk, len);
391 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
398 #define check_block_validity(inode, map) \
399 __check_block_validity((inode), __func__, __LINE__, (map))
401 #ifdef ES_AGGRESSIVE_TEST
402 static void ext4_map_blocks_es_recheck(handle_t *handle,
404 struct ext4_map_blocks *es_map,
405 struct ext4_map_blocks *map,
412 * There is a race window that the result is not the same.
413 * e.g. xfstests #223 when dioread_nolock enables. The reason
414 * is that we lookup a block mapping in extent status tree with
415 * out taking i_data_sem. So at the time the unwritten extent
416 * could be converted.
418 down_read(&EXT4_I(inode)->i_data_sem);
419 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
420 retval = ext4_ext_map_blocks(handle, inode, map, flags &
421 EXT4_GET_BLOCKS_KEEP_SIZE);
423 retval = ext4_ind_map_blocks(handle, inode, map, flags &
424 EXT4_GET_BLOCKS_KEEP_SIZE);
426 up_read((&EXT4_I(inode)->i_data_sem));
429 * We don't check m_len because extent will be collpased in status
430 * tree. So the m_len might not equal.
432 if (es_map->m_lblk != map->m_lblk ||
433 es_map->m_flags != map->m_flags ||
434 es_map->m_pblk != map->m_pblk) {
435 printk("ES cache assertion failed for inode: %lu "
436 "es_cached ex [%d/%d/%llu/%x] != "
437 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
438 inode->i_ino, es_map->m_lblk, es_map->m_len,
439 es_map->m_pblk, es_map->m_flags, map->m_lblk,
440 map->m_len, map->m_pblk, map->m_flags,
444 #endif /* ES_AGGRESSIVE_TEST */
447 * The ext4_map_blocks() function tries to look up the requested blocks,
448 * and returns if the blocks are already mapped.
450 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
451 * and store the allocated blocks in the result buffer head and mark it
454 * If file type is extents based, it will call ext4_ext_map_blocks(),
455 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
458 * On success, it returns the number of blocks being mapped or allocated. if
459 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
460 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
462 * It returns 0 if plain look up failed (blocks have not been allocated), in
463 * that case, @map is returned as unmapped but we still do fill map->m_len to
464 * indicate the length of a hole starting at map->m_lblk.
466 * It returns the error in case of allocation failure.
468 int ext4_map_blocks(handle_t *handle, struct inode *inode,
469 struct ext4_map_blocks *map, int flags)
471 struct extent_status es;
474 #ifdef ES_AGGRESSIVE_TEST
475 struct ext4_map_blocks orig_map;
477 memcpy(&orig_map, map, sizeof(*map));
481 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
482 "logical block %lu\n", inode->i_ino, flags, map->m_len,
483 (unsigned long) map->m_lblk);
486 * ext4_map_blocks returns an int, and m_len is an unsigned int
488 if (unlikely(map->m_len > INT_MAX))
489 map->m_len = INT_MAX;
491 /* We can handle the block number less than EXT_MAX_BLOCKS */
492 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
493 return -EFSCORRUPTED;
495 /* Lookup extent status tree firstly */
496 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
497 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
498 map->m_pblk = ext4_es_pblock(&es) +
499 map->m_lblk - es.es_lblk;
500 map->m_flags |= ext4_es_is_written(&es) ?
501 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
502 retval = es.es_len - (map->m_lblk - es.es_lblk);
503 if (retval > map->m_len)
506 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
508 retval = es.es_len - (map->m_lblk - es.es_lblk);
509 if (retval > map->m_len)
516 #ifdef ES_AGGRESSIVE_TEST
517 ext4_map_blocks_es_recheck(handle, inode, map,
524 * Try to see if we can get the block without requesting a new
527 down_read(&EXT4_I(inode)->i_data_sem);
528 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
529 retval = ext4_ext_map_blocks(handle, inode, map, flags &
530 EXT4_GET_BLOCKS_KEEP_SIZE);
532 retval = ext4_ind_map_blocks(handle, inode, map, flags &
533 EXT4_GET_BLOCKS_KEEP_SIZE);
538 if (unlikely(retval != map->m_len)) {
539 ext4_warning(inode->i_sb,
540 "ES len assertion failed for inode "
541 "%lu: retval %d != map->m_len %d",
542 inode->i_ino, retval, map->m_len);
546 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
547 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
548 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
549 !(status & EXTENT_STATUS_WRITTEN) &&
550 ext4_find_delalloc_range(inode, map->m_lblk,
551 map->m_lblk + map->m_len - 1))
552 status |= EXTENT_STATUS_DELAYED;
553 ret = ext4_es_insert_extent(inode, map->m_lblk,
554 map->m_len, map->m_pblk, status);
558 up_read((&EXT4_I(inode)->i_data_sem));
561 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
562 ret = check_block_validity(inode, map);
567 /* If it is only a block(s) look up */
568 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
572 * Returns if the blocks have already allocated
574 * Note that if blocks have been preallocated
575 * ext4_ext_get_block() returns the create = 0
576 * with buffer head unmapped.
578 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
580 * If we need to convert extent to unwritten
581 * we continue and do the actual work in
582 * ext4_ext_map_blocks()
584 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
588 * Here we clear m_flags because after allocating an new extent,
589 * it will be set again.
591 map->m_flags &= ~EXT4_MAP_FLAGS;
594 * New blocks allocate and/or writing to unwritten extent
595 * will possibly result in updating i_data, so we take
596 * the write lock of i_data_sem, and call get_block()
597 * with create == 1 flag.
599 down_write(&EXT4_I(inode)->i_data_sem);
602 * We need to check for EXT4 here because migrate
603 * could have changed the inode type in between
605 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
606 retval = ext4_ext_map_blocks(handle, inode, map, flags);
608 retval = ext4_ind_map_blocks(handle, inode, map, flags);
610 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
612 * We allocated new blocks which will result in
613 * i_data's format changing. Force the migrate
614 * to fail by clearing migrate flags
616 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
620 * Update reserved blocks/metadata blocks after successful
621 * block allocation which had been deferred till now. We don't
622 * support fallocate for non extent files. So we can update
623 * reserve space here.
626 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
627 ext4_da_update_reserve_space(inode, retval, 1);
633 if (unlikely(retval != map->m_len)) {
634 ext4_warning(inode->i_sb,
635 "ES len assertion failed for inode "
636 "%lu: retval %d != map->m_len %d",
637 inode->i_ino, retval, map->m_len);
642 * We have to zeroout blocks before inserting them into extent
643 * status tree. Otherwise someone could look them up there and
644 * use them before they are really zeroed.
646 if (flags & EXT4_GET_BLOCKS_ZERO &&
647 map->m_flags & EXT4_MAP_MAPPED &&
648 map->m_flags & EXT4_MAP_NEW) {
649 ret = ext4_issue_zeroout(inode, map->m_lblk,
650 map->m_pblk, map->m_len);
658 * If the extent has been zeroed out, we don't need to update
659 * extent status tree.
661 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
662 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
663 if (ext4_es_is_written(&es))
666 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
667 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
668 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
669 !(status & EXTENT_STATUS_WRITTEN) &&
670 ext4_find_delalloc_range(inode, map->m_lblk,
671 map->m_lblk + map->m_len - 1))
672 status |= EXTENT_STATUS_DELAYED;
673 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
674 map->m_pblk, status);
682 up_write((&EXT4_I(inode)->i_data_sem));
683 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
684 ret = check_block_validity(inode, map);
689 * Inodes with freshly allocated blocks where contents will be
690 * visible after transaction commit must be on transaction's
693 if (map->m_flags & EXT4_MAP_NEW &&
694 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
695 !(flags & EXT4_GET_BLOCKS_ZERO) &&
696 !IS_NOQUOTA(inode) &&
697 ext4_should_order_data(inode)) {
698 ret = ext4_jbd2_file_inode(handle, inode);
707 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
708 * we have to be careful as someone else may be manipulating b_state as well.
710 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
712 unsigned long old_state;
713 unsigned long new_state;
715 flags &= EXT4_MAP_FLAGS;
717 /* Dummy buffer_head? Set non-atomically. */
719 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
723 * Someone else may be modifying b_state. Be careful! This is ugly but
724 * once we get rid of using bh as a container for mapping information
725 * to pass to / from get_block functions, this can go away.
728 old_state = READ_ONCE(bh->b_state);
729 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
731 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
734 static int _ext4_get_block(struct inode *inode, sector_t iblock,
735 struct buffer_head *bh, int flags)
737 struct ext4_map_blocks map;
740 if (ext4_has_inline_data(inode))
744 map.m_len = bh->b_size >> inode->i_blkbits;
746 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
749 map_bh(bh, inode->i_sb, map.m_pblk);
750 ext4_update_bh_state(bh, map.m_flags);
751 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
757 int ext4_get_block(struct inode *inode, sector_t iblock,
758 struct buffer_head *bh, int create)
760 return _ext4_get_block(inode, iblock, bh,
761 create ? EXT4_GET_BLOCKS_CREATE : 0);
765 * Get block function used when preparing for buffered write if we require
766 * creating an unwritten extent if blocks haven't been allocated. The extent
767 * will be converted to written after the IO is complete.
769 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
770 struct buffer_head *bh_result, int create)
772 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
773 inode->i_ino, create);
774 return _ext4_get_block(inode, iblock, bh_result,
775 EXT4_GET_BLOCKS_IO_CREATE_EXT);
778 /* Maximum number of blocks we map for direct IO at once. */
779 #define DIO_MAX_BLOCKS 4096
782 * Get blocks function for the cases that need to start a transaction -
783 * generally difference cases of direct IO and DAX IO. It also handles retries
786 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
787 struct buffer_head *bh_result, int flags)
794 /* Trim mapping request to maximum we can map at once for DIO */
795 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
796 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
797 dio_credits = ext4_chunk_trans_blocks(inode,
798 bh_result->b_size >> inode->i_blkbits);
800 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
802 return PTR_ERR(handle);
804 ret = _ext4_get_block(inode, iblock, bh_result, flags);
805 ext4_journal_stop(handle);
807 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
812 /* Get block function for DIO reads and writes to inodes without extents */
813 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
814 struct buffer_head *bh, int create)
816 /* We don't expect handle for direct IO */
817 WARN_ON_ONCE(ext4_journal_current_handle());
820 return _ext4_get_block(inode, iblock, bh, 0);
821 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
825 * Get block function for AIO DIO writes when we create unwritten extent if
826 * blocks are not allocated yet. The extent will be converted to written
827 * after IO is complete.
829 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
830 sector_t iblock, struct buffer_head *bh_result, int create)
834 /* We don't expect handle for direct IO */
835 WARN_ON_ONCE(ext4_journal_current_handle());
837 ret = ext4_get_block_trans(inode, iblock, bh_result,
838 EXT4_GET_BLOCKS_IO_CREATE_EXT);
841 * When doing DIO using unwritten extents, we need io_end to convert
842 * unwritten extents to written on IO completion. We allocate io_end
843 * once we spot unwritten extent and store it in b_private. Generic
844 * DIO code keeps b_private set and furthermore passes the value to
845 * our completion callback in 'private' argument.
847 if (!ret && buffer_unwritten(bh_result)) {
848 if (!bh_result->b_private) {
849 ext4_io_end_t *io_end;
851 io_end = ext4_init_io_end(inode, GFP_KERNEL);
854 bh_result->b_private = io_end;
855 ext4_set_io_unwritten_flag(inode, io_end);
857 set_buffer_defer_completion(bh_result);
864 * Get block function for non-AIO DIO writes when we create unwritten extent if
865 * blocks are not allocated yet. The extent will be converted to written
866 * after IO is complete from ext4_ext_direct_IO() function.
868 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
869 sector_t iblock, struct buffer_head *bh_result, int create)
873 /* We don't expect handle for direct IO */
874 WARN_ON_ONCE(ext4_journal_current_handle());
876 ret = ext4_get_block_trans(inode, iblock, bh_result,
877 EXT4_GET_BLOCKS_IO_CREATE_EXT);
880 * Mark inode as having pending DIO writes to unwritten extents.
881 * ext4_ext_direct_IO() checks this flag and converts extents to
884 if (!ret && buffer_unwritten(bh_result))
885 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
890 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
891 struct buffer_head *bh_result, int create)
895 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
896 inode->i_ino, create);
897 /* We don't expect handle for direct IO */
898 WARN_ON_ONCE(ext4_journal_current_handle());
900 ret = _ext4_get_block(inode, iblock, bh_result, 0);
902 * Blocks should have been preallocated! ext4_file_write_iter() checks
905 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
912 * `handle' can be NULL if create is zero
914 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
915 ext4_lblk_t block, int map_flags)
917 struct ext4_map_blocks map;
918 struct buffer_head *bh;
919 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
922 J_ASSERT(handle != NULL || create == 0);
926 err = ext4_map_blocks(handle, inode, &map, map_flags);
929 return create ? ERR_PTR(-ENOSPC) : NULL;
933 bh = sb_getblk(inode->i_sb, map.m_pblk);
935 return ERR_PTR(-ENOMEM);
936 if (map.m_flags & EXT4_MAP_NEW) {
937 J_ASSERT(create != 0);
938 J_ASSERT(handle != NULL);
941 * Now that we do not always journal data, we should
942 * keep in mind whether this should always journal the
943 * new buffer as metadata. For now, regular file
944 * writes use ext4_get_block instead, so it's not a
948 BUFFER_TRACE(bh, "call get_create_access");
949 err = ext4_journal_get_create_access(handle, bh);
954 if (!buffer_uptodate(bh)) {
955 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
956 set_buffer_uptodate(bh);
959 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
960 err = ext4_handle_dirty_metadata(handle, inode, bh);
964 BUFFER_TRACE(bh, "not a new buffer");
971 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
972 ext4_lblk_t block, int map_flags)
974 struct buffer_head *bh;
976 bh = ext4_getblk(handle, inode, block, map_flags);
979 if (!bh || buffer_uptodate(bh))
981 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
983 if (buffer_uptodate(bh))
986 return ERR_PTR(-EIO);
989 int ext4_walk_page_buffers(handle_t *handle,
990 struct buffer_head *head,
994 int (*fn)(handle_t *handle,
995 struct buffer_head *bh))
997 struct buffer_head *bh;
998 unsigned block_start, block_end;
999 unsigned blocksize = head->b_size;
1001 struct buffer_head *next;
1003 for (bh = head, block_start = 0;
1004 ret == 0 && (bh != head || !block_start);
1005 block_start = block_end, bh = next) {
1006 next = bh->b_this_page;
1007 block_end = block_start + blocksize;
1008 if (block_end <= from || block_start >= to) {
1009 if (partial && !buffer_uptodate(bh))
1013 err = (*fn)(handle, bh);
1021 * To preserve ordering, it is essential that the hole instantiation and
1022 * the data write be encapsulated in a single transaction. We cannot
1023 * close off a transaction and start a new one between the ext4_get_block()
1024 * and the commit_write(). So doing the jbd2_journal_start at the start of
1025 * prepare_write() is the right place.
1027 * Also, this function can nest inside ext4_writepage(). In that case, we
1028 * *know* that ext4_writepage() has generated enough buffer credits to do the
1029 * whole page. So we won't block on the journal in that case, which is good,
1030 * because the caller may be PF_MEMALLOC.
1032 * By accident, ext4 can be reentered when a transaction is open via
1033 * quota file writes. If we were to commit the transaction while thus
1034 * reentered, there can be a deadlock - we would be holding a quota
1035 * lock, and the commit would never complete if another thread had a
1036 * transaction open and was blocking on the quota lock - a ranking
1039 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1040 * will _not_ run commit under these circumstances because handle->h_ref
1041 * is elevated. We'll still have enough credits for the tiny quotafile
1044 int do_journal_get_write_access(handle_t *handle,
1045 struct buffer_head *bh)
1047 int dirty = buffer_dirty(bh);
1050 if (!buffer_mapped(bh) || buffer_freed(bh))
1053 * __block_write_begin() could have dirtied some buffers. Clean
1054 * the dirty bit as jbd2_journal_get_write_access() could complain
1055 * otherwise about fs integrity issues. Setting of the dirty bit
1056 * by __block_write_begin() isn't a real problem here as we clear
1057 * the bit before releasing a page lock and thus writeback cannot
1058 * ever write the buffer.
1061 clear_buffer_dirty(bh);
1062 BUFFER_TRACE(bh, "get write access");
1063 ret = ext4_journal_get_write_access(handle, bh);
1065 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1069 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1070 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1071 get_block_t *get_block)
1073 unsigned from = pos & (PAGE_SIZE - 1);
1074 unsigned to = from + len;
1075 struct inode *inode = page->mapping->host;
1076 unsigned block_start, block_end;
1079 unsigned blocksize = inode->i_sb->s_blocksize;
1081 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1082 bool decrypt = false;
1084 BUG_ON(!PageLocked(page));
1085 BUG_ON(from > PAGE_SIZE);
1086 BUG_ON(to > PAGE_SIZE);
1089 if (!page_has_buffers(page))
1090 create_empty_buffers(page, blocksize, 0);
1091 head = page_buffers(page);
1092 bbits = ilog2(blocksize);
1093 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1095 for (bh = head, block_start = 0; bh != head || !block_start;
1096 block++, block_start = block_end, bh = bh->b_this_page) {
1097 block_end = block_start + blocksize;
1098 if (block_end <= from || block_start >= to) {
1099 if (PageUptodate(page)) {
1100 if (!buffer_uptodate(bh))
1101 set_buffer_uptodate(bh);
1106 clear_buffer_new(bh);
1107 if (!buffer_mapped(bh)) {
1108 WARN_ON(bh->b_size != blocksize);
1109 err = get_block(inode, block, bh, 1);
1112 if (buffer_new(bh)) {
1113 unmap_underlying_metadata(bh->b_bdev,
1115 if (PageUptodate(page)) {
1116 clear_buffer_new(bh);
1117 set_buffer_uptodate(bh);
1118 mark_buffer_dirty(bh);
1121 if (block_end > to || block_start < from)
1122 zero_user_segments(page, to, block_end,
1127 if (PageUptodate(page)) {
1128 if (!buffer_uptodate(bh))
1129 set_buffer_uptodate(bh);
1132 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1133 !buffer_unwritten(bh) &&
1134 (block_start < from || block_end > to)) {
1135 ll_rw_block(READ, 1, &bh);
1137 decrypt = ext4_encrypted_inode(inode) &&
1138 S_ISREG(inode->i_mode);
1142 * If we issued read requests, let them complete.
1144 while (wait_bh > wait) {
1145 wait_on_buffer(*--wait_bh);
1146 if (!buffer_uptodate(*wait_bh))
1150 page_zero_new_buffers(page, from, to);
1152 err = ext4_decrypt(page);
1157 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1158 loff_t pos, unsigned len, unsigned flags,
1159 struct page **pagep, void **fsdata)
1161 struct inode *inode = mapping->host;
1162 int ret, needed_blocks;
1169 trace_ext4_write_begin(inode, pos, len, flags);
1171 * Reserve one block more for addition to orphan list in case
1172 * we allocate blocks but write fails for some reason
1174 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1175 index = pos >> PAGE_SHIFT;
1176 from = pos & (PAGE_SIZE - 1);
1179 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1180 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1189 * grab_cache_page_write_begin() can take a long time if the
1190 * system is thrashing due to memory pressure, or if the page
1191 * is being written back. So grab it first before we start
1192 * the transaction handle. This also allows us to allocate
1193 * the page (if needed) without using GFP_NOFS.
1196 page = grab_cache_page_write_begin(mapping, index, flags);
1202 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1203 if (IS_ERR(handle)) {
1205 return PTR_ERR(handle);
1209 if (page->mapping != mapping) {
1210 /* The page got truncated from under us */
1213 ext4_journal_stop(handle);
1216 /* In case writeback began while the page was unlocked */
1217 wait_for_stable_page(page);
1219 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1220 if (ext4_should_dioread_nolock(inode))
1221 ret = ext4_block_write_begin(page, pos, len,
1222 ext4_get_block_unwritten);
1224 ret = ext4_block_write_begin(page, pos, len,
1227 if (ext4_should_dioread_nolock(inode))
1228 ret = __block_write_begin(page, pos, len,
1229 ext4_get_block_unwritten);
1231 ret = __block_write_begin(page, pos, len, ext4_get_block);
1233 if (!ret && ext4_should_journal_data(inode)) {
1234 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1236 do_journal_get_write_access);
1242 * __block_write_begin may have instantiated a few blocks
1243 * outside i_size. Trim these off again. Don't need
1244 * i_size_read because we hold i_mutex.
1246 * Add inode to orphan list in case we crash before
1249 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1250 ext4_orphan_add(handle, inode);
1252 ext4_journal_stop(handle);
1253 if (pos + len > inode->i_size) {
1254 ext4_truncate_failed_write(inode);
1256 * If truncate failed early the inode might
1257 * still be on the orphan list; we need to
1258 * make sure the inode is removed from the
1259 * orphan list in that case.
1262 ext4_orphan_del(NULL, inode);
1265 if (ret == -ENOSPC &&
1266 ext4_should_retry_alloc(inode->i_sb, &retries))
1275 /* For write_end() in data=journal mode */
1276 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1279 if (!buffer_mapped(bh) || buffer_freed(bh))
1281 set_buffer_uptodate(bh);
1282 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1283 clear_buffer_meta(bh);
1284 clear_buffer_prio(bh);
1289 * We need to pick up the new inode size which generic_commit_write gave us
1290 * `file' can be NULL - eg, when called from page_symlink().
1292 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1293 * buffers are managed internally.
1295 static int ext4_write_end(struct file *file,
1296 struct address_space *mapping,
1297 loff_t pos, unsigned len, unsigned copied,
1298 struct page *page, void *fsdata)
1300 handle_t *handle = ext4_journal_current_handle();
1301 struct inode *inode = mapping->host;
1302 loff_t old_size = inode->i_size;
1304 int i_size_changed = 0;
1306 trace_ext4_write_end(inode, pos, len, copied);
1307 if (ext4_has_inline_data(inode)) {
1308 ret = ext4_write_inline_data_end(inode, pos, len,
1314 copied = block_write_end(file, mapping, pos,
1315 len, copied, page, fsdata);
1317 * it's important to update i_size while still holding page lock:
1318 * page writeout could otherwise come in and zero beyond i_size.
1320 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1325 pagecache_isize_extended(inode, old_size, pos);
1327 * Don't mark the inode dirty under page lock. First, it unnecessarily
1328 * makes the holding time of page lock longer. Second, it forces lock
1329 * ordering of page lock and transaction start for journaling
1333 ext4_mark_inode_dirty(handle, inode);
1335 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1336 /* if we have allocated more blocks and copied
1337 * less. We will have blocks allocated outside
1338 * inode->i_size. So truncate them
1340 ext4_orphan_add(handle, inode);
1342 ret2 = ext4_journal_stop(handle);
1346 if (pos + len > inode->i_size) {
1347 ext4_truncate_failed_write(inode);
1349 * If truncate failed early the inode might still be
1350 * on the orphan list; we need to make sure the inode
1351 * is removed from the orphan list in that case.
1354 ext4_orphan_del(NULL, inode);
1357 return ret ? ret : copied;
1361 * This is a private version of page_zero_new_buffers() which doesn't
1362 * set the buffer to be dirty, since in data=journalled mode we need
1363 * to call ext4_handle_dirty_metadata() instead.
1365 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1367 unsigned int block_start = 0, block_end;
1368 struct buffer_head *head, *bh;
1370 bh = head = page_buffers(page);
1372 block_end = block_start + bh->b_size;
1373 if (buffer_new(bh)) {
1374 if (block_end > from && block_start < to) {
1375 if (!PageUptodate(page)) {
1376 unsigned start, size;
1378 start = max(from, block_start);
1379 size = min(to, block_end) - start;
1381 zero_user(page, start, size);
1382 set_buffer_uptodate(bh);
1384 clear_buffer_new(bh);
1387 block_start = block_end;
1388 bh = bh->b_this_page;
1389 } while (bh != head);
1392 static int ext4_journalled_write_end(struct file *file,
1393 struct address_space *mapping,
1394 loff_t pos, unsigned len, unsigned copied,
1395 struct page *page, void *fsdata)
1397 handle_t *handle = ext4_journal_current_handle();
1398 struct inode *inode = mapping->host;
1399 loff_t old_size = inode->i_size;
1403 int size_changed = 0;
1405 trace_ext4_journalled_write_end(inode, pos, len, copied);
1406 from = pos & (PAGE_SIZE - 1);
1409 BUG_ON(!ext4_handle_valid(handle));
1411 if (ext4_has_inline_data(inode))
1412 copied = ext4_write_inline_data_end(inode, pos, len,
1416 if (!PageUptodate(page))
1418 zero_new_buffers(page, from+copied, to);
1421 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1422 to, &partial, write_end_fn);
1424 SetPageUptodate(page);
1426 size_changed = ext4_update_inode_size(inode, pos + copied);
1427 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1428 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1433 pagecache_isize_extended(inode, old_size, pos);
1436 ret2 = ext4_mark_inode_dirty(handle, inode);
1441 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1442 /* if we have allocated more blocks and copied
1443 * less. We will have blocks allocated outside
1444 * inode->i_size. So truncate them
1446 ext4_orphan_add(handle, inode);
1448 ret2 = ext4_journal_stop(handle);
1451 if (pos + len > inode->i_size) {
1452 ext4_truncate_failed_write(inode);
1454 * If truncate failed early the inode might still be
1455 * on the orphan list; we need to make sure the inode
1456 * is removed from the orphan list in that case.
1459 ext4_orphan_del(NULL, inode);
1462 return ret ? ret : copied;
1466 * Reserve space for a single cluster
1468 static int ext4_da_reserve_space(struct inode *inode)
1470 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1471 struct ext4_inode_info *ei = EXT4_I(inode);
1475 * We will charge metadata quota at writeout time; this saves
1476 * us from metadata over-estimation, though we may go over by
1477 * a small amount in the end. Here we just reserve for data.
1479 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1483 spin_lock(&ei->i_block_reservation_lock);
1484 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1485 spin_unlock(&ei->i_block_reservation_lock);
1486 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1489 ei->i_reserved_data_blocks++;
1490 trace_ext4_da_reserve_space(inode);
1491 spin_unlock(&ei->i_block_reservation_lock);
1493 return 0; /* success */
1496 static void ext4_da_release_space(struct inode *inode, int to_free)
1498 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1499 struct ext4_inode_info *ei = EXT4_I(inode);
1502 return; /* Nothing to release, exit */
1504 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1506 trace_ext4_da_release_space(inode, to_free);
1507 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1509 * if there aren't enough reserved blocks, then the
1510 * counter is messed up somewhere. Since this
1511 * function is called from invalidate page, it's
1512 * harmless to return without any action.
1514 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1515 "ino %lu, to_free %d with only %d reserved "
1516 "data blocks", inode->i_ino, to_free,
1517 ei->i_reserved_data_blocks);
1519 to_free = ei->i_reserved_data_blocks;
1521 ei->i_reserved_data_blocks -= to_free;
1523 /* update fs dirty data blocks counter */
1524 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1526 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1528 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1531 static void ext4_da_page_release_reservation(struct page *page,
1532 unsigned int offset,
1533 unsigned int length)
1535 int to_release = 0, contiguous_blks = 0;
1536 struct buffer_head *head, *bh;
1537 unsigned int curr_off = 0;
1538 struct inode *inode = page->mapping->host;
1539 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1540 unsigned int stop = offset + length;
1544 BUG_ON(stop > PAGE_SIZE || stop < length);
1546 head = page_buffers(page);
1549 unsigned int next_off = curr_off + bh->b_size;
1551 if (next_off > stop)
1554 if ((offset <= curr_off) && (buffer_delay(bh))) {
1557 clear_buffer_delay(bh);
1558 } else if (contiguous_blks) {
1559 lblk = page->index <<
1560 (PAGE_SHIFT - inode->i_blkbits);
1561 lblk += (curr_off >> inode->i_blkbits) -
1563 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1564 contiguous_blks = 0;
1566 curr_off = next_off;
1567 } while ((bh = bh->b_this_page) != head);
1569 if (contiguous_blks) {
1570 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1571 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1572 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1575 /* If we have released all the blocks belonging to a cluster, then we
1576 * need to release the reserved space for that cluster. */
1577 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1578 while (num_clusters > 0) {
1579 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1580 ((num_clusters - 1) << sbi->s_cluster_bits);
1581 if (sbi->s_cluster_ratio == 1 ||
1582 !ext4_find_delalloc_cluster(inode, lblk))
1583 ext4_da_release_space(inode, 1);
1590 * Delayed allocation stuff
1593 struct mpage_da_data {
1594 struct inode *inode;
1595 struct writeback_control *wbc;
1597 pgoff_t first_page; /* The first page to write */
1598 pgoff_t next_page; /* Current page to examine */
1599 pgoff_t last_page; /* Last page to examine */
1601 * Extent to map - this can be after first_page because that can be
1602 * fully mapped. We somewhat abuse m_flags to store whether the extent
1603 * is delalloc or unwritten.
1605 struct ext4_map_blocks map;
1606 struct ext4_io_submit io_submit; /* IO submission data */
1609 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1614 struct pagevec pvec;
1615 struct inode *inode = mpd->inode;
1616 struct address_space *mapping = inode->i_mapping;
1618 /* This is necessary when next_page == 0. */
1619 if (mpd->first_page >= mpd->next_page)
1622 index = mpd->first_page;
1623 end = mpd->next_page - 1;
1625 ext4_lblk_t start, last;
1626 start = index << (PAGE_SHIFT - inode->i_blkbits);
1627 last = end << (PAGE_SHIFT - inode->i_blkbits);
1628 ext4_es_remove_extent(inode, start, last - start + 1);
1631 pagevec_init(&pvec, 0);
1632 while (index <= end) {
1633 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1636 for (i = 0; i < nr_pages; i++) {
1637 struct page *page = pvec.pages[i];
1638 if (page->index > end)
1640 BUG_ON(!PageLocked(page));
1641 BUG_ON(PageWriteback(page));
1643 block_invalidatepage(page, 0, PAGE_SIZE);
1644 ClearPageUptodate(page);
1648 index = pvec.pages[nr_pages - 1]->index + 1;
1649 pagevec_release(&pvec);
1653 static void ext4_print_free_blocks(struct inode *inode)
1655 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1656 struct super_block *sb = inode->i_sb;
1657 struct ext4_inode_info *ei = EXT4_I(inode);
1659 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1660 EXT4_C2B(EXT4_SB(inode->i_sb),
1661 ext4_count_free_clusters(sb)));
1662 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1663 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1664 (long long) EXT4_C2B(EXT4_SB(sb),
1665 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1666 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1667 (long long) EXT4_C2B(EXT4_SB(sb),
1668 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1669 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1670 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1671 ei->i_reserved_data_blocks);
1675 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1677 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1681 * This function is grabs code from the very beginning of
1682 * ext4_map_blocks, but assumes that the caller is from delayed write
1683 * time. This function looks up the requested blocks and sets the
1684 * buffer delay bit under the protection of i_data_sem.
1686 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1687 struct ext4_map_blocks *map,
1688 struct buffer_head *bh)
1690 struct extent_status es;
1692 sector_t invalid_block = ~((sector_t) 0xffff);
1693 #ifdef ES_AGGRESSIVE_TEST
1694 struct ext4_map_blocks orig_map;
1696 memcpy(&orig_map, map, sizeof(*map));
1699 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1703 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1704 "logical block %lu\n", inode->i_ino, map->m_len,
1705 (unsigned long) map->m_lblk);
1707 /* Lookup extent status tree firstly */
1708 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1709 if (ext4_es_is_hole(&es)) {
1711 down_read(&EXT4_I(inode)->i_data_sem);
1716 * Delayed extent could be allocated by fallocate.
1717 * So we need to check it.
1719 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1720 map_bh(bh, inode->i_sb, invalid_block);
1722 set_buffer_delay(bh);
1726 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1727 retval = es.es_len - (iblock - es.es_lblk);
1728 if (retval > map->m_len)
1729 retval = map->m_len;
1730 map->m_len = retval;
1731 if (ext4_es_is_written(&es))
1732 map->m_flags |= EXT4_MAP_MAPPED;
1733 else if (ext4_es_is_unwritten(&es))
1734 map->m_flags |= EXT4_MAP_UNWRITTEN;
1738 #ifdef ES_AGGRESSIVE_TEST
1739 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1745 * Try to see if we can get the block without requesting a new
1746 * file system block.
1748 down_read(&EXT4_I(inode)->i_data_sem);
1749 if (ext4_has_inline_data(inode))
1751 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1752 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1754 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1760 * XXX: __block_prepare_write() unmaps passed block,
1764 * If the block was allocated from previously allocated cluster,
1765 * then we don't need to reserve it again. However we still need
1766 * to reserve metadata for every block we're going to write.
1768 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1769 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1770 ret = ext4_da_reserve_space(inode);
1772 /* not enough space to reserve */
1778 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1779 ~0, EXTENT_STATUS_DELAYED);
1785 map_bh(bh, inode->i_sb, invalid_block);
1787 set_buffer_delay(bh);
1788 } else if (retval > 0) {
1790 unsigned int status;
1792 if (unlikely(retval != map->m_len)) {
1793 ext4_warning(inode->i_sb,
1794 "ES len assertion failed for inode "
1795 "%lu: retval %d != map->m_len %d",
1796 inode->i_ino, retval, map->m_len);
1800 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1801 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1802 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1803 map->m_pblk, status);
1809 up_read((&EXT4_I(inode)->i_data_sem));
1815 * This is a special get_block_t callback which is used by
1816 * ext4_da_write_begin(). It will either return mapped block or
1817 * reserve space for a single block.
1819 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1820 * We also have b_blocknr = -1 and b_bdev initialized properly
1822 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1823 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1824 * initialized properly.
1826 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1827 struct buffer_head *bh, int create)
1829 struct ext4_map_blocks map;
1832 BUG_ON(create == 0);
1833 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1835 map.m_lblk = iblock;
1839 * first, we need to know whether the block is allocated already
1840 * preallocated blocks are unmapped but should treated
1841 * the same as allocated blocks.
1843 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1847 map_bh(bh, inode->i_sb, map.m_pblk);
1848 ext4_update_bh_state(bh, map.m_flags);
1850 if (buffer_unwritten(bh)) {
1851 /* A delayed write to unwritten bh should be marked
1852 * new and mapped. Mapped ensures that we don't do
1853 * get_block multiple times when we write to the same
1854 * offset and new ensures that we do proper zero out
1855 * for partial write.
1858 set_buffer_mapped(bh);
1863 static int bget_one(handle_t *handle, struct buffer_head *bh)
1869 static int bput_one(handle_t *handle, struct buffer_head *bh)
1875 static int __ext4_journalled_writepage(struct page *page,
1878 struct address_space *mapping = page->mapping;
1879 struct inode *inode = mapping->host;
1880 struct buffer_head *page_bufs = NULL;
1881 handle_t *handle = NULL;
1882 int ret = 0, err = 0;
1883 int inline_data = ext4_has_inline_data(inode);
1884 struct buffer_head *inode_bh = NULL;
1886 ClearPageChecked(page);
1889 BUG_ON(page->index != 0);
1890 BUG_ON(len > ext4_get_max_inline_size(inode));
1891 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1892 if (inode_bh == NULL)
1895 page_bufs = page_buffers(page);
1900 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1904 * We need to release the page lock before we start the
1905 * journal, so grab a reference so the page won't disappear
1906 * out from under us.
1911 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1912 ext4_writepage_trans_blocks(inode));
1913 if (IS_ERR(handle)) {
1914 ret = PTR_ERR(handle);
1916 goto out_no_pagelock;
1918 BUG_ON(!ext4_handle_valid(handle));
1922 if (page->mapping != mapping) {
1923 /* The page got truncated from under us */
1924 ext4_journal_stop(handle);
1930 BUFFER_TRACE(inode_bh, "get write access");
1931 ret = ext4_journal_get_write_access(handle, inode_bh);
1933 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1936 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1937 do_journal_get_write_access);
1939 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1944 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1945 err = ext4_journal_stop(handle);
1949 if (!ext4_has_inline_data(inode))
1950 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1952 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1961 * Note that we don't need to start a transaction unless we're journaling data
1962 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1963 * need to file the inode to the transaction's list in ordered mode because if
1964 * we are writing back data added by write(), the inode is already there and if
1965 * we are writing back data modified via mmap(), no one guarantees in which
1966 * transaction the data will hit the disk. In case we are journaling data, we
1967 * cannot start transaction directly because transaction start ranks above page
1968 * lock so we have to do some magic.
1970 * This function can get called via...
1971 * - ext4_writepages after taking page lock (have journal handle)
1972 * - journal_submit_inode_data_buffers (no journal handle)
1973 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1974 * - grab_page_cache when doing write_begin (have journal handle)
1976 * We don't do any block allocation in this function. If we have page with
1977 * multiple blocks we need to write those buffer_heads that are mapped. This
1978 * is important for mmaped based write. So if we do with blocksize 1K
1979 * truncate(f, 1024);
1980 * a = mmap(f, 0, 4096);
1982 * truncate(f, 4096);
1983 * we have in the page first buffer_head mapped via page_mkwrite call back
1984 * but other buffer_heads would be unmapped but dirty (dirty done via the
1985 * do_wp_page). So writepage should write the first block. If we modify
1986 * the mmap area beyond 1024 we will again get a page_fault and the
1987 * page_mkwrite callback will do the block allocation and mark the
1988 * buffer_heads mapped.
1990 * We redirty the page if we have any buffer_heads that is either delay or
1991 * unwritten in the page.
1993 * We can get recursively called as show below.
1995 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1998 * But since we don't do any block allocation we should not deadlock.
1999 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2001 static int ext4_writepage(struct page *page,
2002 struct writeback_control *wbc)
2007 struct buffer_head *page_bufs = NULL;
2008 struct inode *inode = page->mapping->host;
2009 struct ext4_io_submit io_submit;
2010 bool keep_towrite = false;
2012 trace_ext4_writepage(page);
2013 size = i_size_read(inode);
2014 if (page->index == size >> PAGE_SHIFT)
2015 len = size & ~PAGE_MASK;
2019 page_bufs = page_buffers(page);
2021 * We cannot do block allocation or other extent handling in this
2022 * function. If there are buffers needing that, we have to redirty
2023 * the page. But we may reach here when we do a journal commit via
2024 * journal_submit_inode_data_buffers() and in that case we must write
2025 * allocated buffers to achieve data=ordered mode guarantees.
2027 * Also, if there is only one buffer per page (the fs block
2028 * size == the page size), if one buffer needs block
2029 * allocation or needs to modify the extent tree to clear the
2030 * unwritten flag, we know that the page can't be written at
2031 * all, so we might as well refuse the write immediately.
2032 * Unfortunately if the block size != page size, we can't as
2033 * easily detect this case using ext4_walk_page_buffers(), but
2034 * for the extremely common case, this is an optimization that
2035 * skips a useless round trip through ext4_bio_write_page().
2037 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2038 ext4_bh_delay_or_unwritten)) {
2039 redirty_page_for_writepage(wbc, page);
2040 if ((current->flags & PF_MEMALLOC) ||
2041 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2043 * For memory cleaning there's no point in writing only
2044 * some buffers. So just bail out. Warn if we came here
2045 * from direct reclaim.
2047 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2052 keep_towrite = true;
2055 if (PageChecked(page) && ext4_should_journal_data(inode))
2057 * It's mmapped pagecache. Add buffers and journal it. There
2058 * doesn't seem much point in redirtying the page here.
2060 return __ext4_journalled_writepage(page, len);
2062 ext4_io_submit_init(&io_submit, wbc);
2063 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2064 if (!io_submit.io_end) {
2065 redirty_page_for_writepage(wbc, page);
2069 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2070 ext4_io_submit(&io_submit);
2071 /* Drop io_end reference we got from init */
2072 ext4_put_io_end_defer(io_submit.io_end);
2076 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2079 loff_t size = i_size_read(mpd->inode);
2082 BUG_ON(page->index != mpd->first_page);
2083 if (page->index == size >> PAGE_SHIFT)
2084 len = size & ~PAGE_MASK;
2087 clear_page_dirty_for_io(page);
2088 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2090 mpd->wbc->nr_to_write--;
2096 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2099 * mballoc gives us at most this number of blocks...
2100 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2101 * The rest of mballoc seems to handle chunks up to full group size.
2103 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2106 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2108 * @mpd - extent of blocks
2109 * @lblk - logical number of the block in the file
2110 * @bh - buffer head we want to add to the extent
2112 * The function is used to collect contig. blocks in the same state. If the
2113 * buffer doesn't require mapping for writeback and we haven't started the
2114 * extent of buffers to map yet, the function returns 'true' immediately - the
2115 * caller can write the buffer right away. Otherwise the function returns true
2116 * if the block has been added to the extent, false if the block couldn't be
2119 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2120 struct buffer_head *bh)
2122 struct ext4_map_blocks *map = &mpd->map;
2124 /* Buffer that doesn't need mapping for writeback? */
2125 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2126 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2127 /* So far no extent to map => we write the buffer right away */
2128 if (map->m_len == 0)
2133 /* First block in the extent? */
2134 if (map->m_len == 0) {
2137 map->m_flags = bh->b_state & BH_FLAGS;
2141 /* Don't go larger than mballoc is willing to allocate */
2142 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2145 /* Can we merge the block to our big extent? */
2146 if (lblk == map->m_lblk + map->m_len &&
2147 (bh->b_state & BH_FLAGS) == map->m_flags) {
2155 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2157 * @mpd - extent of blocks for mapping
2158 * @head - the first buffer in the page
2159 * @bh - buffer we should start processing from
2160 * @lblk - logical number of the block in the file corresponding to @bh
2162 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2163 * the page for IO if all buffers in this page were mapped and there's no
2164 * accumulated extent of buffers to map or add buffers in the page to the
2165 * extent of buffers to map. The function returns 1 if the caller can continue
2166 * by processing the next page, 0 if it should stop adding buffers to the
2167 * extent to map because we cannot extend it anymore. It can also return value
2168 * < 0 in case of error during IO submission.
2170 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2171 struct buffer_head *head,
2172 struct buffer_head *bh,
2175 struct inode *inode = mpd->inode;
2177 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2178 >> inode->i_blkbits;
2181 BUG_ON(buffer_locked(bh));
2183 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2184 /* Found extent to map? */
2187 /* Everything mapped so far and we hit EOF */
2190 } while (lblk++, (bh = bh->b_this_page) != head);
2191 /* So far everything mapped? Submit the page for IO. */
2192 if (mpd->map.m_len == 0) {
2193 err = mpage_submit_page(mpd, head->b_page);
2197 return lblk < blocks;
2201 * mpage_map_buffers - update buffers corresponding to changed extent and
2202 * submit fully mapped pages for IO
2204 * @mpd - description of extent to map, on return next extent to map
2206 * Scan buffers corresponding to changed extent (we expect corresponding pages
2207 * to be already locked) and update buffer state according to new extent state.
2208 * We map delalloc buffers to their physical location, clear unwritten bits,
2209 * and mark buffers as uninit when we perform writes to unwritten extents
2210 * and do extent conversion after IO is finished. If the last page is not fully
2211 * mapped, we update @map to the next extent in the last page that needs
2212 * mapping. Otherwise we submit the page for IO.
2214 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2216 struct pagevec pvec;
2218 struct inode *inode = mpd->inode;
2219 struct buffer_head *head, *bh;
2220 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2226 start = mpd->map.m_lblk >> bpp_bits;
2227 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2228 lblk = start << bpp_bits;
2229 pblock = mpd->map.m_pblk;
2231 pagevec_init(&pvec, 0);
2232 while (start <= end) {
2233 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2237 for (i = 0; i < nr_pages; i++) {
2238 struct page *page = pvec.pages[i];
2240 if (page->index > end)
2242 /* Up to 'end' pages must be contiguous */
2243 BUG_ON(page->index != start);
2244 bh = head = page_buffers(page);
2246 if (lblk < mpd->map.m_lblk)
2248 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2250 * Buffer after end of mapped extent.
2251 * Find next buffer in the page to map.
2254 mpd->map.m_flags = 0;
2256 * FIXME: If dioread_nolock supports
2257 * blocksize < pagesize, we need to make
2258 * sure we add size mapped so far to
2259 * io_end->size as the following call
2260 * can submit the page for IO.
2262 err = mpage_process_page_bufs(mpd, head,
2264 pagevec_release(&pvec);
2269 if (buffer_delay(bh)) {
2270 clear_buffer_delay(bh);
2271 bh->b_blocknr = pblock++;
2273 clear_buffer_unwritten(bh);
2274 } while (lblk++, (bh = bh->b_this_page) != head);
2277 * FIXME: This is going to break if dioread_nolock
2278 * supports blocksize < pagesize as we will try to
2279 * convert potentially unmapped parts of inode.
2281 mpd->io_submit.io_end->size += PAGE_SIZE;
2282 /* Page fully mapped - let IO run! */
2283 err = mpage_submit_page(mpd, page);
2285 pagevec_release(&pvec);
2290 pagevec_release(&pvec);
2292 /* Extent fully mapped and matches with page boundary. We are done. */
2294 mpd->map.m_flags = 0;
2298 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2300 struct inode *inode = mpd->inode;
2301 struct ext4_map_blocks *map = &mpd->map;
2302 int get_blocks_flags;
2303 int err, dioread_nolock;
2305 trace_ext4_da_write_pages_extent(inode, map);
2307 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2308 * to convert an unwritten extent to be initialized (in the case
2309 * where we have written into one or more preallocated blocks). It is
2310 * possible that we're going to need more metadata blocks than
2311 * previously reserved. However we must not fail because we're in
2312 * writeback and there is nothing we can do about it so it might result
2313 * in data loss. So use reserved blocks to allocate metadata if
2316 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2317 * the blocks in question are delalloc blocks. This indicates
2318 * that the blocks and quotas has already been checked when
2319 * the data was copied into the page cache.
2321 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2322 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2323 dioread_nolock = ext4_should_dioread_nolock(inode);
2325 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2326 if (map->m_flags & (1 << BH_Delay))
2327 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2329 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2332 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2333 if (!mpd->io_submit.io_end->handle &&
2334 ext4_handle_valid(handle)) {
2335 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2336 handle->h_rsv_handle = NULL;
2338 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2341 BUG_ON(map->m_len == 0);
2342 if (map->m_flags & EXT4_MAP_NEW) {
2343 struct block_device *bdev = inode->i_sb->s_bdev;
2346 for (i = 0; i < map->m_len; i++)
2347 unmap_underlying_metadata(bdev, map->m_pblk + i);
2353 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2354 * mpd->len and submit pages underlying it for IO
2356 * @handle - handle for journal operations
2357 * @mpd - extent to map
2358 * @give_up_on_write - we set this to true iff there is a fatal error and there
2359 * is no hope of writing the data. The caller should discard
2360 * dirty pages to avoid infinite loops.
2362 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2363 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2364 * them to initialized or split the described range from larger unwritten
2365 * extent. Note that we need not map all the described range since allocation
2366 * can return less blocks or the range is covered by more unwritten extents. We
2367 * cannot map more because we are limited by reserved transaction credits. On
2368 * the other hand we always make sure that the last touched page is fully
2369 * mapped so that it can be written out (and thus forward progress is
2370 * guaranteed). After mapping we submit all mapped pages for IO.
2372 static int mpage_map_and_submit_extent(handle_t *handle,
2373 struct mpage_da_data *mpd,
2374 bool *give_up_on_write)
2376 struct inode *inode = mpd->inode;
2377 struct ext4_map_blocks *map = &mpd->map;
2382 mpd->io_submit.io_end->offset =
2383 ((loff_t)map->m_lblk) << inode->i_blkbits;
2385 err = mpage_map_one_extent(handle, mpd);
2387 struct super_block *sb = inode->i_sb;
2389 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2390 goto invalidate_dirty_pages;
2392 * Let the uper layers retry transient errors.
2393 * In the case of ENOSPC, if ext4_count_free_blocks()
2394 * is non-zero, a commit should free up blocks.
2396 if ((err == -ENOMEM) ||
2397 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2399 goto update_disksize;
2402 ext4_msg(sb, KERN_CRIT,
2403 "Delayed block allocation failed for "
2404 "inode %lu at logical offset %llu with"
2405 " max blocks %u with error %d",
2407 (unsigned long long)map->m_lblk,
2408 (unsigned)map->m_len, -err);
2409 ext4_msg(sb, KERN_CRIT,
2410 "This should not happen!! Data will "
2413 ext4_print_free_blocks(inode);
2414 invalidate_dirty_pages:
2415 *give_up_on_write = true;
2420 * Update buffer state, submit mapped pages, and get us new
2423 err = mpage_map_and_submit_buffers(mpd);
2425 goto update_disksize;
2426 } while (map->m_len);
2430 * Update on-disk size after IO is submitted. Races with
2431 * truncate are avoided by checking i_size under i_data_sem.
2433 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2434 if (disksize > EXT4_I(inode)->i_disksize) {
2438 down_write(&EXT4_I(inode)->i_data_sem);
2439 i_size = i_size_read(inode);
2440 if (disksize > i_size)
2442 if (disksize > EXT4_I(inode)->i_disksize)
2443 EXT4_I(inode)->i_disksize = disksize;
2444 err2 = ext4_mark_inode_dirty(handle, inode);
2445 up_write(&EXT4_I(inode)->i_data_sem);
2447 ext4_error(inode->i_sb,
2448 "Failed to mark inode %lu dirty",
2457 * Calculate the total number of credits to reserve for one writepages
2458 * iteration. This is called from ext4_writepages(). We map an extent of
2459 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2460 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2461 * bpp - 1 blocks in bpp different extents.
2463 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2465 int bpp = ext4_journal_blocks_per_page(inode);
2467 return ext4_meta_trans_blocks(inode,
2468 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2472 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2473 * and underlying extent to map
2475 * @mpd - where to look for pages
2477 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2478 * IO immediately. When we find a page which isn't mapped we start accumulating
2479 * extent of buffers underlying these pages that needs mapping (formed by
2480 * either delayed or unwritten buffers). We also lock the pages containing
2481 * these buffers. The extent found is returned in @mpd structure (starting at
2482 * mpd->lblk with length mpd->len blocks).
2484 * Note that this function can attach bios to one io_end structure which are
2485 * neither logically nor physically contiguous. Although it may seem as an
2486 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2487 * case as we need to track IO to all buffers underlying a page in one io_end.
2489 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2491 struct address_space *mapping = mpd->inode->i_mapping;
2492 struct pagevec pvec;
2493 unsigned int nr_pages;
2494 long left = mpd->wbc->nr_to_write;
2495 pgoff_t index = mpd->first_page;
2496 pgoff_t end = mpd->last_page;
2499 int blkbits = mpd->inode->i_blkbits;
2501 struct buffer_head *head;
2503 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2504 tag = PAGECACHE_TAG_TOWRITE;
2506 tag = PAGECACHE_TAG_DIRTY;
2508 pagevec_init(&pvec, 0);
2510 mpd->next_page = index;
2511 while (index <= end) {
2512 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2513 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2517 for (i = 0; i < nr_pages; i++) {
2518 struct page *page = pvec.pages[i];
2521 * At this point, the page may be truncated or
2522 * invalidated (changing page->mapping to NULL), or
2523 * even swizzled back from swapper_space to tmpfs file
2524 * mapping. However, page->index will not change
2525 * because we have a reference on the page.
2527 if (page->index > end)
2531 * Accumulated enough dirty pages? This doesn't apply
2532 * to WB_SYNC_ALL mode. For integrity sync we have to
2533 * keep going because someone may be concurrently
2534 * dirtying pages, and we might have synced a lot of
2535 * newly appeared dirty pages, but have not synced all
2536 * of the old dirty pages.
2538 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2541 /* If we can't merge this page, we are done. */
2542 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2547 * If the page is no longer dirty, or its mapping no
2548 * longer corresponds to inode we are writing (which
2549 * means it has been truncated or invalidated), or the
2550 * page is already under writeback and we are not doing
2551 * a data integrity writeback, skip the page
2553 if (!PageDirty(page) ||
2554 (PageWriteback(page) &&
2555 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2556 unlikely(page->mapping != mapping)) {
2561 wait_on_page_writeback(page);
2562 BUG_ON(PageWriteback(page));
2564 if (mpd->map.m_len == 0)
2565 mpd->first_page = page->index;
2566 mpd->next_page = page->index + 1;
2567 /* Add all dirty buffers to mpd */
2568 lblk = ((ext4_lblk_t)page->index) <<
2569 (PAGE_SHIFT - blkbits);
2570 head = page_buffers(page);
2571 err = mpage_process_page_bufs(mpd, head, head, lblk);
2577 pagevec_release(&pvec);
2582 pagevec_release(&pvec);
2586 static int __writepage(struct page *page, struct writeback_control *wbc,
2589 struct address_space *mapping = data;
2590 int ret = ext4_writepage(page, wbc);
2591 mapping_set_error(mapping, ret);
2595 static int ext4_writepages(struct address_space *mapping,
2596 struct writeback_control *wbc)
2598 pgoff_t writeback_index = 0;
2599 long nr_to_write = wbc->nr_to_write;
2600 int range_whole = 0;
2602 handle_t *handle = NULL;
2603 struct mpage_da_data mpd;
2604 struct inode *inode = mapping->host;
2605 int needed_blocks, rsv_blocks = 0, ret = 0;
2606 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2608 struct blk_plug plug;
2609 bool give_up_on_write = false;
2611 trace_ext4_writepages(inode, wbc);
2613 if (dax_mapping(mapping))
2614 return dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2618 * No pages to write? This is mainly a kludge to avoid starting
2619 * a transaction for special inodes like journal inode on last iput()
2620 * because that could violate lock ordering on umount
2622 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2623 goto out_writepages;
2625 if (ext4_should_journal_data(inode)) {
2626 struct blk_plug plug;
2628 blk_start_plug(&plug);
2629 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2630 blk_finish_plug(&plug);
2631 goto out_writepages;
2635 * If the filesystem has aborted, it is read-only, so return
2636 * right away instead of dumping stack traces later on that
2637 * will obscure the real source of the problem. We test
2638 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2639 * the latter could be true if the filesystem is mounted
2640 * read-only, and in that case, ext4_writepages should
2641 * *never* be called, so if that ever happens, we would want
2644 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2646 goto out_writepages;
2649 if (ext4_should_dioread_nolock(inode)) {
2651 * We may need to convert up to one extent per block in
2652 * the page and we may dirty the inode.
2654 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2658 * If we have inline data and arrive here, it means that
2659 * we will soon create the block for the 1st page, so
2660 * we'd better clear the inline data here.
2662 if (ext4_has_inline_data(inode)) {
2663 /* Just inode will be modified... */
2664 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2665 if (IS_ERR(handle)) {
2666 ret = PTR_ERR(handle);
2667 goto out_writepages;
2669 BUG_ON(ext4_test_inode_state(inode,
2670 EXT4_STATE_MAY_INLINE_DATA));
2671 ext4_destroy_inline_data(handle, inode);
2672 ext4_journal_stop(handle);
2675 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2678 if (wbc->range_cyclic) {
2679 writeback_index = mapping->writeback_index;
2680 if (writeback_index)
2682 mpd.first_page = writeback_index;
2685 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2686 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2691 ext4_io_submit_init(&mpd.io_submit, wbc);
2693 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2694 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2696 blk_start_plug(&plug);
2697 while (!done && mpd.first_page <= mpd.last_page) {
2698 /* For each extent of pages we use new io_end */
2699 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2700 if (!mpd.io_submit.io_end) {
2706 * We have two constraints: We find one extent to map and we
2707 * must always write out whole page (makes a difference when
2708 * blocksize < pagesize) so that we don't block on IO when we
2709 * try to write out the rest of the page. Journalled mode is
2710 * not supported by delalloc.
2712 BUG_ON(ext4_should_journal_data(inode));
2713 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2715 /* start a new transaction */
2716 handle = ext4_journal_start_with_reserve(inode,
2717 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2718 if (IS_ERR(handle)) {
2719 ret = PTR_ERR(handle);
2720 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2721 "%ld pages, ino %lu; err %d", __func__,
2722 wbc->nr_to_write, inode->i_ino, ret);
2723 /* Release allocated io_end */
2724 ext4_put_io_end(mpd.io_submit.io_end);
2728 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2729 ret = mpage_prepare_extent_to_map(&mpd);
2732 ret = mpage_map_and_submit_extent(handle, &mpd,
2736 * We scanned the whole range (or exhausted
2737 * nr_to_write), submitted what was mapped and
2738 * didn't find anything needing mapping. We are
2744 ext4_journal_stop(handle);
2745 /* Submit prepared bio */
2746 ext4_io_submit(&mpd.io_submit);
2747 /* Unlock pages we didn't use */
2748 mpage_release_unused_pages(&mpd, give_up_on_write);
2749 /* Drop our io_end reference we got from init */
2750 ext4_put_io_end(mpd.io_submit.io_end);
2752 if (ret == -ENOSPC && sbi->s_journal) {
2754 * Commit the transaction which would
2755 * free blocks released in the transaction
2758 jbd2_journal_force_commit_nested(sbi->s_journal);
2762 /* Fatal error - ENOMEM, EIO... */
2766 blk_finish_plug(&plug);
2767 if (!ret && !cycled && wbc->nr_to_write > 0) {
2769 mpd.last_page = writeback_index - 1;
2775 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2777 * Set the writeback_index so that range_cyclic
2778 * mode will write it back later
2780 mapping->writeback_index = mpd.first_page;
2783 trace_ext4_writepages_result(inode, wbc, ret,
2784 nr_to_write - wbc->nr_to_write);
2788 static int ext4_nonda_switch(struct super_block *sb)
2790 s64 free_clusters, dirty_clusters;
2791 struct ext4_sb_info *sbi = EXT4_SB(sb);
2794 * switch to non delalloc mode if we are running low
2795 * on free block. The free block accounting via percpu
2796 * counters can get slightly wrong with percpu_counter_batch getting
2797 * accumulated on each CPU without updating global counters
2798 * Delalloc need an accurate free block accounting. So switch
2799 * to non delalloc when we are near to error range.
2802 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2804 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2806 * Start pushing delalloc when 1/2 of free blocks are dirty.
2808 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2809 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2811 if (2 * free_clusters < 3 * dirty_clusters ||
2812 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2814 * free block count is less than 150% of dirty blocks
2815 * or free blocks is less than watermark
2822 /* We always reserve for an inode update; the superblock could be there too */
2823 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2825 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2828 if (pos + len <= 0x7fffffffULL)
2831 /* We might need to update the superblock to set LARGE_FILE */
2835 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2836 loff_t pos, unsigned len, unsigned flags,
2837 struct page **pagep, void **fsdata)
2839 int ret, retries = 0;
2842 struct inode *inode = mapping->host;
2845 index = pos >> PAGE_SHIFT;
2847 if (ext4_nonda_switch(inode->i_sb)) {
2848 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2849 return ext4_write_begin(file, mapping, pos,
2850 len, flags, pagep, fsdata);
2852 *fsdata = (void *)0;
2853 trace_ext4_da_write_begin(inode, pos, len, flags);
2855 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2856 ret = ext4_da_write_inline_data_begin(mapping, inode,
2866 * grab_cache_page_write_begin() can take a long time if the
2867 * system is thrashing due to memory pressure, or if the page
2868 * is being written back. So grab it first before we start
2869 * the transaction handle. This also allows us to allocate
2870 * the page (if needed) without using GFP_NOFS.
2873 page = grab_cache_page_write_begin(mapping, index, flags);
2879 * With delayed allocation, we don't log the i_disksize update
2880 * if there is delayed block allocation. But we still need
2881 * to journalling the i_disksize update if writes to the end
2882 * of file which has an already mapped buffer.
2885 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2886 ext4_da_write_credits(inode, pos, len));
2887 if (IS_ERR(handle)) {
2889 return PTR_ERR(handle);
2893 if (page->mapping != mapping) {
2894 /* The page got truncated from under us */
2897 ext4_journal_stop(handle);
2900 /* In case writeback began while the page was unlocked */
2901 wait_for_stable_page(page);
2903 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2904 ret = ext4_block_write_begin(page, pos, len,
2905 ext4_da_get_block_prep);
2907 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2911 ext4_journal_stop(handle);
2913 * block_write_begin may have instantiated a few blocks
2914 * outside i_size. Trim these off again. Don't need
2915 * i_size_read because we hold i_mutex.
2917 if (pos + len > inode->i_size)
2918 ext4_truncate_failed_write(inode);
2920 if (ret == -ENOSPC &&
2921 ext4_should_retry_alloc(inode->i_sb, &retries))
2933 * Check if we should update i_disksize
2934 * when write to the end of file but not require block allocation
2936 static int ext4_da_should_update_i_disksize(struct page *page,
2937 unsigned long offset)
2939 struct buffer_head *bh;
2940 struct inode *inode = page->mapping->host;
2944 bh = page_buffers(page);
2945 idx = offset >> inode->i_blkbits;
2947 for (i = 0; i < idx; i++)
2948 bh = bh->b_this_page;
2950 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2955 static int ext4_da_write_end(struct file *file,
2956 struct address_space *mapping,
2957 loff_t pos, unsigned len, unsigned copied,
2958 struct page *page, void *fsdata)
2960 struct inode *inode = mapping->host;
2962 handle_t *handle = ext4_journal_current_handle();
2964 unsigned long start, end;
2965 int write_mode = (int)(unsigned long)fsdata;
2967 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2968 return ext4_write_end(file, mapping, pos,
2969 len, copied, page, fsdata);
2971 trace_ext4_da_write_end(inode, pos, len, copied);
2972 start = pos & (PAGE_SIZE - 1);
2973 end = start + copied - 1;
2976 * generic_write_end() will run mark_inode_dirty() if i_size
2977 * changes. So let's piggyback the i_disksize mark_inode_dirty
2980 new_i_size = pos + copied;
2981 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2982 if (ext4_has_inline_data(inode) ||
2983 ext4_da_should_update_i_disksize(page, end)) {
2984 ext4_update_i_disksize(inode, new_i_size);
2985 /* We need to mark inode dirty even if
2986 * new_i_size is less that inode->i_size
2987 * bu greater than i_disksize.(hint delalloc)
2989 ext4_mark_inode_dirty(handle, inode);
2993 if (write_mode != CONVERT_INLINE_DATA &&
2994 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2995 ext4_has_inline_data(inode))
2996 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2999 ret2 = generic_write_end(file, mapping, pos, len, copied,
3005 ret2 = ext4_journal_stop(handle);
3009 return ret ? ret : copied;
3012 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3013 unsigned int length)
3016 * Drop reserved blocks
3018 BUG_ON(!PageLocked(page));
3019 if (!page_has_buffers(page))
3022 ext4_da_page_release_reservation(page, offset, length);
3025 ext4_invalidatepage(page, offset, length);
3031 * Force all delayed allocation blocks to be allocated for a given inode.
3033 int ext4_alloc_da_blocks(struct inode *inode)
3035 trace_ext4_alloc_da_blocks(inode);
3037 if (!EXT4_I(inode)->i_reserved_data_blocks)
3041 * We do something simple for now. The filemap_flush() will
3042 * also start triggering a write of the data blocks, which is
3043 * not strictly speaking necessary (and for users of
3044 * laptop_mode, not even desirable). However, to do otherwise
3045 * would require replicating code paths in:
3047 * ext4_writepages() ->
3048 * write_cache_pages() ---> (via passed in callback function)
3049 * __mpage_da_writepage() -->
3050 * mpage_add_bh_to_extent()
3051 * mpage_da_map_blocks()
3053 * The problem is that write_cache_pages(), located in
3054 * mm/page-writeback.c, marks pages clean in preparation for
3055 * doing I/O, which is not desirable if we're not planning on
3058 * We could call write_cache_pages(), and then redirty all of
3059 * the pages by calling redirty_page_for_writepage() but that
3060 * would be ugly in the extreme. So instead we would need to
3061 * replicate parts of the code in the above functions,
3062 * simplifying them because we wouldn't actually intend to
3063 * write out the pages, but rather only collect contiguous
3064 * logical block extents, call the multi-block allocator, and
3065 * then update the buffer heads with the block allocations.
3067 * For now, though, we'll cheat by calling filemap_flush(),
3068 * which will map the blocks, and start the I/O, but not
3069 * actually wait for the I/O to complete.
3071 return filemap_flush(inode->i_mapping);
3075 * bmap() is special. It gets used by applications such as lilo and by
3076 * the swapper to find the on-disk block of a specific piece of data.
3078 * Naturally, this is dangerous if the block concerned is still in the
3079 * journal. If somebody makes a swapfile on an ext4 data-journaling
3080 * filesystem and enables swap, then they may get a nasty shock when the
3081 * data getting swapped to that swapfile suddenly gets overwritten by
3082 * the original zero's written out previously to the journal and
3083 * awaiting writeback in the kernel's buffer cache.
3085 * So, if we see any bmap calls here on a modified, data-journaled file,
3086 * take extra steps to flush any blocks which might be in the cache.
3088 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3090 struct inode *inode = mapping->host;
3095 * We can get here for an inline file via the FIBMAP ioctl
3097 if (ext4_has_inline_data(inode))
3100 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3101 test_opt(inode->i_sb, DELALLOC)) {
3103 * With delalloc we want to sync the file
3104 * so that we can make sure we allocate
3107 filemap_write_and_wait(mapping);
3110 if (EXT4_JOURNAL(inode) &&
3111 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3113 * This is a REALLY heavyweight approach, but the use of
3114 * bmap on dirty files is expected to be extremely rare:
3115 * only if we run lilo or swapon on a freshly made file
3116 * do we expect this to happen.
3118 * (bmap requires CAP_SYS_RAWIO so this does not
3119 * represent an unprivileged user DOS attack --- we'd be
3120 * in trouble if mortal users could trigger this path at
3123 * NB. EXT4_STATE_JDATA is not set on files other than
3124 * regular files. If somebody wants to bmap a directory
3125 * or symlink and gets confused because the buffer
3126 * hasn't yet been flushed to disk, they deserve
3127 * everything they get.
3130 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3131 journal = EXT4_JOURNAL(inode);
3132 jbd2_journal_lock_updates(journal);
3133 err = jbd2_journal_flush(journal);
3134 jbd2_journal_unlock_updates(journal);
3140 return generic_block_bmap(mapping, block, ext4_get_block);
3143 static int ext4_readpage(struct file *file, struct page *page)
3146 struct inode *inode = page->mapping->host;
3148 trace_ext4_readpage(page);
3150 if (ext4_has_inline_data(inode))
3151 ret = ext4_readpage_inline(inode, page);
3154 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3160 ext4_readpages(struct file *file, struct address_space *mapping,
3161 struct list_head *pages, unsigned nr_pages)
3163 struct inode *inode = mapping->host;
3165 /* If the file has inline data, no need to do readpages. */
3166 if (ext4_has_inline_data(inode))
3169 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3172 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3173 unsigned int length)
3175 trace_ext4_invalidatepage(page, offset, length);
3177 /* No journalling happens on data buffers when this function is used */
3178 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3180 block_invalidatepage(page, offset, length);
3183 static int __ext4_journalled_invalidatepage(struct page *page,
3184 unsigned int offset,
3185 unsigned int length)
3187 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3189 trace_ext4_journalled_invalidatepage(page, offset, length);
3192 * If it's a full truncate we just forget about the pending dirtying
3194 if (offset == 0 && length == PAGE_SIZE)
3195 ClearPageChecked(page);
3197 return jbd2_journal_invalidatepage(journal, page, offset, length);
3200 /* Wrapper for aops... */
3201 static void ext4_journalled_invalidatepage(struct page *page,
3202 unsigned int offset,
3203 unsigned int length)
3205 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3208 static int ext4_releasepage(struct page *page, gfp_t wait)
3210 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3212 trace_ext4_releasepage(page);
3214 /* Page has dirty journalled data -> cannot release */
3215 if (PageChecked(page))
3218 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3220 return try_to_free_buffers(page);
3223 #ifdef CONFIG_FS_DAX
3224 int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
3225 struct buffer_head *bh_result, int create)
3229 struct ext4_map_blocks map;
3230 handle_t *handle = NULL;
3233 ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n",
3234 inode->i_ino, create);
3235 map.m_lblk = iblock;
3236 map.m_len = bh_result->b_size >> inode->i_blkbits;
3237 credits = ext4_chunk_trans_blocks(inode, map.m_len);
3239 flags |= EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_CREATE_ZERO;
3240 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
3241 if (IS_ERR(handle)) {
3242 ret = PTR_ERR(handle);
3247 ret = ext4_map_blocks(handle, inode, &map, flags);
3249 err = ext4_journal_stop(handle);
3250 if (ret >= 0 && err < 0)
3255 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3259 * We are protected by i_mmap_sem so we know block cannot go
3260 * away from under us even though we dropped i_data_sem.
3261 * Convert extent to written and write zeros there.
3263 * Note: We may get here even when create == 0.
3265 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
3266 if (IS_ERR(handle)) {
3267 ret = PTR_ERR(handle);
3271 err = ext4_map_blocks(handle, inode, &map,
3272 EXT4_GET_BLOCKS_CONVERT | EXT4_GET_BLOCKS_CREATE_ZERO);
3275 err2 = ext4_journal_stop(handle);
3276 if (err2 < 0 && ret > 0)
3280 WARN_ON_ONCE(ret == 0 && create);
3282 map_bh(bh_result, inode->i_sb, map.m_pblk);
3284 * At least for now we have to clear BH_New so that DAX code
3285 * doesn't attempt to zero blocks again in a racy way.
3287 map.m_flags &= ~EXT4_MAP_NEW;
3288 ext4_update_bh_state(bh_result, map.m_flags);
3289 bh_result->b_size = map.m_len << inode->i_blkbits;
3296 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3297 ssize_t size, void *private)
3299 ext4_io_end_t *io_end = private;
3301 /* if not async direct IO just return */
3305 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3306 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3307 io_end, io_end->inode->i_ino, iocb, offset, size);
3310 * Error during AIO DIO. We cannot convert unwritten extents as the
3311 * data was not written. Just clear the unwritten flag and drop io_end.
3314 ext4_clear_io_unwritten_flag(io_end);
3317 io_end->offset = offset;
3318 io_end->size = size;
3319 ext4_put_io_end(io_end);
3325 * For ext4 extent files, ext4 will do direct-io write to holes,
3326 * preallocated extents, and those write extend the file, no need to
3327 * fall back to buffered IO.
3329 * For holes, we fallocate those blocks, mark them as unwritten
3330 * If those blocks were preallocated, we mark sure they are split, but
3331 * still keep the range to write as unwritten.
3333 * The unwritten extents will be converted to written when DIO is completed.
3334 * For async direct IO, since the IO may still pending when return, we
3335 * set up an end_io call back function, which will do the conversion
3336 * when async direct IO completed.
3338 * If the O_DIRECT write will extend the file then add this inode to the
3339 * orphan list. So recovery will truncate it back to the original size
3340 * if the machine crashes during the write.
3343 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3346 struct file *file = iocb->ki_filp;
3347 struct inode *inode = file->f_mapping->host;
3349 size_t count = iov_iter_count(iter);
3351 get_block_t *get_block_func = NULL;
3353 loff_t final_size = offset + count;
3355 /* Use the old path for reads and writes beyond i_size. */
3356 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3357 return ext4_ind_direct_IO(iocb, iter, offset);
3359 BUG_ON(iocb->private == NULL);
3362 * Make all waiters for direct IO properly wait also for extent
3363 * conversion. This also disallows race between truncate() and
3364 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3366 if (iov_iter_rw(iter) == WRITE)
3367 inode_dio_begin(inode);
3369 /* If we do a overwrite dio, i_mutex locking can be released */
3370 overwrite = *((int *)iocb->private);
3373 inode_unlock(inode);
3376 * We could direct write to holes and fallocate.
3378 * Allocated blocks to fill the hole are marked as unwritten to prevent
3379 * parallel buffered read to expose the stale data before DIO complete
3382 * As to previously fallocated extents, ext4 get_block will just simply
3383 * mark the buffer mapped but still keep the extents unwritten.
3385 * For non AIO case, we will convert those unwritten extents to written
3386 * after return back from blockdev_direct_IO. That way we save us from
3387 * allocating io_end structure and also the overhead of offloading
3388 * the extent convertion to a workqueue.
3390 * For async DIO, the conversion needs to be deferred when the
3391 * IO is completed. The ext4 end_io callback function will be
3392 * called to take care of the conversion work. Here for async
3393 * case, we allocate an io_end structure to hook to the iocb.
3395 iocb->private = NULL;
3397 get_block_func = ext4_dio_get_block_overwrite;
3398 else if (is_sync_kiocb(iocb)) {
3399 get_block_func = ext4_dio_get_block_unwritten_sync;
3400 dio_flags = DIO_LOCKING;
3402 get_block_func = ext4_dio_get_block_unwritten_async;
3403 dio_flags = DIO_LOCKING;
3405 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3406 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3409 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3410 ext4_end_io_dio, dio_flags);
3412 ret = __blockdev_direct_IO(iocb, inode,
3413 inode->i_sb->s_bdev, iter, offset,
3415 ext4_end_io_dio, NULL, dio_flags);
3417 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3418 EXT4_STATE_DIO_UNWRITTEN)) {
3421 * for non AIO case, since the IO is already
3422 * completed, we could do the conversion right here
3424 err = ext4_convert_unwritten_extents(NULL, inode,
3428 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3431 if (iov_iter_rw(iter) == WRITE)
3432 inode_dio_end(inode);
3433 /* take i_mutex locking again if we do a ovewrite dio */
3440 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3443 struct file *file = iocb->ki_filp;
3444 struct inode *inode = file->f_mapping->host;
3445 size_t count = iov_iter_count(iter);
3448 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3449 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3454 * If we are doing data journalling we don't support O_DIRECT
3456 if (ext4_should_journal_data(inode))
3459 /* Let buffer I/O handle the inline data case. */
3460 if (ext4_has_inline_data(inode))
3463 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3464 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3465 ret = ext4_ext_direct_IO(iocb, iter, offset);
3467 ret = ext4_ind_direct_IO(iocb, iter, offset);
3468 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3473 * Pages can be marked dirty completely asynchronously from ext4's journalling
3474 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3475 * much here because ->set_page_dirty is called under VFS locks. The page is
3476 * not necessarily locked.
3478 * We cannot just dirty the page and leave attached buffers clean, because the
3479 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3480 * or jbddirty because all the journalling code will explode.
3482 * So what we do is to mark the page "pending dirty" and next time writepage
3483 * is called, propagate that into the buffers appropriately.
3485 static int ext4_journalled_set_page_dirty(struct page *page)
3487 SetPageChecked(page);
3488 return __set_page_dirty_nobuffers(page);
3491 static const struct address_space_operations ext4_aops = {
3492 .readpage = ext4_readpage,
3493 .readpages = ext4_readpages,
3494 .writepage = ext4_writepage,
3495 .writepages = ext4_writepages,
3496 .write_begin = ext4_write_begin,
3497 .write_end = ext4_write_end,
3499 .invalidatepage = ext4_invalidatepage,
3500 .releasepage = ext4_releasepage,
3501 .direct_IO = ext4_direct_IO,
3502 .migratepage = buffer_migrate_page,
3503 .is_partially_uptodate = block_is_partially_uptodate,
3504 .error_remove_page = generic_error_remove_page,
3507 static const struct address_space_operations ext4_journalled_aops = {
3508 .readpage = ext4_readpage,
3509 .readpages = ext4_readpages,
3510 .writepage = ext4_writepage,
3511 .writepages = ext4_writepages,
3512 .write_begin = ext4_write_begin,
3513 .write_end = ext4_journalled_write_end,
3514 .set_page_dirty = ext4_journalled_set_page_dirty,
3516 .invalidatepage = ext4_journalled_invalidatepage,
3517 .releasepage = ext4_releasepage,
3518 .direct_IO = ext4_direct_IO,
3519 .is_partially_uptodate = block_is_partially_uptodate,
3520 .error_remove_page = generic_error_remove_page,
3523 static const struct address_space_operations ext4_da_aops = {
3524 .readpage = ext4_readpage,
3525 .readpages = ext4_readpages,
3526 .writepage = ext4_writepage,
3527 .writepages = ext4_writepages,
3528 .write_begin = ext4_da_write_begin,
3529 .write_end = ext4_da_write_end,
3531 .invalidatepage = ext4_da_invalidatepage,
3532 .releasepage = ext4_releasepage,
3533 .direct_IO = ext4_direct_IO,
3534 .migratepage = buffer_migrate_page,
3535 .is_partially_uptodate = block_is_partially_uptodate,
3536 .error_remove_page = generic_error_remove_page,
3539 void ext4_set_aops(struct inode *inode)
3541 switch (ext4_inode_journal_mode(inode)) {
3542 case EXT4_INODE_ORDERED_DATA_MODE:
3543 case EXT4_INODE_WRITEBACK_DATA_MODE:
3545 case EXT4_INODE_JOURNAL_DATA_MODE:
3546 inode->i_mapping->a_ops = &ext4_journalled_aops;
3551 if (test_opt(inode->i_sb, DELALLOC))
3552 inode->i_mapping->a_ops = &ext4_da_aops;
3554 inode->i_mapping->a_ops = &ext4_aops;
3557 static int __ext4_block_zero_page_range(handle_t *handle,
3558 struct address_space *mapping, loff_t from, loff_t length)
3560 ext4_fsblk_t index = from >> PAGE_SHIFT;
3561 unsigned offset = from & (PAGE_SIZE-1);
3562 unsigned blocksize, pos;
3564 struct inode *inode = mapping->host;
3565 struct buffer_head *bh;
3569 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3570 mapping_gfp_constraint(mapping, ~__GFP_FS));
3574 blocksize = inode->i_sb->s_blocksize;
3576 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3578 if (!page_has_buffers(page))
3579 create_empty_buffers(page, blocksize, 0);
3581 /* Find the buffer that contains "offset" */
3582 bh = page_buffers(page);
3584 while (offset >= pos) {
3585 bh = bh->b_this_page;
3589 if (buffer_freed(bh)) {
3590 BUFFER_TRACE(bh, "freed: skip");
3593 if (!buffer_mapped(bh)) {
3594 BUFFER_TRACE(bh, "unmapped");
3595 ext4_get_block(inode, iblock, bh, 0);
3596 /* unmapped? It's a hole - nothing to do */
3597 if (!buffer_mapped(bh)) {
3598 BUFFER_TRACE(bh, "still unmapped");
3603 /* Ok, it's mapped. Make sure it's up-to-date */
3604 if (PageUptodate(page))
3605 set_buffer_uptodate(bh);
3607 if (!buffer_uptodate(bh)) {
3609 ll_rw_block(READ, 1, &bh);
3611 /* Uhhuh. Read error. Complain and punt. */
3612 if (!buffer_uptodate(bh))
3614 if (S_ISREG(inode->i_mode) &&
3615 ext4_encrypted_inode(inode)) {
3616 /* We expect the key to be set. */
3617 BUG_ON(!ext4_has_encryption_key(inode));
3618 BUG_ON(blocksize != PAGE_SIZE);
3619 WARN_ON_ONCE(ext4_decrypt(page));
3622 if (ext4_should_journal_data(inode)) {
3623 BUFFER_TRACE(bh, "get write access");
3624 err = ext4_journal_get_write_access(handle, bh);
3628 zero_user(page, offset, length);
3629 BUFFER_TRACE(bh, "zeroed end of block");
3631 if (ext4_should_journal_data(inode)) {
3632 err = ext4_handle_dirty_metadata(handle, inode, bh);
3635 mark_buffer_dirty(bh);
3636 if (ext4_should_order_data(inode))
3637 err = ext4_jbd2_file_inode(handle, inode);
3647 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3648 * starting from file offset 'from'. The range to be zero'd must
3649 * be contained with in one block. If the specified range exceeds
3650 * the end of the block it will be shortened to end of the block
3651 * that cooresponds to 'from'
3653 static int ext4_block_zero_page_range(handle_t *handle,
3654 struct address_space *mapping, loff_t from, loff_t length)
3656 struct inode *inode = mapping->host;
3657 unsigned offset = from & (PAGE_SIZE-1);
3658 unsigned blocksize = inode->i_sb->s_blocksize;
3659 unsigned max = blocksize - (offset & (blocksize - 1));
3662 * correct length if it does not fall between
3663 * 'from' and the end of the block
3665 if (length > max || length < 0)
3669 return dax_zero_page_range(inode, from, length, ext4_get_block);
3670 return __ext4_block_zero_page_range(handle, mapping, from, length);
3674 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3675 * up to the end of the block which corresponds to `from'.
3676 * This required during truncate. We need to physically zero the tail end
3677 * of that block so it doesn't yield old data if the file is later grown.
3679 static int ext4_block_truncate_page(handle_t *handle,
3680 struct address_space *mapping, loff_t from)
3682 unsigned offset = from & (PAGE_SIZE-1);
3685 struct inode *inode = mapping->host;
3687 blocksize = inode->i_sb->s_blocksize;
3688 length = blocksize - (offset & (blocksize - 1));
3690 return ext4_block_zero_page_range(handle, mapping, from, length);
3693 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3694 loff_t lstart, loff_t length)
3696 struct super_block *sb = inode->i_sb;
3697 struct address_space *mapping = inode->i_mapping;
3698 unsigned partial_start, partial_end;
3699 ext4_fsblk_t start, end;
3700 loff_t byte_end = (lstart + length - 1);
3703 partial_start = lstart & (sb->s_blocksize - 1);
3704 partial_end = byte_end & (sb->s_blocksize - 1);
3706 start = lstart >> sb->s_blocksize_bits;
3707 end = byte_end >> sb->s_blocksize_bits;
3709 /* Handle partial zero within the single block */
3711 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3712 err = ext4_block_zero_page_range(handle, mapping,
3716 /* Handle partial zero out on the start of the range */
3717 if (partial_start) {
3718 err = ext4_block_zero_page_range(handle, mapping,
3719 lstart, sb->s_blocksize);
3723 /* Handle partial zero out on the end of the range */
3724 if (partial_end != sb->s_blocksize - 1)
3725 err = ext4_block_zero_page_range(handle, mapping,
3726 byte_end - partial_end,
3731 int ext4_can_truncate(struct inode *inode)
3733 if (S_ISREG(inode->i_mode))
3735 if (S_ISDIR(inode->i_mode))
3737 if (S_ISLNK(inode->i_mode))
3738 return !ext4_inode_is_fast_symlink(inode);
3743 * We have to make sure i_disksize gets properly updated before we truncate
3744 * page cache due to hole punching or zero range. Otherwise i_disksize update
3745 * can get lost as it may have been postponed to submission of writeback but
3746 * that will never happen after we truncate page cache.
3748 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3752 loff_t size = i_size_read(inode);
3754 WARN_ON(!inode_is_locked(inode));
3755 if (offset > size || offset + len < size)
3758 if (EXT4_I(inode)->i_disksize >= size)
3761 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3763 return PTR_ERR(handle);
3764 ext4_update_i_disksize(inode, size);
3765 ext4_mark_inode_dirty(handle, inode);
3766 ext4_journal_stop(handle);
3772 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3773 * associated with the given offset and length
3775 * @inode: File inode
3776 * @offset: The offset where the hole will begin
3777 * @len: The length of the hole
3779 * Returns: 0 on success or negative on failure
3782 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3784 struct super_block *sb = inode->i_sb;
3785 ext4_lblk_t first_block, stop_block;
3786 struct address_space *mapping = inode->i_mapping;
3787 loff_t first_block_offset, last_block_offset;
3789 unsigned int credits;
3792 if (!S_ISREG(inode->i_mode))
3795 trace_ext4_punch_hole(inode, offset, length, 0);
3798 * Write out all dirty pages to avoid race conditions
3799 * Then release them.
3801 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3802 ret = filemap_write_and_wait_range(mapping, offset,
3803 offset + length - 1);
3810 /* No need to punch hole beyond i_size */
3811 if (offset >= inode->i_size)
3815 * If the hole extends beyond i_size, set the hole
3816 * to end after the page that contains i_size
3818 if (offset + length > inode->i_size) {
3819 length = inode->i_size +
3820 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3824 if (offset & (sb->s_blocksize - 1) ||
3825 (offset + length) & (sb->s_blocksize - 1)) {
3827 * Attach jinode to inode for jbd2 if we do any zeroing of
3830 ret = ext4_inode_attach_jinode(inode);
3836 /* Wait all existing dio workers, newcomers will block on i_mutex */
3837 ext4_inode_block_unlocked_dio(inode);
3838 inode_dio_wait(inode);
3841 * Prevent page faults from reinstantiating pages we have released from
3844 down_write(&EXT4_I(inode)->i_mmap_sem);
3845 first_block_offset = round_up(offset, sb->s_blocksize);
3846 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3848 /* Now release the pages and zero block aligned part of pages*/
3849 if (last_block_offset > first_block_offset) {
3850 ret = ext4_update_disksize_before_punch(inode, offset, length);
3853 truncate_pagecache_range(inode, first_block_offset,
3857 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3858 credits = ext4_writepage_trans_blocks(inode);
3860 credits = ext4_blocks_for_truncate(inode);
3861 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3862 if (IS_ERR(handle)) {
3863 ret = PTR_ERR(handle);
3864 ext4_std_error(sb, ret);
3868 ret = ext4_zero_partial_blocks(handle, inode, offset,
3873 first_block = (offset + sb->s_blocksize - 1) >>
3874 EXT4_BLOCK_SIZE_BITS(sb);
3875 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3877 /* If there are no blocks to remove, return now */
3878 if (first_block >= stop_block)
3881 down_write(&EXT4_I(inode)->i_data_sem);
3882 ext4_discard_preallocations(inode);
3884 ret = ext4_es_remove_extent(inode, first_block,
3885 stop_block - first_block);
3887 up_write(&EXT4_I(inode)->i_data_sem);
3891 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3892 ret = ext4_ext_remove_space(inode, first_block,
3895 ret = ext4_ind_remove_space(handle, inode, first_block,
3898 up_write(&EXT4_I(inode)->i_data_sem);
3900 ext4_handle_sync(handle);
3902 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3903 ext4_mark_inode_dirty(handle, inode);
3905 ext4_journal_stop(handle);
3907 up_write(&EXT4_I(inode)->i_mmap_sem);
3908 ext4_inode_resume_unlocked_dio(inode);
3910 inode_unlock(inode);
3914 int ext4_inode_attach_jinode(struct inode *inode)
3916 struct ext4_inode_info *ei = EXT4_I(inode);
3917 struct jbd2_inode *jinode;
3919 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3922 jinode = jbd2_alloc_inode(GFP_KERNEL);
3923 spin_lock(&inode->i_lock);
3926 spin_unlock(&inode->i_lock);
3929 ei->jinode = jinode;
3930 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3933 spin_unlock(&inode->i_lock);
3934 if (unlikely(jinode != NULL))
3935 jbd2_free_inode(jinode);
3942 * We block out ext4_get_block() block instantiations across the entire
3943 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3944 * simultaneously on behalf of the same inode.
3946 * As we work through the truncate and commit bits of it to the journal there
3947 * is one core, guiding principle: the file's tree must always be consistent on
3948 * disk. We must be able to restart the truncate after a crash.
3950 * The file's tree may be transiently inconsistent in memory (although it
3951 * probably isn't), but whenever we close off and commit a journal transaction,
3952 * the contents of (the filesystem + the journal) must be consistent and
3953 * restartable. It's pretty simple, really: bottom up, right to left (although
3954 * left-to-right works OK too).
3956 * Note that at recovery time, journal replay occurs *before* the restart of
3957 * truncate against the orphan inode list.
3959 * The committed inode has the new, desired i_size (which is the same as
3960 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3961 * that this inode's truncate did not complete and it will again call
3962 * ext4_truncate() to have another go. So there will be instantiated blocks
3963 * to the right of the truncation point in a crashed ext4 filesystem. But
3964 * that's fine - as long as they are linked from the inode, the post-crash
3965 * ext4_truncate() run will find them and release them.
3967 void ext4_truncate(struct inode *inode)
3969 struct ext4_inode_info *ei = EXT4_I(inode);
3970 unsigned int credits;
3972 struct address_space *mapping = inode->i_mapping;
3975 * There is a possibility that we're either freeing the inode
3976 * or it's a completely new inode. In those cases we might not
3977 * have i_mutex locked because it's not necessary.
3979 if (!(inode->i_state & (I_NEW|I_FREEING)))
3980 WARN_ON(!inode_is_locked(inode));
3981 trace_ext4_truncate_enter(inode);
3983 if (!ext4_can_truncate(inode))
3986 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3988 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3989 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3991 if (ext4_has_inline_data(inode)) {
3994 ext4_inline_data_truncate(inode, &has_inline);
3999 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4000 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4001 if (ext4_inode_attach_jinode(inode) < 0)
4005 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4006 credits = ext4_writepage_trans_blocks(inode);
4008 credits = ext4_blocks_for_truncate(inode);
4010 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4011 if (IS_ERR(handle)) {
4012 ext4_std_error(inode->i_sb, PTR_ERR(handle));
4016 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4017 ext4_block_truncate_page(handle, mapping, inode->i_size);
4020 * We add the inode to the orphan list, so that if this
4021 * truncate spans multiple transactions, and we crash, we will
4022 * resume the truncate when the filesystem recovers. It also
4023 * marks the inode dirty, to catch the new size.
4025 * Implication: the file must always be in a sane, consistent
4026 * truncatable state while each transaction commits.
4028 if (ext4_orphan_add(handle, inode))
4031 down_write(&EXT4_I(inode)->i_data_sem);
4033 ext4_discard_preallocations(inode);
4035 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4036 ext4_ext_truncate(handle, inode);
4038 ext4_ind_truncate(handle, inode);
4040 up_write(&ei->i_data_sem);
4043 ext4_handle_sync(handle);
4047 * If this was a simple ftruncate() and the file will remain alive,
4048 * then we need to clear up the orphan record which we created above.
4049 * However, if this was a real unlink then we were called by
4050 * ext4_evict_inode(), and we allow that function to clean up the
4051 * orphan info for us.
4054 ext4_orphan_del(handle, inode);
4056 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4057 ext4_mark_inode_dirty(handle, inode);
4058 ext4_journal_stop(handle);
4060 trace_ext4_truncate_exit(inode);
4064 * ext4_get_inode_loc returns with an extra refcount against the inode's
4065 * underlying buffer_head on success. If 'in_mem' is true, we have all
4066 * data in memory that is needed to recreate the on-disk version of this
4069 static int __ext4_get_inode_loc(struct inode *inode,
4070 struct ext4_iloc *iloc, int in_mem)
4072 struct ext4_group_desc *gdp;
4073 struct buffer_head *bh;
4074 struct super_block *sb = inode->i_sb;
4076 int inodes_per_block, inode_offset;
4079 if (!ext4_valid_inum(sb, inode->i_ino))
4080 return -EFSCORRUPTED;
4082 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4083 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4088 * Figure out the offset within the block group inode table
4090 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4091 inode_offset = ((inode->i_ino - 1) %
4092 EXT4_INODES_PER_GROUP(sb));
4093 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4094 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4096 bh = sb_getblk(sb, block);
4099 if (!buffer_uptodate(bh)) {
4103 * If the buffer has the write error flag, we have failed
4104 * to write out another inode in the same block. In this
4105 * case, we don't have to read the block because we may
4106 * read the old inode data successfully.
4108 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4109 set_buffer_uptodate(bh);
4111 if (buffer_uptodate(bh)) {
4112 /* someone brought it uptodate while we waited */
4118 * If we have all information of the inode in memory and this
4119 * is the only valid inode in the block, we need not read the
4123 struct buffer_head *bitmap_bh;
4126 start = inode_offset & ~(inodes_per_block - 1);
4128 /* Is the inode bitmap in cache? */
4129 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4130 if (unlikely(!bitmap_bh))
4134 * If the inode bitmap isn't in cache then the
4135 * optimisation may end up performing two reads instead
4136 * of one, so skip it.
4138 if (!buffer_uptodate(bitmap_bh)) {
4142 for (i = start; i < start + inodes_per_block; i++) {
4143 if (i == inode_offset)
4145 if (ext4_test_bit(i, bitmap_bh->b_data))
4149 if (i == start + inodes_per_block) {
4150 /* all other inodes are free, so skip I/O */
4151 memset(bh->b_data, 0, bh->b_size);
4152 set_buffer_uptodate(bh);
4160 * If we need to do any I/O, try to pre-readahead extra
4161 * blocks from the inode table.
4163 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4164 ext4_fsblk_t b, end, table;
4166 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4168 table = ext4_inode_table(sb, gdp);
4169 /* s_inode_readahead_blks is always a power of 2 */
4170 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4174 num = EXT4_INODES_PER_GROUP(sb);
4175 if (ext4_has_group_desc_csum(sb))
4176 num -= ext4_itable_unused_count(sb, gdp);
4177 table += num / inodes_per_block;
4181 sb_breadahead(sb, b++);
4185 * There are other valid inodes in the buffer, this inode
4186 * has in-inode xattrs, or we don't have this inode in memory.
4187 * Read the block from disk.
4189 trace_ext4_load_inode(inode);
4191 bh->b_end_io = end_buffer_read_sync;
4192 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4194 if (!buffer_uptodate(bh)) {
4195 EXT4_ERROR_INODE_BLOCK(inode, block,
4196 "unable to read itable block");
4206 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4208 /* We have all inode data except xattrs in memory here. */
4209 return __ext4_get_inode_loc(inode, iloc,
4210 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4213 void ext4_set_inode_flags(struct inode *inode)
4215 unsigned int flags = EXT4_I(inode)->i_flags;
4216 unsigned int new_fl = 0;
4218 if (flags & EXT4_SYNC_FL)
4220 if (flags & EXT4_APPEND_FL)
4222 if (flags & EXT4_IMMUTABLE_FL)
4223 new_fl |= S_IMMUTABLE;
4224 if (flags & EXT4_NOATIME_FL)
4225 new_fl |= S_NOATIME;
4226 if (flags & EXT4_DIRSYNC_FL)
4227 new_fl |= S_DIRSYNC;
4228 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
4230 inode_set_flags(inode, new_fl,
4231 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4234 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4235 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4237 unsigned int vfs_fl;
4238 unsigned long old_fl, new_fl;
4241 vfs_fl = ei->vfs_inode.i_flags;
4242 old_fl = ei->i_flags;
4243 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4244 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4246 if (vfs_fl & S_SYNC)
4247 new_fl |= EXT4_SYNC_FL;
4248 if (vfs_fl & S_APPEND)
4249 new_fl |= EXT4_APPEND_FL;
4250 if (vfs_fl & S_IMMUTABLE)
4251 new_fl |= EXT4_IMMUTABLE_FL;
4252 if (vfs_fl & S_NOATIME)
4253 new_fl |= EXT4_NOATIME_FL;
4254 if (vfs_fl & S_DIRSYNC)
4255 new_fl |= EXT4_DIRSYNC_FL;
4256 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4259 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4260 struct ext4_inode_info *ei)
4263 struct inode *inode = &(ei->vfs_inode);
4264 struct super_block *sb = inode->i_sb;
4266 if (ext4_has_feature_huge_file(sb)) {
4267 /* we are using combined 48 bit field */
4268 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4269 le32_to_cpu(raw_inode->i_blocks_lo);
4270 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4271 /* i_blocks represent file system block size */
4272 return i_blocks << (inode->i_blkbits - 9);
4277 return le32_to_cpu(raw_inode->i_blocks_lo);
4281 static inline void ext4_iget_extra_inode(struct inode *inode,
4282 struct ext4_inode *raw_inode,
4283 struct ext4_inode_info *ei)
4285 __le32 *magic = (void *)raw_inode +
4286 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4287 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4288 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4289 ext4_find_inline_data_nolock(inode);
4291 EXT4_I(inode)->i_inline_off = 0;
4294 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4296 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, EXT4_FEATURE_RO_COMPAT_PROJECT))
4298 *projid = EXT4_I(inode)->i_projid;
4302 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4304 struct ext4_iloc iloc;
4305 struct ext4_inode *raw_inode;
4306 struct ext4_inode_info *ei;
4307 struct inode *inode;
4308 journal_t *journal = EXT4_SB(sb)->s_journal;
4315 inode = iget_locked(sb, ino);
4317 return ERR_PTR(-ENOMEM);
4318 if (!(inode->i_state & I_NEW))
4324 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4327 raw_inode = ext4_raw_inode(&iloc);
4329 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4330 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4331 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4332 EXT4_INODE_SIZE(inode->i_sb)) {
4333 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4334 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4335 EXT4_INODE_SIZE(inode->i_sb));
4336 ret = -EFSCORRUPTED;
4340 ei->i_extra_isize = 0;
4342 /* Precompute checksum seed for inode metadata */
4343 if (ext4_has_metadata_csum(sb)) {
4344 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4346 __le32 inum = cpu_to_le32(inode->i_ino);
4347 __le32 gen = raw_inode->i_generation;
4348 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4350 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4354 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4355 EXT4_ERROR_INODE(inode, "checksum invalid");
4360 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4361 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4362 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4363 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4364 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4365 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4366 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4368 i_projid = EXT4_DEF_PROJID;
4370 if (!(test_opt(inode->i_sb, NO_UID32))) {
4371 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4372 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4374 i_uid_write(inode, i_uid);
4375 i_gid_write(inode, i_gid);
4376 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4377 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4379 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4380 ei->i_inline_off = 0;
4381 ei->i_dir_start_lookup = 0;
4382 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4383 /* We now have enough fields to check if the inode was active or not.
4384 * This is needed because nfsd might try to access dead inodes
4385 * the test is that same one that e2fsck uses
4386 * NeilBrown 1999oct15
4388 if (inode->i_nlink == 0) {
4389 if ((inode->i_mode == 0 ||
4390 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4391 ino != EXT4_BOOT_LOADER_INO) {
4392 /* this inode is deleted */
4396 /* The only unlinked inodes we let through here have
4397 * valid i_mode and are being read by the orphan
4398 * recovery code: that's fine, we're about to complete
4399 * the process of deleting those.
4400 * OR it is the EXT4_BOOT_LOADER_INO which is
4401 * not initialized on a new filesystem. */
4403 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4404 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4405 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4406 if (ext4_has_feature_64bit(sb))
4408 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4409 inode->i_size = ext4_isize(raw_inode);
4410 ei->i_disksize = inode->i_size;
4412 ei->i_reserved_quota = 0;
4414 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4415 ei->i_block_group = iloc.block_group;
4416 ei->i_last_alloc_group = ~0;
4418 * NOTE! The in-memory inode i_data array is in little-endian order
4419 * even on big-endian machines: we do NOT byteswap the block numbers!
4421 for (block = 0; block < EXT4_N_BLOCKS; block++)
4422 ei->i_data[block] = raw_inode->i_block[block];
4423 INIT_LIST_HEAD(&ei->i_orphan);
4426 * Set transaction id's of transactions that have to be committed
4427 * to finish f[data]sync. We set them to currently running transaction
4428 * as we cannot be sure that the inode or some of its metadata isn't
4429 * part of the transaction - the inode could have been reclaimed and
4430 * now it is reread from disk.
4433 transaction_t *transaction;
4436 read_lock(&journal->j_state_lock);
4437 if (journal->j_running_transaction)
4438 transaction = journal->j_running_transaction;
4440 transaction = journal->j_committing_transaction;
4442 tid = transaction->t_tid;
4444 tid = journal->j_commit_sequence;
4445 read_unlock(&journal->j_state_lock);
4446 ei->i_sync_tid = tid;
4447 ei->i_datasync_tid = tid;
4450 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4451 if (ei->i_extra_isize == 0) {
4452 /* The extra space is currently unused. Use it. */
4453 ei->i_extra_isize = sizeof(struct ext4_inode) -
4454 EXT4_GOOD_OLD_INODE_SIZE;
4456 ext4_iget_extra_inode(inode, raw_inode, ei);
4460 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4461 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4462 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4463 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4465 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4466 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4467 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4468 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4470 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4475 if (ei->i_file_acl &&
4476 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4477 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4479 ret = -EFSCORRUPTED;
4481 } else if (!ext4_has_inline_data(inode)) {
4482 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4483 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4484 (S_ISLNK(inode->i_mode) &&
4485 !ext4_inode_is_fast_symlink(inode))))
4486 /* Validate extent which is part of inode */
4487 ret = ext4_ext_check_inode(inode);
4488 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4489 (S_ISLNK(inode->i_mode) &&
4490 !ext4_inode_is_fast_symlink(inode))) {
4491 /* Validate block references which are part of inode */
4492 ret = ext4_ind_check_inode(inode);
4498 if (S_ISREG(inode->i_mode)) {
4499 inode->i_op = &ext4_file_inode_operations;
4500 inode->i_fop = &ext4_file_operations;
4501 ext4_set_aops(inode);
4502 } else if (S_ISDIR(inode->i_mode)) {
4503 inode->i_op = &ext4_dir_inode_operations;
4504 inode->i_fop = &ext4_dir_operations;
4505 } else if (S_ISLNK(inode->i_mode)) {
4506 if (ext4_encrypted_inode(inode)) {
4507 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4508 ext4_set_aops(inode);
4509 } else if (ext4_inode_is_fast_symlink(inode)) {
4510 inode->i_link = (char *)ei->i_data;
4511 inode->i_op = &ext4_fast_symlink_inode_operations;
4512 nd_terminate_link(ei->i_data, inode->i_size,
4513 sizeof(ei->i_data) - 1);
4515 inode->i_op = &ext4_symlink_inode_operations;
4516 ext4_set_aops(inode);
4518 inode_nohighmem(inode);
4519 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4520 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4521 inode->i_op = &ext4_special_inode_operations;
4522 if (raw_inode->i_block[0])
4523 init_special_inode(inode, inode->i_mode,
4524 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4526 init_special_inode(inode, inode->i_mode,
4527 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4528 } else if (ino == EXT4_BOOT_LOADER_INO) {
4529 make_bad_inode(inode);
4531 ret = -EFSCORRUPTED;
4532 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4536 ext4_set_inode_flags(inode);
4537 unlock_new_inode(inode);
4543 return ERR_PTR(ret);
4546 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4548 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4549 return ERR_PTR(-EFSCORRUPTED);
4550 return ext4_iget(sb, ino);
4553 static int ext4_inode_blocks_set(handle_t *handle,
4554 struct ext4_inode *raw_inode,
4555 struct ext4_inode_info *ei)
4557 struct inode *inode = &(ei->vfs_inode);
4558 u64 i_blocks = inode->i_blocks;
4559 struct super_block *sb = inode->i_sb;
4561 if (i_blocks <= ~0U) {
4563 * i_blocks can be represented in a 32 bit variable
4564 * as multiple of 512 bytes
4566 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4567 raw_inode->i_blocks_high = 0;
4568 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4571 if (!ext4_has_feature_huge_file(sb))
4574 if (i_blocks <= 0xffffffffffffULL) {
4576 * i_blocks can be represented in a 48 bit variable
4577 * as multiple of 512 bytes
4579 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4580 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4581 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4583 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4584 /* i_block is stored in file system block size */
4585 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4586 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4587 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4592 struct other_inode {
4593 unsigned long orig_ino;
4594 struct ext4_inode *raw_inode;
4597 static int other_inode_match(struct inode * inode, unsigned long ino,
4600 struct other_inode *oi = (struct other_inode *) data;
4602 if ((inode->i_ino != ino) ||
4603 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4604 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4605 ((inode->i_state & I_DIRTY_TIME) == 0))
4607 spin_lock(&inode->i_lock);
4608 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4609 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4610 (inode->i_state & I_DIRTY_TIME)) {
4611 struct ext4_inode_info *ei = EXT4_I(inode);
4613 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4614 spin_unlock(&inode->i_lock);
4616 spin_lock(&ei->i_raw_lock);
4617 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4618 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4619 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4620 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4621 spin_unlock(&ei->i_raw_lock);
4622 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4625 spin_unlock(&inode->i_lock);
4630 * Opportunistically update the other time fields for other inodes in
4631 * the same inode table block.
4633 static void ext4_update_other_inodes_time(struct super_block *sb,
4634 unsigned long orig_ino, char *buf)
4636 struct other_inode oi;
4638 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4639 int inode_size = EXT4_INODE_SIZE(sb);
4641 oi.orig_ino = orig_ino;
4643 * Calculate the first inode in the inode table block. Inode
4644 * numbers are one-based. That is, the first inode in a block
4645 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4647 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4648 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4649 if (ino == orig_ino)
4651 oi.raw_inode = (struct ext4_inode *) buf;
4652 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4657 * Post the struct inode info into an on-disk inode location in the
4658 * buffer-cache. This gobbles the caller's reference to the
4659 * buffer_head in the inode location struct.
4661 * The caller must have write access to iloc->bh.
4663 static int ext4_do_update_inode(handle_t *handle,
4664 struct inode *inode,
4665 struct ext4_iloc *iloc)
4667 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4668 struct ext4_inode_info *ei = EXT4_I(inode);
4669 struct buffer_head *bh = iloc->bh;
4670 struct super_block *sb = inode->i_sb;
4671 int err = 0, rc, block;
4672 int need_datasync = 0, set_large_file = 0;
4677 spin_lock(&ei->i_raw_lock);
4679 /* For fields not tracked in the in-memory inode,
4680 * initialise them to zero for new inodes. */
4681 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4682 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4684 ext4_get_inode_flags(ei);
4685 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4686 i_uid = i_uid_read(inode);
4687 i_gid = i_gid_read(inode);
4688 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4689 if (!(test_opt(inode->i_sb, NO_UID32))) {
4690 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4691 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4693 * Fix up interoperability with old kernels. Otherwise, old inodes get
4694 * re-used with the upper 16 bits of the uid/gid intact
4697 raw_inode->i_uid_high =
4698 cpu_to_le16(high_16_bits(i_uid));
4699 raw_inode->i_gid_high =
4700 cpu_to_le16(high_16_bits(i_gid));
4702 raw_inode->i_uid_high = 0;
4703 raw_inode->i_gid_high = 0;
4706 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4707 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4708 raw_inode->i_uid_high = 0;
4709 raw_inode->i_gid_high = 0;
4711 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4713 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4714 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4715 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4716 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4718 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4720 spin_unlock(&ei->i_raw_lock);
4723 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4724 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4725 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4726 raw_inode->i_file_acl_high =
4727 cpu_to_le16(ei->i_file_acl >> 32);
4728 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4729 if (ei->i_disksize != ext4_isize(raw_inode)) {
4730 ext4_isize_set(raw_inode, ei->i_disksize);
4733 if (ei->i_disksize > 0x7fffffffULL) {
4734 if (!ext4_has_feature_large_file(sb) ||
4735 EXT4_SB(sb)->s_es->s_rev_level ==
4736 cpu_to_le32(EXT4_GOOD_OLD_REV))
4739 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4740 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4741 if (old_valid_dev(inode->i_rdev)) {
4742 raw_inode->i_block[0] =
4743 cpu_to_le32(old_encode_dev(inode->i_rdev));
4744 raw_inode->i_block[1] = 0;
4746 raw_inode->i_block[0] = 0;
4747 raw_inode->i_block[1] =
4748 cpu_to_le32(new_encode_dev(inode->i_rdev));
4749 raw_inode->i_block[2] = 0;
4751 } else if (!ext4_has_inline_data(inode)) {
4752 for (block = 0; block < EXT4_N_BLOCKS; block++)
4753 raw_inode->i_block[block] = ei->i_data[block];
4756 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4757 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4758 if (ei->i_extra_isize) {
4759 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4760 raw_inode->i_version_hi =
4761 cpu_to_le32(inode->i_version >> 32);
4762 raw_inode->i_extra_isize =
4763 cpu_to_le16(ei->i_extra_isize);
4767 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
4768 EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4769 i_projid != EXT4_DEF_PROJID);
4771 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4772 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4773 raw_inode->i_projid = cpu_to_le32(i_projid);
4775 ext4_inode_csum_set(inode, raw_inode, ei);
4776 spin_unlock(&ei->i_raw_lock);
4777 if (inode->i_sb->s_flags & MS_LAZYTIME)
4778 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4781 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4782 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4785 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4786 if (set_large_file) {
4787 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4788 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4791 ext4_update_dynamic_rev(sb);
4792 ext4_set_feature_large_file(sb);
4793 ext4_handle_sync(handle);
4794 err = ext4_handle_dirty_super(handle, sb);
4796 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4799 ext4_std_error(inode->i_sb, err);
4804 * ext4_write_inode()
4806 * We are called from a few places:
4808 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4809 * Here, there will be no transaction running. We wait for any running
4810 * transaction to commit.
4812 * - Within flush work (sys_sync(), kupdate and such).
4813 * We wait on commit, if told to.
4815 * - Within iput_final() -> write_inode_now()
4816 * We wait on commit, if told to.
4818 * In all cases it is actually safe for us to return without doing anything,
4819 * because the inode has been copied into a raw inode buffer in
4820 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4823 * Note that we are absolutely dependent upon all inode dirtiers doing the
4824 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4825 * which we are interested.
4827 * It would be a bug for them to not do this. The code:
4829 * mark_inode_dirty(inode)
4831 * inode->i_size = expr;
4833 * is in error because write_inode() could occur while `stuff()' is running,
4834 * and the new i_size will be lost. Plus the inode will no longer be on the
4835 * superblock's dirty inode list.
4837 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4841 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4844 if (EXT4_SB(inode->i_sb)->s_journal) {
4845 if (ext4_journal_current_handle()) {
4846 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4852 * No need to force transaction in WB_SYNC_NONE mode. Also
4853 * ext4_sync_fs() will force the commit after everything is
4856 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4859 err = ext4_force_commit(inode->i_sb);
4861 struct ext4_iloc iloc;
4863 err = __ext4_get_inode_loc(inode, &iloc, 0);
4867 * sync(2) will flush the whole buffer cache. No need to do
4868 * it here separately for each inode.
4870 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4871 sync_dirty_buffer(iloc.bh);
4872 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4873 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4874 "IO error syncing inode");
4883 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4884 * buffers that are attached to a page stradding i_size and are undergoing
4885 * commit. In that case we have to wait for commit to finish and try again.
4887 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4891 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4892 tid_t commit_tid = 0;
4895 offset = inode->i_size & (PAGE_SIZE - 1);
4897 * All buffers in the last page remain valid? Then there's nothing to
4898 * do. We do the check mainly to optimize the common PAGE_SIZE ==
4901 if (offset > PAGE_SIZE - (1 << inode->i_blkbits))
4904 page = find_lock_page(inode->i_mapping,
4905 inode->i_size >> PAGE_SHIFT);
4908 ret = __ext4_journalled_invalidatepage(page, offset,
4909 PAGE_SIZE - offset);
4915 read_lock(&journal->j_state_lock);
4916 if (journal->j_committing_transaction)
4917 commit_tid = journal->j_committing_transaction->t_tid;
4918 read_unlock(&journal->j_state_lock);
4920 jbd2_log_wait_commit(journal, commit_tid);
4927 * Called from notify_change.
4929 * We want to trap VFS attempts to truncate the file as soon as
4930 * possible. In particular, we want to make sure that when the VFS
4931 * shrinks i_size, we put the inode on the orphan list and modify
4932 * i_disksize immediately, so that during the subsequent flushing of
4933 * dirty pages and freeing of disk blocks, we can guarantee that any
4934 * commit will leave the blocks being flushed in an unused state on
4935 * disk. (On recovery, the inode will get truncated and the blocks will
4936 * be freed, so we have a strong guarantee that no future commit will
4937 * leave these blocks visible to the user.)
4939 * Another thing we have to assure is that if we are in ordered mode
4940 * and inode is still attached to the committing transaction, we must
4941 * we start writeout of all the dirty pages which are being truncated.
4942 * This way we are sure that all the data written in the previous
4943 * transaction are already on disk (truncate waits for pages under
4946 * Called with inode->i_mutex down.
4948 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4950 struct inode *inode = d_inode(dentry);
4953 const unsigned int ia_valid = attr->ia_valid;
4955 error = inode_change_ok(inode, attr);
4959 if (is_quota_modification(inode, attr)) {
4960 error = dquot_initialize(inode);
4964 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4965 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4968 /* (user+group)*(old+new) structure, inode write (sb,
4969 * inode block, ? - but truncate inode update has it) */
4970 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4971 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4972 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4973 if (IS_ERR(handle)) {
4974 error = PTR_ERR(handle);
4977 error = dquot_transfer(inode, attr);
4979 ext4_journal_stop(handle);
4982 /* Update corresponding info in inode so that everything is in
4983 * one transaction */
4984 if (attr->ia_valid & ATTR_UID)
4985 inode->i_uid = attr->ia_uid;
4986 if (attr->ia_valid & ATTR_GID)
4987 inode->i_gid = attr->ia_gid;
4988 error = ext4_mark_inode_dirty(handle, inode);
4989 ext4_journal_stop(handle);
4992 if (attr->ia_valid & ATTR_SIZE) {
4994 loff_t oldsize = inode->i_size;
4995 int shrink = (attr->ia_size <= inode->i_size);
4997 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4998 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5000 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5003 if (!S_ISREG(inode->i_mode))
5006 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5007 inode_inc_iversion(inode);
5009 if (ext4_should_order_data(inode) &&
5010 (attr->ia_size < inode->i_size)) {
5011 error = ext4_begin_ordered_truncate(inode,
5016 if (attr->ia_size != inode->i_size) {
5017 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5018 if (IS_ERR(handle)) {
5019 error = PTR_ERR(handle);
5022 if (ext4_handle_valid(handle) && shrink) {
5023 error = ext4_orphan_add(handle, inode);
5027 * Update c/mtime on truncate up, ext4_truncate() will
5028 * update c/mtime in shrink case below
5031 inode->i_mtime = ext4_current_time(inode);
5032 inode->i_ctime = inode->i_mtime;
5034 down_write(&EXT4_I(inode)->i_data_sem);
5035 EXT4_I(inode)->i_disksize = attr->ia_size;
5036 rc = ext4_mark_inode_dirty(handle, inode);
5040 * We have to update i_size under i_data_sem together
5041 * with i_disksize to avoid races with writeback code
5042 * running ext4_wb_update_i_disksize().
5045 i_size_write(inode, attr->ia_size);
5046 up_write(&EXT4_I(inode)->i_data_sem);
5047 ext4_journal_stop(handle);
5050 ext4_orphan_del(NULL, inode);
5055 pagecache_isize_extended(inode, oldsize, inode->i_size);
5058 * Blocks are going to be removed from the inode. Wait
5059 * for dio in flight. Temporarily disable
5060 * dioread_nolock to prevent livelock.
5063 if (!ext4_should_journal_data(inode)) {
5064 ext4_inode_block_unlocked_dio(inode);
5065 inode_dio_wait(inode);
5066 ext4_inode_resume_unlocked_dio(inode);
5068 ext4_wait_for_tail_page_commit(inode);
5070 down_write(&EXT4_I(inode)->i_mmap_sem);
5072 * Truncate pagecache after we've waited for commit
5073 * in data=journal mode to make pages freeable.
5075 truncate_pagecache(inode, inode->i_size);
5077 ext4_truncate(inode);
5078 up_write(&EXT4_I(inode)->i_mmap_sem);
5082 setattr_copy(inode, attr);
5083 mark_inode_dirty(inode);
5087 * If the call to ext4_truncate failed to get a transaction handle at
5088 * all, we need to clean up the in-core orphan list manually.
5090 if (orphan && inode->i_nlink)
5091 ext4_orphan_del(NULL, inode);
5093 if (!rc && (ia_valid & ATTR_MODE))
5094 rc = posix_acl_chmod(inode, inode->i_mode);
5097 ext4_std_error(inode->i_sb, error);
5103 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5106 struct inode *inode;
5107 unsigned long long delalloc_blocks;
5109 inode = d_inode(dentry);
5110 generic_fillattr(inode, stat);
5113 * If there is inline data in the inode, the inode will normally not
5114 * have data blocks allocated (it may have an external xattr block).
5115 * Report at least one sector for such files, so tools like tar, rsync,
5116 * others doen't incorrectly think the file is completely sparse.
5118 if (unlikely(ext4_has_inline_data(inode)))
5119 stat->blocks += (stat->size + 511) >> 9;
5122 * We can't update i_blocks if the block allocation is delayed
5123 * otherwise in the case of system crash before the real block
5124 * allocation is done, we will have i_blocks inconsistent with
5125 * on-disk file blocks.
5126 * We always keep i_blocks updated together with real
5127 * allocation. But to not confuse with user, stat
5128 * will return the blocks that include the delayed allocation
5129 * blocks for this file.
5131 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5132 EXT4_I(inode)->i_reserved_data_blocks);
5133 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5137 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5140 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5141 return ext4_ind_trans_blocks(inode, lblocks);
5142 return ext4_ext_index_trans_blocks(inode, pextents);
5146 * Account for index blocks, block groups bitmaps and block group
5147 * descriptor blocks if modify datablocks and index blocks
5148 * worse case, the indexs blocks spread over different block groups
5150 * If datablocks are discontiguous, they are possible to spread over
5151 * different block groups too. If they are contiguous, with flexbg,
5152 * they could still across block group boundary.
5154 * Also account for superblock, inode, quota and xattr blocks
5156 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5159 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5165 * How many index blocks need to touch to map @lblocks logical blocks
5166 * to @pextents physical extents?
5168 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5173 * Now let's see how many group bitmaps and group descriptors need
5176 groups = idxblocks + pextents;
5178 if (groups > ngroups)
5180 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5181 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5183 /* bitmaps and block group descriptor blocks */
5184 ret += groups + gdpblocks;
5186 /* Blocks for super block, inode, quota and xattr blocks */
5187 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5193 * Calculate the total number of credits to reserve to fit
5194 * the modification of a single pages into a single transaction,
5195 * which may include multiple chunks of block allocations.
5197 * This could be called via ext4_write_begin()
5199 * We need to consider the worse case, when
5200 * one new block per extent.
5202 int ext4_writepage_trans_blocks(struct inode *inode)
5204 int bpp = ext4_journal_blocks_per_page(inode);
5207 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5209 /* Account for data blocks for journalled mode */
5210 if (ext4_should_journal_data(inode))
5216 * Calculate the journal credits for a chunk of data modification.
5218 * This is called from DIO, fallocate or whoever calling
5219 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5221 * journal buffers for data blocks are not included here, as DIO
5222 * and fallocate do no need to journal data buffers.
5224 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5226 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5230 * The caller must have previously called ext4_reserve_inode_write().
5231 * Give this, we know that the caller already has write access to iloc->bh.
5233 int ext4_mark_iloc_dirty(handle_t *handle,
5234 struct inode *inode, struct ext4_iloc *iloc)
5238 if (IS_I_VERSION(inode))
5239 inode_inc_iversion(inode);
5241 /* the do_update_inode consumes one bh->b_count */
5244 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5245 err = ext4_do_update_inode(handle, inode, iloc);
5251 * On success, We end up with an outstanding reference count against
5252 * iloc->bh. This _must_ be cleaned up later.
5256 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5257 struct ext4_iloc *iloc)
5261 err = ext4_get_inode_loc(inode, iloc);
5263 BUFFER_TRACE(iloc->bh, "get_write_access");
5264 err = ext4_journal_get_write_access(handle, iloc->bh);
5270 ext4_std_error(inode->i_sb, err);
5275 * Expand an inode by new_extra_isize bytes.
5276 * Returns 0 on success or negative error number on failure.
5278 static int ext4_expand_extra_isize(struct inode *inode,
5279 unsigned int new_extra_isize,
5280 struct ext4_iloc iloc,
5283 struct ext4_inode *raw_inode;
5284 struct ext4_xattr_ibody_header *header;
5286 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5289 raw_inode = ext4_raw_inode(&iloc);
5291 header = IHDR(inode, raw_inode);
5293 /* No extended attributes present */
5294 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5295 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5296 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5298 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5302 /* try to expand with EAs present */
5303 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5308 * What we do here is to mark the in-core inode as clean with respect to inode
5309 * dirtiness (it may still be data-dirty).
5310 * This means that the in-core inode may be reaped by prune_icache
5311 * without having to perform any I/O. This is a very good thing,
5312 * because *any* task may call prune_icache - even ones which
5313 * have a transaction open against a different journal.
5315 * Is this cheating? Not really. Sure, we haven't written the
5316 * inode out, but prune_icache isn't a user-visible syncing function.
5317 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5318 * we start and wait on commits.
5320 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5322 struct ext4_iloc iloc;
5323 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5324 static unsigned int mnt_count;
5328 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5329 err = ext4_reserve_inode_write(handle, inode, &iloc);
5332 if (ext4_handle_valid(handle) &&
5333 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5334 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5336 * We need extra buffer credits since we may write into EA block
5337 * with this same handle. If journal_extend fails, then it will
5338 * only result in a minor loss of functionality for that inode.
5339 * If this is felt to be critical, then e2fsck should be run to
5340 * force a large enough s_min_extra_isize.
5342 if ((jbd2_journal_extend(handle,
5343 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5344 ret = ext4_expand_extra_isize(inode,
5345 sbi->s_want_extra_isize,
5348 ext4_set_inode_state(inode,
5349 EXT4_STATE_NO_EXPAND);
5351 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5352 ext4_warning(inode->i_sb,
5353 "Unable to expand inode %lu. Delete"
5354 " some EAs or run e2fsck.",
5357 le16_to_cpu(sbi->s_es->s_mnt_count);
5362 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5366 * ext4_dirty_inode() is called from __mark_inode_dirty()
5368 * We're really interested in the case where a file is being extended.
5369 * i_size has been changed by generic_commit_write() and we thus need
5370 * to include the updated inode in the current transaction.
5372 * Also, dquot_alloc_block() will always dirty the inode when blocks
5373 * are allocated to the file.
5375 * If the inode is marked synchronous, we don't honour that here - doing
5376 * so would cause a commit on atime updates, which we don't bother doing.
5377 * We handle synchronous inodes at the highest possible level.
5379 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5380 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5381 * to copy into the on-disk inode structure are the timestamp files.
5383 void ext4_dirty_inode(struct inode *inode, int flags)
5387 if (flags == I_DIRTY_TIME)
5389 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5393 ext4_mark_inode_dirty(handle, inode);
5395 ext4_journal_stop(handle);
5402 * Bind an inode's backing buffer_head into this transaction, to prevent
5403 * it from being flushed to disk early. Unlike
5404 * ext4_reserve_inode_write, this leaves behind no bh reference and
5405 * returns no iloc structure, so the caller needs to repeat the iloc
5406 * lookup to mark the inode dirty later.
5408 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5410 struct ext4_iloc iloc;
5414 err = ext4_get_inode_loc(inode, &iloc);
5416 BUFFER_TRACE(iloc.bh, "get_write_access");
5417 err = jbd2_journal_get_write_access(handle, iloc.bh);
5419 err = ext4_handle_dirty_metadata(handle,
5425 ext4_std_error(inode->i_sb, err);
5430 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5437 * We have to be very careful here: changing a data block's
5438 * journaling status dynamically is dangerous. If we write a
5439 * data block to the journal, change the status and then delete
5440 * that block, we risk forgetting to revoke the old log record
5441 * from the journal and so a subsequent replay can corrupt data.
5442 * So, first we make sure that the journal is empty and that
5443 * nobody is changing anything.
5446 journal = EXT4_JOURNAL(inode);
5449 if (is_journal_aborted(journal))
5451 /* We have to allocate physical blocks for delalloc blocks
5452 * before flushing journal. otherwise delalloc blocks can not
5453 * be allocated any more. even more truncate on delalloc blocks
5454 * could trigger BUG by flushing delalloc blocks in journal.
5455 * There is no delalloc block in non-journal data mode.
5457 if (val && test_opt(inode->i_sb, DELALLOC)) {
5458 err = ext4_alloc_da_blocks(inode);
5463 /* Wait for all existing dio workers */
5464 ext4_inode_block_unlocked_dio(inode);
5465 inode_dio_wait(inode);
5467 jbd2_journal_lock_updates(journal);
5470 * OK, there are no updates running now, and all cached data is
5471 * synced to disk. We are now in a completely consistent state
5472 * which doesn't have anything in the journal, and we know that
5473 * no filesystem updates are running, so it is safe to modify
5474 * the inode's in-core data-journaling state flag now.
5478 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5480 err = jbd2_journal_flush(journal);
5482 jbd2_journal_unlock_updates(journal);
5483 ext4_inode_resume_unlocked_dio(inode);
5486 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5488 ext4_set_aops(inode);
5490 jbd2_journal_unlock_updates(journal);
5491 ext4_inode_resume_unlocked_dio(inode);
5493 /* Finally we can mark the inode as dirty. */
5495 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5497 return PTR_ERR(handle);
5499 err = ext4_mark_inode_dirty(handle, inode);
5500 ext4_handle_sync(handle);
5501 ext4_journal_stop(handle);
5502 ext4_std_error(inode->i_sb, err);
5507 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5509 return !buffer_mapped(bh);
5512 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5514 struct page *page = vmf->page;
5518 struct file *file = vma->vm_file;
5519 struct inode *inode = file_inode(file);
5520 struct address_space *mapping = inode->i_mapping;
5522 get_block_t *get_block;
5525 sb_start_pagefault(inode->i_sb);
5526 file_update_time(vma->vm_file);
5528 down_read(&EXT4_I(inode)->i_mmap_sem);
5529 /* Delalloc case is easy... */
5530 if (test_opt(inode->i_sb, DELALLOC) &&
5531 !ext4_should_journal_data(inode) &&
5532 !ext4_nonda_switch(inode->i_sb)) {
5534 ret = block_page_mkwrite(vma, vmf,
5535 ext4_da_get_block_prep);
5536 } while (ret == -ENOSPC &&
5537 ext4_should_retry_alloc(inode->i_sb, &retries));
5542 size = i_size_read(inode);
5543 /* Page got truncated from under us? */
5544 if (page->mapping != mapping || page_offset(page) > size) {
5546 ret = VM_FAULT_NOPAGE;
5550 if (page->index == size >> PAGE_SHIFT)
5551 len = size & ~PAGE_MASK;
5555 * Return if we have all the buffers mapped. This avoids the need to do
5556 * journal_start/journal_stop which can block and take a long time
5558 if (page_has_buffers(page)) {
5559 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5561 ext4_bh_unmapped)) {
5562 /* Wait so that we don't change page under IO */
5563 wait_for_stable_page(page);
5564 ret = VM_FAULT_LOCKED;
5569 /* OK, we need to fill the hole... */
5570 if (ext4_should_dioread_nolock(inode))
5571 get_block = ext4_get_block_unwritten;
5573 get_block = ext4_get_block;
5575 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5576 ext4_writepage_trans_blocks(inode));
5577 if (IS_ERR(handle)) {
5578 ret = VM_FAULT_SIGBUS;
5581 ret = block_page_mkwrite(vma, vmf, get_block);
5582 if (!ret && ext4_should_journal_data(inode)) {
5583 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5584 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5586 ret = VM_FAULT_SIGBUS;
5587 ext4_journal_stop(handle);
5590 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5592 ext4_journal_stop(handle);
5593 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5596 ret = block_page_mkwrite_return(ret);
5598 up_read(&EXT4_I(inode)->i_mmap_sem);
5599 sb_end_pagefault(inode->i_sb);
5603 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5605 struct inode *inode = file_inode(vma->vm_file);
5608 down_read(&EXT4_I(inode)->i_mmap_sem);
5609 err = filemap_fault(vma, vmf);
5610 up_read(&EXT4_I(inode)->i_mmap_sem);
5616 * Find the first extent at or after @lblk in an inode that is not a hole.
5617 * Search for @map_len blocks at most. The extent is returned in @result.
5619 * The function returns 1 if we found an extent. The function returns 0 in
5620 * case there is no extent at or after @lblk and in that case also sets
5621 * @result->es_len to 0. In case of error, the error code is returned.
5623 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5624 unsigned int map_len, struct extent_status *result)
5626 struct ext4_map_blocks map;
5627 struct extent_status es = {};
5631 map.m_len = map_len;
5634 * For non-extent based files this loop may iterate several times since
5635 * we do not determine full hole size.
5637 while (map.m_len > 0) {
5638 ret = ext4_map_blocks(NULL, inode, &map, 0);
5641 /* There's extent covering m_lblk? Just return it. */
5645 ext4_es_store_pblock(result, map.m_pblk);
5646 result->es_lblk = map.m_lblk;
5647 result->es_len = map.m_len;
5648 if (map.m_flags & EXT4_MAP_UNWRITTEN)
5649 status = EXTENT_STATUS_UNWRITTEN;
5651 status = EXTENT_STATUS_WRITTEN;
5652 ext4_es_store_status(result, status);
5655 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5656 map.m_lblk + map.m_len - 1,
5658 /* Is delalloc data before next block in extent tree? */
5659 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5660 ext4_lblk_t offset = 0;
5662 if (es.es_lblk < lblk)
5663 offset = lblk - es.es_lblk;
5664 result->es_lblk = es.es_lblk + offset;
5665 ext4_es_store_pblock(result,
5666 ext4_es_pblock(&es) + offset);
5667 result->es_len = es.es_len - offset;
5668 ext4_es_store_status(result, ext4_es_status(&es));
5672 /* There's a hole at m_lblk, advance us after it */
5673 map.m_lblk += map.m_len;
5674 map_len -= map.m_len;
5675 map.m_len = map_len;