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);
56 int offset = offsetof(struct ext4_inode, i_checksum_lo);
57 unsigned int csum_size = sizeof(dummy_csum);
59 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
60 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
62 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
63 EXT4_GOOD_OLD_INODE_SIZE - offset);
65 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
66 offset = offsetof(struct ext4_inode, i_checksum_hi);
67 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
68 EXT4_GOOD_OLD_INODE_SIZE,
69 offset - EXT4_GOOD_OLD_INODE_SIZE);
70 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
71 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
74 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
75 EXT4_INODE_SIZE(inode->i_sb) -
83 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
84 struct ext4_inode_info *ei)
86 __u32 provided, calculated;
88 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
89 cpu_to_le32(EXT4_OS_LINUX) ||
90 !ext4_has_metadata_csum(inode->i_sb))
93 provided = le16_to_cpu(raw->i_checksum_lo);
94 calculated = ext4_inode_csum(inode, raw, ei);
95 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
96 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
97 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 return provided == calculated;
104 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
105 struct ext4_inode_info *ei)
109 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
110 cpu_to_le32(EXT4_OS_LINUX) ||
111 !ext4_has_metadata_csum(inode->i_sb))
114 csum = ext4_inode_csum(inode, raw, ei);
115 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
116 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
117 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
118 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 trace_ext4_begin_ordered_truncate(inode, new_size);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode)->jinode)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
134 EXT4_I(inode)->jinode,
138 static void ext4_invalidatepage(struct page *page, unsigned int offset,
139 unsigned int length);
140 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
141 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
142 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 * Test whether an inode is a fast symlink.
148 int ext4_inode_is_fast_symlink(struct inode *inode)
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
153 if (ext4_has_inline_data(inode))
156 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
160 * Restart the transaction associated with *handle. This does a commit,
161 * so before we call here everything must be consistently dirtied against
164 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
170 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
171 * moment, get_block can be called only for blocks inside i_size since
172 * page cache has been already dropped and writes are blocked by
173 * i_mutex. So we can safely drop the i_data_sem here.
175 BUG_ON(EXT4_JOURNAL(inode) == NULL);
176 jbd_debug(2, "restarting handle %p\n", handle);
177 up_write(&EXT4_I(inode)->i_data_sem);
178 ret = ext4_journal_restart(handle, nblocks);
179 down_write(&EXT4_I(inode)->i_data_sem);
180 ext4_discard_preallocations(inode);
186 * Called at the last iput() if i_nlink is zero.
188 void ext4_evict_inode(struct inode *inode)
193 trace_ext4_evict_inode(inode);
195 if (inode->i_nlink) {
197 * When journalling data dirty buffers are tracked only in the
198 * journal. So although mm thinks everything is clean and
199 * ready for reaping the inode might still have some pages to
200 * write in the running transaction or waiting to be
201 * checkpointed. Thus calling jbd2_journal_invalidatepage()
202 * (via truncate_inode_pages()) to discard these buffers can
203 * cause data loss. Also even if we did not discard these
204 * buffers, we would have no way to find them after the inode
205 * is reaped and thus user could see stale data if he tries to
206 * read them before the transaction is checkpointed. So be
207 * careful and force everything to disk here... We use
208 * ei->i_datasync_tid to store the newest transaction
209 * containing inode's data.
211 * Note that directories do not have this problem because they
212 * don't use page cache.
214 if (ext4_should_journal_data(inode) &&
215 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
216 inode->i_ino != EXT4_JOURNAL_INO) {
217 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
218 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
220 jbd2_complete_transaction(journal, commit_tid);
221 filemap_write_and_wait(&inode->i_data);
223 truncate_inode_pages_final(&inode->i_data);
228 if (is_bad_inode(inode))
230 dquot_initialize(inode);
232 if (ext4_should_order_data(inode))
233 ext4_begin_ordered_truncate(inode, 0);
234 truncate_inode_pages_final(&inode->i_data);
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode->i_sb);
241 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
242 ext4_blocks_for_truncate(inode)+3);
243 if (IS_ERR(handle)) {
244 ext4_std_error(inode->i_sb, PTR_ERR(handle));
246 * If we're going to skip the normal cleanup, we still need to
247 * make sure that the in-core orphan linked list is properly
250 ext4_orphan_del(NULL, inode);
251 sb_end_intwrite(inode->i_sb);
256 ext4_handle_sync(handle);
258 err = ext4_mark_inode_dirty(handle, inode);
260 ext4_warning(inode->i_sb,
261 "couldn't mark inode dirty (err %d)", err);
265 ext4_truncate(inode);
268 * ext4_ext_truncate() doesn't reserve any slop when it
269 * restarts journal transactions; therefore there may not be
270 * enough credits left in the handle to remove the inode from
271 * the orphan list and set the dtime field.
273 if (!ext4_handle_has_enough_credits(handle, 3)) {
274 err = ext4_journal_extend(handle, 3);
276 err = ext4_journal_restart(handle, 3);
278 ext4_warning(inode->i_sb,
279 "couldn't extend journal (err %d)", err);
281 ext4_journal_stop(handle);
282 ext4_orphan_del(NULL, inode);
283 sb_end_intwrite(inode->i_sb);
289 * Kill off the orphan record which ext4_truncate created.
290 * AKPM: I think this can be inside the above `if'.
291 * Note that ext4_orphan_del() has to be able to cope with the
292 * deletion of a non-existent orphan - this is because we don't
293 * know if ext4_truncate() actually created an orphan record.
294 * (Well, we could do this if we need to, but heck - it works)
296 ext4_orphan_del(handle, inode);
297 EXT4_I(inode)->i_dtime = get_seconds();
300 * One subtle ordering requirement: if anything has gone wrong
301 * (transaction abort, IO errors, whatever), then we can still
302 * do these next steps (the fs will already have been marked as
303 * having errors), but we can't free the inode if the mark_dirty
306 if (ext4_mark_inode_dirty(handle, inode))
307 /* If that failed, just do the required in-core inode clear. */
308 ext4_clear_inode(inode);
310 ext4_free_inode(handle, inode);
311 ext4_journal_stop(handle);
312 sb_end_intwrite(inode->i_sb);
315 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
319 qsize_t *ext4_get_reserved_space(struct inode *inode)
321 return &EXT4_I(inode)->i_reserved_quota;
326 * Called with i_data_sem down, which is important since we can call
327 * ext4_discard_preallocations() from here.
329 void ext4_da_update_reserve_space(struct inode *inode,
330 int used, int quota_claim)
332 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
333 struct ext4_inode_info *ei = EXT4_I(inode);
335 spin_lock(&ei->i_block_reservation_lock);
336 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
337 if (unlikely(used > ei->i_reserved_data_blocks)) {
338 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
339 "with only %d reserved data blocks",
340 __func__, inode->i_ino, used,
341 ei->i_reserved_data_blocks);
343 used = ei->i_reserved_data_blocks;
346 /* Update per-inode reservations */
347 ei->i_reserved_data_blocks -= used;
348 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
350 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
352 /* Update quota subsystem for data blocks */
354 dquot_claim_block(inode, EXT4_C2B(sbi, used));
357 * We did fallocate with an offset that is already delayed
358 * allocated. So on delayed allocated writeback we should
359 * not re-claim the quota for fallocated blocks.
361 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
365 * If we have done all the pending block allocations and if
366 * there aren't any writers on the inode, we can discard the
367 * inode's preallocations.
369 if ((ei->i_reserved_data_blocks == 0) &&
370 (atomic_read(&inode->i_writecount) == 0))
371 ext4_discard_preallocations(inode);
374 static int __check_block_validity(struct inode *inode, const char *func,
376 struct ext4_map_blocks *map)
378 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
380 ext4_error_inode(inode, func, line, map->m_pblk,
381 "lblock %lu mapped to illegal pblock "
382 "(length %d)", (unsigned long) map->m_lblk,
384 return -EFSCORRUPTED;
389 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
394 if (ext4_encrypted_inode(inode))
395 return ext4_encrypted_zeroout(inode, lblk, pblk, len);
397 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
404 #define check_block_validity(inode, map) \
405 __check_block_validity((inode), __func__, __LINE__, (map))
407 #ifdef ES_AGGRESSIVE_TEST
408 static void ext4_map_blocks_es_recheck(handle_t *handle,
410 struct ext4_map_blocks *es_map,
411 struct ext4_map_blocks *map,
418 * There is a race window that the result is not the same.
419 * e.g. xfstests #223 when dioread_nolock enables. The reason
420 * is that we lookup a block mapping in extent status tree with
421 * out taking i_data_sem. So at the time the unwritten extent
422 * could be converted.
424 down_read(&EXT4_I(inode)->i_data_sem);
425 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
426 retval = ext4_ext_map_blocks(handle, inode, map, flags &
427 EXT4_GET_BLOCKS_KEEP_SIZE);
429 retval = ext4_ind_map_blocks(handle, inode, map, flags &
430 EXT4_GET_BLOCKS_KEEP_SIZE);
432 up_read((&EXT4_I(inode)->i_data_sem));
435 * We don't check m_len because extent will be collpased in status
436 * tree. So the m_len might not equal.
438 if (es_map->m_lblk != map->m_lblk ||
439 es_map->m_flags != map->m_flags ||
440 es_map->m_pblk != map->m_pblk) {
441 printk("ES cache assertion failed for inode: %lu "
442 "es_cached ex [%d/%d/%llu/%x] != "
443 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
444 inode->i_ino, es_map->m_lblk, es_map->m_len,
445 es_map->m_pblk, es_map->m_flags, map->m_lblk,
446 map->m_len, map->m_pblk, map->m_flags,
450 #endif /* ES_AGGRESSIVE_TEST */
453 * The ext4_map_blocks() function tries to look up the requested blocks,
454 * and returns if the blocks are already mapped.
456 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
457 * and store the allocated blocks in the result buffer head and mark it
460 * If file type is extents based, it will call ext4_ext_map_blocks(),
461 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
464 * On success, it returns the number of blocks being mapped or allocated. if
465 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
466 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
468 * It returns 0 if plain look up failed (blocks have not been allocated), in
469 * that case, @map is returned as unmapped but we still do fill map->m_len to
470 * indicate the length of a hole starting at map->m_lblk.
472 * It returns the error in case of allocation failure.
474 int ext4_map_blocks(handle_t *handle, struct inode *inode,
475 struct ext4_map_blocks *map, int flags)
477 struct extent_status es;
480 #ifdef ES_AGGRESSIVE_TEST
481 struct ext4_map_blocks orig_map;
483 memcpy(&orig_map, map, sizeof(*map));
487 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
488 "logical block %lu\n", inode->i_ino, flags, map->m_len,
489 (unsigned long) map->m_lblk);
492 * ext4_map_blocks returns an int, and m_len is an unsigned int
494 if (unlikely(map->m_len > INT_MAX))
495 map->m_len = INT_MAX;
497 /* We can handle the block number less than EXT_MAX_BLOCKS */
498 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
499 return -EFSCORRUPTED;
501 /* Lookup extent status tree firstly */
502 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
503 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
504 map->m_pblk = ext4_es_pblock(&es) +
505 map->m_lblk - es.es_lblk;
506 map->m_flags |= ext4_es_is_written(&es) ?
507 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
508 retval = es.es_len - (map->m_lblk - es.es_lblk);
509 if (retval > map->m_len)
512 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
514 retval = es.es_len - (map->m_lblk - es.es_lblk);
515 if (retval > map->m_len)
522 #ifdef ES_AGGRESSIVE_TEST
523 ext4_map_blocks_es_recheck(handle, inode, map,
530 * Try to see if we can get the block without requesting a new
533 down_read(&EXT4_I(inode)->i_data_sem);
534 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
535 retval = ext4_ext_map_blocks(handle, inode, map, flags &
536 EXT4_GET_BLOCKS_KEEP_SIZE);
538 retval = ext4_ind_map_blocks(handle, inode, map, flags &
539 EXT4_GET_BLOCKS_KEEP_SIZE);
544 if (unlikely(retval != map->m_len)) {
545 ext4_warning(inode->i_sb,
546 "ES len assertion failed for inode "
547 "%lu: retval %d != map->m_len %d",
548 inode->i_ino, retval, map->m_len);
552 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
553 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
554 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
555 !(status & EXTENT_STATUS_WRITTEN) &&
556 ext4_find_delalloc_range(inode, map->m_lblk,
557 map->m_lblk + map->m_len - 1))
558 status |= EXTENT_STATUS_DELAYED;
559 ret = ext4_es_insert_extent(inode, map->m_lblk,
560 map->m_len, map->m_pblk, status);
564 up_read((&EXT4_I(inode)->i_data_sem));
567 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
568 ret = check_block_validity(inode, map);
573 /* If it is only a block(s) look up */
574 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
578 * Returns if the blocks have already allocated
580 * Note that if blocks have been preallocated
581 * ext4_ext_get_block() returns the create = 0
582 * with buffer head unmapped.
584 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
586 * If we need to convert extent to unwritten
587 * we continue and do the actual work in
588 * ext4_ext_map_blocks()
590 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
594 * Here we clear m_flags because after allocating an new extent,
595 * it will be set again.
597 map->m_flags &= ~EXT4_MAP_FLAGS;
600 * New blocks allocate and/or writing to unwritten extent
601 * will possibly result in updating i_data, so we take
602 * the write lock of i_data_sem, and call get_block()
603 * with create == 1 flag.
605 down_write(&EXT4_I(inode)->i_data_sem);
608 * We need to check for EXT4 here because migrate
609 * could have changed the inode type in between
611 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
612 retval = ext4_ext_map_blocks(handle, inode, map, flags);
614 retval = ext4_ind_map_blocks(handle, inode, map, flags);
616 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
618 * We allocated new blocks which will result in
619 * i_data's format changing. Force the migrate
620 * to fail by clearing migrate flags
622 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
626 * Update reserved blocks/metadata blocks after successful
627 * block allocation which had been deferred till now. We don't
628 * support fallocate for non extent files. So we can update
629 * reserve space here.
632 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
633 ext4_da_update_reserve_space(inode, retval, 1);
639 if (unlikely(retval != map->m_len)) {
640 ext4_warning(inode->i_sb,
641 "ES len assertion failed for inode "
642 "%lu: retval %d != map->m_len %d",
643 inode->i_ino, retval, map->m_len);
648 * We have to zeroout blocks before inserting them into extent
649 * status tree. Otherwise someone could look them up there and
650 * use them before they are really zeroed.
652 if (flags & EXT4_GET_BLOCKS_ZERO &&
653 map->m_flags & EXT4_MAP_MAPPED &&
654 map->m_flags & EXT4_MAP_NEW) {
655 ret = ext4_issue_zeroout(inode, map->m_lblk,
656 map->m_pblk, map->m_len);
664 * If the extent has been zeroed out, we don't need to update
665 * extent status tree.
667 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
668 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
669 if (ext4_es_is_written(&es))
672 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
673 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
674 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
675 !(status & EXTENT_STATUS_WRITTEN) &&
676 ext4_find_delalloc_range(inode, map->m_lblk,
677 map->m_lblk + map->m_len - 1))
678 status |= EXTENT_STATUS_DELAYED;
679 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
680 map->m_pblk, status);
688 up_write((&EXT4_I(inode)->i_data_sem));
689 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
690 ret = check_block_validity(inode, map);
695 * Inodes with freshly allocated blocks where contents will be
696 * visible after transaction commit must be on transaction's
699 if (map->m_flags & EXT4_MAP_NEW &&
700 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
701 !(flags & EXT4_GET_BLOCKS_ZERO) &&
702 !IS_NOQUOTA(inode) &&
703 ext4_should_order_data(inode)) {
704 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
705 ret = ext4_jbd2_inode_add_wait(handle, inode);
707 ret = ext4_jbd2_inode_add_write(handle, inode);
716 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
717 * we have to be careful as someone else may be manipulating b_state as well.
719 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
721 unsigned long old_state;
722 unsigned long new_state;
724 flags &= EXT4_MAP_FLAGS;
726 /* Dummy buffer_head? Set non-atomically. */
728 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
732 * Someone else may be modifying b_state. Be careful! This is ugly but
733 * once we get rid of using bh as a container for mapping information
734 * to pass to / from get_block functions, this can go away.
737 old_state = READ_ONCE(bh->b_state);
738 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
740 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
743 static int _ext4_get_block(struct inode *inode, sector_t iblock,
744 struct buffer_head *bh, int flags)
746 struct ext4_map_blocks map;
749 if (ext4_has_inline_data(inode))
753 map.m_len = bh->b_size >> inode->i_blkbits;
755 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
758 map_bh(bh, inode->i_sb, map.m_pblk);
759 ext4_update_bh_state(bh, map.m_flags);
760 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
766 int ext4_get_block(struct inode *inode, sector_t iblock,
767 struct buffer_head *bh, int create)
769 return _ext4_get_block(inode, iblock, bh,
770 create ? EXT4_GET_BLOCKS_CREATE : 0);
774 * Get block function used when preparing for buffered write if we require
775 * creating an unwritten extent if blocks haven't been allocated. The extent
776 * will be converted to written after the IO is complete.
778 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
779 struct buffer_head *bh_result, int create)
781 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
782 inode->i_ino, create);
783 return _ext4_get_block(inode, iblock, bh_result,
784 EXT4_GET_BLOCKS_IO_CREATE_EXT);
787 /* Maximum number of blocks we map for direct IO at once. */
788 #define DIO_MAX_BLOCKS 4096
791 * Get blocks function for the cases that need to start a transaction -
792 * generally difference cases of direct IO and DAX IO. It also handles retries
795 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
796 struct buffer_head *bh_result, int flags)
803 /* Trim mapping request to maximum we can map at once for DIO */
804 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
805 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
806 dio_credits = ext4_chunk_trans_blocks(inode,
807 bh_result->b_size >> inode->i_blkbits);
809 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
811 return PTR_ERR(handle);
813 ret = _ext4_get_block(inode, iblock, bh_result, flags);
814 ext4_journal_stop(handle);
816 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
821 /* Get block function for DIO reads and writes to inodes without extents */
822 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
823 struct buffer_head *bh, int create)
825 /* We don't expect handle for direct IO */
826 WARN_ON_ONCE(ext4_journal_current_handle());
829 return _ext4_get_block(inode, iblock, bh, 0);
830 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
834 * Get block function for AIO DIO writes when we create unwritten extent if
835 * blocks are not allocated yet. The extent will be converted to written
836 * after IO is complete.
838 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
839 sector_t iblock, struct buffer_head *bh_result, int create)
843 /* We don't expect handle for direct IO */
844 WARN_ON_ONCE(ext4_journal_current_handle());
846 ret = ext4_get_block_trans(inode, iblock, bh_result,
847 EXT4_GET_BLOCKS_IO_CREATE_EXT);
850 * When doing DIO using unwritten extents, we need io_end to convert
851 * unwritten extents to written on IO completion. We allocate io_end
852 * once we spot unwritten extent and store it in b_private. Generic
853 * DIO code keeps b_private set and furthermore passes the value to
854 * our completion callback in 'private' argument.
856 if (!ret && buffer_unwritten(bh_result)) {
857 if (!bh_result->b_private) {
858 ext4_io_end_t *io_end;
860 io_end = ext4_init_io_end(inode, GFP_KERNEL);
863 bh_result->b_private = io_end;
864 ext4_set_io_unwritten_flag(inode, io_end);
866 set_buffer_defer_completion(bh_result);
873 * Get block function for non-AIO DIO writes when we create unwritten extent if
874 * blocks are not allocated yet. The extent will be converted to written
875 * after IO is complete from ext4_ext_direct_IO() function.
877 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
878 sector_t iblock, struct buffer_head *bh_result, int create)
882 /* We don't expect handle for direct IO */
883 WARN_ON_ONCE(ext4_journal_current_handle());
885 ret = ext4_get_block_trans(inode, iblock, bh_result,
886 EXT4_GET_BLOCKS_IO_CREATE_EXT);
889 * Mark inode as having pending DIO writes to unwritten extents.
890 * ext4_ext_direct_IO() checks this flag and converts extents to
893 if (!ret && buffer_unwritten(bh_result))
894 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
899 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
900 struct buffer_head *bh_result, int create)
904 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
905 inode->i_ino, create);
906 /* We don't expect handle for direct IO */
907 WARN_ON_ONCE(ext4_journal_current_handle());
909 ret = _ext4_get_block(inode, iblock, bh_result, 0);
911 * Blocks should have been preallocated! ext4_file_write_iter() checks
914 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
921 * `handle' can be NULL if create is zero
923 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
924 ext4_lblk_t block, int map_flags)
926 struct ext4_map_blocks map;
927 struct buffer_head *bh;
928 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
931 J_ASSERT(handle != NULL || create == 0);
935 err = ext4_map_blocks(handle, inode, &map, map_flags);
938 return create ? ERR_PTR(-ENOSPC) : NULL;
942 bh = sb_getblk(inode->i_sb, map.m_pblk);
944 return ERR_PTR(-ENOMEM);
945 if (map.m_flags & EXT4_MAP_NEW) {
946 J_ASSERT(create != 0);
947 J_ASSERT(handle != NULL);
950 * Now that we do not always journal data, we should
951 * keep in mind whether this should always journal the
952 * new buffer as metadata. For now, regular file
953 * writes use ext4_get_block instead, so it's not a
957 BUFFER_TRACE(bh, "call get_create_access");
958 err = ext4_journal_get_create_access(handle, bh);
963 if (!buffer_uptodate(bh)) {
964 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
965 set_buffer_uptodate(bh);
968 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
969 err = ext4_handle_dirty_metadata(handle, inode, bh);
973 BUFFER_TRACE(bh, "not a new buffer");
980 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
981 ext4_lblk_t block, int map_flags)
983 struct buffer_head *bh;
985 bh = ext4_getblk(handle, inode, block, map_flags);
988 if (!bh || buffer_uptodate(bh))
990 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
992 if (buffer_uptodate(bh))
995 return ERR_PTR(-EIO);
998 int ext4_walk_page_buffers(handle_t *handle,
999 struct buffer_head *head,
1003 int (*fn)(handle_t *handle,
1004 struct buffer_head *bh))
1006 struct buffer_head *bh;
1007 unsigned block_start, block_end;
1008 unsigned blocksize = head->b_size;
1010 struct buffer_head *next;
1012 for (bh = head, block_start = 0;
1013 ret == 0 && (bh != head || !block_start);
1014 block_start = block_end, bh = next) {
1015 next = bh->b_this_page;
1016 block_end = block_start + blocksize;
1017 if (block_end <= from || block_start >= to) {
1018 if (partial && !buffer_uptodate(bh))
1022 err = (*fn)(handle, bh);
1030 * To preserve ordering, it is essential that the hole instantiation and
1031 * the data write be encapsulated in a single transaction. We cannot
1032 * close off a transaction and start a new one between the ext4_get_block()
1033 * and the commit_write(). So doing the jbd2_journal_start at the start of
1034 * prepare_write() is the right place.
1036 * Also, this function can nest inside ext4_writepage(). In that case, we
1037 * *know* that ext4_writepage() has generated enough buffer credits to do the
1038 * whole page. So we won't block on the journal in that case, which is good,
1039 * because the caller may be PF_MEMALLOC.
1041 * By accident, ext4 can be reentered when a transaction is open via
1042 * quota file writes. If we were to commit the transaction while thus
1043 * reentered, there can be a deadlock - we would be holding a quota
1044 * lock, and the commit would never complete if another thread had a
1045 * transaction open and was blocking on the quota lock - a ranking
1048 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1049 * will _not_ run commit under these circumstances because handle->h_ref
1050 * is elevated. We'll still have enough credits for the tiny quotafile
1053 int do_journal_get_write_access(handle_t *handle,
1054 struct buffer_head *bh)
1056 int dirty = buffer_dirty(bh);
1059 if (!buffer_mapped(bh) || buffer_freed(bh))
1062 * __block_write_begin() could have dirtied some buffers. Clean
1063 * the dirty bit as jbd2_journal_get_write_access() could complain
1064 * otherwise about fs integrity issues. Setting of the dirty bit
1065 * by __block_write_begin() isn't a real problem here as we clear
1066 * the bit before releasing a page lock and thus writeback cannot
1067 * ever write the buffer.
1070 clear_buffer_dirty(bh);
1071 BUFFER_TRACE(bh, "get write access");
1072 ret = ext4_journal_get_write_access(handle, bh);
1074 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1078 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1079 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1080 get_block_t *get_block)
1082 unsigned from = pos & (PAGE_SIZE - 1);
1083 unsigned to = from + len;
1084 struct inode *inode = page->mapping->host;
1085 unsigned block_start, block_end;
1088 unsigned blocksize = inode->i_sb->s_blocksize;
1090 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1091 bool decrypt = false;
1093 BUG_ON(!PageLocked(page));
1094 BUG_ON(from > PAGE_SIZE);
1095 BUG_ON(to > PAGE_SIZE);
1098 if (!page_has_buffers(page))
1099 create_empty_buffers(page, blocksize, 0);
1100 head = page_buffers(page);
1101 bbits = ilog2(blocksize);
1102 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1104 for (bh = head, block_start = 0; bh != head || !block_start;
1105 block++, block_start = block_end, bh = bh->b_this_page) {
1106 block_end = block_start + blocksize;
1107 if (block_end <= from || block_start >= to) {
1108 if (PageUptodate(page)) {
1109 if (!buffer_uptodate(bh))
1110 set_buffer_uptodate(bh);
1115 clear_buffer_new(bh);
1116 if (!buffer_mapped(bh)) {
1117 WARN_ON(bh->b_size != blocksize);
1118 err = get_block(inode, block, bh, 1);
1121 if (buffer_new(bh)) {
1122 unmap_underlying_metadata(bh->b_bdev,
1124 if (PageUptodate(page)) {
1125 clear_buffer_new(bh);
1126 set_buffer_uptodate(bh);
1127 mark_buffer_dirty(bh);
1130 if (block_end > to || block_start < from)
1131 zero_user_segments(page, to, block_end,
1136 if (PageUptodate(page)) {
1137 if (!buffer_uptodate(bh))
1138 set_buffer_uptodate(bh);
1141 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1142 !buffer_unwritten(bh) &&
1143 (block_start < from || block_end > to)) {
1144 ll_rw_block(READ, 1, &bh);
1146 decrypt = ext4_encrypted_inode(inode) &&
1147 S_ISREG(inode->i_mode);
1151 * If we issued read requests, let them complete.
1153 while (wait_bh > wait) {
1154 wait_on_buffer(*--wait_bh);
1155 if (!buffer_uptodate(*wait_bh))
1159 page_zero_new_buffers(page, from, to);
1161 err = ext4_decrypt(page);
1166 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1167 loff_t pos, unsigned len, unsigned flags,
1168 struct page **pagep, void **fsdata)
1170 struct inode *inode = mapping->host;
1171 int ret, needed_blocks;
1178 trace_ext4_write_begin(inode, pos, len, flags);
1180 * Reserve one block more for addition to orphan list in case
1181 * we allocate blocks but write fails for some reason
1183 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1184 index = pos >> PAGE_SHIFT;
1185 from = pos & (PAGE_SIZE - 1);
1188 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1189 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1198 * grab_cache_page_write_begin() can take a long time if the
1199 * system is thrashing due to memory pressure, or if the page
1200 * is being written back. So grab it first before we start
1201 * the transaction handle. This also allows us to allocate
1202 * the page (if needed) without using GFP_NOFS.
1205 page = grab_cache_page_write_begin(mapping, index, flags);
1211 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1212 if (IS_ERR(handle)) {
1214 return PTR_ERR(handle);
1218 if (page->mapping != mapping) {
1219 /* The page got truncated from under us */
1222 ext4_journal_stop(handle);
1225 /* In case writeback began while the page was unlocked */
1226 wait_for_stable_page(page);
1228 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1229 if (ext4_should_dioread_nolock(inode))
1230 ret = ext4_block_write_begin(page, pos, len,
1231 ext4_get_block_unwritten);
1233 ret = ext4_block_write_begin(page, pos, len,
1236 if (ext4_should_dioread_nolock(inode))
1237 ret = __block_write_begin(page, pos, len,
1238 ext4_get_block_unwritten);
1240 ret = __block_write_begin(page, pos, len, ext4_get_block);
1242 if (!ret && ext4_should_journal_data(inode)) {
1243 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1245 do_journal_get_write_access);
1251 * __block_write_begin may have instantiated a few blocks
1252 * outside i_size. Trim these off again. Don't need
1253 * i_size_read because we hold i_mutex.
1255 * Add inode to orphan list in case we crash before
1258 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1259 ext4_orphan_add(handle, inode);
1261 ext4_journal_stop(handle);
1262 if (pos + len > inode->i_size) {
1263 ext4_truncate_failed_write(inode);
1265 * If truncate failed early the inode might
1266 * still be on the orphan list; we need to
1267 * make sure the inode is removed from the
1268 * orphan list in that case.
1271 ext4_orphan_del(NULL, inode);
1274 if (ret == -ENOSPC &&
1275 ext4_should_retry_alloc(inode->i_sb, &retries))
1284 /* For write_end() in data=journal mode */
1285 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1288 if (!buffer_mapped(bh) || buffer_freed(bh))
1290 set_buffer_uptodate(bh);
1291 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1292 clear_buffer_meta(bh);
1293 clear_buffer_prio(bh);
1298 * We need to pick up the new inode size which generic_commit_write gave us
1299 * `file' can be NULL - eg, when called from page_symlink().
1301 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1302 * buffers are managed internally.
1304 static int ext4_write_end(struct file *file,
1305 struct address_space *mapping,
1306 loff_t pos, unsigned len, unsigned copied,
1307 struct page *page, void *fsdata)
1309 handle_t *handle = ext4_journal_current_handle();
1310 struct inode *inode = mapping->host;
1311 loff_t old_size = inode->i_size;
1313 int i_size_changed = 0;
1315 trace_ext4_write_end(inode, pos, len, copied);
1316 if (ext4_has_inline_data(inode)) {
1317 ret = ext4_write_inline_data_end(inode, pos, len,
1323 copied = block_write_end(file, mapping, pos,
1324 len, copied, page, fsdata);
1326 * it's important to update i_size while still holding page lock:
1327 * page writeout could otherwise come in and zero beyond i_size.
1329 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1334 pagecache_isize_extended(inode, old_size, pos);
1336 * Don't mark the inode dirty under page lock. First, it unnecessarily
1337 * makes the holding time of page lock longer. Second, it forces lock
1338 * ordering of page lock and transaction start for journaling
1342 ext4_mark_inode_dirty(handle, inode);
1344 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1345 /* if we have allocated more blocks and copied
1346 * less. We will have blocks allocated outside
1347 * inode->i_size. So truncate them
1349 ext4_orphan_add(handle, inode);
1351 ret2 = ext4_journal_stop(handle);
1355 if (pos + len > inode->i_size) {
1356 ext4_truncate_failed_write(inode);
1358 * If truncate failed early the inode might still be
1359 * on the orphan list; we need to make sure the inode
1360 * is removed from the orphan list in that case.
1363 ext4_orphan_del(NULL, inode);
1366 return ret ? ret : copied;
1370 * This is a private version of page_zero_new_buffers() which doesn't
1371 * set the buffer to be dirty, since in data=journalled mode we need
1372 * to call ext4_handle_dirty_metadata() instead.
1374 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1376 unsigned int block_start = 0, block_end;
1377 struct buffer_head *head, *bh;
1379 bh = head = page_buffers(page);
1381 block_end = block_start + bh->b_size;
1382 if (buffer_new(bh)) {
1383 if (block_end > from && block_start < to) {
1384 if (!PageUptodate(page)) {
1385 unsigned start, size;
1387 start = max(from, block_start);
1388 size = min(to, block_end) - start;
1390 zero_user(page, start, size);
1391 set_buffer_uptodate(bh);
1393 clear_buffer_new(bh);
1396 block_start = block_end;
1397 bh = bh->b_this_page;
1398 } while (bh != head);
1401 static int ext4_journalled_write_end(struct file *file,
1402 struct address_space *mapping,
1403 loff_t pos, unsigned len, unsigned copied,
1404 struct page *page, void *fsdata)
1406 handle_t *handle = ext4_journal_current_handle();
1407 struct inode *inode = mapping->host;
1408 loff_t old_size = inode->i_size;
1412 int size_changed = 0;
1414 trace_ext4_journalled_write_end(inode, pos, len, copied);
1415 from = pos & (PAGE_SIZE - 1);
1418 BUG_ON(!ext4_handle_valid(handle));
1420 if (ext4_has_inline_data(inode))
1421 copied = ext4_write_inline_data_end(inode, pos, len,
1425 if (!PageUptodate(page))
1427 zero_new_buffers(page, from+copied, to);
1430 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1431 to, &partial, write_end_fn);
1433 SetPageUptodate(page);
1435 size_changed = ext4_update_inode_size(inode, pos + copied);
1436 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1437 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1442 pagecache_isize_extended(inode, old_size, pos);
1445 ret2 = ext4_mark_inode_dirty(handle, inode);
1450 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1451 /* if we have allocated more blocks and copied
1452 * less. We will have blocks allocated outside
1453 * inode->i_size. So truncate them
1455 ext4_orphan_add(handle, inode);
1457 ret2 = ext4_journal_stop(handle);
1460 if (pos + len > inode->i_size) {
1461 ext4_truncate_failed_write(inode);
1463 * If truncate failed early the inode might still be
1464 * on the orphan list; we need to make sure the inode
1465 * is removed from the orphan list in that case.
1468 ext4_orphan_del(NULL, inode);
1471 return ret ? ret : copied;
1475 * Reserve space for a single cluster
1477 static int ext4_da_reserve_space(struct inode *inode)
1479 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1480 struct ext4_inode_info *ei = EXT4_I(inode);
1484 * We will charge metadata quota at writeout time; this saves
1485 * us from metadata over-estimation, though we may go over by
1486 * a small amount in the end. Here we just reserve for data.
1488 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1492 spin_lock(&ei->i_block_reservation_lock);
1493 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1494 spin_unlock(&ei->i_block_reservation_lock);
1495 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1498 ei->i_reserved_data_blocks++;
1499 trace_ext4_da_reserve_space(inode);
1500 spin_unlock(&ei->i_block_reservation_lock);
1502 return 0; /* success */
1505 static void ext4_da_release_space(struct inode *inode, int to_free)
1507 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1508 struct ext4_inode_info *ei = EXT4_I(inode);
1511 return; /* Nothing to release, exit */
1513 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1515 trace_ext4_da_release_space(inode, to_free);
1516 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1518 * if there aren't enough reserved blocks, then the
1519 * counter is messed up somewhere. Since this
1520 * function is called from invalidate page, it's
1521 * harmless to return without any action.
1523 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1524 "ino %lu, to_free %d with only %d reserved "
1525 "data blocks", inode->i_ino, to_free,
1526 ei->i_reserved_data_blocks);
1528 to_free = ei->i_reserved_data_blocks;
1530 ei->i_reserved_data_blocks -= to_free;
1532 /* update fs dirty data blocks counter */
1533 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1535 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1537 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1540 static void ext4_da_page_release_reservation(struct page *page,
1541 unsigned int offset,
1542 unsigned int length)
1544 int to_release = 0, contiguous_blks = 0;
1545 struct buffer_head *head, *bh;
1546 unsigned int curr_off = 0;
1547 struct inode *inode = page->mapping->host;
1548 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1549 unsigned int stop = offset + length;
1553 BUG_ON(stop > PAGE_SIZE || stop < length);
1555 head = page_buffers(page);
1558 unsigned int next_off = curr_off + bh->b_size;
1560 if (next_off > stop)
1563 if ((offset <= curr_off) && (buffer_delay(bh))) {
1566 clear_buffer_delay(bh);
1567 } else if (contiguous_blks) {
1568 lblk = page->index <<
1569 (PAGE_SHIFT - inode->i_blkbits);
1570 lblk += (curr_off >> inode->i_blkbits) -
1572 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1573 contiguous_blks = 0;
1575 curr_off = next_off;
1576 } while ((bh = bh->b_this_page) != head);
1578 if (contiguous_blks) {
1579 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1580 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1581 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1584 /* If we have released all the blocks belonging to a cluster, then we
1585 * need to release the reserved space for that cluster. */
1586 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1587 while (num_clusters > 0) {
1588 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1589 ((num_clusters - 1) << sbi->s_cluster_bits);
1590 if (sbi->s_cluster_ratio == 1 ||
1591 !ext4_find_delalloc_cluster(inode, lblk))
1592 ext4_da_release_space(inode, 1);
1599 * Delayed allocation stuff
1602 struct mpage_da_data {
1603 struct inode *inode;
1604 struct writeback_control *wbc;
1606 pgoff_t first_page; /* The first page to write */
1607 pgoff_t next_page; /* Current page to examine */
1608 pgoff_t last_page; /* Last page to examine */
1610 * Extent to map - this can be after first_page because that can be
1611 * fully mapped. We somewhat abuse m_flags to store whether the extent
1612 * is delalloc or unwritten.
1614 struct ext4_map_blocks map;
1615 struct ext4_io_submit io_submit; /* IO submission data */
1618 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1623 struct pagevec pvec;
1624 struct inode *inode = mpd->inode;
1625 struct address_space *mapping = inode->i_mapping;
1627 /* This is necessary when next_page == 0. */
1628 if (mpd->first_page >= mpd->next_page)
1631 index = mpd->first_page;
1632 end = mpd->next_page - 1;
1634 ext4_lblk_t start, last;
1635 start = index << (PAGE_SHIFT - inode->i_blkbits);
1636 last = end << (PAGE_SHIFT - inode->i_blkbits);
1637 ext4_es_remove_extent(inode, start, last - start + 1);
1640 pagevec_init(&pvec, 0);
1641 while (index <= end) {
1642 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1645 for (i = 0; i < nr_pages; i++) {
1646 struct page *page = pvec.pages[i];
1647 if (page->index > end)
1649 BUG_ON(!PageLocked(page));
1650 BUG_ON(PageWriteback(page));
1652 block_invalidatepage(page, 0, PAGE_SIZE);
1653 ClearPageUptodate(page);
1657 index = pvec.pages[nr_pages - 1]->index + 1;
1658 pagevec_release(&pvec);
1662 static void ext4_print_free_blocks(struct inode *inode)
1664 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1665 struct super_block *sb = inode->i_sb;
1666 struct ext4_inode_info *ei = EXT4_I(inode);
1668 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1669 EXT4_C2B(EXT4_SB(inode->i_sb),
1670 ext4_count_free_clusters(sb)));
1671 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1672 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1673 (long long) EXT4_C2B(EXT4_SB(sb),
1674 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1675 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1676 (long long) EXT4_C2B(EXT4_SB(sb),
1677 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1678 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1679 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1680 ei->i_reserved_data_blocks);
1684 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1686 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1690 * This function is grabs code from the very beginning of
1691 * ext4_map_blocks, but assumes that the caller is from delayed write
1692 * time. This function looks up the requested blocks and sets the
1693 * buffer delay bit under the protection of i_data_sem.
1695 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1696 struct ext4_map_blocks *map,
1697 struct buffer_head *bh)
1699 struct extent_status es;
1701 sector_t invalid_block = ~((sector_t) 0xffff);
1702 #ifdef ES_AGGRESSIVE_TEST
1703 struct ext4_map_blocks orig_map;
1705 memcpy(&orig_map, map, sizeof(*map));
1708 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1712 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1713 "logical block %lu\n", inode->i_ino, map->m_len,
1714 (unsigned long) map->m_lblk);
1716 /* Lookup extent status tree firstly */
1717 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1718 if (ext4_es_is_hole(&es)) {
1720 down_read(&EXT4_I(inode)->i_data_sem);
1725 * Delayed extent could be allocated by fallocate.
1726 * So we need to check it.
1728 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1729 map_bh(bh, inode->i_sb, invalid_block);
1731 set_buffer_delay(bh);
1735 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1736 retval = es.es_len - (iblock - es.es_lblk);
1737 if (retval > map->m_len)
1738 retval = map->m_len;
1739 map->m_len = retval;
1740 if (ext4_es_is_written(&es))
1741 map->m_flags |= EXT4_MAP_MAPPED;
1742 else if (ext4_es_is_unwritten(&es))
1743 map->m_flags |= EXT4_MAP_UNWRITTEN;
1747 #ifdef ES_AGGRESSIVE_TEST
1748 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1754 * Try to see if we can get the block without requesting a new
1755 * file system block.
1757 down_read(&EXT4_I(inode)->i_data_sem);
1758 if (ext4_has_inline_data(inode))
1760 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1761 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1763 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1769 * XXX: __block_prepare_write() unmaps passed block,
1773 * If the block was allocated from previously allocated cluster,
1774 * then we don't need to reserve it again. However we still need
1775 * to reserve metadata for every block we're going to write.
1777 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1778 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1779 ret = ext4_da_reserve_space(inode);
1781 /* not enough space to reserve */
1787 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1788 ~0, EXTENT_STATUS_DELAYED);
1794 map_bh(bh, inode->i_sb, invalid_block);
1796 set_buffer_delay(bh);
1797 } else if (retval > 0) {
1799 unsigned int status;
1801 if (unlikely(retval != map->m_len)) {
1802 ext4_warning(inode->i_sb,
1803 "ES len assertion failed for inode "
1804 "%lu: retval %d != map->m_len %d",
1805 inode->i_ino, retval, map->m_len);
1809 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1810 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1811 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1812 map->m_pblk, status);
1818 up_read((&EXT4_I(inode)->i_data_sem));
1824 * This is a special get_block_t callback which is used by
1825 * ext4_da_write_begin(). It will either return mapped block or
1826 * reserve space for a single block.
1828 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1829 * We also have b_blocknr = -1 and b_bdev initialized properly
1831 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1832 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1833 * initialized properly.
1835 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1836 struct buffer_head *bh, int create)
1838 struct ext4_map_blocks map;
1841 BUG_ON(create == 0);
1842 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1844 map.m_lblk = iblock;
1848 * first, we need to know whether the block is allocated already
1849 * preallocated blocks are unmapped but should treated
1850 * the same as allocated blocks.
1852 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1856 map_bh(bh, inode->i_sb, map.m_pblk);
1857 ext4_update_bh_state(bh, map.m_flags);
1859 if (buffer_unwritten(bh)) {
1860 /* A delayed write to unwritten bh should be marked
1861 * new and mapped. Mapped ensures that we don't do
1862 * get_block multiple times when we write to the same
1863 * offset and new ensures that we do proper zero out
1864 * for partial write.
1867 set_buffer_mapped(bh);
1872 static int bget_one(handle_t *handle, struct buffer_head *bh)
1878 static int bput_one(handle_t *handle, struct buffer_head *bh)
1884 static int __ext4_journalled_writepage(struct page *page,
1887 struct address_space *mapping = page->mapping;
1888 struct inode *inode = mapping->host;
1889 struct buffer_head *page_bufs = NULL;
1890 handle_t *handle = NULL;
1891 int ret = 0, err = 0;
1892 int inline_data = ext4_has_inline_data(inode);
1893 struct buffer_head *inode_bh = NULL;
1895 ClearPageChecked(page);
1898 BUG_ON(page->index != 0);
1899 BUG_ON(len > ext4_get_max_inline_size(inode));
1900 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1901 if (inode_bh == NULL)
1904 page_bufs = page_buffers(page);
1909 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1913 * We need to release the page lock before we start the
1914 * journal, so grab a reference so the page won't disappear
1915 * out from under us.
1920 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1921 ext4_writepage_trans_blocks(inode));
1922 if (IS_ERR(handle)) {
1923 ret = PTR_ERR(handle);
1925 goto out_no_pagelock;
1927 BUG_ON(!ext4_handle_valid(handle));
1931 if (page->mapping != mapping) {
1932 /* The page got truncated from under us */
1933 ext4_journal_stop(handle);
1939 BUFFER_TRACE(inode_bh, "get write access");
1940 ret = ext4_journal_get_write_access(handle, inode_bh);
1942 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1945 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1946 do_journal_get_write_access);
1948 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1953 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1954 err = ext4_journal_stop(handle);
1958 if (!ext4_has_inline_data(inode))
1959 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1961 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1970 * Note that we don't need to start a transaction unless we're journaling data
1971 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1972 * need to file the inode to the transaction's list in ordered mode because if
1973 * we are writing back data added by write(), the inode is already there and if
1974 * we are writing back data modified via mmap(), no one guarantees in which
1975 * transaction the data will hit the disk. In case we are journaling data, we
1976 * cannot start transaction directly because transaction start ranks above page
1977 * lock so we have to do some magic.
1979 * This function can get called via...
1980 * - ext4_writepages after taking page lock (have journal handle)
1981 * - journal_submit_inode_data_buffers (no journal handle)
1982 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1983 * - grab_page_cache when doing write_begin (have journal handle)
1985 * We don't do any block allocation in this function. If we have page with
1986 * multiple blocks we need to write those buffer_heads that are mapped. This
1987 * is important for mmaped based write. So if we do with blocksize 1K
1988 * truncate(f, 1024);
1989 * a = mmap(f, 0, 4096);
1991 * truncate(f, 4096);
1992 * we have in the page first buffer_head mapped via page_mkwrite call back
1993 * but other buffer_heads would be unmapped but dirty (dirty done via the
1994 * do_wp_page). So writepage should write the first block. If we modify
1995 * the mmap area beyond 1024 we will again get a page_fault and the
1996 * page_mkwrite callback will do the block allocation and mark the
1997 * buffer_heads mapped.
1999 * We redirty the page if we have any buffer_heads that is either delay or
2000 * unwritten in the page.
2002 * We can get recursively called as show below.
2004 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2007 * But since we don't do any block allocation we should not deadlock.
2008 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2010 static int ext4_writepage(struct page *page,
2011 struct writeback_control *wbc)
2016 struct buffer_head *page_bufs = NULL;
2017 struct inode *inode = page->mapping->host;
2018 struct ext4_io_submit io_submit;
2019 bool keep_towrite = false;
2021 trace_ext4_writepage(page);
2022 size = i_size_read(inode);
2023 if (page->index == size >> PAGE_SHIFT)
2024 len = size & ~PAGE_MASK;
2028 page_bufs = page_buffers(page);
2030 * We cannot do block allocation or other extent handling in this
2031 * function. If there are buffers needing that, we have to redirty
2032 * the page. But we may reach here when we do a journal commit via
2033 * journal_submit_inode_data_buffers() and in that case we must write
2034 * allocated buffers to achieve data=ordered mode guarantees.
2036 * Also, if there is only one buffer per page (the fs block
2037 * size == the page size), if one buffer needs block
2038 * allocation or needs to modify the extent tree to clear the
2039 * unwritten flag, we know that the page can't be written at
2040 * all, so we might as well refuse the write immediately.
2041 * Unfortunately if the block size != page size, we can't as
2042 * easily detect this case using ext4_walk_page_buffers(), but
2043 * for the extremely common case, this is an optimization that
2044 * skips a useless round trip through ext4_bio_write_page().
2046 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2047 ext4_bh_delay_or_unwritten)) {
2048 redirty_page_for_writepage(wbc, page);
2049 if ((current->flags & PF_MEMALLOC) ||
2050 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2052 * For memory cleaning there's no point in writing only
2053 * some buffers. So just bail out. Warn if we came here
2054 * from direct reclaim.
2056 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2061 keep_towrite = true;
2064 if (PageChecked(page) && ext4_should_journal_data(inode))
2066 * It's mmapped pagecache. Add buffers and journal it. There
2067 * doesn't seem much point in redirtying the page here.
2069 return __ext4_journalled_writepage(page, len);
2071 ext4_io_submit_init(&io_submit, wbc);
2072 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2073 if (!io_submit.io_end) {
2074 redirty_page_for_writepage(wbc, page);
2078 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2079 ext4_io_submit(&io_submit);
2080 /* Drop io_end reference we got from init */
2081 ext4_put_io_end_defer(io_submit.io_end);
2085 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2088 loff_t size = i_size_read(mpd->inode);
2091 BUG_ON(page->index != mpd->first_page);
2092 if (page->index == size >> PAGE_SHIFT)
2093 len = size & ~PAGE_MASK;
2096 clear_page_dirty_for_io(page);
2097 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2099 mpd->wbc->nr_to_write--;
2105 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2108 * mballoc gives us at most this number of blocks...
2109 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2110 * The rest of mballoc seems to handle chunks up to full group size.
2112 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2115 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2117 * @mpd - extent of blocks
2118 * @lblk - logical number of the block in the file
2119 * @bh - buffer head we want to add to the extent
2121 * The function is used to collect contig. blocks in the same state. If the
2122 * buffer doesn't require mapping for writeback and we haven't started the
2123 * extent of buffers to map yet, the function returns 'true' immediately - the
2124 * caller can write the buffer right away. Otherwise the function returns true
2125 * if the block has been added to the extent, false if the block couldn't be
2128 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2129 struct buffer_head *bh)
2131 struct ext4_map_blocks *map = &mpd->map;
2133 /* Buffer that doesn't need mapping for writeback? */
2134 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2135 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2136 /* So far no extent to map => we write the buffer right away */
2137 if (map->m_len == 0)
2142 /* First block in the extent? */
2143 if (map->m_len == 0) {
2146 map->m_flags = bh->b_state & BH_FLAGS;
2150 /* Don't go larger than mballoc is willing to allocate */
2151 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2154 /* Can we merge the block to our big extent? */
2155 if (lblk == map->m_lblk + map->m_len &&
2156 (bh->b_state & BH_FLAGS) == map->m_flags) {
2164 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2166 * @mpd - extent of blocks for mapping
2167 * @head - the first buffer in the page
2168 * @bh - buffer we should start processing from
2169 * @lblk - logical number of the block in the file corresponding to @bh
2171 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2172 * the page for IO if all buffers in this page were mapped and there's no
2173 * accumulated extent of buffers to map or add buffers in the page to the
2174 * extent of buffers to map. The function returns 1 if the caller can continue
2175 * by processing the next page, 0 if it should stop adding buffers to the
2176 * extent to map because we cannot extend it anymore. It can also return value
2177 * < 0 in case of error during IO submission.
2179 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2180 struct buffer_head *head,
2181 struct buffer_head *bh,
2184 struct inode *inode = mpd->inode;
2186 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2187 >> inode->i_blkbits;
2190 BUG_ON(buffer_locked(bh));
2192 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2193 /* Found extent to map? */
2196 /* Everything mapped so far and we hit EOF */
2199 } while (lblk++, (bh = bh->b_this_page) != head);
2200 /* So far everything mapped? Submit the page for IO. */
2201 if (mpd->map.m_len == 0) {
2202 err = mpage_submit_page(mpd, head->b_page);
2206 return lblk < blocks;
2210 * mpage_map_buffers - update buffers corresponding to changed extent and
2211 * submit fully mapped pages for IO
2213 * @mpd - description of extent to map, on return next extent to map
2215 * Scan buffers corresponding to changed extent (we expect corresponding pages
2216 * to be already locked) and update buffer state according to new extent state.
2217 * We map delalloc buffers to their physical location, clear unwritten bits,
2218 * and mark buffers as uninit when we perform writes to unwritten extents
2219 * and do extent conversion after IO is finished. If the last page is not fully
2220 * mapped, we update @map to the next extent in the last page that needs
2221 * mapping. Otherwise we submit the page for IO.
2223 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2225 struct pagevec pvec;
2227 struct inode *inode = mpd->inode;
2228 struct buffer_head *head, *bh;
2229 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2235 start = mpd->map.m_lblk >> bpp_bits;
2236 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2237 lblk = start << bpp_bits;
2238 pblock = mpd->map.m_pblk;
2240 pagevec_init(&pvec, 0);
2241 while (start <= end) {
2242 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2246 for (i = 0; i < nr_pages; i++) {
2247 struct page *page = pvec.pages[i];
2249 if (page->index > end)
2251 /* Up to 'end' pages must be contiguous */
2252 BUG_ON(page->index != start);
2253 bh = head = page_buffers(page);
2255 if (lblk < mpd->map.m_lblk)
2257 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2259 * Buffer after end of mapped extent.
2260 * Find next buffer in the page to map.
2263 mpd->map.m_flags = 0;
2265 * FIXME: If dioread_nolock supports
2266 * blocksize < pagesize, we need to make
2267 * sure we add size mapped so far to
2268 * io_end->size as the following call
2269 * can submit the page for IO.
2271 err = mpage_process_page_bufs(mpd, head,
2273 pagevec_release(&pvec);
2278 if (buffer_delay(bh)) {
2279 clear_buffer_delay(bh);
2280 bh->b_blocknr = pblock++;
2282 clear_buffer_unwritten(bh);
2283 } while (lblk++, (bh = bh->b_this_page) != head);
2286 * FIXME: This is going to break if dioread_nolock
2287 * supports blocksize < pagesize as we will try to
2288 * convert potentially unmapped parts of inode.
2290 mpd->io_submit.io_end->size += PAGE_SIZE;
2291 /* Page fully mapped - let IO run! */
2292 err = mpage_submit_page(mpd, page);
2294 pagevec_release(&pvec);
2299 pagevec_release(&pvec);
2301 /* Extent fully mapped and matches with page boundary. We are done. */
2303 mpd->map.m_flags = 0;
2307 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2309 struct inode *inode = mpd->inode;
2310 struct ext4_map_blocks *map = &mpd->map;
2311 int get_blocks_flags;
2312 int err, dioread_nolock;
2314 trace_ext4_da_write_pages_extent(inode, map);
2316 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2317 * to convert an unwritten extent to be initialized (in the case
2318 * where we have written into one or more preallocated blocks). It is
2319 * possible that we're going to need more metadata blocks than
2320 * previously reserved. However we must not fail because we're in
2321 * writeback and there is nothing we can do about it so it might result
2322 * in data loss. So use reserved blocks to allocate metadata if
2325 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2326 * the blocks in question are delalloc blocks. This indicates
2327 * that the blocks and quotas has already been checked when
2328 * the data was copied into the page cache.
2330 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2331 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2332 EXT4_GET_BLOCKS_IO_SUBMIT;
2333 dioread_nolock = ext4_should_dioread_nolock(inode);
2335 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2336 if (map->m_flags & (1 << BH_Delay))
2337 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2339 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2342 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2343 if (!mpd->io_submit.io_end->handle &&
2344 ext4_handle_valid(handle)) {
2345 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2346 handle->h_rsv_handle = NULL;
2348 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2351 BUG_ON(map->m_len == 0);
2352 if (map->m_flags & EXT4_MAP_NEW) {
2353 struct block_device *bdev = inode->i_sb->s_bdev;
2356 for (i = 0; i < map->m_len; i++)
2357 unmap_underlying_metadata(bdev, map->m_pblk + i);
2363 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2364 * mpd->len and submit pages underlying it for IO
2366 * @handle - handle for journal operations
2367 * @mpd - extent to map
2368 * @give_up_on_write - we set this to true iff there is a fatal error and there
2369 * is no hope of writing the data. The caller should discard
2370 * dirty pages to avoid infinite loops.
2372 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2373 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2374 * them to initialized or split the described range from larger unwritten
2375 * extent. Note that we need not map all the described range since allocation
2376 * can return less blocks or the range is covered by more unwritten extents. We
2377 * cannot map more because we are limited by reserved transaction credits. On
2378 * the other hand we always make sure that the last touched page is fully
2379 * mapped so that it can be written out (and thus forward progress is
2380 * guaranteed). After mapping we submit all mapped pages for IO.
2382 static int mpage_map_and_submit_extent(handle_t *handle,
2383 struct mpage_da_data *mpd,
2384 bool *give_up_on_write)
2386 struct inode *inode = mpd->inode;
2387 struct ext4_map_blocks *map = &mpd->map;
2392 mpd->io_submit.io_end->offset =
2393 ((loff_t)map->m_lblk) << inode->i_blkbits;
2395 err = mpage_map_one_extent(handle, mpd);
2397 struct super_block *sb = inode->i_sb;
2399 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2400 goto invalidate_dirty_pages;
2402 * Let the uper layers retry transient errors.
2403 * In the case of ENOSPC, if ext4_count_free_blocks()
2404 * is non-zero, a commit should free up blocks.
2406 if ((err == -ENOMEM) ||
2407 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2409 goto update_disksize;
2412 ext4_msg(sb, KERN_CRIT,
2413 "Delayed block allocation failed for "
2414 "inode %lu at logical offset %llu with"
2415 " max blocks %u with error %d",
2417 (unsigned long long)map->m_lblk,
2418 (unsigned)map->m_len, -err);
2419 ext4_msg(sb, KERN_CRIT,
2420 "This should not happen!! Data will "
2423 ext4_print_free_blocks(inode);
2424 invalidate_dirty_pages:
2425 *give_up_on_write = true;
2430 * Update buffer state, submit mapped pages, and get us new
2433 err = mpage_map_and_submit_buffers(mpd);
2435 goto update_disksize;
2436 } while (map->m_len);
2440 * Update on-disk size after IO is submitted. Races with
2441 * truncate are avoided by checking i_size under i_data_sem.
2443 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2444 if (disksize > EXT4_I(inode)->i_disksize) {
2448 down_write(&EXT4_I(inode)->i_data_sem);
2449 i_size = i_size_read(inode);
2450 if (disksize > i_size)
2452 if (disksize > EXT4_I(inode)->i_disksize)
2453 EXT4_I(inode)->i_disksize = disksize;
2454 err2 = ext4_mark_inode_dirty(handle, inode);
2455 up_write(&EXT4_I(inode)->i_data_sem);
2457 ext4_error(inode->i_sb,
2458 "Failed to mark inode %lu dirty",
2467 * Calculate the total number of credits to reserve for one writepages
2468 * iteration. This is called from ext4_writepages(). We map an extent of
2469 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2470 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2471 * bpp - 1 blocks in bpp different extents.
2473 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2475 int bpp = ext4_journal_blocks_per_page(inode);
2477 return ext4_meta_trans_blocks(inode,
2478 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2482 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2483 * and underlying extent to map
2485 * @mpd - where to look for pages
2487 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2488 * IO immediately. When we find a page which isn't mapped we start accumulating
2489 * extent of buffers underlying these pages that needs mapping (formed by
2490 * either delayed or unwritten buffers). We also lock the pages containing
2491 * these buffers. The extent found is returned in @mpd structure (starting at
2492 * mpd->lblk with length mpd->len blocks).
2494 * Note that this function can attach bios to one io_end structure which are
2495 * neither logically nor physically contiguous. Although it may seem as an
2496 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2497 * case as we need to track IO to all buffers underlying a page in one io_end.
2499 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2501 struct address_space *mapping = mpd->inode->i_mapping;
2502 struct pagevec pvec;
2503 unsigned int nr_pages;
2504 long left = mpd->wbc->nr_to_write;
2505 pgoff_t index = mpd->first_page;
2506 pgoff_t end = mpd->last_page;
2509 int blkbits = mpd->inode->i_blkbits;
2511 struct buffer_head *head;
2513 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2514 tag = PAGECACHE_TAG_TOWRITE;
2516 tag = PAGECACHE_TAG_DIRTY;
2518 pagevec_init(&pvec, 0);
2520 mpd->next_page = index;
2521 while (index <= end) {
2522 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2523 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2527 for (i = 0; i < nr_pages; i++) {
2528 struct page *page = pvec.pages[i];
2531 * At this point, the page may be truncated or
2532 * invalidated (changing page->mapping to NULL), or
2533 * even swizzled back from swapper_space to tmpfs file
2534 * mapping. However, page->index will not change
2535 * because we have a reference on the page.
2537 if (page->index > end)
2541 * Accumulated enough dirty pages? This doesn't apply
2542 * to WB_SYNC_ALL mode. For integrity sync we have to
2543 * keep going because someone may be concurrently
2544 * dirtying pages, and we might have synced a lot of
2545 * newly appeared dirty pages, but have not synced all
2546 * of the old dirty pages.
2548 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2551 /* If we can't merge this page, we are done. */
2552 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2557 * If the page is no longer dirty, or its mapping no
2558 * longer corresponds to inode we are writing (which
2559 * means it has been truncated or invalidated), or the
2560 * page is already under writeback and we are not doing
2561 * a data integrity writeback, skip the page
2563 if (!PageDirty(page) ||
2564 (PageWriteback(page) &&
2565 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2566 unlikely(page->mapping != mapping)) {
2571 wait_on_page_writeback(page);
2572 BUG_ON(PageWriteback(page));
2574 if (mpd->map.m_len == 0)
2575 mpd->first_page = page->index;
2576 mpd->next_page = page->index + 1;
2577 /* Add all dirty buffers to mpd */
2578 lblk = ((ext4_lblk_t)page->index) <<
2579 (PAGE_SHIFT - blkbits);
2580 head = page_buffers(page);
2581 err = mpage_process_page_bufs(mpd, head, head, lblk);
2587 pagevec_release(&pvec);
2592 pagevec_release(&pvec);
2596 static int __writepage(struct page *page, struct writeback_control *wbc,
2599 struct address_space *mapping = data;
2600 int ret = ext4_writepage(page, wbc);
2601 mapping_set_error(mapping, ret);
2605 static int ext4_writepages(struct address_space *mapping,
2606 struct writeback_control *wbc)
2608 pgoff_t writeback_index = 0;
2609 long nr_to_write = wbc->nr_to_write;
2610 int range_whole = 0;
2612 handle_t *handle = NULL;
2613 struct mpage_da_data mpd;
2614 struct inode *inode = mapping->host;
2615 int needed_blocks, rsv_blocks = 0, ret = 0;
2616 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2618 struct blk_plug plug;
2619 bool give_up_on_write = false;
2621 percpu_down_read(&sbi->s_journal_flag_rwsem);
2622 trace_ext4_writepages(inode, wbc);
2624 if (dax_mapping(mapping)) {
2625 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2627 goto out_writepages;
2631 * No pages to write? This is mainly a kludge to avoid starting
2632 * a transaction for special inodes like journal inode on last iput()
2633 * because that could violate lock ordering on umount
2635 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2636 goto out_writepages;
2638 if (ext4_should_journal_data(inode)) {
2639 struct blk_plug plug;
2641 blk_start_plug(&plug);
2642 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2643 blk_finish_plug(&plug);
2644 goto out_writepages;
2648 * If the filesystem has aborted, it is read-only, so return
2649 * right away instead of dumping stack traces later on that
2650 * will obscure the real source of the problem. We test
2651 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2652 * the latter could be true if the filesystem is mounted
2653 * read-only, and in that case, ext4_writepages should
2654 * *never* be called, so if that ever happens, we would want
2657 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2659 goto out_writepages;
2662 if (ext4_should_dioread_nolock(inode)) {
2664 * We may need to convert up to one extent per block in
2665 * the page and we may dirty the inode.
2667 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2671 * If we have inline data and arrive here, it means that
2672 * we will soon create the block for the 1st page, so
2673 * we'd better clear the inline data here.
2675 if (ext4_has_inline_data(inode)) {
2676 /* Just inode will be modified... */
2677 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2678 if (IS_ERR(handle)) {
2679 ret = PTR_ERR(handle);
2680 goto out_writepages;
2682 BUG_ON(ext4_test_inode_state(inode,
2683 EXT4_STATE_MAY_INLINE_DATA));
2684 ext4_destroy_inline_data(handle, inode);
2685 ext4_journal_stop(handle);
2688 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2691 if (wbc->range_cyclic) {
2692 writeback_index = mapping->writeback_index;
2693 if (writeback_index)
2695 mpd.first_page = writeback_index;
2698 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2699 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2704 ext4_io_submit_init(&mpd.io_submit, wbc);
2706 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2707 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2709 blk_start_plug(&plug);
2710 while (!done && mpd.first_page <= mpd.last_page) {
2711 /* For each extent of pages we use new io_end */
2712 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2713 if (!mpd.io_submit.io_end) {
2719 * We have two constraints: We find one extent to map and we
2720 * must always write out whole page (makes a difference when
2721 * blocksize < pagesize) so that we don't block on IO when we
2722 * try to write out the rest of the page. Journalled mode is
2723 * not supported by delalloc.
2725 BUG_ON(ext4_should_journal_data(inode));
2726 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2728 /* start a new transaction */
2729 handle = ext4_journal_start_with_reserve(inode,
2730 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2731 if (IS_ERR(handle)) {
2732 ret = PTR_ERR(handle);
2733 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2734 "%ld pages, ino %lu; err %d", __func__,
2735 wbc->nr_to_write, inode->i_ino, ret);
2736 /* Release allocated io_end */
2737 ext4_put_io_end(mpd.io_submit.io_end);
2741 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2742 ret = mpage_prepare_extent_to_map(&mpd);
2745 ret = mpage_map_and_submit_extent(handle, &mpd,
2749 * We scanned the whole range (or exhausted
2750 * nr_to_write), submitted what was mapped and
2751 * didn't find anything needing mapping. We are
2757 ext4_journal_stop(handle);
2758 /* Submit prepared bio */
2759 ext4_io_submit(&mpd.io_submit);
2760 /* Unlock pages we didn't use */
2761 mpage_release_unused_pages(&mpd, give_up_on_write);
2762 /* Drop our io_end reference we got from init */
2763 ext4_put_io_end(mpd.io_submit.io_end);
2765 if (ret == -ENOSPC && sbi->s_journal) {
2767 * Commit the transaction which would
2768 * free blocks released in the transaction
2771 jbd2_journal_force_commit_nested(sbi->s_journal);
2775 /* Fatal error - ENOMEM, EIO... */
2779 blk_finish_plug(&plug);
2780 if (!ret && !cycled && wbc->nr_to_write > 0) {
2782 mpd.last_page = writeback_index - 1;
2788 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2790 * Set the writeback_index so that range_cyclic
2791 * mode will write it back later
2793 mapping->writeback_index = mpd.first_page;
2796 trace_ext4_writepages_result(inode, wbc, ret,
2797 nr_to_write - wbc->nr_to_write);
2798 percpu_up_read(&sbi->s_journal_flag_rwsem);
2802 static int ext4_nonda_switch(struct super_block *sb)
2804 s64 free_clusters, dirty_clusters;
2805 struct ext4_sb_info *sbi = EXT4_SB(sb);
2808 * switch to non delalloc mode if we are running low
2809 * on free block. The free block accounting via percpu
2810 * counters can get slightly wrong with percpu_counter_batch getting
2811 * accumulated on each CPU without updating global counters
2812 * Delalloc need an accurate free block accounting. So switch
2813 * to non delalloc when we are near to error range.
2816 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2818 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2820 * Start pushing delalloc when 1/2 of free blocks are dirty.
2822 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2823 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2825 if (2 * free_clusters < 3 * dirty_clusters ||
2826 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2828 * free block count is less than 150% of dirty blocks
2829 * or free blocks is less than watermark
2836 /* We always reserve for an inode update; the superblock could be there too */
2837 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2839 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2842 if (pos + len <= 0x7fffffffULL)
2845 /* We might need to update the superblock to set LARGE_FILE */
2849 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2850 loff_t pos, unsigned len, unsigned flags,
2851 struct page **pagep, void **fsdata)
2853 int ret, retries = 0;
2856 struct inode *inode = mapping->host;
2859 index = pos >> PAGE_SHIFT;
2861 if (ext4_nonda_switch(inode->i_sb)) {
2862 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2863 return ext4_write_begin(file, mapping, pos,
2864 len, flags, pagep, fsdata);
2866 *fsdata = (void *)0;
2867 trace_ext4_da_write_begin(inode, pos, len, flags);
2869 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2870 ret = ext4_da_write_inline_data_begin(mapping, inode,
2880 * grab_cache_page_write_begin() can take a long time if the
2881 * system is thrashing due to memory pressure, or if the page
2882 * is being written back. So grab it first before we start
2883 * the transaction handle. This also allows us to allocate
2884 * the page (if needed) without using GFP_NOFS.
2887 page = grab_cache_page_write_begin(mapping, index, flags);
2893 * With delayed allocation, we don't log the i_disksize update
2894 * if there is delayed block allocation. But we still need
2895 * to journalling the i_disksize update if writes to the end
2896 * of file which has an already mapped buffer.
2899 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2900 ext4_da_write_credits(inode, pos, len));
2901 if (IS_ERR(handle)) {
2903 return PTR_ERR(handle);
2907 if (page->mapping != mapping) {
2908 /* The page got truncated from under us */
2911 ext4_journal_stop(handle);
2914 /* In case writeback began while the page was unlocked */
2915 wait_for_stable_page(page);
2917 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2918 ret = ext4_block_write_begin(page, pos, len,
2919 ext4_da_get_block_prep);
2921 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2925 ext4_journal_stop(handle);
2927 * block_write_begin may have instantiated a few blocks
2928 * outside i_size. Trim these off again. Don't need
2929 * i_size_read because we hold i_mutex.
2931 if (pos + len > inode->i_size)
2932 ext4_truncate_failed_write(inode);
2934 if (ret == -ENOSPC &&
2935 ext4_should_retry_alloc(inode->i_sb, &retries))
2947 * Check if we should update i_disksize
2948 * when write to the end of file but not require block allocation
2950 static int ext4_da_should_update_i_disksize(struct page *page,
2951 unsigned long offset)
2953 struct buffer_head *bh;
2954 struct inode *inode = page->mapping->host;
2958 bh = page_buffers(page);
2959 idx = offset >> inode->i_blkbits;
2961 for (i = 0; i < idx; i++)
2962 bh = bh->b_this_page;
2964 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2969 static int ext4_da_write_end(struct file *file,
2970 struct address_space *mapping,
2971 loff_t pos, unsigned len, unsigned copied,
2972 struct page *page, void *fsdata)
2974 struct inode *inode = mapping->host;
2976 handle_t *handle = ext4_journal_current_handle();
2978 unsigned long start, end;
2979 int write_mode = (int)(unsigned long)fsdata;
2981 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2982 return ext4_write_end(file, mapping, pos,
2983 len, copied, page, fsdata);
2985 trace_ext4_da_write_end(inode, pos, len, copied);
2986 start = pos & (PAGE_SIZE - 1);
2987 end = start + copied - 1;
2990 * generic_write_end() will run mark_inode_dirty() if i_size
2991 * changes. So let's piggyback the i_disksize mark_inode_dirty
2994 new_i_size = pos + copied;
2995 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2996 if (ext4_has_inline_data(inode) ||
2997 ext4_da_should_update_i_disksize(page, end)) {
2998 ext4_update_i_disksize(inode, new_i_size);
2999 /* We need to mark inode dirty even if
3000 * new_i_size is less that inode->i_size
3001 * bu greater than i_disksize.(hint delalloc)
3003 ext4_mark_inode_dirty(handle, inode);
3007 if (write_mode != CONVERT_INLINE_DATA &&
3008 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3009 ext4_has_inline_data(inode))
3010 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3013 ret2 = generic_write_end(file, mapping, pos, len, copied,
3019 ret2 = ext4_journal_stop(handle);
3023 return ret ? ret : copied;
3026 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3027 unsigned int length)
3030 * Drop reserved blocks
3032 BUG_ON(!PageLocked(page));
3033 if (!page_has_buffers(page))
3036 ext4_da_page_release_reservation(page, offset, length);
3039 ext4_invalidatepage(page, offset, length);
3045 * Force all delayed allocation blocks to be allocated for a given inode.
3047 int ext4_alloc_da_blocks(struct inode *inode)
3049 trace_ext4_alloc_da_blocks(inode);
3051 if (!EXT4_I(inode)->i_reserved_data_blocks)
3055 * We do something simple for now. The filemap_flush() will
3056 * also start triggering a write of the data blocks, which is
3057 * not strictly speaking necessary (and for users of
3058 * laptop_mode, not even desirable). However, to do otherwise
3059 * would require replicating code paths in:
3061 * ext4_writepages() ->
3062 * write_cache_pages() ---> (via passed in callback function)
3063 * __mpage_da_writepage() -->
3064 * mpage_add_bh_to_extent()
3065 * mpage_da_map_blocks()
3067 * The problem is that write_cache_pages(), located in
3068 * mm/page-writeback.c, marks pages clean in preparation for
3069 * doing I/O, which is not desirable if we're not planning on
3072 * We could call write_cache_pages(), and then redirty all of
3073 * the pages by calling redirty_page_for_writepage() but that
3074 * would be ugly in the extreme. So instead we would need to
3075 * replicate parts of the code in the above functions,
3076 * simplifying them because we wouldn't actually intend to
3077 * write out the pages, but rather only collect contiguous
3078 * logical block extents, call the multi-block allocator, and
3079 * then update the buffer heads with the block allocations.
3081 * For now, though, we'll cheat by calling filemap_flush(),
3082 * which will map the blocks, and start the I/O, but not
3083 * actually wait for the I/O to complete.
3085 return filemap_flush(inode->i_mapping);
3089 * bmap() is special. It gets used by applications such as lilo and by
3090 * the swapper to find the on-disk block of a specific piece of data.
3092 * Naturally, this is dangerous if the block concerned is still in the
3093 * journal. If somebody makes a swapfile on an ext4 data-journaling
3094 * filesystem and enables swap, then they may get a nasty shock when the
3095 * data getting swapped to that swapfile suddenly gets overwritten by
3096 * the original zero's written out previously to the journal and
3097 * awaiting writeback in the kernel's buffer cache.
3099 * So, if we see any bmap calls here on a modified, data-journaled file,
3100 * take extra steps to flush any blocks which might be in the cache.
3102 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3104 struct inode *inode = mapping->host;
3109 * We can get here for an inline file via the FIBMAP ioctl
3111 if (ext4_has_inline_data(inode))
3114 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3115 test_opt(inode->i_sb, DELALLOC)) {
3117 * With delalloc we want to sync the file
3118 * so that we can make sure we allocate
3121 filemap_write_and_wait(mapping);
3124 if (EXT4_JOURNAL(inode) &&
3125 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3127 * This is a REALLY heavyweight approach, but the use of
3128 * bmap on dirty files is expected to be extremely rare:
3129 * only if we run lilo or swapon on a freshly made file
3130 * do we expect this to happen.
3132 * (bmap requires CAP_SYS_RAWIO so this does not
3133 * represent an unprivileged user DOS attack --- we'd be
3134 * in trouble if mortal users could trigger this path at
3137 * NB. EXT4_STATE_JDATA is not set on files other than
3138 * regular files. If somebody wants to bmap a directory
3139 * or symlink and gets confused because the buffer
3140 * hasn't yet been flushed to disk, they deserve
3141 * everything they get.
3144 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3145 journal = EXT4_JOURNAL(inode);
3146 jbd2_journal_lock_updates(journal);
3147 err = jbd2_journal_flush(journal);
3148 jbd2_journal_unlock_updates(journal);
3154 return generic_block_bmap(mapping, block, ext4_get_block);
3157 static int ext4_readpage(struct file *file, struct page *page)
3160 struct inode *inode = page->mapping->host;
3162 trace_ext4_readpage(page);
3164 if (ext4_has_inline_data(inode))
3165 ret = ext4_readpage_inline(inode, page);
3168 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3174 ext4_readpages(struct file *file, struct address_space *mapping,
3175 struct list_head *pages, unsigned nr_pages)
3177 struct inode *inode = mapping->host;
3179 /* If the file has inline data, no need to do readpages. */
3180 if (ext4_has_inline_data(inode))
3183 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3186 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3187 unsigned int length)
3189 trace_ext4_invalidatepage(page, offset, length);
3191 /* No journalling happens on data buffers when this function is used */
3192 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3194 block_invalidatepage(page, offset, length);
3197 static int __ext4_journalled_invalidatepage(struct page *page,
3198 unsigned int offset,
3199 unsigned int length)
3201 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3203 trace_ext4_journalled_invalidatepage(page, offset, length);
3206 * If it's a full truncate we just forget about the pending dirtying
3208 if (offset == 0 && length == PAGE_SIZE)
3209 ClearPageChecked(page);
3211 return jbd2_journal_invalidatepage(journal, page, offset, length);
3214 /* Wrapper for aops... */
3215 static void ext4_journalled_invalidatepage(struct page *page,
3216 unsigned int offset,
3217 unsigned int length)
3219 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3222 static int ext4_releasepage(struct page *page, gfp_t wait)
3224 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3226 trace_ext4_releasepage(page);
3228 /* Page has dirty journalled data -> cannot release */
3229 if (PageChecked(page))
3232 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3234 return try_to_free_buffers(page);
3237 #ifdef CONFIG_FS_DAX
3239 * Get block function for DAX IO and mmap faults. It takes care of converting
3240 * unwritten extents to written ones and initializes new / converted blocks
3243 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3244 struct buffer_head *bh_result, int create)
3248 ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create);
3250 return _ext4_get_block(inode, iblock, bh_result, 0);
3252 ret = ext4_get_block_trans(inode, iblock, bh_result,
3253 EXT4_GET_BLOCKS_PRE_IO |
3254 EXT4_GET_BLOCKS_CREATE_ZERO);
3258 if (buffer_unwritten(bh_result)) {
3260 * We are protected by i_mmap_sem or i_mutex so we know block
3261 * cannot go away from under us even though we dropped
3262 * i_data_sem. Convert extent to written and write zeros there.
3264 ret = ext4_get_block_trans(inode, iblock, bh_result,
3265 EXT4_GET_BLOCKS_CONVERT |
3266 EXT4_GET_BLOCKS_CREATE_ZERO);
3271 * At least for now we have to clear BH_New so that DAX code
3272 * doesn't attempt to zero blocks again in a racy way.
3274 clear_buffer_new(bh_result);
3278 /* Just define empty function, it will never get called. */
3279 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3280 struct buffer_head *bh_result, int create)
3287 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3288 ssize_t size, void *private)
3290 ext4_io_end_t *io_end = private;
3292 /* if not async direct IO just return */
3296 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3297 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3298 io_end, io_end->inode->i_ino, iocb, offset, size);
3301 * Error during AIO DIO. We cannot convert unwritten extents as the
3302 * data was not written. Just clear the unwritten flag and drop io_end.
3305 ext4_clear_io_unwritten_flag(io_end);
3308 io_end->offset = offset;
3309 io_end->size = size;
3310 ext4_put_io_end(io_end);
3316 * Handling of direct IO writes.
3318 * For ext4 extent files, ext4 will do direct-io write even to holes,
3319 * preallocated extents, and those write extend the file, no need to
3320 * fall back to buffered IO.
3322 * For holes, we fallocate those blocks, mark them as unwritten
3323 * If those blocks were preallocated, we mark sure they are split, but
3324 * still keep the range to write as unwritten.
3326 * The unwritten extents will be converted to written when DIO is completed.
3327 * For async direct IO, since the IO may still pending when return, we
3328 * set up an end_io call back function, which will do the conversion
3329 * when async direct IO completed.
3331 * If the O_DIRECT write will extend the file then add this inode to the
3332 * orphan list. So recovery will truncate it back to the original size
3333 * if the machine crashes during the write.
3336 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3338 struct file *file = iocb->ki_filp;
3339 struct inode *inode = file->f_mapping->host;
3340 struct ext4_inode_info *ei = EXT4_I(inode);
3342 loff_t offset = iocb->ki_pos;
3343 size_t count = iov_iter_count(iter);
3345 get_block_t *get_block_func = NULL;
3347 loff_t final_size = offset + count;
3351 if (final_size > inode->i_size) {
3352 /* Credits for sb + inode write */
3353 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3354 if (IS_ERR(handle)) {
3355 ret = PTR_ERR(handle);
3358 ret = ext4_orphan_add(handle, inode);
3360 ext4_journal_stop(handle);
3364 ei->i_disksize = inode->i_size;
3365 ext4_journal_stop(handle);
3368 BUG_ON(iocb->private == NULL);
3371 * Make all waiters for direct IO properly wait also for extent
3372 * conversion. This also disallows race between truncate() and
3373 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3375 inode_dio_begin(inode);
3377 /* If we do a overwrite dio, i_mutex locking can be released */
3378 overwrite = *((int *)iocb->private);
3381 inode_unlock(inode);
3384 * For extent mapped files we could direct write to holes and fallocate.
3386 * Allocated blocks to fill the hole are marked as unwritten to prevent
3387 * parallel buffered read to expose the stale data before DIO complete
3390 * As to previously fallocated extents, ext4 get_block will just simply
3391 * mark the buffer mapped but still keep the extents unwritten.
3393 * For non AIO case, we will convert those unwritten extents to written
3394 * after return back from blockdev_direct_IO. That way we save us from
3395 * allocating io_end structure and also the overhead of offloading
3396 * the extent convertion to a workqueue.
3398 * For async DIO, the conversion needs to be deferred when the
3399 * IO is completed. The ext4 end_io callback function will be
3400 * called to take care of the conversion work. Here for async
3401 * case, we allocate an io_end structure to hook to the iocb.
3403 iocb->private = NULL;
3405 get_block_func = ext4_dio_get_block_overwrite;
3406 else if (IS_DAX(inode)) {
3408 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3409 * writes need zeroing either because they can race with page
3410 * faults or because they use partial blocks.
3412 if (round_down(offset, 1<<inode->i_blkbits) >= inode->i_size &&
3413 ext4_aligned_io(inode, offset, count))
3414 get_block_func = ext4_dio_get_block;
3416 get_block_func = ext4_dax_get_block;
3417 dio_flags = DIO_LOCKING;
3418 } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3419 round_down(offset, 1 << inode->i_blkbits) >= inode->i_size) {
3420 get_block_func = ext4_dio_get_block;
3421 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3422 } else if (is_sync_kiocb(iocb)) {
3423 get_block_func = ext4_dio_get_block_unwritten_sync;
3424 dio_flags = DIO_LOCKING;
3426 get_block_func = ext4_dio_get_block_unwritten_async;
3427 dio_flags = DIO_LOCKING;
3429 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3430 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3432 if (IS_DAX(inode)) {
3433 ret = dax_do_io(iocb, inode, iter, get_block_func,
3434 ext4_end_io_dio, dio_flags);
3436 ret = __blockdev_direct_IO(iocb, inode,
3437 inode->i_sb->s_bdev, iter,
3439 ext4_end_io_dio, NULL, dio_flags);
3441 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3442 EXT4_STATE_DIO_UNWRITTEN)) {
3445 * for non AIO case, since the IO is already
3446 * completed, we could do the conversion right here
3448 err = ext4_convert_unwritten_extents(NULL, inode,
3452 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3455 inode_dio_end(inode);
3456 /* take i_mutex locking again if we do a ovewrite dio */
3460 if (ret < 0 && final_size > inode->i_size)
3461 ext4_truncate_failed_write(inode);
3463 /* Handle extending of i_size after direct IO write */
3467 /* Credits for sb + inode write */
3468 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3469 if (IS_ERR(handle)) {
3470 /* This is really bad luck. We've written the data
3471 * but cannot extend i_size. Bail out and pretend
3472 * the write failed... */
3473 ret = PTR_ERR(handle);
3475 ext4_orphan_del(NULL, inode);
3480 ext4_orphan_del(handle, inode);
3482 loff_t end = offset + ret;
3483 if (end > inode->i_size) {
3484 ei->i_disksize = end;
3485 i_size_write(inode, end);
3487 * We're going to return a positive `ret'
3488 * here due to non-zero-length I/O, so there's
3489 * no way of reporting error returns from
3490 * ext4_mark_inode_dirty() to userspace. So
3493 ext4_mark_inode_dirty(handle, inode);
3496 err = ext4_journal_stop(handle);
3504 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3507 struct inode *inode = iocb->ki_filp->f_mapping->host;
3510 if (ext4_should_dioread_nolock(inode)) {
3512 * Nolock dioread optimization may be dynamically disabled
3513 * via ext4_inode_block_unlocked_dio(). Check inode's state
3514 * while holding extra i_dio_count ref.
3516 inode_dio_begin(inode);
3518 if (unlikely(ext4_test_inode_state(inode,
3519 EXT4_STATE_DIOREAD_LOCK)))
3520 inode_dio_end(inode);
3524 if (IS_DAX(inode)) {
3525 ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block,
3526 NULL, unlocked ? 0 : DIO_LOCKING);
3528 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3529 iter, ext4_dio_get_block,
3531 unlocked ? 0 : DIO_LOCKING);
3534 inode_dio_end(inode);
3538 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3540 struct file *file = iocb->ki_filp;
3541 struct inode *inode = file->f_mapping->host;
3542 size_t count = iov_iter_count(iter);
3543 loff_t offset = iocb->ki_pos;
3546 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3547 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3552 * If we are doing data journalling we don't support O_DIRECT
3554 if (ext4_should_journal_data(inode))
3557 /* Let buffer I/O handle the inline data case. */
3558 if (ext4_has_inline_data(inode))
3561 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3562 if (iov_iter_rw(iter) == READ)
3563 ret = ext4_direct_IO_read(iocb, iter);
3565 ret = ext4_direct_IO_write(iocb, iter);
3566 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3571 * Pages can be marked dirty completely asynchronously from ext4's journalling
3572 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3573 * much here because ->set_page_dirty is called under VFS locks. The page is
3574 * not necessarily locked.
3576 * We cannot just dirty the page and leave attached buffers clean, because the
3577 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3578 * or jbddirty because all the journalling code will explode.
3580 * So what we do is to mark the page "pending dirty" and next time writepage
3581 * is called, propagate that into the buffers appropriately.
3583 static int ext4_journalled_set_page_dirty(struct page *page)
3585 SetPageChecked(page);
3586 return __set_page_dirty_nobuffers(page);
3589 static const struct address_space_operations ext4_aops = {
3590 .readpage = ext4_readpage,
3591 .readpages = ext4_readpages,
3592 .writepage = ext4_writepage,
3593 .writepages = ext4_writepages,
3594 .write_begin = ext4_write_begin,
3595 .write_end = ext4_write_end,
3597 .invalidatepage = ext4_invalidatepage,
3598 .releasepage = ext4_releasepage,
3599 .direct_IO = ext4_direct_IO,
3600 .migratepage = buffer_migrate_page,
3601 .is_partially_uptodate = block_is_partially_uptodate,
3602 .error_remove_page = generic_error_remove_page,
3605 static const struct address_space_operations ext4_journalled_aops = {
3606 .readpage = ext4_readpage,
3607 .readpages = ext4_readpages,
3608 .writepage = ext4_writepage,
3609 .writepages = ext4_writepages,
3610 .write_begin = ext4_write_begin,
3611 .write_end = ext4_journalled_write_end,
3612 .set_page_dirty = ext4_journalled_set_page_dirty,
3614 .invalidatepage = ext4_journalled_invalidatepage,
3615 .releasepage = ext4_releasepage,
3616 .direct_IO = ext4_direct_IO,
3617 .is_partially_uptodate = block_is_partially_uptodate,
3618 .error_remove_page = generic_error_remove_page,
3621 static const struct address_space_operations ext4_da_aops = {
3622 .readpage = ext4_readpage,
3623 .readpages = ext4_readpages,
3624 .writepage = ext4_writepage,
3625 .writepages = ext4_writepages,
3626 .write_begin = ext4_da_write_begin,
3627 .write_end = ext4_da_write_end,
3629 .invalidatepage = ext4_da_invalidatepage,
3630 .releasepage = ext4_releasepage,
3631 .direct_IO = ext4_direct_IO,
3632 .migratepage = buffer_migrate_page,
3633 .is_partially_uptodate = block_is_partially_uptodate,
3634 .error_remove_page = generic_error_remove_page,
3637 void ext4_set_aops(struct inode *inode)
3639 switch (ext4_inode_journal_mode(inode)) {
3640 case EXT4_INODE_ORDERED_DATA_MODE:
3641 case EXT4_INODE_WRITEBACK_DATA_MODE:
3643 case EXT4_INODE_JOURNAL_DATA_MODE:
3644 inode->i_mapping->a_ops = &ext4_journalled_aops;
3649 if (test_opt(inode->i_sb, DELALLOC))
3650 inode->i_mapping->a_ops = &ext4_da_aops;
3652 inode->i_mapping->a_ops = &ext4_aops;
3655 static int __ext4_block_zero_page_range(handle_t *handle,
3656 struct address_space *mapping, loff_t from, loff_t length)
3658 ext4_fsblk_t index = from >> PAGE_SHIFT;
3659 unsigned offset = from & (PAGE_SIZE-1);
3660 unsigned blocksize, pos;
3662 struct inode *inode = mapping->host;
3663 struct buffer_head *bh;
3667 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3668 mapping_gfp_constraint(mapping, ~__GFP_FS));
3672 blocksize = inode->i_sb->s_blocksize;
3674 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3676 if (!page_has_buffers(page))
3677 create_empty_buffers(page, blocksize, 0);
3679 /* Find the buffer that contains "offset" */
3680 bh = page_buffers(page);
3682 while (offset >= pos) {
3683 bh = bh->b_this_page;
3687 if (buffer_freed(bh)) {
3688 BUFFER_TRACE(bh, "freed: skip");
3691 if (!buffer_mapped(bh)) {
3692 BUFFER_TRACE(bh, "unmapped");
3693 ext4_get_block(inode, iblock, bh, 0);
3694 /* unmapped? It's a hole - nothing to do */
3695 if (!buffer_mapped(bh)) {
3696 BUFFER_TRACE(bh, "still unmapped");
3701 /* Ok, it's mapped. Make sure it's up-to-date */
3702 if (PageUptodate(page))
3703 set_buffer_uptodate(bh);
3705 if (!buffer_uptodate(bh)) {
3707 ll_rw_block(READ, 1, &bh);
3709 /* Uhhuh. Read error. Complain and punt. */
3710 if (!buffer_uptodate(bh))
3712 if (S_ISREG(inode->i_mode) &&
3713 ext4_encrypted_inode(inode)) {
3714 /* We expect the key to be set. */
3715 BUG_ON(!ext4_has_encryption_key(inode));
3716 BUG_ON(blocksize != PAGE_SIZE);
3717 WARN_ON_ONCE(ext4_decrypt(page));
3720 if (ext4_should_journal_data(inode)) {
3721 BUFFER_TRACE(bh, "get write access");
3722 err = ext4_journal_get_write_access(handle, bh);
3726 zero_user(page, offset, length);
3727 BUFFER_TRACE(bh, "zeroed end of block");
3729 if (ext4_should_journal_data(inode)) {
3730 err = ext4_handle_dirty_metadata(handle, inode, bh);
3733 mark_buffer_dirty(bh);
3734 if (ext4_should_order_data(inode))
3735 err = ext4_jbd2_inode_add_write(handle, inode);
3745 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3746 * starting from file offset 'from'. The range to be zero'd must
3747 * be contained with in one block. If the specified range exceeds
3748 * the end of the block it will be shortened to end of the block
3749 * that cooresponds to 'from'
3751 static int ext4_block_zero_page_range(handle_t *handle,
3752 struct address_space *mapping, loff_t from, loff_t length)
3754 struct inode *inode = mapping->host;
3755 unsigned offset = from & (PAGE_SIZE-1);
3756 unsigned blocksize = inode->i_sb->s_blocksize;
3757 unsigned max = blocksize - (offset & (blocksize - 1));
3760 * correct length if it does not fall between
3761 * 'from' and the end of the block
3763 if (length > max || length < 0)
3767 return dax_zero_page_range(inode, from, length, ext4_get_block);
3768 return __ext4_block_zero_page_range(handle, mapping, from, length);
3772 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3773 * up to the end of the block which corresponds to `from'.
3774 * This required during truncate. We need to physically zero the tail end
3775 * of that block so it doesn't yield old data if the file is later grown.
3777 static int ext4_block_truncate_page(handle_t *handle,
3778 struct address_space *mapping, loff_t from)
3780 unsigned offset = from & (PAGE_SIZE-1);
3783 struct inode *inode = mapping->host;
3785 blocksize = inode->i_sb->s_blocksize;
3786 length = blocksize - (offset & (blocksize - 1));
3788 return ext4_block_zero_page_range(handle, mapping, from, length);
3791 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3792 loff_t lstart, loff_t length)
3794 struct super_block *sb = inode->i_sb;
3795 struct address_space *mapping = inode->i_mapping;
3796 unsigned partial_start, partial_end;
3797 ext4_fsblk_t start, end;
3798 loff_t byte_end = (lstart + length - 1);
3801 partial_start = lstart & (sb->s_blocksize - 1);
3802 partial_end = byte_end & (sb->s_blocksize - 1);
3804 start = lstart >> sb->s_blocksize_bits;
3805 end = byte_end >> sb->s_blocksize_bits;
3807 /* Handle partial zero within the single block */
3809 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3810 err = ext4_block_zero_page_range(handle, mapping,
3814 /* Handle partial zero out on the start of the range */
3815 if (partial_start) {
3816 err = ext4_block_zero_page_range(handle, mapping,
3817 lstart, sb->s_blocksize);
3821 /* Handle partial zero out on the end of the range */
3822 if (partial_end != sb->s_blocksize - 1)
3823 err = ext4_block_zero_page_range(handle, mapping,
3824 byte_end - partial_end,
3829 int ext4_can_truncate(struct inode *inode)
3831 if (S_ISREG(inode->i_mode))
3833 if (S_ISDIR(inode->i_mode))
3835 if (S_ISLNK(inode->i_mode))
3836 return !ext4_inode_is_fast_symlink(inode);
3841 * We have to make sure i_disksize gets properly updated before we truncate
3842 * page cache due to hole punching or zero range. Otherwise i_disksize update
3843 * can get lost as it may have been postponed to submission of writeback but
3844 * that will never happen after we truncate page cache.
3846 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3850 loff_t size = i_size_read(inode);
3852 WARN_ON(!inode_is_locked(inode));
3853 if (offset > size || offset + len < size)
3856 if (EXT4_I(inode)->i_disksize >= size)
3859 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3861 return PTR_ERR(handle);
3862 ext4_update_i_disksize(inode, size);
3863 ext4_mark_inode_dirty(handle, inode);
3864 ext4_journal_stop(handle);
3870 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3871 * associated with the given offset and length
3873 * @inode: File inode
3874 * @offset: The offset where the hole will begin
3875 * @len: The length of the hole
3877 * Returns: 0 on success or negative on failure
3880 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3882 struct super_block *sb = inode->i_sb;
3883 ext4_lblk_t first_block, stop_block;
3884 struct address_space *mapping = inode->i_mapping;
3885 loff_t first_block_offset, last_block_offset;
3887 unsigned int credits;
3890 if (!S_ISREG(inode->i_mode))
3893 trace_ext4_punch_hole(inode, offset, length, 0);
3896 * Write out all dirty pages to avoid race conditions
3897 * Then release them.
3899 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3900 ret = filemap_write_and_wait_range(mapping, offset,
3901 offset + length - 1);
3908 /* No need to punch hole beyond i_size */
3909 if (offset >= inode->i_size)
3913 * If the hole extends beyond i_size, set the hole
3914 * to end after the page that contains i_size
3916 if (offset + length > inode->i_size) {
3917 length = inode->i_size +
3918 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3922 if (offset & (sb->s_blocksize - 1) ||
3923 (offset + length) & (sb->s_blocksize - 1)) {
3925 * Attach jinode to inode for jbd2 if we do any zeroing of
3928 ret = ext4_inode_attach_jinode(inode);
3934 /* Wait all existing dio workers, newcomers will block on i_mutex */
3935 ext4_inode_block_unlocked_dio(inode);
3936 inode_dio_wait(inode);
3939 * Prevent page faults from reinstantiating pages we have released from
3942 down_write(&EXT4_I(inode)->i_mmap_sem);
3943 first_block_offset = round_up(offset, sb->s_blocksize);
3944 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3946 /* Now release the pages and zero block aligned part of pages*/
3947 if (last_block_offset > first_block_offset) {
3948 ret = ext4_update_disksize_before_punch(inode, offset, length);
3951 truncate_pagecache_range(inode, first_block_offset,
3955 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3956 credits = ext4_writepage_trans_blocks(inode);
3958 credits = ext4_blocks_for_truncate(inode);
3959 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3960 if (IS_ERR(handle)) {
3961 ret = PTR_ERR(handle);
3962 ext4_std_error(sb, ret);
3966 ret = ext4_zero_partial_blocks(handle, inode, offset,
3971 first_block = (offset + sb->s_blocksize - 1) >>
3972 EXT4_BLOCK_SIZE_BITS(sb);
3973 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3975 /* If there are no blocks to remove, return now */
3976 if (first_block >= stop_block)
3979 down_write(&EXT4_I(inode)->i_data_sem);
3980 ext4_discard_preallocations(inode);
3982 ret = ext4_es_remove_extent(inode, first_block,
3983 stop_block - first_block);
3985 up_write(&EXT4_I(inode)->i_data_sem);
3989 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3990 ret = ext4_ext_remove_space(inode, first_block,
3993 ret = ext4_ind_remove_space(handle, inode, first_block,
3996 up_write(&EXT4_I(inode)->i_data_sem);
3998 ext4_handle_sync(handle);
4000 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4001 ext4_mark_inode_dirty(handle, inode);
4003 ext4_journal_stop(handle);
4005 up_write(&EXT4_I(inode)->i_mmap_sem);
4006 ext4_inode_resume_unlocked_dio(inode);
4008 inode_unlock(inode);
4012 int ext4_inode_attach_jinode(struct inode *inode)
4014 struct ext4_inode_info *ei = EXT4_I(inode);
4015 struct jbd2_inode *jinode;
4017 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4020 jinode = jbd2_alloc_inode(GFP_KERNEL);
4021 spin_lock(&inode->i_lock);
4024 spin_unlock(&inode->i_lock);
4027 ei->jinode = jinode;
4028 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4031 spin_unlock(&inode->i_lock);
4032 if (unlikely(jinode != NULL))
4033 jbd2_free_inode(jinode);
4040 * We block out ext4_get_block() block instantiations across the entire
4041 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4042 * simultaneously on behalf of the same inode.
4044 * As we work through the truncate and commit bits of it to the journal there
4045 * is one core, guiding principle: the file's tree must always be consistent on
4046 * disk. We must be able to restart the truncate after a crash.
4048 * The file's tree may be transiently inconsistent in memory (although it
4049 * probably isn't), but whenever we close off and commit a journal transaction,
4050 * the contents of (the filesystem + the journal) must be consistent and
4051 * restartable. It's pretty simple, really: bottom up, right to left (although
4052 * left-to-right works OK too).
4054 * Note that at recovery time, journal replay occurs *before* the restart of
4055 * truncate against the orphan inode list.
4057 * The committed inode has the new, desired i_size (which is the same as
4058 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4059 * that this inode's truncate did not complete and it will again call
4060 * ext4_truncate() to have another go. So there will be instantiated blocks
4061 * to the right of the truncation point in a crashed ext4 filesystem. But
4062 * that's fine - as long as they are linked from the inode, the post-crash
4063 * ext4_truncate() run will find them and release them.
4065 void ext4_truncate(struct inode *inode)
4067 struct ext4_inode_info *ei = EXT4_I(inode);
4068 unsigned int credits;
4070 struct address_space *mapping = inode->i_mapping;
4073 * There is a possibility that we're either freeing the inode
4074 * or it's a completely new inode. In those cases we might not
4075 * have i_mutex locked because it's not necessary.
4077 if (!(inode->i_state & (I_NEW|I_FREEING)))
4078 WARN_ON(!inode_is_locked(inode));
4079 trace_ext4_truncate_enter(inode);
4081 if (!ext4_can_truncate(inode))
4084 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4086 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4087 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4089 if (ext4_has_inline_data(inode)) {
4092 ext4_inline_data_truncate(inode, &has_inline);
4097 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4098 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4099 if (ext4_inode_attach_jinode(inode) < 0)
4103 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4104 credits = ext4_writepage_trans_blocks(inode);
4106 credits = ext4_blocks_for_truncate(inode);
4108 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4109 if (IS_ERR(handle)) {
4110 ext4_std_error(inode->i_sb, PTR_ERR(handle));
4114 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4115 ext4_block_truncate_page(handle, mapping, inode->i_size);
4118 * We add the inode to the orphan list, so that if this
4119 * truncate spans multiple transactions, and we crash, we will
4120 * resume the truncate when the filesystem recovers. It also
4121 * marks the inode dirty, to catch the new size.
4123 * Implication: the file must always be in a sane, consistent
4124 * truncatable state while each transaction commits.
4126 if (ext4_orphan_add(handle, inode))
4129 down_write(&EXT4_I(inode)->i_data_sem);
4131 ext4_discard_preallocations(inode);
4133 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4134 ext4_ext_truncate(handle, inode);
4136 ext4_ind_truncate(handle, inode);
4138 up_write(&ei->i_data_sem);
4141 ext4_handle_sync(handle);
4145 * If this was a simple ftruncate() and the file will remain alive,
4146 * then we need to clear up the orphan record which we created above.
4147 * However, if this was a real unlink then we were called by
4148 * ext4_evict_inode(), and we allow that function to clean up the
4149 * orphan info for us.
4152 ext4_orphan_del(handle, inode);
4154 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4155 ext4_mark_inode_dirty(handle, inode);
4156 ext4_journal_stop(handle);
4158 trace_ext4_truncate_exit(inode);
4162 * ext4_get_inode_loc returns with an extra refcount against the inode's
4163 * underlying buffer_head on success. If 'in_mem' is true, we have all
4164 * data in memory that is needed to recreate the on-disk version of this
4167 static int __ext4_get_inode_loc(struct inode *inode,
4168 struct ext4_iloc *iloc, int in_mem)
4170 struct ext4_group_desc *gdp;
4171 struct buffer_head *bh;
4172 struct super_block *sb = inode->i_sb;
4174 int inodes_per_block, inode_offset;
4177 if (!ext4_valid_inum(sb, inode->i_ino))
4178 return -EFSCORRUPTED;
4180 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4181 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4186 * Figure out the offset within the block group inode table
4188 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4189 inode_offset = ((inode->i_ino - 1) %
4190 EXT4_INODES_PER_GROUP(sb));
4191 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4192 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4194 bh = sb_getblk(sb, block);
4197 if (!buffer_uptodate(bh)) {
4201 * If the buffer has the write error flag, we have failed
4202 * to write out another inode in the same block. In this
4203 * case, we don't have to read the block because we may
4204 * read the old inode data successfully.
4206 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4207 set_buffer_uptodate(bh);
4209 if (buffer_uptodate(bh)) {
4210 /* someone brought it uptodate while we waited */
4216 * If we have all information of the inode in memory and this
4217 * is the only valid inode in the block, we need not read the
4221 struct buffer_head *bitmap_bh;
4224 start = inode_offset & ~(inodes_per_block - 1);
4226 /* Is the inode bitmap in cache? */
4227 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4228 if (unlikely(!bitmap_bh))
4232 * If the inode bitmap isn't in cache then the
4233 * optimisation may end up performing two reads instead
4234 * of one, so skip it.
4236 if (!buffer_uptodate(bitmap_bh)) {
4240 for (i = start; i < start + inodes_per_block; i++) {
4241 if (i == inode_offset)
4243 if (ext4_test_bit(i, bitmap_bh->b_data))
4247 if (i == start + inodes_per_block) {
4248 /* all other inodes are free, so skip I/O */
4249 memset(bh->b_data, 0, bh->b_size);
4250 set_buffer_uptodate(bh);
4258 * If we need to do any I/O, try to pre-readahead extra
4259 * blocks from the inode table.
4261 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4262 ext4_fsblk_t b, end, table;
4264 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4266 table = ext4_inode_table(sb, gdp);
4267 /* s_inode_readahead_blks is always a power of 2 */
4268 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4272 num = EXT4_INODES_PER_GROUP(sb);
4273 if (ext4_has_group_desc_csum(sb))
4274 num -= ext4_itable_unused_count(sb, gdp);
4275 table += num / inodes_per_block;
4279 sb_breadahead(sb, b++);
4283 * There are other valid inodes in the buffer, this inode
4284 * has in-inode xattrs, or we don't have this inode in memory.
4285 * Read the block from disk.
4287 trace_ext4_load_inode(inode);
4289 bh->b_end_io = end_buffer_read_sync;
4290 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4292 if (!buffer_uptodate(bh)) {
4293 EXT4_ERROR_INODE_BLOCK(inode, block,
4294 "unable to read itable block");
4304 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4306 /* We have all inode data except xattrs in memory here. */
4307 return __ext4_get_inode_loc(inode, iloc,
4308 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4311 void ext4_set_inode_flags(struct inode *inode)
4313 unsigned int flags = EXT4_I(inode)->i_flags;
4314 unsigned int new_fl = 0;
4316 if (flags & EXT4_SYNC_FL)
4318 if (flags & EXT4_APPEND_FL)
4320 if (flags & EXT4_IMMUTABLE_FL)
4321 new_fl |= S_IMMUTABLE;
4322 if (flags & EXT4_NOATIME_FL)
4323 new_fl |= S_NOATIME;
4324 if (flags & EXT4_DIRSYNC_FL)
4325 new_fl |= S_DIRSYNC;
4326 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
4328 inode_set_flags(inode, new_fl,
4329 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4332 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4333 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4335 unsigned int vfs_fl;
4336 unsigned long old_fl, new_fl;
4339 vfs_fl = ei->vfs_inode.i_flags;
4340 old_fl = ei->i_flags;
4341 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4342 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4344 if (vfs_fl & S_SYNC)
4345 new_fl |= EXT4_SYNC_FL;
4346 if (vfs_fl & S_APPEND)
4347 new_fl |= EXT4_APPEND_FL;
4348 if (vfs_fl & S_IMMUTABLE)
4349 new_fl |= EXT4_IMMUTABLE_FL;
4350 if (vfs_fl & S_NOATIME)
4351 new_fl |= EXT4_NOATIME_FL;
4352 if (vfs_fl & S_DIRSYNC)
4353 new_fl |= EXT4_DIRSYNC_FL;
4354 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4357 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4358 struct ext4_inode_info *ei)
4361 struct inode *inode = &(ei->vfs_inode);
4362 struct super_block *sb = inode->i_sb;
4364 if (ext4_has_feature_huge_file(sb)) {
4365 /* we are using combined 48 bit field */
4366 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4367 le32_to_cpu(raw_inode->i_blocks_lo);
4368 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4369 /* i_blocks represent file system block size */
4370 return i_blocks << (inode->i_blkbits - 9);
4375 return le32_to_cpu(raw_inode->i_blocks_lo);
4379 static inline void ext4_iget_extra_inode(struct inode *inode,
4380 struct ext4_inode *raw_inode,
4381 struct ext4_inode_info *ei)
4383 __le32 *magic = (void *)raw_inode +
4384 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4385 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4386 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4387 ext4_find_inline_data_nolock(inode);
4389 EXT4_I(inode)->i_inline_off = 0;
4392 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4394 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, EXT4_FEATURE_RO_COMPAT_PROJECT))
4396 *projid = EXT4_I(inode)->i_projid;
4400 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4402 struct ext4_iloc iloc;
4403 struct ext4_inode *raw_inode;
4404 struct ext4_inode_info *ei;
4405 struct inode *inode;
4406 journal_t *journal = EXT4_SB(sb)->s_journal;
4413 inode = iget_locked(sb, ino);
4415 return ERR_PTR(-ENOMEM);
4416 if (!(inode->i_state & I_NEW))
4422 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4425 raw_inode = ext4_raw_inode(&iloc);
4427 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4428 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4429 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4430 EXT4_INODE_SIZE(inode->i_sb)) {
4431 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4432 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4433 EXT4_INODE_SIZE(inode->i_sb));
4434 ret = -EFSCORRUPTED;
4438 ei->i_extra_isize = 0;
4440 /* Precompute checksum seed for inode metadata */
4441 if (ext4_has_metadata_csum(sb)) {
4442 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4444 __le32 inum = cpu_to_le32(inode->i_ino);
4445 __le32 gen = raw_inode->i_generation;
4446 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4448 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4452 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4453 EXT4_ERROR_INODE(inode, "checksum invalid");
4458 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4459 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4460 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4461 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4462 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4463 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4464 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4466 i_projid = EXT4_DEF_PROJID;
4468 if (!(test_opt(inode->i_sb, NO_UID32))) {
4469 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4470 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4472 i_uid_write(inode, i_uid);
4473 i_gid_write(inode, i_gid);
4474 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4475 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4477 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4478 ei->i_inline_off = 0;
4479 ei->i_dir_start_lookup = 0;
4480 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4481 /* We now have enough fields to check if the inode was active or not.
4482 * This is needed because nfsd might try to access dead inodes
4483 * the test is that same one that e2fsck uses
4484 * NeilBrown 1999oct15
4486 if (inode->i_nlink == 0) {
4487 if ((inode->i_mode == 0 ||
4488 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4489 ino != EXT4_BOOT_LOADER_INO) {
4490 /* this inode is deleted */
4494 /* The only unlinked inodes we let through here have
4495 * valid i_mode and are being read by the orphan
4496 * recovery code: that's fine, we're about to complete
4497 * the process of deleting those.
4498 * OR it is the EXT4_BOOT_LOADER_INO which is
4499 * not initialized on a new filesystem. */
4501 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4502 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4503 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4504 if (ext4_has_feature_64bit(sb))
4506 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4507 inode->i_size = ext4_isize(raw_inode);
4508 ei->i_disksize = inode->i_size;
4510 ei->i_reserved_quota = 0;
4512 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4513 ei->i_block_group = iloc.block_group;
4514 ei->i_last_alloc_group = ~0;
4516 * NOTE! The in-memory inode i_data array is in little-endian order
4517 * even on big-endian machines: we do NOT byteswap the block numbers!
4519 for (block = 0; block < EXT4_N_BLOCKS; block++)
4520 ei->i_data[block] = raw_inode->i_block[block];
4521 INIT_LIST_HEAD(&ei->i_orphan);
4524 * Set transaction id's of transactions that have to be committed
4525 * to finish f[data]sync. We set them to currently running transaction
4526 * as we cannot be sure that the inode or some of its metadata isn't
4527 * part of the transaction - the inode could have been reclaimed and
4528 * now it is reread from disk.
4531 transaction_t *transaction;
4534 read_lock(&journal->j_state_lock);
4535 if (journal->j_running_transaction)
4536 transaction = journal->j_running_transaction;
4538 transaction = journal->j_committing_transaction;
4540 tid = transaction->t_tid;
4542 tid = journal->j_commit_sequence;
4543 read_unlock(&journal->j_state_lock);
4544 ei->i_sync_tid = tid;
4545 ei->i_datasync_tid = tid;
4548 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4549 if (ei->i_extra_isize == 0) {
4550 /* The extra space is currently unused. Use it. */
4551 ei->i_extra_isize = sizeof(struct ext4_inode) -
4552 EXT4_GOOD_OLD_INODE_SIZE;
4554 ext4_iget_extra_inode(inode, raw_inode, ei);
4558 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4559 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4560 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4561 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4563 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4564 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4565 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4566 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4568 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4573 if (ei->i_file_acl &&
4574 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4575 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4577 ret = -EFSCORRUPTED;
4579 } else if (!ext4_has_inline_data(inode)) {
4580 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4581 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4582 (S_ISLNK(inode->i_mode) &&
4583 !ext4_inode_is_fast_symlink(inode))))
4584 /* Validate extent which is part of inode */
4585 ret = ext4_ext_check_inode(inode);
4586 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4587 (S_ISLNK(inode->i_mode) &&
4588 !ext4_inode_is_fast_symlink(inode))) {
4589 /* Validate block references which are part of inode */
4590 ret = ext4_ind_check_inode(inode);
4596 if (S_ISREG(inode->i_mode)) {
4597 inode->i_op = &ext4_file_inode_operations;
4598 inode->i_fop = &ext4_file_operations;
4599 ext4_set_aops(inode);
4600 } else if (S_ISDIR(inode->i_mode)) {
4601 inode->i_op = &ext4_dir_inode_operations;
4602 inode->i_fop = &ext4_dir_operations;
4603 } else if (S_ISLNK(inode->i_mode)) {
4604 if (ext4_encrypted_inode(inode)) {
4605 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4606 ext4_set_aops(inode);
4607 } else if (ext4_inode_is_fast_symlink(inode)) {
4608 inode->i_link = (char *)ei->i_data;
4609 inode->i_op = &ext4_fast_symlink_inode_operations;
4610 nd_terminate_link(ei->i_data, inode->i_size,
4611 sizeof(ei->i_data) - 1);
4613 inode->i_op = &ext4_symlink_inode_operations;
4614 ext4_set_aops(inode);
4616 inode_nohighmem(inode);
4617 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4618 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4619 inode->i_op = &ext4_special_inode_operations;
4620 if (raw_inode->i_block[0])
4621 init_special_inode(inode, inode->i_mode,
4622 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4624 init_special_inode(inode, inode->i_mode,
4625 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4626 } else if (ino == EXT4_BOOT_LOADER_INO) {
4627 make_bad_inode(inode);
4629 ret = -EFSCORRUPTED;
4630 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4634 ext4_set_inode_flags(inode);
4635 unlock_new_inode(inode);
4641 return ERR_PTR(ret);
4644 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4646 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4647 return ERR_PTR(-EFSCORRUPTED);
4648 return ext4_iget(sb, ino);
4651 static int ext4_inode_blocks_set(handle_t *handle,
4652 struct ext4_inode *raw_inode,
4653 struct ext4_inode_info *ei)
4655 struct inode *inode = &(ei->vfs_inode);
4656 u64 i_blocks = inode->i_blocks;
4657 struct super_block *sb = inode->i_sb;
4659 if (i_blocks <= ~0U) {
4661 * i_blocks can be represented in a 32 bit variable
4662 * as multiple of 512 bytes
4664 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4665 raw_inode->i_blocks_high = 0;
4666 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4669 if (!ext4_has_feature_huge_file(sb))
4672 if (i_blocks <= 0xffffffffffffULL) {
4674 * i_blocks can be represented in a 48 bit variable
4675 * as multiple of 512 bytes
4677 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4678 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4679 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4681 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4682 /* i_block is stored in file system block size */
4683 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4684 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4685 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4690 struct other_inode {
4691 unsigned long orig_ino;
4692 struct ext4_inode *raw_inode;
4695 static int other_inode_match(struct inode * inode, unsigned long ino,
4698 struct other_inode *oi = (struct other_inode *) data;
4700 if ((inode->i_ino != ino) ||
4701 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4702 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4703 ((inode->i_state & I_DIRTY_TIME) == 0))
4705 spin_lock(&inode->i_lock);
4706 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4707 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4708 (inode->i_state & I_DIRTY_TIME)) {
4709 struct ext4_inode_info *ei = EXT4_I(inode);
4711 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4712 spin_unlock(&inode->i_lock);
4714 spin_lock(&ei->i_raw_lock);
4715 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4716 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4717 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4718 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4719 spin_unlock(&ei->i_raw_lock);
4720 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4723 spin_unlock(&inode->i_lock);
4728 * Opportunistically update the other time fields for other inodes in
4729 * the same inode table block.
4731 static void ext4_update_other_inodes_time(struct super_block *sb,
4732 unsigned long orig_ino, char *buf)
4734 struct other_inode oi;
4736 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4737 int inode_size = EXT4_INODE_SIZE(sb);
4739 oi.orig_ino = orig_ino;
4741 * Calculate the first inode in the inode table block. Inode
4742 * numbers are one-based. That is, the first inode in a block
4743 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4745 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4746 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4747 if (ino == orig_ino)
4749 oi.raw_inode = (struct ext4_inode *) buf;
4750 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4755 * Post the struct inode info into an on-disk inode location in the
4756 * buffer-cache. This gobbles the caller's reference to the
4757 * buffer_head in the inode location struct.
4759 * The caller must have write access to iloc->bh.
4761 static int ext4_do_update_inode(handle_t *handle,
4762 struct inode *inode,
4763 struct ext4_iloc *iloc)
4765 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4766 struct ext4_inode_info *ei = EXT4_I(inode);
4767 struct buffer_head *bh = iloc->bh;
4768 struct super_block *sb = inode->i_sb;
4769 int err = 0, rc, block;
4770 int need_datasync = 0, set_large_file = 0;
4775 spin_lock(&ei->i_raw_lock);
4777 /* For fields not tracked in the in-memory inode,
4778 * initialise them to zero for new inodes. */
4779 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4780 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4782 ext4_get_inode_flags(ei);
4783 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4784 i_uid = i_uid_read(inode);
4785 i_gid = i_gid_read(inode);
4786 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4787 if (!(test_opt(inode->i_sb, NO_UID32))) {
4788 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4789 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4791 * Fix up interoperability with old kernels. Otherwise, old inodes get
4792 * re-used with the upper 16 bits of the uid/gid intact
4795 raw_inode->i_uid_high =
4796 cpu_to_le16(high_16_bits(i_uid));
4797 raw_inode->i_gid_high =
4798 cpu_to_le16(high_16_bits(i_gid));
4800 raw_inode->i_uid_high = 0;
4801 raw_inode->i_gid_high = 0;
4804 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4805 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4806 raw_inode->i_uid_high = 0;
4807 raw_inode->i_gid_high = 0;
4809 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4811 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4812 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4813 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4814 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4816 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4818 spin_unlock(&ei->i_raw_lock);
4821 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4822 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4823 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4824 raw_inode->i_file_acl_high =
4825 cpu_to_le16(ei->i_file_acl >> 32);
4826 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4827 if (ei->i_disksize != ext4_isize(raw_inode)) {
4828 ext4_isize_set(raw_inode, ei->i_disksize);
4831 if (ei->i_disksize > 0x7fffffffULL) {
4832 if (!ext4_has_feature_large_file(sb) ||
4833 EXT4_SB(sb)->s_es->s_rev_level ==
4834 cpu_to_le32(EXT4_GOOD_OLD_REV))
4837 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4838 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4839 if (old_valid_dev(inode->i_rdev)) {
4840 raw_inode->i_block[0] =
4841 cpu_to_le32(old_encode_dev(inode->i_rdev));
4842 raw_inode->i_block[1] = 0;
4844 raw_inode->i_block[0] = 0;
4845 raw_inode->i_block[1] =
4846 cpu_to_le32(new_encode_dev(inode->i_rdev));
4847 raw_inode->i_block[2] = 0;
4849 } else if (!ext4_has_inline_data(inode)) {
4850 for (block = 0; block < EXT4_N_BLOCKS; block++)
4851 raw_inode->i_block[block] = ei->i_data[block];
4854 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4855 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4856 if (ei->i_extra_isize) {
4857 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4858 raw_inode->i_version_hi =
4859 cpu_to_le32(inode->i_version >> 32);
4860 raw_inode->i_extra_isize =
4861 cpu_to_le16(ei->i_extra_isize);
4865 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
4866 EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4867 i_projid != EXT4_DEF_PROJID);
4869 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4870 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4871 raw_inode->i_projid = cpu_to_le32(i_projid);
4873 ext4_inode_csum_set(inode, raw_inode, ei);
4874 spin_unlock(&ei->i_raw_lock);
4875 if (inode->i_sb->s_flags & MS_LAZYTIME)
4876 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4879 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4880 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4883 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4884 if (set_large_file) {
4885 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4886 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4889 ext4_update_dynamic_rev(sb);
4890 ext4_set_feature_large_file(sb);
4891 ext4_handle_sync(handle);
4892 err = ext4_handle_dirty_super(handle, sb);
4894 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4897 ext4_std_error(inode->i_sb, err);
4902 * ext4_write_inode()
4904 * We are called from a few places:
4906 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4907 * Here, there will be no transaction running. We wait for any running
4908 * transaction to commit.
4910 * - Within flush work (sys_sync(), kupdate and such).
4911 * We wait on commit, if told to.
4913 * - Within iput_final() -> write_inode_now()
4914 * We wait on commit, if told to.
4916 * In all cases it is actually safe for us to return without doing anything,
4917 * because the inode has been copied into a raw inode buffer in
4918 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4921 * Note that we are absolutely dependent upon all inode dirtiers doing the
4922 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4923 * which we are interested.
4925 * It would be a bug for them to not do this. The code:
4927 * mark_inode_dirty(inode)
4929 * inode->i_size = expr;
4931 * is in error because write_inode() could occur while `stuff()' is running,
4932 * and the new i_size will be lost. Plus the inode will no longer be on the
4933 * superblock's dirty inode list.
4935 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4939 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4942 if (EXT4_SB(inode->i_sb)->s_journal) {
4943 if (ext4_journal_current_handle()) {
4944 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4950 * No need to force transaction in WB_SYNC_NONE mode. Also
4951 * ext4_sync_fs() will force the commit after everything is
4954 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4957 err = ext4_force_commit(inode->i_sb);
4959 struct ext4_iloc iloc;
4961 err = __ext4_get_inode_loc(inode, &iloc, 0);
4965 * sync(2) will flush the whole buffer cache. No need to do
4966 * it here separately for each inode.
4968 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4969 sync_dirty_buffer(iloc.bh);
4970 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4971 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4972 "IO error syncing inode");
4981 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4982 * buffers that are attached to a page stradding i_size and are undergoing
4983 * commit. In that case we have to wait for commit to finish and try again.
4985 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4989 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4990 tid_t commit_tid = 0;
4993 offset = inode->i_size & (PAGE_SIZE - 1);
4995 * All buffers in the last page remain valid? Then there's nothing to
4996 * do. We do the check mainly to optimize the common PAGE_SIZE ==
4999 if (offset > PAGE_SIZE - (1 << inode->i_blkbits))
5002 page = find_lock_page(inode->i_mapping,
5003 inode->i_size >> PAGE_SHIFT);
5006 ret = __ext4_journalled_invalidatepage(page, offset,
5007 PAGE_SIZE - offset);
5013 read_lock(&journal->j_state_lock);
5014 if (journal->j_committing_transaction)
5015 commit_tid = journal->j_committing_transaction->t_tid;
5016 read_unlock(&journal->j_state_lock);
5018 jbd2_log_wait_commit(journal, commit_tid);
5025 * Called from notify_change.
5027 * We want to trap VFS attempts to truncate the file as soon as
5028 * possible. In particular, we want to make sure that when the VFS
5029 * shrinks i_size, we put the inode on the orphan list and modify
5030 * i_disksize immediately, so that during the subsequent flushing of
5031 * dirty pages and freeing of disk blocks, we can guarantee that any
5032 * commit will leave the blocks being flushed in an unused state on
5033 * disk. (On recovery, the inode will get truncated and the blocks will
5034 * be freed, so we have a strong guarantee that no future commit will
5035 * leave these blocks visible to the user.)
5037 * Another thing we have to assure is that if we are in ordered mode
5038 * and inode is still attached to the committing transaction, we must
5039 * we start writeout of all the dirty pages which are being truncated.
5040 * This way we are sure that all the data written in the previous
5041 * transaction are already on disk (truncate waits for pages under
5044 * Called with inode->i_mutex down.
5046 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5048 struct inode *inode = d_inode(dentry);
5051 const unsigned int ia_valid = attr->ia_valid;
5053 error = inode_change_ok(inode, attr);
5057 if (is_quota_modification(inode, attr)) {
5058 error = dquot_initialize(inode);
5062 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5063 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5066 /* (user+group)*(old+new) structure, inode write (sb,
5067 * inode block, ? - but truncate inode update has it) */
5068 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5069 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5070 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5071 if (IS_ERR(handle)) {
5072 error = PTR_ERR(handle);
5075 error = dquot_transfer(inode, attr);
5077 ext4_journal_stop(handle);
5080 /* Update corresponding info in inode so that everything is in
5081 * one transaction */
5082 if (attr->ia_valid & ATTR_UID)
5083 inode->i_uid = attr->ia_uid;
5084 if (attr->ia_valid & ATTR_GID)
5085 inode->i_gid = attr->ia_gid;
5086 error = ext4_mark_inode_dirty(handle, inode);
5087 ext4_journal_stop(handle);
5090 if (attr->ia_valid & ATTR_SIZE) {
5092 loff_t oldsize = inode->i_size;
5093 int shrink = (attr->ia_size <= inode->i_size);
5095 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5096 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5098 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5101 if (!S_ISREG(inode->i_mode))
5104 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5105 inode_inc_iversion(inode);
5107 if (ext4_should_order_data(inode) &&
5108 (attr->ia_size < inode->i_size)) {
5109 error = ext4_begin_ordered_truncate(inode,
5114 if (attr->ia_size != inode->i_size) {
5115 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5116 if (IS_ERR(handle)) {
5117 error = PTR_ERR(handle);
5120 if (ext4_handle_valid(handle) && shrink) {
5121 error = ext4_orphan_add(handle, inode);
5125 * Update c/mtime on truncate up, ext4_truncate() will
5126 * update c/mtime in shrink case below
5129 inode->i_mtime = ext4_current_time(inode);
5130 inode->i_ctime = inode->i_mtime;
5132 down_write(&EXT4_I(inode)->i_data_sem);
5133 EXT4_I(inode)->i_disksize = attr->ia_size;
5134 rc = ext4_mark_inode_dirty(handle, inode);
5138 * We have to update i_size under i_data_sem together
5139 * with i_disksize to avoid races with writeback code
5140 * running ext4_wb_update_i_disksize().
5143 i_size_write(inode, attr->ia_size);
5144 up_write(&EXT4_I(inode)->i_data_sem);
5145 ext4_journal_stop(handle);
5148 ext4_orphan_del(NULL, inode);
5153 pagecache_isize_extended(inode, oldsize, inode->i_size);
5156 * Blocks are going to be removed from the inode. Wait
5157 * for dio in flight. Temporarily disable
5158 * dioread_nolock to prevent livelock.
5161 if (!ext4_should_journal_data(inode)) {
5162 ext4_inode_block_unlocked_dio(inode);
5163 inode_dio_wait(inode);
5164 ext4_inode_resume_unlocked_dio(inode);
5166 ext4_wait_for_tail_page_commit(inode);
5168 down_write(&EXT4_I(inode)->i_mmap_sem);
5170 * Truncate pagecache after we've waited for commit
5171 * in data=journal mode to make pages freeable.
5173 truncate_pagecache(inode, inode->i_size);
5175 ext4_truncate(inode);
5176 up_write(&EXT4_I(inode)->i_mmap_sem);
5180 setattr_copy(inode, attr);
5181 mark_inode_dirty(inode);
5185 * If the call to ext4_truncate failed to get a transaction handle at
5186 * all, we need to clean up the in-core orphan list manually.
5188 if (orphan && inode->i_nlink)
5189 ext4_orphan_del(NULL, inode);
5191 if (!rc && (ia_valid & ATTR_MODE))
5192 rc = posix_acl_chmod(inode, inode->i_mode);
5195 ext4_std_error(inode->i_sb, error);
5201 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5204 struct inode *inode;
5205 unsigned long long delalloc_blocks;
5207 inode = d_inode(dentry);
5208 generic_fillattr(inode, stat);
5211 * If there is inline data in the inode, the inode will normally not
5212 * have data blocks allocated (it may have an external xattr block).
5213 * Report at least one sector for such files, so tools like tar, rsync,
5214 * others doen't incorrectly think the file is completely sparse.
5216 if (unlikely(ext4_has_inline_data(inode)))
5217 stat->blocks += (stat->size + 511) >> 9;
5220 * We can't update i_blocks if the block allocation is delayed
5221 * otherwise in the case of system crash before the real block
5222 * allocation is done, we will have i_blocks inconsistent with
5223 * on-disk file blocks.
5224 * We always keep i_blocks updated together with real
5225 * allocation. But to not confuse with user, stat
5226 * will return the blocks that include the delayed allocation
5227 * blocks for this file.
5229 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5230 EXT4_I(inode)->i_reserved_data_blocks);
5231 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5235 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5238 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5239 return ext4_ind_trans_blocks(inode, lblocks);
5240 return ext4_ext_index_trans_blocks(inode, pextents);
5244 * Account for index blocks, block groups bitmaps and block group
5245 * descriptor blocks if modify datablocks and index blocks
5246 * worse case, the indexs blocks spread over different block groups
5248 * If datablocks are discontiguous, they are possible to spread over
5249 * different block groups too. If they are contiguous, with flexbg,
5250 * they could still across block group boundary.
5252 * Also account for superblock, inode, quota and xattr blocks
5254 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5257 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5263 * How many index blocks need to touch to map @lblocks logical blocks
5264 * to @pextents physical extents?
5266 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5271 * Now let's see how many group bitmaps and group descriptors need
5274 groups = idxblocks + pextents;
5276 if (groups > ngroups)
5278 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5279 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5281 /* bitmaps and block group descriptor blocks */
5282 ret += groups + gdpblocks;
5284 /* Blocks for super block, inode, quota and xattr blocks */
5285 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5291 * Calculate the total number of credits to reserve to fit
5292 * the modification of a single pages into a single transaction,
5293 * which may include multiple chunks of block allocations.
5295 * This could be called via ext4_write_begin()
5297 * We need to consider the worse case, when
5298 * one new block per extent.
5300 int ext4_writepage_trans_blocks(struct inode *inode)
5302 int bpp = ext4_journal_blocks_per_page(inode);
5305 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5307 /* Account for data blocks for journalled mode */
5308 if (ext4_should_journal_data(inode))
5314 * Calculate the journal credits for a chunk of data modification.
5316 * This is called from DIO, fallocate or whoever calling
5317 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5319 * journal buffers for data blocks are not included here, as DIO
5320 * and fallocate do no need to journal data buffers.
5322 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5324 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5328 * The caller must have previously called ext4_reserve_inode_write().
5329 * Give this, we know that the caller already has write access to iloc->bh.
5331 int ext4_mark_iloc_dirty(handle_t *handle,
5332 struct inode *inode, struct ext4_iloc *iloc)
5336 if (IS_I_VERSION(inode))
5337 inode_inc_iversion(inode);
5339 /* the do_update_inode consumes one bh->b_count */
5342 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5343 err = ext4_do_update_inode(handle, inode, iloc);
5349 * On success, We end up with an outstanding reference count against
5350 * iloc->bh. This _must_ be cleaned up later.
5354 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5355 struct ext4_iloc *iloc)
5359 err = ext4_get_inode_loc(inode, iloc);
5361 BUFFER_TRACE(iloc->bh, "get_write_access");
5362 err = ext4_journal_get_write_access(handle, iloc->bh);
5368 ext4_std_error(inode->i_sb, err);
5373 * Expand an inode by new_extra_isize bytes.
5374 * Returns 0 on success or negative error number on failure.
5376 static int ext4_expand_extra_isize(struct inode *inode,
5377 unsigned int new_extra_isize,
5378 struct ext4_iloc iloc,
5381 struct ext4_inode *raw_inode;
5382 struct ext4_xattr_ibody_header *header;
5384 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5387 raw_inode = ext4_raw_inode(&iloc);
5389 header = IHDR(inode, raw_inode);
5391 /* No extended attributes present */
5392 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5393 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5394 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5396 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5400 /* try to expand with EAs present */
5401 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5406 * What we do here is to mark the in-core inode as clean with respect to inode
5407 * dirtiness (it may still be data-dirty).
5408 * This means that the in-core inode may be reaped by prune_icache
5409 * without having to perform any I/O. This is a very good thing,
5410 * because *any* task may call prune_icache - even ones which
5411 * have a transaction open against a different journal.
5413 * Is this cheating? Not really. Sure, we haven't written the
5414 * inode out, but prune_icache isn't a user-visible syncing function.
5415 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5416 * we start and wait on commits.
5418 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5420 struct ext4_iloc iloc;
5421 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5422 static unsigned int mnt_count;
5426 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5427 err = ext4_reserve_inode_write(handle, inode, &iloc);
5430 if (ext4_handle_valid(handle) &&
5431 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5432 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5434 * We need extra buffer credits since we may write into EA block
5435 * with this same handle. If journal_extend fails, then it will
5436 * only result in a minor loss of functionality for that inode.
5437 * If this is felt to be critical, then e2fsck should be run to
5438 * force a large enough s_min_extra_isize.
5440 if ((jbd2_journal_extend(handle,
5441 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5442 ret = ext4_expand_extra_isize(inode,
5443 sbi->s_want_extra_isize,
5446 ext4_set_inode_state(inode,
5447 EXT4_STATE_NO_EXPAND);
5449 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5450 ext4_warning(inode->i_sb,
5451 "Unable to expand inode %lu. Delete"
5452 " some EAs or run e2fsck.",
5455 le16_to_cpu(sbi->s_es->s_mnt_count);
5460 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5464 * ext4_dirty_inode() is called from __mark_inode_dirty()
5466 * We're really interested in the case where a file is being extended.
5467 * i_size has been changed by generic_commit_write() and we thus need
5468 * to include the updated inode in the current transaction.
5470 * Also, dquot_alloc_block() will always dirty the inode when blocks
5471 * are allocated to the file.
5473 * If the inode is marked synchronous, we don't honour that here - doing
5474 * so would cause a commit on atime updates, which we don't bother doing.
5475 * We handle synchronous inodes at the highest possible level.
5477 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5478 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5479 * to copy into the on-disk inode structure are the timestamp files.
5481 void ext4_dirty_inode(struct inode *inode, int flags)
5485 if (flags == I_DIRTY_TIME)
5487 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5491 ext4_mark_inode_dirty(handle, inode);
5493 ext4_journal_stop(handle);
5500 * Bind an inode's backing buffer_head into this transaction, to prevent
5501 * it from being flushed to disk early. Unlike
5502 * ext4_reserve_inode_write, this leaves behind no bh reference and
5503 * returns no iloc structure, so the caller needs to repeat the iloc
5504 * lookup to mark the inode dirty later.
5506 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5508 struct ext4_iloc iloc;
5512 err = ext4_get_inode_loc(inode, &iloc);
5514 BUFFER_TRACE(iloc.bh, "get_write_access");
5515 err = jbd2_journal_get_write_access(handle, iloc.bh);
5517 err = ext4_handle_dirty_metadata(handle,
5523 ext4_std_error(inode->i_sb, err);
5528 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5533 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5536 * We have to be very careful here: changing a data block's
5537 * journaling status dynamically is dangerous. If we write a
5538 * data block to the journal, change the status and then delete
5539 * that block, we risk forgetting to revoke the old log record
5540 * from the journal and so a subsequent replay can corrupt data.
5541 * So, first we make sure that the journal is empty and that
5542 * nobody is changing anything.
5545 journal = EXT4_JOURNAL(inode);
5548 if (is_journal_aborted(journal))
5551 /* Wait for all existing dio workers */
5552 ext4_inode_block_unlocked_dio(inode);
5553 inode_dio_wait(inode);
5556 * Before flushing the journal and switching inode's aops, we have
5557 * to flush all dirty data the inode has. There can be outstanding
5558 * delayed allocations, there can be unwritten extents created by
5559 * fallocate or buffered writes in dioread_nolock mode covered by
5560 * dirty data which can be converted only after flushing the dirty
5561 * data (and journalled aops don't know how to handle these cases).
5564 down_write(&EXT4_I(inode)->i_mmap_sem);
5565 err = filemap_write_and_wait(inode->i_mapping);
5567 up_write(&EXT4_I(inode)->i_mmap_sem);
5568 ext4_inode_resume_unlocked_dio(inode);
5573 percpu_down_write(&sbi->s_journal_flag_rwsem);
5574 jbd2_journal_lock_updates(journal);
5577 * OK, there are no updates running now, and all cached data is
5578 * synced to disk. We are now in a completely consistent state
5579 * which doesn't have anything in the journal, and we know that
5580 * no filesystem updates are running, so it is safe to modify
5581 * the inode's in-core data-journaling state flag now.
5585 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5587 err = jbd2_journal_flush(journal);
5589 jbd2_journal_unlock_updates(journal);
5590 percpu_up_write(&sbi->s_journal_flag_rwsem);
5591 ext4_inode_resume_unlocked_dio(inode);
5594 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5596 ext4_set_aops(inode);
5598 jbd2_journal_unlock_updates(journal);
5599 percpu_up_write(&sbi->s_journal_flag_rwsem);
5602 up_write(&EXT4_I(inode)->i_mmap_sem);
5603 ext4_inode_resume_unlocked_dio(inode);
5605 /* Finally we can mark the inode as dirty. */
5607 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5609 return PTR_ERR(handle);
5611 err = ext4_mark_inode_dirty(handle, inode);
5612 ext4_handle_sync(handle);
5613 ext4_journal_stop(handle);
5614 ext4_std_error(inode->i_sb, err);
5619 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5621 return !buffer_mapped(bh);
5624 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5626 struct page *page = vmf->page;
5630 struct file *file = vma->vm_file;
5631 struct inode *inode = file_inode(file);
5632 struct address_space *mapping = inode->i_mapping;
5634 get_block_t *get_block;
5637 sb_start_pagefault(inode->i_sb);
5638 file_update_time(vma->vm_file);
5640 down_read(&EXT4_I(inode)->i_mmap_sem);
5641 /* Delalloc case is easy... */
5642 if (test_opt(inode->i_sb, DELALLOC) &&
5643 !ext4_should_journal_data(inode) &&
5644 !ext4_nonda_switch(inode->i_sb)) {
5646 ret = block_page_mkwrite(vma, vmf,
5647 ext4_da_get_block_prep);
5648 } while (ret == -ENOSPC &&
5649 ext4_should_retry_alloc(inode->i_sb, &retries));
5654 size = i_size_read(inode);
5655 /* Page got truncated from under us? */
5656 if (page->mapping != mapping || page_offset(page) > size) {
5658 ret = VM_FAULT_NOPAGE;
5662 if (page->index == size >> PAGE_SHIFT)
5663 len = size & ~PAGE_MASK;
5667 * Return if we have all the buffers mapped. This avoids the need to do
5668 * journal_start/journal_stop which can block and take a long time
5670 if (page_has_buffers(page)) {
5671 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5673 ext4_bh_unmapped)) {
5674 /* Wait so that we don't change page under IO */
5675 wait_for_stable_page(page);
5676 ret = VM_FAULT_LOCKED;
5681 /* OK, we need to fill the hole... */
5682 if (ext4_should_dioread_nolock(inode))
5683 get_block = ext4_get_block_unwritten;
5685 get_block = ext4_get_block;
5687 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5688 ext4_writepage_trans_blocks(inode));
5689 if (IS_ERR(handle)) {
5690 ret = VM_FAULT_SIGBUS;
5693 ret = block_page_mkwrite(vma, vmf, get_block);
5694 if (!ret && ext4_should_journal_data(inode)) {
5695 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5696 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5698 ret = VM_FAULT_SIGBUS;
5699 ext4_journal_stop(handle);
5702 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5704 ext4_journal_stop(handle);
5705 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5708 ret = block_page_mkwrite_return(ret);
5710 up_read(&EXT4_I(inode)->i_mmap_sem);
5711 sb_end_pagefault(inode->i_sb);
5715 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5717 struct inode *inode = file_inode(vma->vm_file);
5720 down_read(&EXT4_I(inode)->i_mmap_sem);
5721 err = filemap_fault(vma, vmf);
5722 up_read(&EXT4_I(inode)->i_mmap_sem);
5728 * Find the first extent at or after @lblk in an inode that is not a hole.
5729 * Search for @map_len blocks at most. The extent is returned in @result.
5731 * The function returns 1 if we found an extent. The function returns 0 in
5732 * case there is no extent at or after @lblk and in that case also sets
5733 * @result->es_len to 0. In case of error, the error code is returned.
5735 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5736 unsigned int map_len, struct extent_status *result)
5738 struct ext4_map_blocks map;
5739 struct extent_status es = {};
5743 map.m_len = map_len;
5746 * For non-extent based files this loop may iterate several times since
5747 * we do not determine full hole size.
5749 while (map.m_len > 0) {
5750 ret = ext4_map_blocks(NULL, inode, &map, 0);
5753 /* There's extent covering m_lblk? Just return it. */
5757 ext4_es_store_pblock(result, map.m_pblk);
5758 result->es_lblk = map.m_lblk;
5759 result->es_len = map.m_len;
5760 if (map.m_flags & EXT4_MAP_UNWRITTEN)
5761 status = EXTENT_STATUS_UNWRITTEN;
5763 status = EXTENT_STATUS_WRITTEN;
5764 ext4_es_store_status(result, status);
5767 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5768 map.m_lblk + map.m_len - 1,
5770 /* Is delalloc data before next block in extent tree? */
5771 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5772 ext4_lblk_t offset = 0;
5774 if (es.es_lblk < lblk)
5775 offset = lblk - es.es_lblk;
5776 result->es_lblk = es.es_lblk + offset;
5777 ext4_es_store_pblock(result,
5778 ext4_es_pblock(&es) + offset);
5779 result->es_len = es.es_len - offset;
5780 ext4_es_store_status(result, ext4_es_status(&es));
5784 /* There's a hole at m_lblk, advance us after it */
5785 map.m_lblk += map.m_len;
5786 map_len -= map.m_len;
5787 map.m_len = map_len;