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 (inode->i_ino != EXT4_JOURNAL_INO &&
215 ext4_should_journal_data(inode) &&
216 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
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);
264 if (inode->i_blocks) {
265 err = ext4_truncate(inode);
267 ext4_error(inode->i_sb,
268 "couldn't truncate inode %lu (err %d)",
275 * ext4_ext_truncate() doesn't reserve any slop when it
276 * restarts journal transactions; therefore there may not be
277 * enough credits left in the handle to remove the inode from
278 * the orphan list and set the dtime field.
280 if (!ext4_handle_has_enough_credits(handle, 3)) {
281 err = ext4_journal_extend(handle, 3);
283 err = ext4_journal_restart(handle, 3);
285 ext4_warning(inode->i_sb,
286 "couldn't extend journal (err %d)", err);
288 ext4_journal_stop(handle);
289 ext4_orphan_del(NULL, inode);
290 sb_end_intwrite(inode->i_sb);
296 * Kill off the orphan record which ext4_truncate created.
297 * AKPM: I think this can be inside the above `if'.
298 * Note that ext4_orphan_del() has to be able to cope with the
299 * deletion of a non-existent orphan - this is because we don't
300 * know if ext4_truncate() actually created an orphan record.
301 * (Well, we could do this if we need to, but heck - it works)
303 ext4_orphan_del(handle, inode);
304 EXT4_I(inode)->i_dtime = get_seconds();
307 * One subtle ordering requirement: if anything has gone wrong
308 * (transaction abort, IO errors, whatever), then we can still
309 * do these next steps (the fs will already have been marked as
310 * having errors), but we can't free the inode if the mark_dirty
313 if (ext4_mark_inode_dirty(handle, inode))
314 /* If that failed, just do the required in-core inode clear. */
315 ext4_clear_inode(inode);
317 ext4_free_inode(handle, inode);
318 ext4_journal_stop(handle);
319 sb_end_intwrite(inode->i_sb);
322 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
326 qsize_t *ext4_get_reserved_space(struct inode *inode)
328 return &EXT4_I(inode)->i_reserved_quota;
333 * Called with i_data_sem down, which is important since we can call
334 * ext4_discard_preallocations() from here.
336 void ext4_da_update_reserve_space(struct inode *inode,
337 int used, int quota_claim)
339 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
340 struct ext4_inode_info *ei = EXT4_I(inode);
342 spin_lock(&ei->i_block_reservation_lock);
343 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
344 if (unlikely(used > ei->i_reserved_data_blocks)) {
345 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
346 "with only %d reserved data blocks",
347 __func__, inode->i_ino, used,
348 ei->i_reserved_data_blocks);
350 used = ei->i_reserved_data_blocks;
353 /* Update per-inode reservations */
354 ei->i_reserved_data_blocks -= used;
355 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
357 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
359 /* Update quota subsystem for data blocks */
361 dquot_claim_block(inode, EXT4_C2B(sbi, used));
364 * We did fallocate with an offset that is already delayed
365 * allocated. So on delayed allocated writeback we should
366 * not re-claim the quota for fallocated blocks.
368 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
372 * If we have done all the pending block allocations and if
373 * there aren't any writers on the inode, we can discard the
374 * inode's preallocations.
376 if ((ei->i_reserved_data_blocks == 0) &&
377 (atomic_read(&inode->i_writecount) == 0))
378 ext4_discard_preallocations(inode);
381 static int __check_block_validity(struct inode *inode, const char *func,
383 struct ext4_map_blocks *map)
385 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
387 ext4_error_inode(inode, func, line, map->m_pblk,
388 "lblock %lu mapped to illegal pblock "
389 "(length %d)", (unsigned long) map->m_lblk,
391 return -EFSCORRUPTED;
396 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
401 if (ext4_encrypted_inode(inode))
402 return fscrypt_zeroout_range(inode, lblk, pblk, len);
404 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
411 #define check_block_validity(inode, map) \
412 __check_block_validity((inode), __func__, __LINE__, (map))
414 #ifdef ES_AGGRESSIVE_TEST
415 static void ext4_map_blocks_es_recheck(handle_t *handle,
417 struct ext4_map_blocks *es_map,
418 struct ext4_map_blocks *map,
425 * There is a race window that the result is not the same.
426 * e.g. xfstests #223 when dioread_nolock enables. The reason
427 * is that we lookup a block mapping in extent status tree with
428 * out taking i_data_sem. So at the time the unwritten extent
429 * could be converted.
431 down_read(&EXT4_I(inode)->i_data_sem);
432 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
433 retval = ext4_ext_map_blocks(handle, inode, map, flags &
434 EXT4_GET_BLOCKS_KEEP_SIZE);
436 retval = ext4_ind_map_blocks(handle, inode, map, flags &
437 EXT4_GET_BLOCKS_KEEP_SIZE);
439 up_read((&EXT4_I(inode)->i_data_sem));
442 * We don't check m_len because extent will be collpased in status
443 * tree. So the m_len might not equal.
445 if (es_map->m_lblk != map->m_lblk ||
446 es_map->m_flags != map->m_flags ||
447 es_map->m_pblk != map->m_pblk) {
448 printk("ES cache assertion failed for inode: %lu "
449 "es_cached ex [%d/%d/%llu/%x] != "
450 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
451 inode->i_ino, es_map->m_lblk, es_map->m_len,
452 es_map->m_pblk, es_map->m_flags, map->m_lblk,
453 map->m_len, map->m_pblk, map->m_flags,
457 #endif /* ES_AGGRESSIVE_TEST */
460 * The ext4_map_blocks() function tries to look up the requested blocks,
461 * and returns if the blocks are already mapped.
463 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
464 * and store the allocated blocks in the result buffer head and mark it
467 * If file type is extents based, it will call ext4_ext_map_blocks(),
468 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
471 * On success, it returns the number of blocks being mapped or allocated. if
472 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
473 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
475 * It returns 0 if plain look up failed (blocks have not been allocated), in
476 * that case, @map is returned as unmapped but we still do fill map->m_len to
477 * indicate the length of a hole starting at map->m_lblk.
479 * It returns the error in case of allocation failure.
481 int ext4_map_blocks(handle_t *handle, struct inode *inode,
482 struct ext4_map_blocks *map, int flags)
484 struct extent_status es;
487 #ifdef ES_AGGRESSIVE_TEST
488 struct ext4_map_blocks orig_map;
490 memcpy(&orig_map, map, sizeof(*map));
494 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
495 "logical block %lu\n", inode->i_ino, flags, map->m_len,
496 (unsigned long) map->m_lblk);
499 * ext4_map_blocks returns an int, and m_len is an unsigned int
501 if (unlikely(map->m_len > INT_MAX))
502 map->m_len = INT_MAX;
504 /* We can handle the block number less than EXT_MAX_BLOCKS */
505 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
506 return -EFSCORRUPTED;
508 /* Lookup extent status tree firstly */
509 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
510 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
511 map->m_pblk = ext4_es_pblock(&es) +
512 map->m_lblk - es.es_lblk;
513 map->m_flags |= ext4_es_is_written(&es) ?
514 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
515 retval = es.es_len - (map->m_lblk - es.es_lblk);
516 if (retval > map->m_len)
519 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
521 retval = es.es_len - (map->m_lblk - es.es_lblk);
522 if (retval > map->m_len)
529 #ifdef ES_AGGRESSIVE_TEST
530 ext4_map_blocks_es_recheck(handle, inode, map,
537 * Try to see if we can get the block without requesting a new
540 down_read(&EXT4_I(inode)->i_data_sem);
541 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
542 retval = ext4_ext_map_blocks(handle, inode, map, flags &
543 EXT4_GET_BLOCKS_KEEP_SIZE);
545 retval = ext4_ind_map_blocks(handle, inode, map, flags &
546 EXT4_GET_BLOCKS_KEEP_SIZE);
551 if (unlikely(retval != map->m_len)) {
552 ext4_warning(inode->i_sb,
553 "ES len assertion failed for inode "
554 "%lu: retval %d != map->m_len %d",
555 inode->i_ino, retval, map->m_len);
559 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
560 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
561 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
562 !(status & EXTENT_STATUS_WRITTEN) &&
563 ext4_find_delalloc_range(inode, map->m_lblk,
564 map->m_lblk + map->m_len - 1))
565 status |= EXTENT_STATUS_DELAYED;
566 ret = ext4_es_insert_extent(inode, map->m_lblk,
567 map->m_len, map->m_pblk, status);
571 up_read((&EXT4_I(inode)->i_data_sem));
574 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
575 ret = check_block_validity(inode, map);
580 /* If it is only a block(s) look up */
581 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
585 * Returns if the blocks have already allocated
587 * Note that if blocks have been preallocated
588 * ext4_ext_get_block() returns the create = 0
589 * with buffer head unmapped.
591 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
593 * If we need to convert extent to unwritten
594 * we continue and do the actual work in
595 * ext4_ext_map_blocks()
597 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
601 * Here we clear m_flags because after allocating an new extent,
602 * it will be set again.
604 map->m_flags &= ~EXT4_MAP_FLAGS;
607 * New blocks allocate and/or writing to unwritten extent
608 * will possibly result in updating i_data, so we take
609 * the write lock of i_data_sem, and call get_block()
610 * with create == 1 flag.
612 down_write(&EXT4_I(inode)->i_data_sem);
615 * We need to check for EXT4 here because migrate
616 * could have changed the inode type in between
618 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
619 retval = ext4_ext_map_blocks(handle, inode, map, flags);
621 retval = ext4_ind_map_blocks(handle, inode, map, flags);
623 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
625 * We allocated new blocks which will result in
626 * i_data's format changing. Force the migrate
627 * to fail by clearing migrate flags
629 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
633 * Update reserved blocks/metadata blocks after successful
634 * block allocation which had been deferred till now. We don't
635 * support fallocate for non extent files. So we can update
636 * reserve space here.
639 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
640 ext4_da_update_reserve_space(inode, retval, 1);
646 if (unlikely(retval != map->m_len)) {
647 ext4_warning(inode->i_sb,
648 "ES len assertion failed for inode "
649 "%lu: retval %d != map->m_len %d",
650 inode->i_ino, retval, map->m_len);
655 * We have to zeroout blocks before inserting them into extent
656 * status tree. Otherwise someone could look them up there and
657 * use them before they are really zeroed. We also have to
658 * unmap metadata before zeroing as otherwise writeback can
659 * overwrite zeros with stale data from block device.
661 if (flags & EXT4_GET_BLOCKS_ZERO &&
662 map->m_flags & EXT4_MAP_MAPPED &&
663 map->m_flags & EXT4_MAP_NEW) {
666 for (i = 0; i < map->m_len; i++) {
667 unmap_underlying_metadata(inode->i_sb->s_bdev,
670 ret = ext4_issue_zeroout(inode, map->m_lblk,
671 map->m_pblk, map->m_len);
679 * If the extent has been zeroed out, we don't need to update
680 * extent status tree.
682 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
683 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
684 if (ext4_es_is_written(&es))
687 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
688 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
689 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
690 !(status & EXTENT_STATUS_WRITTEN) &&
691 ext4_find_delalloc_range(inode, map->m_lblk,
692 map->m_lblk + map->m_len - 1))
693 status |= EXTENT_STATUS_DELAYED;
694 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
695 map->m_pblk, status);
703 up_write((&EXT4_I(inode)->i_data_sem));
704 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
705 ret = check_block_validity(inode, map);
710 * Inodes with freshly allocated blocks where contents will be
711 * visible after transaction commit must be on transaction's
714 if (map->m_flags & EXT4_MAP_NEW &&
715 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
716 !(flags & EXT4_GET_BLOCKS_ZERO) &&
717 !IS_NOQUOTA(inode) &&
718 ext4_should_order_data(inode)) {
719 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
720 ret = ext4_jbd2_inode_add_wait(handle, inode);
722 ret = ext4_jbd2_inode_add_write(handle, inode);
731 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
732 * we have to be careful as someone else may be manipulating b_state as well.
734 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
736 unsigned long old_state;
737 unsigned long new_state;
739 flags &= EXT4_MAP_FLAGS;
741 /* Dummy buffer_head? Set non-atomically. */
743 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
747 * Someone else may be modifying b_state. Be careful! This is ugly but
748 * once we get rid of using bh as a container for mapping information
749 * to pass to / from get_block functions, this can go away.
752 old_state = READ_ONCE(bh->b_state);
753 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
755 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
758 static int _ext4_get_block(struct inode *inode, sector_t iblock,
759 struct buffer_head *bh, int flags)
761 struct ext4_map_blocks map;
764 if (ext4_has_inline_data(inode))
768 map.m_len = bh->b_size >> inode->i_blkbits;
770 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
773 map_bh(bh, inode->i_sb, map.m_pblk);
774 ext4_update_bh_state(bh, map.m_flags);
775 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
777 } else if (ret == 0) {
778 /* hole case, need to fill in bh->b_size */
779 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
784 int ext4_get_block(struct inode *inode, sector_t iblock,
785 struct buffer_head *bh, int create)
787 return _ext4_get_block(inode, iblock, bh,
788 create ? EXT4_GET_BLOCKS_CREATE : 0);
792 * Get block function used when preparing for buffered write if we require
793 * creating an unwritten extent if blocks haven't been allocated. The extent
794 * will be converted to written after the IO is complete.
796 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
797 struct buffer_head *bh_result, int create)
799 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
800 inode->i_ino, create);
801 return _ext4_get_block(inode, iblock, bh_result,
802 EXT4_GET_BLOCKS_IO_CREATE_EXT);
805 /* Maximum number of blocks we map for direct IO at once. */
806 #define DIO_MAX_BLOCKS 4096
809 * Get blocks function for the cases that need to start a transaction -
810 * generally difference cases of direct IO and DAX IO. It also handles retries
813 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
814 struct buffer_head *bh_result, int flags)
821 /* Trim mapping request to maximum we can map at once for DIO */
822 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
823 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
824 dio_credits = ext4_chunk_trans_blocks(inode,
825 bh_result->b_size >> inode->i_blkbits);
827 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
829 return PTR_ERR(handle);
831 ret = _ext4_get_block(inode, iblock, bh_result, flags);
832 ext4_journal_stop(handle);
834 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
839 /* Get block function for DIO reads and writes to inodes without extents */
840 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
841 struct buffer_head *bh, int create)
843 /* We don't expect handle for direct IO */
844 WARN_ON_ONCE(ext4_journal_current_handle());
847 return _ext4_get_block(inode, iblock, bh, 0);
848 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
852 * Get block function for AIO DIO writes when we create unwritten extent if
853 * blocks are not allocated yet. The extent will be converted to written
854 * after IO is complete.
856 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
857 sector_t iblock, struct buffer_head *bh_result, int create)
861 /* We don't expect handle for direct IO */
862 WARN_ON_ONCE(ext4_journal_current_handle());
864 ret = ext4_get_block_trans(inode, iblock, bh_result,
865 EXT4_GET_BLOCKS_IO_CREATE_EXT);
868 * When doing DIO using unwritten extents, we need io_end to convert
869 * unwritten extents to written on IO completion. We allocate io_end
870 * once we spot unwritten extent and store it in b_private. Generic
871 * DIO code keeps b_private set and furthermore passes the value to
872 * our completion callback in 'private' argument.
874 if (!ret && buffer_unwritten(bh_result)) {
875 if (!bh_result->b_private) {
876 ext4_io_end_t *io_end;
878 io_end = ext4_init_io_end(inode, GFP_KERNEL);
881 bh_result->b_private = io_end;
882 ext4_set_io_unwritten_flag(inode, io_end);
884 set_buffer_defer_completion(bh_result);
891 * Get block function for non-AIO DIO writes when we create unwritten extent if
892 * blocks are not allocated yet. The extent will be converted to written
893 * after IO is complete from ext4_ext_direct_IO() function.
895 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
896 sector_t iblock, struct buffer_head *bh_result, int create)
900 /* We don't expect handle for direct IO */
901 WARN_ON_ONCE(ext4_journal_current_handle());
903 ret = ext4_get_block_trans(inode, iblock, bh_result,
904 EXT4_GET_BLOCKS_IO_CREATE_EXT);
907 * Mark inode as having pending DIO writes to unwritten extents.
908 * ext4_ext_direct_IO() checks this flag and converts extents to
911 if (!ret && buffer_unwritten(bh_result))
912 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
917 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
918 struct buffer_head *bh_result, int create)
922 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
923 inode->i_ino, create);
924 /* We don't expect handle for direct IO */
925 WARN_ON_ONCE(ext4_journal_current_handle());
927 ret = _ext4_get_block(inode, iblock, bh_result, 0);
929 * Blocks should have been preallocated! ext4_file_write_iter() checks
932 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
939 * `handle' can be NULL if create is zero
941 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
942 ext4_lblk_t block, int map_flags)
944 struct ext4_map_blocks map;
945 struct buffer_head *bh;
946 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
949 J_ASSERT(handle != NULL || create == 0);
953 err = ext4_map_blocks(handle, inode, &map, map_flags);
956 return create ? ERR_PTR(-ENOSPC) : NULL;
960 bh = sb_getblk(inode->i_sb, map.m_pblk);
962 return ERR_PTR(-ENOMEM);
963 if (map.m_flags & EXT4_MAP_NEW) {
964 J_ASSERT(create != 0);
965 J_ASSERT(handle != NULL);
968 * Now that we do not always journal data, we should
969 * keep in mind whether this should always journal the
970 * new buffer as metadata. For now, regular file
971 * writes use ext4_get_block instead, so it's not a
975 BUFFER_TRACE(bh, "call get_create_access");
976 err = ext4_journal_get_create_access(handle, bh);
981 if (!buffer_uptodate(bh)) {
982 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
983 set_buffer_uptodate(bh);
986 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
987 err = ext4_handle_dirty_metadata(handle, inode, bh);
991 BUFFER_TRACE(bh, "not a new buffer");
998 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
999 ext4_lblk_t block, int map_flags)
1001 struct buffer_head *bh;
1003 bh = ext4_getblk(handle, inode, block, map_flags);
1006 if (!bh || buffer_uptodate(bh))
1008 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1010 if (buffer_uptodate(bh))
1013 return ERR_PTR(-EIO);
1016 int ext4_walk_page_buffers(handle_t *handle,
1017 struct buffer_head *head,
1021 int (*fn)(handle_t *handle,
1022 struct buffer_head *bh))
1024 struct buffer_head *bh;
1025 unsigned block_start, block_end;
1026 unsigned blocksize = head->b_size;
1028 struct buffer_head *next;
1030 for (bh = head, block_start = 0;
1031 ret == 0 && (bh != head || !block_start);
1032 block_start = block_end, bh = next) {
1033 next = bh->b_this_page;
1034 block_end = block_start + blocksize;
1035 if (block_end <= from || block_start >= to) {
1036 if (partial && !buffer_uptodate(bh))
1040 err = (*fn)(handle, bh);
1048 * To preserve ordering, it is essential that the hole instantiation and
1049 * the data write be encapsulated in a single transaction. We cannot
1050 * close off a transaction and start a new one between the ext4_get_block()
1051 * and the commit_write(). So doing the jbd2_journal_start at the start of
1052 * prepare_write() is the right place.
1054 * Also, this function can nest inside ext4_writepage(). In that case, we
1055 * *know* that ext4_writepage() has generated enough buffer credits to do the
1056 * whole page. So we won't block on the journal in that case, which is good,
1057 * because the caller may be PF_MEMALLOC.
1059 * By accident, ext4 can be reentered when a transaction is open via
1060 * quota file writes. If we were to commit the transaction while thus
1061 * reentered, there can be a deadlock - we would be holding a quota
1062 * lock, and the commit would never complete if another thread had a
1063 * transaction open and was blocking on the quota lock - a ranking
1066 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1067 * will _not_ run commit under these circumstances because handle->h_ref
1068 * is elevated. We'll still have enough credits for the tiny quotafile
1071 int do_journal_get_write_access(handle_t *handle,
1072 struct buffer_head *bh)
1074 int dirty = buffer_dirty(bh);
1077 if (!buffer_mapped(bh) || buffer_freed(bh))
1080 * __block_write_begin() could have dirtied some buffers. Clean
1081 * the dirty bit as jbd2_journal_get_write_access() could complain
1082 * otherwise about fs integrity issues. Setting of the dirty bit
1083 * by __block_write_begin() isn't a real problem here as we clear
1084 * the bit before releasing a page lock and thus writeback cannot
1085 * ever write the buffer.
1088 clear_buffer_dirty(bh);
1089 BUFFER_TRACE(bh, "get write access");
1090 ret = ext4_journal_get_write_access(handle, bh);
1092 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1096 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1097 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1098 get_block_t *get_block)
1100 unsigned from = pos & (PAGE_SIZE - 1);
1101 unsigned to = from + len;
1102 struct inode *inode = page->mapping->host;
1103 unsigned block_start, block_end;
1106 unsigned blocksize = inode->i_sb->s_blocksize;
1108 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1109 bool decrypt = false;
1111 BUG_ON(!PageLocked(page));
1112 BUG_ON(from > PAGE_SIZE);
1113 BUG_ON(to > PAGE_SIZE);
1116 if (!page_has_buffers(page))
1117 create_empty_buffers(page, blocksize, 0);
1118 head = page_buffers(page);
1119 bbits = ilog2(blocksize);
1120 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1122 for (bh = head, block_start = 0; bh != head || !block_start;
1123 block++, block_start = block_end, bh = bh->b_this_page) {
1124 block_end = block_start + blocksize;
1125 if (block_end <= from || block_start >= to) {
1126 if (PageUptodate(page)) {
1127 if (!buffer_uptodate(bh))
1128 set_buffer_uptodate(bh);
1133 clear_buffer_new(bh);
1134 if (!buffer_mapped(bh)) {
1135 WARN_ON(bh->b_size != blocksize);
1136 err = get_block(inode, block, bh, 1);
1139 if (buffer_new(bh)) {
1140 unmap_underlying_metadata(bh->b_bdev,
1142 if (PageUptodate(page)) {
1143 clear_buffer_new(bh);
1144 set_buffer_uptodate(bh);
1145 mark_buffer_dirty(bh);
1148 if (block_end > to || block_start < from)
1149 zero_user_segments(page, to, block_end,
1154 if (PageUptodate(page)) {
1155 if (!buffer_uptodate(bh))
1156 set_buffer_uptodate(bh);
1159 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1160 !buffer_unwritten(bh) &&
1161 (block_start < from || block_end > to)) {
1162 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1164 decrypt = ext4_encrypted_inode(inode) &&
1165 S_ISREG(inode->i_mode);
1169 * If we issued read requests, let them complete.
1171 while (wait_bh > wait) {
1172 wait_on_buffer(*--wait_bh);
1173 if (!buffer_uptodate(*wait_bh))
1177 page_zero_new_buffers(page, from, to);
1179 err = fscrypt_decrypt_page(page->mapping->host, page,
1180 PAGE_SIZE, 0, page->index);
1185 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1186 loff_t pos, unsigned len, unsigned flags,
1187 struct page **pagep, void **fsdata)
1189 struct inode *inode = mapping->host;
1190 int ret, needed_blocks;
1197 trace_ext4_write_begin(inode, pos, len, flags);
1199 * Reserve one block more for addition to orphan list in case
1200 * we allocate blocks but write fails for some reason
1202 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1203 index = pos >> PAGE_SHIFT;
1204 from = pos & (PAGE_SIZE - 1);
1207 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1208 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1217 * grab_cache_page_write_begin() can take a long time if the
1218 * system is thrashing due to memory pressure, or if the page
1219 * is being written back. So grab it first before we start
1220 * the transaction handle. This also allows us to allocate
1221 * the page (if needed) without using GFP_NOFS.
1224 page = grab_cache_page_write_begin(mapping, index, flags);
1230 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1231 if (IS_ERR(handle)) {
1233 return PTR_ERR(handle);
1237 if (page->mapping != mapping) {
1238 /* The page got truncated from under us */
1241 ext4_journal_stop(handle);
1244 /* In case writeback began while the page was unlocked */
1245 wait_for_stable_page(page);
1247 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1248 if (ext4_should_dioread_nolock(inode))
1249 ret = ext4_block_write_begin(page, pos, len,
1250 ext4_get_block_unwritten);
1252 ret = ext4_block_write_begin(page, pos, len,
1255 if (ext4_should_dioread_nolock(inode))
1256 ret = __block_write_begin(page, pos, len,
1257 ext4_get_block_unwritten);
1259 ret = __block_write_begin(page, pos, len, ext4_get_block);
1261 if (!ret && ext4_should_journal_data(inode)) {
1262 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1264 do_journal_get_write_access);
1270 * __block_write_begin may have instantiated a few blocks
1271 * outside i_size. Trim these off again. Don't need
1272 * i_size_read because we hold i_mutex.
1274 * Add inode to orphan list in case we crash before
1277 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1278 ext4_orphan_add(handle, inode);
1280 ext4_journal_stop(handle);
1281 if (pos + len > inode->i_size) {
1282 ext4_truncate_failed_write(inode);
1284 * If truncate failed early the inode might
1285 * still be on the orphan list; we need to
1286 * make sure the inode is removed from the
1287 * orphan list in that case.
1290 ext4_orphan_del(NULL, inode);
1293 if (ret == -ENOSPC &&
1294 ext4_should_retry_alloc(inode->i_sb, &retries))
1303 /* For write_end() in data=journal mode */
1304 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1307 if (!buffer_mapped(bh) || buffer_freed(bh))
1309 set_buffer_uptodate(bh);
1310 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1311 clear_buffer_meta(bh);
1312 clear_buffer_prio(bh);
1317 * We need to pick up the new inode size which generic_commit_write gave us
1318 * `file' can be NULL - eg, when called from page_symlink().
1320 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1321 * buffers are managed internally.
1323 static int ext4_write_end(struct file *file,
1324 struct address_space *mapping,
1325 loff_t pos, unsigned len, unsigned copied,
1326 struct page *page, void *fsdata)
1328 handle_t *handle = ext4_journal_current_handle();
1329 struct inode *inode = mapping->host;
1330 loff_t old_size = inode->i_size;
1332 int i_size_changed = 0;
1334 trace_ext4_write_end(inode, pos, len, copied);
1335 if (ext4_has_inline_data(inode)) {
1336 ret = ext4_write_inline_data_end(inode, pos, len,
1342 copied = block_write_end(file, mapping, pos,
1343 len, copied, page, fsdata);
1345 * it's important to update i_size while still holding page lock:
1346 * page writeout could otherwise come in and zero beyond i_size.
1348 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1353 pagecache_isize_extended(inode, old_size, pos);
1355 * Don't mark the inode dirty under page lock. First, it unnecessarily
1356 * makes the holding time of page lock longer. Second, it forces lock
1357 * ordering of page lock and transaction start for journaling
1361 ext4_mark_inode_dirty(handle, inode);
1363 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1364 /* if we have allocated more blocks and copied
1365 * less. We will have blocks allocated outside
1366 * inode->i_size. So truncate them
1368 ext4_orphan_add(handle, inode);
1370 ret2 = ext4_journal_stop(handle);
1374 if (pos + len > inode->i_size) {
1375 ext4_truncate_failed_write(inode);
1377 * If truncate failed early the inode might still be
1378 * on the orphan list; we need to make sure the inode
1379 * is removed from the orphan list in that case.
1382 ext4_orphan_del(NULL, inode);
1385 return ret ? ret : copied;
1389 * This is a private version of page_zero_new_buffers() which doesn't
1390 * set the buffer to be dirty, since in data=journalled mode we need
1391 * to call ext4_handle_dirty_metadata() instead.
1393 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1395 unsigned int block_start = 0, block_end;
1396 struct buffer_head *head, *bh;
1398 bh = head = page_buffers(page);
1400 block_end = block_start + bh->b_size;
1401 if (buffer_new(bh)) {
1402 if (block_end > from && block_start < to) {
1403 if (!PageUptodate(page)) {
1404 unsigned start, size;
1406 start = max(from, block_start);
1407 size = min(to, block_end) - start;
1409 zero_user(page, start, size);
1410 set_buffer_uptodate(bh);
1412 clear_buffer_new(bh);
1415 block_start = block_end;
1416 bh = bh->b_this_page;
1417 } while (bh != head);
1420 static int ext4_journalled_write_end(struct file *file,
1421 struct address_space *mapping,
1422 loff_t pos, unsigned len, unsigned copied,
1423 struct page *page, void *fsdata)
1425 handle_t *handle = ext4_journal_current_handle();
1426 struct inode *inode = mapping->host;
1427 loff_t old_size = inode->i_size;
1431 int size_changed = 0;
1433 trace_ext4_journalled_write_end(inode, pos, len, copied);
1434 from = pos & (PAGE_SIZE - 1);
1437 BUG_ON(!ext4_handle_valid(handle));
1439 if (ext4_has_inline_data(inode))
1440 copied = ext4_write_inline_data_end(inode, pos, len,
1444 if (!PageUptodate(page))
1446 zero_new_buffers(page, from+copied, to);
1449 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1450 to, &partial, write_end_fn);
1452 SetPageUptodate(page);
1454 size_changed = ext4_update_inode_size(inode, pos + copied);
1455 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1456 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1461 pagecache_isize_extended(inode, old_size, pos);
1464 ret2 = ext4_mark_inode_dirty(handle, inode);
1469 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1470 /* if we have allocated more blocks and copied
1471 * less. We will have blocks allocated outside
1472 * inode->i_size. So truncate them
1474 ext4_orphan_add(handle, inode);
1476 ret2 = ext4_journal_stop(handle);
1479 if (pos + len > inode->i_size) {
1480 ext4_truncate_failed_write(inode);
1482 * If truncate failed early the inode might still be
1483 * on the orphan list; we need to make sure the inode
1484 * is removed from the orphan list in that case.
1487 ext4_orphan_del(NULL, inode);
1490 return ret ? ret : copied;
1494 * Reserve space for a single cluster
1496 static int ext4_da_reserve_space(struct inode *inode)
1498 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1499 struct ext4_inode_info *ei = EXT4_I(inode);
1503 * We will charge metadata quota at writeout time; this saves
1504 * us from metadata over-estimation, though we may go over by
1505 * a small amount in the end. Here we just reserve for data.
1507 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1511 spin_lock(&ei->i_block_reservation_lock);
1512 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1513 spin_unlock(&ei->i_block_reservation_lock);
1514 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1517 ei->i_reserved_data_blocks++;
1518 trace_ext4_da_reserve_space(inode);
1519 spin_unlock(&ei->i_block_reservation_lock);
1521 return 0; /* success */
1524 static void ext4_da_release_space(struct inode *inode, int to_free)
1526 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1527 struct ext4_inode_info *ei = EXT4_I(inode);
1530 return; /* Nothing to release, exit */
1532 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1534 trace_ext4_da_release_space(inode, to_free);
1535 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1537 * if there aren't enough reserved blocks, then the
1538 * counter is messed up somewhere. Since this
1539 * function is called from invalidate page, it's
1540 * harmless to return without any action.
1542 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1543 "ino %lu, to_free %d with only %d reserved "
1544 "data blocks", inode->i_ino, to_free,
1545 ei->i_reserved_data_blocks);
1547 to_free = ei->i_reserved_data_blocks;
1549 ei->i_reserved_data_blocks -= to_free;
1551 /* update fs dirty data blocks counter */
1552 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1554 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1556 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1559 static void ext4_da_page_release_reservation(struct page *page,
1560 unsigned int offset,
1561 unsigned int length)
1563 int to_release = 0, contiguous_blks = 0;
1564 struct buffer_head *head, *bh;
1565 unsigned int curr_off = 0;
1566 struct inode *inode = page->mapping->host;
1567 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1568 unsigned int stop = offset + length;
1572 BUG_ON(stop > PAGE_SIZE || stop < length);
1574 head = page_buffers(page);
1577 unsigned int next_off = curr_off + bh->b_size;
1579 if (next_off > stop)
1582 if ((offset <= curr_off) && (buffer_delay(bh))) {
1585 clear_buffer_delay(bh);
1586 } else if (contiguous_blks) {
1587 lblk = page->index <<
1588 (PAGE_SHIFT - inode->i_blkbits);
1589 lblk += (curr_off >> inode->i_blkbits) -
1591 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1592 contiguous_blks = 0;
1594 curr_off = next_off;
1595 } while ((bh = bh->b_this_page) != head);
1597 if (contiguous_blks) {
1598 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1599 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1600 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1603 /* If we have released all the blocks belonging to a cluster, then we
1604 * need to release the reserved space for that cluster. */
1605 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1606 while (num_clusters > 0) {
1607 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1608 ((num_clusters - 1) << sbi->s_cluster_bits);
1609 if (sbi->s_cluster_ratio == 1 ||
1610 !ext4_find_delalloc_cluster(inode, lblk))
1611 ext4_da_release_space(inode, 1);
1618 * Delayed allocation stuff
1621 struct mpage_da_data {
1622 struct inode *inode;
1623 struct writeback_control *wbc;
1625 pgoff_t first_page; /* The first page to write */
1626 pgoff_t next_page; /* Current page to examine */
1627 pgoff_t last_page; /* Last page to examine */
1629 * Extent to map - this can be after first_page because that can be
1630 * fully mapped. We somewhat abuse m_flags to store whether the extent
1631 * is delalloc or unwritten.
1633 struct ext4_map_blocks map;
1634 struct ext4_io_submit io_submit; /* IO submission data */
1637 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1642 struct pagevec pvec;
1643 struct inode *inode = mpd->inode;
1644 struct address_space *mapping = inode->i_mapping;
1646 /* This is necessary when next_page == 0. */
1647 if (mpd->first_page >= mpd->next_page)
1650 index = mpd->first_page;
1651 end = mpd->next_page - 1;
1653 ext4_lblk_t start, last;
1654 start = index << (PAGE_SHIFT - inode->i_blkbits);
1655 last = end << (PAGE_SHIFT - inode->i_blkbits);
1656 ext4_es_remove_extent(inode, start, last - start + 1);
1659 pagevec_init(&pvec, 0);
1660 while (index <= end) {
1661 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1664 for (i = 0; i < nr_pages; i++) {
1665 struct page *page = pvec.pages[i];
1666 if (page->index > end)
1668 BUG_ON(!PageLocked(page));
1669 BUG_ON(PageWriteback(page));
1671 if (page_mapped(page))
1672 clear_page_dirty_for_io(page);
1673 block_invalidatepage(page, 0, PAGE_SIZE);
1674 ClearPageUptodate(page);
1678 index = pvec.pages[nr_pages - 1]->index + 1;
1679 pagevec_release(&pvec);
1683 static void ext4_print_free_blocks(struct inode *inode)
1685 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1686 struct super_block *sb = inode->i_sb;
1687 struct ext4_inode_info *ei = EXT4_I(inode);
1689 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1690 EXT4_C2B(EXT4_SB(inode->i_sb),
1691 ext4_count_free_clusters(sb)));
1692 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1693 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1694 (long long) EXT4_C2B(EXT4_SB(sb),
1695 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1696 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1697 (long long) EXT4_C2B(EXT4_SB(sb),
1698 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1699 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1700 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1701 ei->i_reserved_data_blocks);
1705 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1707 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1711 * This function is grabs code from the very beginning of
1712 * ext4_map_blocks, but assumes that the caller is from delayed write
1713 * time. This function looks up the requested blocks and sets the
1714 * buffer delay bit under the protection of i_data_sem.
1716 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1717 struct ext4_map_blocks *map,
1718 struct buffer_head *bh)
1720 struct extent_status es;
1722 sector_t invalid_block = ~((sector_t) 0xffff);
1723 #ifdef ES_AGGRESSIVE_TEST
1724 struct ext4_map_blocks orig_map;
1726 memcpy(&orig_map, map, sizeof(*map));
1729 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1733 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1734 "logical block %lu\n", inode->i_ino, map->m_len,
1735 (unsigned long) map->m_lblk);
1737 /* Lookup extent status tree firstly */
1738 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1739 if (ext4_es_is_hole(&es)) {
1741 down_read(&EXT4_I(inode)->i_data_sem);
1746 * Delayed extent could be allocated by fallocate.
1747 * So we need to check it.
1749 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1750 map_bh(bh, inode->i_sb, invalid_block);
1752 set_buffer_delay(bh);
1756 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1757 retval = es.es_len - (iblock - es.es_lblk);
1758 if (retval > map->m_len)
1759 retval = map->m_len;
1760 map->m_len = retval;
1761 if (ext4_es_is_written(&es))
1762 map->m_flags |= EXT4_MAP_MAPPED;
1763 else if (ext4_es_is_unwritten(&es))
1764 map->m_flags |= EXT4_MAP_UNWRITTEN;
1768 #ifdef ES_AGGRESSIVE_TEST
1769 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1775 * Try to see if we can get the block without requesting a new
1776 * file system block.
1778 down_read(&EXT4_I(inode)->i_data_sem);
1779 if (ext4_has_inline_data(inode))
1781 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1782 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1784 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1790 * XXX: __block_prepare_write() unmaps passed block,
1794 * If the block was allocated from previously allocated cluster,
1795 * then we don't need to reserve it again. However we still need
1796 * to reserve metadata for every block we're going to write.
1798 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1799 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1800 ret = ext4_da_reserve_space(inode);
1802 /* not enough space to reserve */
1808 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1809 ~0, EXTENT_STATUS_DELAYED);
1815 map_bh(bh, inode->i_sb, invalid_block);
1817 set_buffer_delay(bh);
1818 } else if (retval > 0) {
1820 unsigned int status;
1822 if (unlikely(retval != map->m_len)) {
1823 ext4_warning(inode->i_sb,
1824 "ES len assertion failed for inode "
1825 "%lu: retval %d != map->m_len %d",
1826 inode->i_ino, retval, map->m_len);
1830 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1831 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1832 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1833 map->m_pblk, status);
1839 up_read((&EXT4_I(inode)->i_data_sem));
1845 * This is a special get_block_t callback which is used by
1846 * ext4_da_write_begin(). It will either return mapped block or
1847 * reserve space for a single block.
1849 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1850 * We also have b_blocknr = -1 and b_bdev initialized properly
1852 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1853 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1854 * initialized properly.
1856 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1857 struct buffer_head *bh, int create)
1859 struct ext4_map_blocks map;
1862 BUG_ON(create == 0);
1863 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1865 map.m_lblk = iblock;
1869 * first, we need to know whether the block is allocated already
1870 * preallocated blocks are unmapped but should treated
1871 * the same as allocated blocks.
1873 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1877 map_bh(bh, inode->i_sb, map.m_pblk);
1878 ext4_update_bh_state(bh, map.m_flags);
1880 if (buffer_unwritten(bh)) {
1881 /* A delayed write to unwritten bh should be marked
1882 * new and mapped. Mapped ensures that we don't do
1883 * get_block multiple times when we write to the same
1884 * offset and new ensures that we do proper zero out
1885 * for partial write.
1888 set_buffer_mapped(bh);
1893 static int bget_one(handle_t *handle, struct buffer_head *bh)
1899 static int bput_one(handle_t *handle, struct buffer_head *bh)
1905 static int __ext4_journalled_writepage(struct page *page,
1908 struct address_space *mapping = page->mapping;
1909 struct inode *inode = mapping->host;
1910 struct buffer_head *page_bufs = NULL;
1911 handle_t *handle = NULL;
1912 int ret = 0, err = 0;
1913 int inline_data = ext4_has_inline_data(inode);
1914 struct buffer_head *inode_bh = NULL;
1916 ClearPageChecked(page);
1919 BUG_ON(page->index != 0);
1920 BUG_ON(len > ext4_get_max_inline_size(inode));
1921 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1922 if (inode_bh == NULL)
1925 page_bufs = page_buffers(page);
1930 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1934 * We need to release the page lock before we start the
1935 * journal, so grab a reference so the page won't disappear
1936 * out from under us.
1941 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1942 ext4_writepage_trans_blocks(inode));
1943 if (IS_ERR(handle)) {
1944 ret = PTR_ERR(handle);
1946 goto out_no_pagelock;
1948 BUG_ON(!ext4_handle_valid(handle));
1952 if (page->mapping != mapping) {
1953 /* The page got truncated from under us */
1954 ext4_journal_stop(handle);
1960 BUFFER_TRACE(inode_bh, "get write access");
1961 ret = ext4_journal_get_write_access(handle, inode_bh);
1963 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1966 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1967 do_journal_get_write_access);
1969 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1974 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1975 err = ext4_journal_stop(handle);
1979 if (!ext4_has_inline_data(inode))
1980 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1982 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1991 * Note that we don't need to start a transaction unless we're journaling data
1992 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1993 * need to file the inode to the transaction's list in ordered mode because if
1994 * we are writing back data added by write(), the inode is already there and if
1995 * we are writing back data modified via mmap(), no one guarantees in which
1996 * transaction the data will hit the disk. In case we are journaling data, we
1997 * cannot start transaction directly because transaction start ranks above page
1998 * lock so we have to do some magic.
2000 * This function can get called via...
2001 * - ext4_writepages after taking page lock (have journal handle)
2002 * - journal_submit_inode_data_buffers (no journal handle)
2003 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2004 * - grab_page_cache when doing write_begin (have journal handle)
2006 * We don't do any block allocation in this function. If we have page with
2007 * multiple blocks we need to write those buffer_heads that are mapped. This
2008 * is important for mmaped based write. So if we do with blocksize 1K
2009 * truncate(f, 1024);
2010 * a = mmap(f, 0, 4096);
2012 * truncate(f, 4096);
2013 * we have in the page first buffer_head mapped via page_mkwrite call back
2014 * but other buffer_heads would be unmapped but dirty (dirty done via the
2015 * do_wp_page). So writepage should write the first block. If we modify
2016 * the mmap area beyond 1024 we will again get a page_fault and the
2017 * page_mkwrite callback will do the block allocation and mark the
2018 * buffer_heads mapped.
2020 * We redirty the page if we have any buffer_heads that is either delay or
2021 * unwritten in the page.
2023 * We can get recursively called as show below.
2025 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2028 * But since we don't do any block allocation we should not deadlock.
2029 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2031 static int ext4_writepage(struct page *page,
2032 struct writeback_control *wbc)
2037 struct buffer_head *page_bufs = NULL;
2038 struct inode *inode = page->mapping->host;
2039 struct ext4_io_submit io_submit;
2040 bool keep_towrite = false;
2042 trace_ext4_writepage(page);
2043 size = i_size_read(inode);
2044 if (page->index == size >> PAGE_SHIFT)
2045 len = size & ~PAGE_MASK;
2049 page_bufs = page_buffers(page);
2051 * We cannot do block allocation or other extent handling in this
2052 * function. If there are buffers needing that, we have to redirty
2053 * the page. But we may reach here when we do a journal commit via
2054 * journal_submit_inode_data_buffers() and in that case we must write
2055 * allocated buffers to achieve data=ordered mode guarantees.
2057 * Also, if there is only one buffer per page (the fs block
2058 * size == the page size), if one buffer needs block
2059 * allocation or needs to modify the extent tree to clear the
2060 * unwritten flag, we know that the page can't be written at
2061 * all, so we might as well refuse the write immediately.
2062 * Unfortunately if the block size != page size, we can't as
2063 * easily detect this case using ext4_walk_page_buffers(), but
2064 * for the extremely common case, this is an optimization that
2065 * skips a useless round trip through ext4_bio_write_page().
2067 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2068 ext4_bh_delay_or_unwritten)) {
2069 redirty_page_for_writepage(wbc, page);
2070 if ((current->flags & PF_MEMALLOC) ||
2071 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2073 * For memory cleaning there's no point in writing only
2074 * some buffers. So just bail out. Warn if we came here
2075 * from direct reclaim.
2077 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2082 keep_towrite = true;
2085 if (PageChecked(page) && ext4_should_journal_data(inode))
2087 * It's mmapped pagecache. Add buffers and journal it. There
2088 * doesn't seem much point in redirtying the page here.
2090 return __ext4_journalled_writepage(page, len);
2092 ext4_io_submit_init(&io_submit, wbc);
2093 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2094 if (!io_submit.io_end) {
2095 redirty_page_for_writepage(wbc, page);
2099 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2100 ext4_io_submit(&io_submit);
2101 /* Drop io_end reference we got from init */
2102 ext4_put_io_end_defer(io_submit.io_end);
2106 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2109 loff_t size = i_size_read(mpd->inode);
2112 BUG_ON(page->index != mpd->first_page);
2113 if (page->index == size >> PAGE_SHIFT)
2114 len = size & ~PAGE_MASK;
2117 clear_page_dirty_for_io(page);
2118 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2120 mpd->wbc->nr_to_write--;
2126 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2129 * mballoc gives us at most this number of blocks...
2130 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2131 * The rest of mballoc seems to handle chunks up to full group size.
2133 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2136 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2138 * @mpd - extent of blocks
2139 * @lblk - logical number of the block in the file
2140 * @bh - buffer head we want to add to the extent
2142 * The function is used to collect contig. blocks in the same state. If the
2143 * buffer doesn't require mapping for writeback and we haven't started the
2144 * extent of buffers to map yet, the function returns 'true' immediately - the
2145 * caller can write the buffer right away. Otherwise the function returns true
2146 * if the block has been added to the extent, false if the block couldn't be
2149 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2150 struct buffer_head *bh)
2152 struct ext4_map_blocks *map = &mpd->map;
2154 /* Buffer that doesn't need mapping for writeback? */
2155 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2156 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2157 /* So far no extent to map => we write the buffer right away */
2158 if (map->m_len == 0)
2163 /* First block in the extent? */
2164 if (map->m_len == 0) {
2167 map->m_flags = bh->b_state & BH_FLAGS;
2171 /* Don't go larger than mballoc is willing to allocate */
2172 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2175 /* Can we merge the block to our big extent? */
2176 if (lblk == map->m_lblk + map->m_len &&
2177 (bh->b_state & BH_FLAGS) == map->m_flags) {
2185 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2187 * @mpd - extent of blocks for mapping
2188 * @head - the first buffer in the page
2189 * @bh - buffer we should start processing from
2190 * @lblk - logical number of the block in the file corresponding to @bh
2192 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2193 * the page for IO if all buffers in this page were mapped and there's no
2194 * accumulated extent of buffers to map or add buffers in the page to the
2195 * extent of buffers to map. The function returns 1 if the caller can continue
2196 * by processing the next page, 0 if it should stop adding buffers to the
2197 * extent to map because we cannot extend it anymore. It can also return value
2198 * < 0 in case of error during IO submission.
2200 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2201 struct buffer_head *head,
2202 struct buffer_head *bh,
2205 struct inode *inode = mpd->inode;
2207 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2208 >> inode->i_blkbits;
2211 BUG_ON(buffer_locked(bh));
2213 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2214 /* Found extent to map? */
2217 /* Everything mapped so far and we hit EOF */
2220 } while (lblk++, (bh = bh->b_this_page) != head);
2221 /* So far everything mapped? Submit the page for IO. */
2222 if (mpd->map.m_len == 0) {
2223 err = mpage_submit_page(mpd, head->b_page);
2227 return lblk < blocks;
2231 * mpage_map_buffers - update buffers corresponding to changed extent and
2232 * submit fully mapped pages for IO
2234 * @mpd - description of extent to map, on return next extent to map
2236 * Scan buffers corresponding to changed extent (we expect corresponding pages
2237 * to be already locked) and update buffer state according to new extent state.
2238 * We map delalloc buffers to their physical location, clear unwritten bits,
2239 * and mark buffers as uninit when we perform writes to unwritten extents
2240 * and do extent conversion after IO is finished. If the last page is not fully
2241 * mapped, we update @map to the next extent in the last page that needs
2242 * mapping. Otherwise we submit the page for IO.
2244 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2246 struct pagevec pvec;
2248 struct inode *inode = mpd->inode;
2249 struct buffer_head *head, *bh;
2250 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2256 start = mpd->map.m_lblk >> bpp_bits;
2257 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2258 lblk = start << bpp_bits;
2259 pblock = mpd->map.m_pblk;
2261 pagevec_init(&pvec, 0);
2262 while (start <= end) {
2263 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2267 for (i = 0; i < nr_pages; i++) {
2268 struct page *page = pvec.pages[i];
2270 if (page->index > end)
2272 /* Up to 'end' pages must be contiguous */
2273 BUG_ON(page->index != start);
2274 bh = head = page_buffers(page);
2276 if (lblk < mpd->map.m_lblk)
2278 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2280 * Buffer after end of mapped extent.
2281 * Find next buffer in the page to map.
2284 mpd->map.m_flags = 0;
2286 * FIXME: If dioread_nolock supports
2287 * blocksize < pagesize, we need to make
2288 * sure we add size mapped so far to
2289 * io_end->size as the following call
2290 * can submit the page for IO.
2292 err = mpage_process_page_bufs(mpd, head,
2294 pagevec_release(&pvec);
2299 if (buffer_delay(bh)) {
2300 clear_buffer_delay(bh);
2301 bh->b_blocknr = pblock++;
2303 clear_buffer_unwritten(bh);
2304 } while (lblk++, (bh = bh->b_this_page) != head);
2307 * FIXME: This is going to break if dioread_nolock
2308 * supports blocksize < pagesize as we will try to
2309 * convert potentially unmapped parts of inode.
2311 mpd->io_submit.io_end->size += PAGE_SIZE;
2312 /* Page fully mapped - let IO run! */
2313 err = mpage_submit_page(mpd, page);
2315 pagevec_release(&pvec);
2320 pagevec_release(&pvec);
2322 /* Extent fully mapped and matches with page boundary. We are done. */
2324 mpd->map.m_flags = 0;
2328 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2330 struct inode *inode = mpd->inode;
2331 struct ext4_map_blocks *map = &mpd->map;
2332 int get_blocks_flags;
2333 int err, dioread_nolock;
2335 trace_ext4_da_write_pages_extent(inode, map);
2337 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2338 * to convert an unwritten extent to be initialized (in the case
2339 * where we have written into one or more preallocated blocks). It is
2340 * possible that we're going to need more metadata blocks than
2341 * previously reserved. However we must not fail because we're in
2342 * writeback and there is nothing we can do about it so it might result
2343 * in data loss. So use reserved blocks to allocate metadata if
2346 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2347 * the blocks in question are delalloc blocks. This indicates
2348 * that the blocks and quotas has already been checked when
2349 * the data was copied into the page cache.
2351 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2352 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2353 EXT4_GET_BLOCKS_IO_SUBMIT;
2354 dioread_nolock = ext4_should_dioread_nolock(inode);
2356 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2357 if (map->m_flags & (1 << BH_Delay))
2358 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2360 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2363 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2364 if (!mpd->io_submit.io_end->handle &&
2365 ext4_handle_valid(handle)) {
2366 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2367 handle->h_rsv_handle = NULL;
2369 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2372 BUG_ON(map->m_len == 0);
2373 if (map->m_flags & EXT4_MAP_NEW) {
2374 struct block_device *bdev = inode->i_sb->s_bdev;
2377 for (i = 0; i < map->m_len; i++)
2378 unmap_underlying_metadata(bdev, map->m_pblk + i);
2384 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2385 * mpd->len and submit pages underlying it for IO
2387 * @handle - handle for journal operations
2388 * @mpd - extent to map
2389 * @give_up_on_write - we set this to true iff there is a fatal error and there
2390 * is no hope of writing the data. The caller should discard
2391 * dirty pages to avoid infinite loops.
2393 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2394 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2395 * them to initialized or split the described range from larger unwritten
2396 * extent. Note that we need not map all the described range since allocation
2397 * can return less blocks or the range is covered by more unwritten extents. We
2398 * cannot map more because we are limited by reserved transaction credits. On
2399 * the other hand we always make sure that the last touched page is fully
2400 * mapped so that it can be written out (and thus forward progress is
2401 * guaranteed). After mapping we submit all mapped pages for IO.
2403 static int mpage_map_and_submit_extent(handle_t *handle,
2404 struct mpage_da_data *mpd,
2405 bool *give_up_on_write)
2407 struct inode *inode = mpd->inode;
2408 struct ext4_map_blocks *map = &mpd->map;
2413 mpd->io_submit.io_end->offset =
2414 ((loff_t)map->m_lblk) << inode->i_blkbits;
2416 err = mpage_map_one_extent(handle, mpd);
2418 struct super_block *sb = inode->i_sb;
2420 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2421 goto invalidate_dirty_pages;
2423 * Let the uper layers retry transient errors.
2424 * In the case of ENOSPC, if ext4_count_free_blocks()
2425 * is non-zero, a commit should free up blocks.
2427 if ((err == -ENOMEM) ||
2428 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2430 goto update_disksize;
2433 ext4_msg(sb, KERN_CRIT,
2434 "Delayed block allocation failed for "
2435 "inode %lu at logical offset %llu with"
2436 " max blocks %u with error %d",
2438 (unsigned long long)map->m_lblk,
2439 (unsigned)map->m_len, -err);
2440 ext4_msg(sb, KERN_CRIT,
2441 "This should not happen!! Data will "
2444 ext4_print_free_blocks(inode);
2445 invalidate_dirty_pages:
2446 *give_up_on_write = true;
2451 * Update buffer state, submit mapped pages, and get us new
2454 err = mpage_map_and_submit_buffers(mpd);
2456 goto update_disksize;
2457 } while (map->m_len);
2461 * Update on-disk size after IO is submitted. Races with
2462 * truncate are avoided by checking i_size under i_data_sem.
2464 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2465 if (disksize > EXT4_I(inode)->i_disksize) {
2469 down_write(&EXT4_I(inode)->i_data_sem);
2470 i_size = i_size_read(inode);
2471 if (disksize > i_size)
2473 if (disksize > EXT4_I(inode)->i_disksize)
2474 EXT4_I(inode)->i_disksize = disksize;
2475 err2 = ext4_mark_inode_dirty(handle, inode);
2476 up_write(&EXT4_I(inode)->i_data_sem);
2478 ext4_error(inode->i_sb,
2479 "Failed to mark inode %lu dirty",
2488 * Calculate the total number of credits to reserve for one writepages
2489 * iteration. This is called from ext4_writepages(). We map an extent of
2490 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2491 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2492 * bpp - 1 blocks in bpp different extents.
2494 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2496 int bpp = ext4_journal_blocks_per_page(inode);
2498 return ext4_meta_trans_blocks(inode,
2499 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2503 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2504 * and underlying extent to map
2506 * @mpd - where to look for pages
2508 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2509 * IO immediately. When we find a page which isn't mapped we start accumulating
2510 * extent of buffers underlying these pages that needs mapping (formed by
2511 * either delayed or unwritten buffers). We also lock the pages containing
2512 * these buffers. The extent found is returned in @mpd structure (starting at
2513 * mpd->lblk with length mpd->len blocks).
2515 * Note that this function can attach bios to one io_end structure which are
2516 * neither logically nor physically contiguous. Although it may seem as an
2517 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2518 * case as we need to track IO to all buffers underlying a page in one io_end.
2520 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2522 struct address_space *mapping = mpd->inode->i_mapping;
2523 struct pagevec pvec;
2524 unsigned int nr_pages;
2525 long left = mpd->wbc->nr_to_write;
2526 pgoff_t index = mpd->first_page;
2527 pgoff_t end = mpd->last_page;
2530 int blkbits = mpd->inode->i_blkbits;
2532 struct buffer_head *head;
2534 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2535 tag = PAGECACHE_TAG_TOWRITE;
2537 tag = PAGECACHE_TAG_DIRTY;
2539 pagevec_init(&pvec, 0);
2541 mpd->next_page = index;
2542 while (index <= end) {
2543 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2544 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2548 for (i = 0; i < nr_pages; i++) {
2549 struct page *page = pvec.pages[i];
2552 * At this point, the page may be truncated or
2553 * invalidated (changing page->mapping to NULL), or
2554 * even swizzled back from swapper_space to tmpfs file
2555 * mapping. However, page->index will not change
2556 * because we have a reference on the page.
2558 if (page->index > end)
2562 * Accumulated enough dirty pages? This doesn't apply
2563 * to WB_SYNC_ALL mode. For integrity sync we have to
2564 * keep going because someone may be concurrently
2565 * dirtying pages, and we might have synced a lot of
2566 * newly appeared dirty pages, but have not synced all
2567 * of the old dirty pages.
2569 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2572 /* If we can't merge this page, we are done. */
2573 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2578 * If the page is no longer dirty, or its mapping no
2579 * longer corresponds to inode we are writing (which
2580 * means it has been truncated or invalidated), or the
2581 * page is already under writeback and we are not doing
2582 * a data integrity writeback, skip the page
2584 if (!PageDirty(page) ||
2585 (PageWriteback(page) &&
2586 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2587 unlikely(page->mapping != mapping)) {
2592 wait_on_page_writeback(page);
2593 BUG_ON(PageWriteback(page));
2595 if (mpd->map.m_len == 0)
2596 mpd->first_page = page->index;
2597 mpd->next_page = page->index + 1;
2598 /* Add all dirty buffers to mpd */
2599 lblk = ((ext4_lblk_t)page->index) <<
2600 (PAGE_SHIFT - blkbits);
2601 head = page_buffers(page);
2602 err = mpage_process_page_bufs(mpd, head, head, lblk);
2608 pagevec_release(&pvec);
2613 pagevec_release(&pvec);
2617 static int __writepage(struct page *page, struct writeback_control *wbc,
2620 struct address_space *mapping = data;
2621 int ret = ext4_writepage(page, wbc);
2622 mapping_set_error(mapping, ret);
2626 static int ext4_writepages(struct address_space *mapping,
2627 struct writeback_control *wbc)
2629 pgoff_t writeback_index = 0;
2630 long nr_to_write = wbc->nr_to_write;
2631 int range_whole = 0;
2633 handle_t *handle = NULL;
2634 struct mpage_da_data mpd;
2635 struct inode *inode = mapping->host;
2636 int needed_blocks, rsv_blocks = 0, ret = 0;
2637 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2639 struct blk_plug plug;
2640 bool give_up_on_write = false;
2642 percpu_down_read(&sbi->s_journal_flag_rwsem);
2643 trace_ext4_writepages(inode, wbc);
2645 if (dax_mapping(mapping)) {
2646 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2648 goto out_writepages;
2652 * No pages to write? This is mainly a kludge to avoid starting
2653 * a transaction for special inodes like journal inode on last iput()
2654 * because that could violate lock ordering on umount
2656 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2657 goto out_writepages;
2659 if (ext4_should_journal_data(inode)) {
2660 struct blk_plug plug;
2662 blk_start_plug(&plug);
2663 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2664 blk_finish_plug(&plug);
2665 goto out_writepages;
2669 * If the filesystem has aborted, it is read-only, so return
2670 * right away instead of dumping stack traces later on that
2671 * will obscure the real source of the problem. We test
2672 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2673 * the latter could be true if the filesystem is mounted
2674 * read-only, and in that case, ext4_writepages should
2675 * *never* be called, so if that ever happens, we would want
2678 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2680 goto out_writepages;
2683 if (ext4_should_dioread_nolock(inode)) {
2685 * We may need to convert up to one extent per block in
2686 * the page and we may dirty the inode.
2688 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2692 * If we have inline data and arrive here, it means that
2693 * we will soon create the block for the 1st page, so
2694 * we'd better clear the inline data here.
2696 if (ext4_has_inline_data(inode)) {
2697 /* Just inode will be modified... */
2698 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2699 if (IS_ERR(handle)) {
2700 ret = PTR_ERR(handle);
2701 goto out_writepages;
2703 BUG_ON(ext4_test_inode_state(inode,
2704 EXT4_STATE_MAY_INLINE_DATA));
2705 ext4_destroy_inline_data(handle, inode);
2706 ext4_journal_stop(handle);
2709 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2712 if (wbc->range_cyclic) {
2713 writeback_index = mapping->writeback_index;
2714 if (writeback_index)
2716 mpd.first_page = writeback_index;
2719 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2720 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2725 ext4_io_submit_init(&mpd.io_submit, wbc);
2727 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2728 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2730 blk_start_plug(&plug);
2731 while (!done && mpd.first_page <= mpd.last_page) {
2732 /* For each extent of pages we use new io_end */
2733 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2734 if (!mpd.io_submit.io_end) {
2740 * We have two constraints: We find one extent to map and we
2741 * must always write out whole page (makes a difference when
2742 * blocksize < pagesize) so that we don't block on IO when we
2743 * try to write out the rest of the page. Journalled mode is
2744 * not supported by delalloc.
2746 BUG_ON(ext4_should_journal_data(inode));
2747 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2749 /* start a new transaction */
2750 handle = ext4_journal_start_with_reserve(inode,
2751 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2752 if (IS_ERR(handle)) {
2753 ret = PTR_ERR(handle);
2754 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2755 "%ld pages, ino %lu; err %d", __func__,
2756 wbc->nr_to_write, inode->i_ino, ret);
2757 /* Release allocated io_end */
2758 ext4_put_io_end(mpd.io_submit.io_end);
2762 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2763 ret = mpage_prepare_extent_to_map(&mpd);
2766 ret = mpage_map_and_submit_extent(handle, &mpd,
2770 * We scanned the whole range (or exhausted
2771 * nr_to_write), submitted what was mapped and
2772 * didn't find anything needing mapping. We are
2779 * Caution: If the handle is synchronous,
2780 * ext4_journal_stop() can wait for transaction commit
2781 * to finish which may depend on writeback of pages to
2782 * complete or on page lock to be released. In that
2783 * case, we have to wait until after after we have
2784 * submitted all the IO, released page locks we hold,
2785 * and dropped io_end reference (for extent conversion
2786 * to be able to complete) before stopping the handle.
2788 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2789 ext4_journal_stop(handle);
2792 /* Submit prepared bio */
2793 ext4_io_submit(&mpd.io_submit);
2794 /* Unlock pages we didn't use */
2795 mpage_release_unused_pages(&mpd, give_up_on_write);
2797 * Drop our io_end reference we got from init. We have
2798 * to be careful and use deferred io_end finishing if
2799 * we are still holding the transaction as we can
2800 * release the last reference to io_end which may end
2801 * up doing unwritten extent conversion.
2804 ext4_put_io_end_defer(mpd.io_submit.io_end);
2805 ext4_journal_stop(handle);
2807 ext4_put_io_end(mpd.io_submit.io_end);
2809 if (ret == -ENOSPC && sbi->s_journal) {
2811 * Commit the transaction which would
2812 * free blocks released in the transaction
2815 jbd2_journal_force_commit_nested(sbi->s_journal);
2819 /* Fatal error - ENOMEM, EIO... */
2823 blk_finish_plug(&plug);
2824 if (!ret && !cycled && wbc->nr_to_write > 0) {
2826 mpd.last_page = writeback_index - 1;
2832 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2834 * Set the writeback_index so that range_cyclic
2835 * mode will write it back later
2837 mapping->writeback_index = mpd.first_page;
2840 trace_ext4_writepages_result(inode, wbc, ret,
2841 nr_to_write - wbc->nr_to_write);
2842 percpu_up_read(&sbi->s_journal_flag_rwsem);
2846 static int ext4_nonda_switch(struct super_block *sb)
2848 s64 free_clusters, dirty_clusters;
2849 struct ext4_sb_info *sbi = EXT4_SB(sb);
2852 * switch to non delalloc mode if we are running low
2853 * on free block. The free block accounting via percpu
2854 * counters can get slightly wrong with percpu_counter_batch getting
2855 * accumulated on each CPU without updating global counters
2856 * Delalloc need an accurate free block accounting. So switch
2857 * to non delalloc when we are near to error range.
2860 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2862 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2864 * Start pushing delalloc when 1/2 of free blocks are dirty.
2866 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2867 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2869 if (2 * free_clusters < 3 * dirty_clusters ||
2870 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2872 * free block count is less than 150% of dirty blocks
2873 * or free blocks is less than watermark
2880 /* We always reserve for an inode update; the superblock could be there too */
2881 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2883 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2886 if (pos + len <= 0x7fffffffULL)
2889 /* We might need to update the superblock to set LARGE_FILE */
2893 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2894 loff_t pos, unsigned len, unsigned flags,
2895 struct page **pagep, void **fsdata)
2897 int ret, retries = 0;
2900 struct inode *inode = mapping->host;
2903 index = pos >> PAGE_SHIFT;
2905 if (ext4_nonda_switch(inode->i_sb)) {
2906 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2907 return ext4_write_begin(file, mapping, pos,
2908 len, flags, pagep, fsdata);
2910 *fsdata = (void *)0;
2911 trace_ext4_da_write_begin(inode, pos, len, flags);
2913 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2914 ret = ext4_da_write_inline_data_begin(mapping, inode,
2924 * grab_cache_page_write_begin() can take a long time if the
2925 * system is thrashing due to memory pressure, or if the page
2926 * is being written back. So grab it first before we start
2927 * the transaction handle. This also allows us to allocate
2928 * the page (if needed) without using GFP_NOFS.
2931 page = grab_cache_page_write_begin(mapping, index, flags);
2937 * With delayed allocation, we don't log the i_disksize update
2938 * if there is delayed block allocation. But we still need
2939 * to journalling the i_disksize update if writes to the end
2940 * of file which has an already mapped buffer.
2943 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2944 ext4_da_write_credits(inode, pos, len));
2945 if (IS_ERR(handle)) {
2947 return PTR_ERR(handle);
2951 if (page->mapping != mapping) {
2952 /* The page got truncated from under us */
2955 ext4_journal_stop(handle);
2958 /* In case writeback began while the page was unlocked */
2959 wait_for_stable_page(page);
2961 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2962 ret = ext4_block_write_begin(page, pos, len,
2963 ext4_da_get_block_prep);
2965 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2969 ext4_journal_stop(handle);
2971 * block_write_begin may have instantiated a few blocks
2972 * outside i_size. Trim these off again. Don't need
2973 * i_size_read because we hold i_mutex.
2975 if (pos + len > inode->i_size)
2976 ext4_truncate_failed_write(inode);
2978 if (ret == -ENOSPC &&
2979 ext4_should_retry_alloc(inode->i_sb, &retries))
2991 * Check if we should update i_disksize
2992 * when write to the end of file but not require block allocation
2994 static int ext4_da_should_update_i_disksize(struct page *page,
2995 unsigned long offset)
2997 struct buffer_head *bh;
2998 struct inode *inode = page->mapping->host;
3002 bh = page_buffers(page);
3003 idx = offset >> inode->i_blkbits;
3005 for (i = 0; i < idx; i++)
3006 bh = bh->b_this_page;
3008 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3013 static int ext4_da_write_end(struct file *file,
3014 struct address_space *mapping,
3015 loff_t pos, unsigned len, unsigned copied,
3016 struct page *page, void *fsdata)
3018 struct inode *inode = mapping->host;
3020 handle_t *handle = ext4_journal_current_handle();
3022 unsigned long start, end;
3023 int write_mode = (int)(unsigned long)fsdata;
3025 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3026 return ext4_write_end(file, mapping, pos,
3027 len, copied, page, fsdata);
3029 trace_ext4_da_write_end(inode, pos, len, copied);
3030 start = pos & (PAGE_SIZE - 1);
3031 end = start + copied - 1;
3034 * generic_write_end() will run mark_inode_dirty() if i_size
3035 * changes. So let's piggyback the i_disksize mark_inode_dirty
3038 new_i_size = pos + copied;
3039 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3040 if (ext4_has_inline_data(inode) ||
3041 ext4_da_should_update_i_disksize(page, end)) {
3042 ext4_update_i_disksize(inode, new_i_size);
3043 /* We need to mark inode dirty even if
3044 * new_i_size is less that inode->i_size
3045 * bu greater than i_disksize.(hint delalloc)
3047 ext4_mark_inode_dirty(handle, inode);
3051 if (write_mode != CONVERT_INLINE_DATA &&
3052 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3053 ext4_has_inline_data(inode))
3054 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3057 ret2 = generic_write_end(file, mapping, pos, len, copied,
3063 ret2 = ext4_journal_stop(handle);
3067 return ret ? ret : copied;
3070 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3071 unsigned int length)
3074 * Drop reserved blocks
3076 BUG_ON(!PageLocked(page));
3077 if (!page_has_buffers(page))
3080 ext4_da_page_release_reservation(page, offset, length);
3083 ext4_invalidatepage(page, offset, length);
3089 * Force all delayed allocation blocks to be allocated for a given inode.
3091 int ext4_alloc_da_blocks(struct inode *inode)
3093 trace_ext4_alloc_da_blocks(inode);
3095 if (!EXT4_I(inode)->i_reserved_data_blocks)
3099 * We do something simple for now. The filemap_flush() will
3100 * also start triggering a write of the data blocks, which is
3101 * not strictly speaking necessary (and for users of
3102 * laptop_mode, not even desirable). However, to do otherwise
3103 * would require replicating code paths in:
3105 * ext4_writepages() ->
3106 * write_cache_pages() ---> (via passed in callback function)
3107 * __mpage_da_writepage() -->
3108 * mpage_add_bh_to_extent()
3109 * mpage_da_map_blocks()
3111 * The problem is that write_cache_pages(), located in
3112 * mm/page-writeback.c, marks pages clean in preparation for
3113 * doing I/O, which is not desirable if we're not planning on
3116 * We could call write_cache_pages(), and then redirty all of
3117 * the pages by calling redirty_page_for_writepage() but that
3118 * would be ugly in the extreme. So instead we would need to
3119 * replicate parts of the code in the above functions,
3120 * simplifying them because we wouldn't actually intend to
3121 * write out the pages, but rather only collect contiguous
3122 * logical block extents, call the multi-block allocator, and
3123 * then update the buffer heads with the block allocations.
3125 * For now, though, we'll cheat by calling filemap_flush(),
3126 * which will map the blocks, and start the I/O, but not
3127 * actually wait for the I/O to complete.
3129 return filemap_flush(inode->i_mapping);
3133 * bmap() is special. It gets used by applications such as lilo and by
3134 * the swapper to find the on-disk block of a specific piece of data.
3136 * Naturally, this is dangerous if the block concerned is still in the
3137 * journal. If somebody makes a swapfile on an ext4 data-journaling
3138 * filesystem and enables swap, then they may get a nasty shock when the
3139 * data getting swapped to that swapfile suddenly gets overwritten by
3140 * the original zero's written out previously to the journal and
3141 * awaiting writeback in the kernel's buffer cache.
3143 * So, if we see any bmap calls here on a modified, data-journaled file,
3144 * take extra steps to flush any blocks which might be in the cache.
3146 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3148 struct inode *inode = mapping->host;
3153 * We can get here for an inline file via the FIBMAP ioctl
3155 if (ext4_has_inline_data(inode))
3158 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3159 test_opt(inode->i_sb, DELALLOC)) {
3161 * With delalloc we want to sync the file
3162 * so that we can make sure we allocate
3165 filemap_write_and_wait(mapping);
3168 if (EXT4_JOURNAL(inode) &&
3169 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3171 * This is a REALLY heavyweight approach, but the use of
3172 * bmap on dirty files is expected to be extremely rare:
3173 * only if we run lilo or swapon on a freshly made file
3174 * do we expect this to happen.
3176 * (bmap requires CAP_SYS_RAWIO so this does not
3177 * represent an unprivileged user DOS attack --- we'd be
3178 * in trouble if mortal users could trigger this path at
3181 * NB. EXT4_STATE_JDATA is not set on files other than
3182 * regular files. If somebody wants to bmap a directory
3183 * or symlink and gets confused because the buffer
3184 * hasn't yet been flushed to disk, they deserve
3185 * everything they get.
3188 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3189 journal = EXT4_JOURNAL(inode);
3190 jbd2_journal_lock_updates(journal);
3191 err = jbd2_journal_flush(journal);
3192 jbd2_journal_unlock_updates(journal);
3198 return generic_block_bmap(mapping, block, ext4_get_block);
3201 static int ext4_readpage(struct file *file, struct page *page)
3204 struct inode *inode = page->mapping->host;
3206 trace_ext4_readpage(page);
3208 if (ext4_has_inline_data(inode))
3209 ret = ext4_readpage_inline(inode, page);
3212 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3218 ext4_readpages(struct file *file, struct address_space *mapping,
3219 struct list_head *pages, unsigned nr_pages)
3221 struct inode *inode = mapping->host;
3223 /* If the file has inline data, no need to do readpages. */
3224 if (ext4_has_inline_data(inode))
3227 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3230 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3231 unsigned int length)
3233 trace_ext4_invalidatepage(page, offset, length);
3235 /* No journalling happens on data buffers when this function is used */
3236 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3238 block_invalidatepage(page, offset, length);
3241 static int __ext4_journalled_invalidatepage(struct page *page,
3242 unsigned int offset,
3243 unsigned int length)
3245 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3247 trace_ext4_journalled_invalidatepage(page, offset, length);
3250 * If it's a full truncate we just forget about the pending dirtying
3252 if (offset == 0 && length == PAGE_SIZE)
3253 ClearPageChecked(page);
3255 return jbd2_journal_invalidatepage(journal, page, offset, length);
3258 /* Wrapper for aops... */
3259 static void ext4_journalled_invalidatepage(struct page *page,
3260 unsigned int offset,
3261 unsigned int length)
3263 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3266 static int ext4_releasepage(struct page *page, gfp_t wait)
3268 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3270 trace_ext4_releasepage(page);
3272 /* Page has dirty journalled data -> cannot release */
3273 if (PageChecked(page))
3276 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3278 return try_to_free_buffers(page);
3281 #ifdef CONFIG_FS_DAX
3283 * Get block function for DAX IO and mmap faults. It takes care of converting
3284 * unwritten extents to written ones and initializes new / converted blocks
3287 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3288 struct buffer_head *bh_result, int create)
3292 ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create);
3294 return _ext4_get_block(inode, iblock, bh_result, 0);
3296 ret = ext4_get_block_trans(inode, iblock, bh_result,
3297 EXT4_GET_BLOCKS_PRE_IO |
3298 EXT4_GET_BLOCKS_CREATE_ZERO);
3302 if (buffer_unwritten(bh_result)) {
3304 * We are protected by i_mmap_sem or i_mutex so we know block
3305 * cannot go away from under us even though we dropped
3306 * i_data_sem. Convert extent to written and write zeros there.
3308 ret = ext4_get_block_trans(inode, iblock, bh_result,
3309 EXT4_GET_BLOCKS_CONVERT |
3310 EXT4_GET_BLOCKS_CREATE_ZERO);
3315 * At least for now we have to clear BH_New so that DAX code
3316 * doesn't attempt to zero blocks again in a racy way.
3318 clear_buffer_new(bh_result);
3322 /* Just define empty function, it will never get called. */
3323 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3324 struct buffer_head *bh_result, int create)
3331 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3332 ssize_t size, void *private)
3334 ext4_io_end_t *io_end = private;
3336 /* if not async direct IO just return */
3340 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3341 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3342 io_end, io_end->inode->i_ino, iocb, offset, size);
3345 * Error during AIO DIO. We cannot convert unwritten extents as the
3346 * data was not written. Just clear the unwritten flag and drop io_end.
3349 ext4_clear_io_unwritten_flag(io_end);
3352 io_end->offset = offset;
3353 io_end->size = size;
3354 ext4_put_io_end(io_end);
3360 * Handling of direct IO writes.
3362 * For ext4 extent files, ext4 will do direct-io write even to holes,
3363 * preallocated extents, and those write extend the file, no need to
3364 * fall back to buffered IO.
3366 * For holes, we fallocate those blocks, mark them as unwritten
3367 * If those blocks were preallocated, we mark sure they are split, but
3368 * still keep the range to write as unwritten.
3370 * The unwritten extents will be converted to written when DIO is completed.
3371 * For async direct IO, since the IO may still pending when return, we
3372 * set up an end_io call back function, which will do the conversion
3373 * when async direct IO completed.
3375 * If the O_DIRECT write will extend the file then add this inode to the
3376 * orphan list. So recovery will truncate it back to the original size
3377 * if the machine crashes during the write.
3380 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3382 struct file *file = iocb->ki_filp;
3383 struct inode *inode = file->f_mapping->host;
3384 struct ext4_inode_info *ei = EXT4_I(inode);
3386 loff_t offset = iocb->ki_pos;
3387 size_t count = iov_iter_count(iter);
3389 get_block_t *get_block_func = NULL;
3391 loff_t final_size = offset + count;
3395 if (final_size > inode->i_size) {
3396 /* Credits for sb + inode write */
3397 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3398 if (IS_ERR(handle)) {
3399 ret = PTR_ERR(handle);
3402 ret = ext4_orphan_add(handle, inode);
3404 ext4_journal_stop(handle);
3408 ei->i_disksize = inode->i_size;
3409 ext4_journal_stop(handle);
3412 BUG_ON(iocb->private == NULL);
3415 * Make all waiters for direct IO properly wait also for extent
3416 * conversion. This also disallows race between truncate() and
3417 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3419 inode_dio_begin(inode);
3421 /* If we do a overwrite dio, i_mutex locking can be released */
3422 overwrite = *((int *)iocb->private);
3425 inode_unlock(inode);
3428 * For extent mapped files we could direct write to holes and fallocate.
3430 * Allocated blocks to fill the hole are marked as unwritten to prevent
3431 * parallel buffered read to expose the stale data before DIO complete
3434 * As to previously fallocated extents, ext4 get_block will just simply
3435 * mark the buffer mapped but still keep the extents unwritten.
3437 * For non AIO case, we will convert those unwritten extents to written
3438 * after return back from blockdev_direct_IO. That way we save us from
3439 * allocating io_end structure and also the overhead of offloading
3440 * the extent convertion to a workqueue.
3442 * For async DIO, the conversion needs to be deferred when the
3443 * IO is completed. The ext4 end_io callback function will be
3444 * called to take care of the conversion work. Here for async
3445 * case, we allocate an io_end structure to hook to the iocb.
3447 iocb->private = NULL;
3449 get_block_func = ext4_dio_get_block_overwrite;
3450 else if (IS_DAX(inode)) {
3452 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3453 * writes need zeroing either because they can race with page
3454 * faults or because they use partial blocks.
3456 if (round_down(offset, 1<<inode->i_blkbits) >= inode->i_size &&
3457 ext4_aligned_io(inode, offset, count))
3458 get_block_func = ext4_dio_get_block;
3460 get_block_func = ext4_dax_get_block;
3461 dio_flags = DIO_LOCKING;
3462 } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3463 round_down(offset, 1 << inode->i_blkbits) >= inode->i_size) {
3464 get_block_func = ext4_dio_get_block;
3465 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3466 } else if (is_sync_kiocb(iocb)) {
3467 get_block_func = ext4_dio_get_block_unwritten_sync;
3468 dio_flags = DIO_LOCKING;
3470 get_block_func = ext4_dio_get_block_unwritten_async;
3471 dio_flags = DIO_LOCKING;
3473 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3474 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3476 if (IS_DAX(inode)) {
3477 ret = dax_do_io(iocb, inode, iter, get_block_func,
3478 ext4_end_io_dio, dio_flags);
3480 ret = __blockdev_direct_IO(iocb, inode,
3481 inode->i_sb->s_bdev, iter,
3483 ext4_end_io_dio, NULL, dio_flags);
3485 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3486 EXT4_STATE_DIO_UNWRITTEN)) {
3489 * for non AIO case, since the IO is already
3490 * completed, we could do the conversion right here
3492 err = ext4_convert_unwritten_extents(NULL, inode,
3496 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3499 inode_dio_end(inode);
3500 /* take i_mutex locking again if we do a ovewrite dio */
3504 if (ret < 0 && final_size > inode->i_size)
3505 ext4_truncate_failed_write(inode);
3507 /* Handle extending of i_size after direct IO write */
3511 /* Credits for sb + inode write */
3512 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3513 if (IS_ERR(handle)) {
3514 /* This is really bad luck. We've written the data
3515 * but cannot extend i_size. Bail out and pretend
3516 * the write failed... */
3517 ret = PTR_ERR(handle);
3519 ext4_orphan_del(NULL, inode);
3524 ext4_orphan_del(handle, inode);
3526 loff_t end = offset + ret;
3527 if (end > inode->i_size) {
3528 ei->i_disksize = end;
3529 i_size_write(inode, end);
3531 * We're going to return a positive `ret'
3532 * here due to non-zero-length I/O, so there's
3533 * no way of reporting error returns from
3534 * ext4_mark_inode_dirty() to userspace. So
3537 ext4_mark_inode_dirty(handle, inode);
3540 err = ext4_journal_stop(handle);
3548 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3550 struct address_space *mapping = iocb->ki_filp->f_mapping;
3551 struct inode *inode = mapping->host;
3555 * Shared inode_lock is enough for us - it protects against concurrent
3556 * writes & truncates and since we take care of writing back page cache,
3557 * we are protected against page writeback as well.
3559 inode_lock_shared(inode);
3560 if (IS_DAX(inode)) {
3561 ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block, NULL, 0);
3563 size_t count = iov_iter_count(iter);
3565 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3566 iocb->ki_pos + count);
3569 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3570 iter, ext4_dio_get_block,
3574 inode_unlock_shared(inode);
3578 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3580 struct file *file = iocb->ki_filp;
3581 struct inode *inode = file->f_mapping->host;
3582 size_t count = iov_iter_count(iter);
3583 loff_t offset = iocb->ki_pos;
3586 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3587 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3592 * If we are doing data journalling we don't support O_DIRECT
3594 if (ext4_should_journal_data(inode))
3597 /* Let buffer I/O handle the inline data case. */
3598 if (ext4_has_inline_data(inode))
3601 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3602 if (iov_iter_rw(iter) == READ)
3603 ret = ext4_direct_IO_read(iocb, iter);
3605 ret = ext4_direct_IO_write(iocb, iter);
3606 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3611 * Pages can be marked dirty completely asynchronously from ext4's journalling
3612 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3613 * much here because ->set_page_dirty is called under VFS locks. The page is
3614 * not necessarily locked.
3616 * We cannot just dirty the page and leave attached buffers clean, because the
3617 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3618 * or jbddirty because all the journalling code will explode.
3620 * So what we do is to mark the page "pending dirty" and next time writepage
3621 * is called, propagate that into the buffers appropriately.
3623 static int ext4_journalled_set_page_dirty(struct page *page)
3625 SetPageChecked(page);
3626 return __set_page_dirty_nobuffers(page);
3629 static const struct address_space_operations ext4_aops = {
3630 .readpage = ext4_readpage,
3631 .readpages = ext4_readpages,
3632 .writepage = ext4_writepage,
3633 .writepages = ext4_writepages,
3634 .write_begin = ext4_write_begin,
3635 .write_end = ext4_write_end,
3637 .invalidatepage = ext4_invalidatepage,
3638 .releasepage = ext4_releasepage,
3639 .direct_IO = ext4_direct_IO,
3640 .migratepage = buffer_migrate_page,
3641 .is_partially_uptodate = block_is_partially_uptodate,
3642 .error_remove_page = generic_error_remove_page,
3645 static const struct address_space_operations ext4_journalled_aops = {
3646 .readpage = ext4_readpage,
3647 .readpages = ext4_readpages,
3648 .writepage = ext4_writepage,
3649 .writepages = ext4_writepages,
3650 .write_begin = ext4_write_begin,
3651 .write_end = ext4_journalled_write_end,
3652 .set_page_dirty = ext4_journalled_set_page_dirty,
3654 .invalidatepage = ext4_journalled_invalidatepage,
3655 .releasepage = ext4_releasepage,
3656 .direct_IO = ext4_direct_IO,
3657 .is_partially_uptodate = block_is_partially_uptodate,
3658 .error_remove_page = generic_error_remove_page,
3661 static const struct address_space_operations ext4_da_aops = {
3662 .readpage = ext4_readpage,
3663 .readpages = ext4_readpages,
3664 .writepage = ext4_writepage,
3665 .writepages = ext4_writepages,
3666 .write_begin = ext4_da_write_begin,
3667 .write_end = ext4_da_write_end,
3669 .invalidatepage = ext4_da_invalidatepage,
3670 .releasepage = ext4_releasepage,
3671 .direct_IO = ext4_direct_IO,
3672 .migratepage = buffer_migrate_page,
3673 .is_partially_uptodate = block_is_partially_uptodate,
3674 .error_remove_page = generic_error_remove_page,
3677 void ext4_set_aops(struct inode *inode)
3679 switch (ext4_inode_journal_mode(inode)) {
3680 case EXT4_INODE_ORDERED_DATA_MODE:
3681 case EXT4_INODE_WRITEBACK_DATA_MODE:
3683 case EXT4_INODE_JOURNAL_DATA_MODE:
3684 inode->i_mapping->a_ops = &ext4_journalled_aops;
3689 if (test_opt(inode->i_sb, DELALLOC))
3690 inode->i_mapping->a_ops = &ext4_da_aops;
3692 inode->i_mapping->a_ops = &ext4_aops;
3695 static int __ext4_block_zero_page_range(handle_t *handle,
3696 struct address_space *mapping, loff_t from, loff_t length)
3698 ext4_fsblk_t index = from >> PAGE_SHIFT;
3699 unsigned offset = from & (PAGE_SIZE-1);
3700 unsigned blocksize, pos;
3702 struct inode *inode = mapping->host;
3703 struct buffer_head *bh;
3707 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3708 mapping_gfp_constraint(mapping, ~__GFP_FS));
3712 blocksize = inode->i_sb->s_blocksize;
3714 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3716 if (!page_has_buffers(page))
3717 create_empty_buffers(page, blocksize, 0);
3719 /* Find the buffer that contains "offset" */
3720 bh = page_buffers(page);
3722 while (offset >= pos) {
3723 bh = bh->b_this_page;
3727 if (buffer_freed(bh)) {
3728 BUFFER_TRACE(bh, "freed: skip");
3731 if (!buffer_mapped(bh)) {
3732 BUFFER_TRACE(bh, "unmapped");
3733 ext4_get_block(inode, iblock, bh, 0);
3734 /* unmapped? It's a hole - nothing to do */
3735 if (!buffer_mapped(bh)) {
3736 BUFFER_TRACE(bh, "still unmapped");
3741 /* Ok, it's mapped. Make sure it's up-to-date */
3742 if (PageUptodate(page))
3743 set_buffer_uptodate(bh);
3745 if (!buffer_uptodate(bh)) {
3747 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3749 /* Uhhuh. Read error. Complain and punt. */
3750 if (!buffer_uptodate(bh))
3752 if (S_ISREG(inode->i_mode) &&
3753 ext4_encrypted_inode(inode)) {
3754 /* We expect the key to be set. */
3755 BUG_ON(!fscrypt_has_encryption_key(inode));
3756 BUG_ON(blocksize != PAGE_SIZE);
3757 BUG_ON(!PageLocked(page));
3758 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
3759 page, PAGE_SIZE, 0, page->index));
3762 if (ext4_should_journal_data(inode)) {
3763 BUFFER_TRACE(bh, "get write access");
3764 err = ext4_journal_get_write_access(handle, bh);
3768 zero_user(page, offset, length);
3769 BUFFER_TRACE(bh, "zeroed end of block");
3771 if (ext4_should_journal_data(inode)) {
3772 err = ext4_handle_dirty_metadata(handle, inode, bh);
3775 mark_buffer_dirty(bh);
3776 if (ext4_should_order_data(inode))
3777 err = ext4_jbd2_inode_add_write(handle, inode);
3787 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3788 * starting from file offset 'from'. The range to be zero'd must
3789 * be contained with in one block. If the specified range exceeds
3790 * the end of the block it will be shortened to end of the block
3791 * that cooresponds to 'from'
3793 static int ext4_block_zero_page_range(handle_t *handle,
3794 struct address_space *mapping, loff_t from, loff_t length)
3796 struct inode *inode = mapping->host;
3797 unsigned offset = from & (PAGE_SIZE-1);
3798 unsigned blocksize = inode->i_sb->s_blocksize;
3799 unsigned max = blocksize - (offset & (blocksize - 1));
3802 * correct length if it does not fall between
3803 * 'from' and the end of the block
3805 if (length > max || length < 0)
3809 return dax_zero_page_range(inode, from, length, ext4_get_block);
3810 return __ext4_block_zero_page_range(handle, mapping, from, length);
3814 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3815 * up to the end of the block which corresponds to `from'.
3816 * This required during truncate. We need to physically zero the tail end
3817 * of that block so it doesn't yield old data if the file is later grown.
3819 static int ext4_block_truncate_page(handle_t *handle,
3820 struct address_space *mapping, loff_t from)
3822 unsigned offset = from & (PAGE_SIZE-1);
3825 struct inode *inode = mapping->host;
3827 blocksize = inode->i_sb->s_blocksize;
3828 length = blocksize - (offset & (blocksize - 1));
3830 return ext4_block_zero_page_range(handle, mapping, from, length);
3833 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3834 loff_t lstart, loff_t length)
3836 struct super_block *sb = inode->i_sb;
3837 struct address_space *mapping = inode->i_mapping;
3838 unsigned partial_start, partial_end;
3839 ext4_fsblk_t start, end;
3840 loff_t byte_end = (lstart + length - 1);
3843 partial_start = lstart & (sb->s_blocksize - 1);
3844 partial_end = byte_end & (sb->s_blocksize - 1);
3846 start = lstart >> sb->s_blocksize_bits;
3847 end = byte_end >> sb->s_blocksize_bits;
3849 /* Handle partial zero within the single block */
3851 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3852 err = ext4_block_zero_page_range(handle, mapping,
3856 /* Handle partial zero out on the start of the range */
3857 if (partial_start) {
3858 err = ext4_block_zero_page_range(handle, mapping,
3859 lstart, sb->s_blocksize);
3863 /* Handle partial zero out on the end of the range */
3864 if (partial_end != sb->s_blocksize - 1)
3865 err = ext4_block_zero_page_range(handle, mapping,
3866 byte_end - partial_end,
3871 int ext4_can_truncate(struct inode *inode)
3873 if (S_ISREG(inode->i_mode))
3875 if (S_ISDIR(inode->i_mode))
3877 if (S_ISLNK(inode->i_mode))
3878 return !ext4_inode_is_fast_symlink(inode);
3883 * We have to make sure i_disksize gets properly updated before we truncate
3884 * page cache due to hole punching or zero range. Otherwise i_disksize update
3885 * can get lost as it may have been postponed to submission of writeback but
3886 * that will never happen after we truncate page cache.
3888 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3892 loff_t size = i_size_read(inode);
3894 WARN_ON(!inode_is_locked(inode));
3895 if (offset > size || offset + len < size)
3898 if (EXT4_I(inode)->i_disksize >= size)
3901 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3903 return PTR_ERR(handle);
3904 ext4_update_i_disksize(inode, size);
3905 ext4_mark_inode_dirty(handle, inode);
3906 ext4_journal_stop(handle);
3912 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3913 * associated with the given offset and length
3915 * @inode: File inode
3916 * @offset: The offset where the hole will begin
3917 * @len: The length of the hole
3919 * Returns: 0 on success or negative on failure
3922 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3924 struct super_block *sb = inode->i_sb;
3925 ext4_lblk_t first_block, stop_block;
3926 struct address_space *mapping = inode->i_mapping;
3927 loff_t first_block_offset, last_block_offset;
3929 unsigned int credits;
3932 if (!S_ISREG(inode->i_mode))
3935 trace_ext4_punch_hole(inode, offset, length, 0);
3938 * Write out all dirty pages to avoid race conditions
3939 * Then release them.
3941 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3942 ret = filemap_write_and_wait_range(mapping, offset,
3943 offset + length - 1);
3950 /* No need to punch hole beyond i_size */
3951 if (offset >= inode->i_size)
3955 * If the hole extends beyond i_size, set the hole
3956 * to end after the page that contains i_size
3958 if (offset + length > inode->i_size) {
3959 length = inode->i_size +
3960 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3964 if (offset & (sb->s_blocksize - 1) ||
3965 (offset + length) & (sb->s_blocksize - 1)) {
3967 * Attach jinode to inode for jbd2 if we do any zeroing of
3970 ret = ext4_inode_attach_jinode(inode);
3976 /* Wait all existing dio workers, newcomers will block on i_mutex */
3977 ext4_inode_block_unlocked_dio(inode);
3978 inode_dio_wait(inode);
3981 * Prevent page faults from reinstantiating pages we have released from
3984 down_write(&EXT4_I(inode)->i_mmap_sem);
3985 first_block_offset = round_up(offset, sb->s_blocksize);
3986 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3988 /* Now release the pages and zero block aligned part of pages*/
3989 if (last_block_offset > first_block_offset) {
3990 ret = ext4_update_disksize_before_punch(inode, offset, length);
3993 truncate_pagecache_range(inode, first_block_offset,
3997 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3998 credits = ext4_writepage_trans_blocks(inode);
4000 credits = ext4_blocks_for_truncate(inode);
4001 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4002 if (IS_ERR(handle)) {
4003 ret = PTR_ERR(handle);
4004 ext4_std_error(sb, ret);
4008 ret = ext4_zero_partial_blocks(handle, inode, offset,
4013 first_block = (offset + sb->s_blocksize - 1) >>
4014 EXT4_BLOCK_SIZE_BITS(sb);
4015 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4017 /* If there are no blocks to remove, return now */
4018 if (first_block >= stop_block)
4021 down_write(&EXT4_I(inode)->i_data_sem);
4022 ext4_discard_preallocations(inode);
4024 ret = ext4_es_remove_extent(inode, first_block,
4025 stop_block - first_block);
4027 up_write(&EXT4_I(inode)->i_data_sem);
4031 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4032 ret = ext4_ext_remove_space(inode, first_block,
4035 ret = ext4_ind_remove_space(handle, inode, first_block,
4038 up_write(&EXT4_I(inode)->i_data_sem);
4040 ext4_handle_sync(handle);
4042 inode->i_mtime = inode->i_ctime = current_time(inode);
4043 ext4_mark_inode_dirty(handle, inode);
4045 ext4_journal_stop(handle);
4047 up_write(&EXT4_I(inode)->i_mmap_sem);
4048 ext4_inode_resume_unlocked_dio(inode);
4050 inode_unlock(inode);
4054 int ext4_inode_attach_jinode(struct inode *inode)
4056 struct ext4_inode_info *ei = EXT4_I(inode);
4057 struct jbd2_inode *jinode;
4059 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4062 jinode = jbd2_alloc_inode(GFP_KERNEL);
4063 spin_lock(&inode->i_lock);
4066 spin_unlock(&inode->i_lock);
4069 ei->jinode = jinode;
4070 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4073 spin_unlock(&inode->i_lock);
4074 if (unlikely(jinode != NULL))
4075 jbd2_free_inode(jinode);
4082 * We block out ext4_get_block() block instantiations across the entire
4083 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4084 * simultaneously on behalf of the same inode.
4086 * As we work through the truncate and commit bits of it to the journal there
4087 * is one core, guiding principle: the file's tree must always be consistent on
4088 * disk. We must be able to restart the truncate after a crash.
4090 * The file's tree may be transiently inconsistent in memory (although it
4091 * probably isn't), but whenever we close off and commit a journal transaction,
4092 * the contents of (the filesystem + the journal) must be consistent and
4093 * restartable. It's pretty simple, really: bottom up, right to left (although
4094 * left-to-right works OK too).
4096 * Note that at recovery time, journal replay occurs *before* the restart of
4097 * truncate against the orphan inode list.
4099 * The committed inode has the new, desired i_size (which is the same as
4100 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4101 * that this inode's truncate did not complete and it will again call
4102 * ext4_truncate() to have another go. So there will be instantiated blocks
4103 * to the right of the truncation point in a crashed ext4 filesystem. But
4104 * that's fine - as long as they are linked from the inode, the post-crash
4105 * ext4_truncate() run will find them and release them.
4107 int ext4_truncate(struct inode *inode)
4109 struct ext4_inode_info *ei = EXT4_I(inode);
4110 unsigned int credits;
4113 struct address_space *mapping = inode->i_mapping;
4116 * There is a possibility that we're either freeing the inode
4117 * or it's a completely new inode. In those cases we might not
4118 * have i_mutex locked because it's not necessary.
4120 if (!(inode->i_state & (I_NEW|I_FREEING)))
4121 WARN_ON(!inode_is_locked(inode));
4122 trace_ext4_truncate_enter(inode);
4124 if (!ext4_can_truncate(inode))
4127 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4129 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4130 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4132 if (ext4_has_inline_data(inode)) {
4135 ext4_inline_data_truncate(inode, &has_inline);
4140 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4141 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4142 if (ext4_inode_attach_jinode(inode) < 0)
4146 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4147 credits = ext4_writepage_trans_blocks(inode);
4149 credits = ext4_blocks_for_truncate(inode);
4151 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4153 return PTR_ERR(handle);
4155 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4156 ext4_block_truncate_page(handle, mapping, inode->i_size);
4159 * We add the inode to the orphan list, so that if this
4160 * truncate spans multiple transactions, and we crash, we will
4161 * resume the truncate when the filesystem recovers. It also
4162 * marks the inode dirty, to catch the new size.
4164 * Implication: the file must always be in a sane, consistent
4165 * truncatable state while each transaction commits.
4167 err = ext4_orphan_add(handle, inode);
4171 down_write(&EXT4_I(inode)->i_data_sem);
4173 ext4_discard_preallocations(inode);
4175 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4176 err = ext4_ext_truncate(handle, inode);
4178 ext4_ind_truncate(handle, inode);
4180 up_write(&ei->i_data_sem);
4185 ext4_handle_sync(handle);
4189 * If this was a simple ftruncate() and the file will remain alive,
4190 * then we need to clear up the orphan record which we created above.
4191 * However, if this was a real unlink then we were called by
4192 * ext4_evict_inode(), and we allow that function to clean up the
4193 * orphan info for us.
4196 ext4_orphan_del(handle, inode);
4198 inode->i_mtime = inode->i_ctime = current_time(inode);
4199 ext4_mark_inode_dirty(handle, inode);
4200 ext4_journal_stop(handle);
4202 trace_ext4_truncate_exit(inode);
4207 * ext4_get_inode_loc returns with an extra refcount against the inode's
4208 * underlying buffer_head on success. If 'in_mem' is true, we have all
4209 * data in memory that is needed to recreate the on-disk version of this
4212 static int __ext4_get_inode_loc(struct inode *inode,
4213 struct ext4_iloc *iloc, int in_mem)
4215 struct ext4_group_desc *gdp;
4216 struct buffer_head *bh;
4217 struct super_block *sb = inode->i_sb;
4219 int inodes_per_block, inode_offset;
4222 if (!ext4_valid_inum(sb, inode->i_ino))
4223 return -EFSCORRUPTED;
4225 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4226 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4231 * Figure out the offset within the block group inode table
4233 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4234 inode_offset = ((inode->i_ino - 1) %
4235 EXT4_INODES_PER_GROUP(sb));
4236 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4237 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4239 bh = sb_getblk(sb, block);
4242 if (!buffer_uptodate(bh)) {
4246 * If the buffer has the write error flag, we have failed
4247 * to write out another inode in the same block. In this
4248 * case, we don't have to read the block because we may
4249 * read the old inode data successfully.
4251 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4252 set_buffer_uptodate(bh);
4254 if (buffer_uptodate(bh)) {
4255 /* someone brought it uptodate while we waited */
4261 * If we have all information of the inode in memory and this
4262 * is the only valid inode in the block, we need not read the
4266 struct buffer_head *bitmap_bh;
4269 start = inode_offset & ~(inodes_per_block - 1);
4271 /* Is the inode bitmap in cache? */
4272 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4273 if (unlikely(!bitmap_bh))
4277 * If the inode bitmap isn't in cache then the
4278 * optimisation may end up performing two reads instead
4279 * of one, so skip it.
4281 if (!buffer_uptodate(bitmap_bh)) {
4285 for (i = start; i < start + inodes_per_block; i++) {
4286 if (i == inode_offset)
4288 if (ext4_test_bit(i, bitmap_bh->b_data))
4292 if (i == start + inodes_per_block) {
4293 /* all other inodes are free, so skip I/O */
4294 memset(bh->b_data, 0, bh->b_size);
4295 set_buffer_uptodate(bh);
4303 * If we need to do any I/O, try to pre-readahead extra
4304 * blocks from the inode table.
4306 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4307 ext4_fsblk_t b, end, table;
4309 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4311 table = ext4_inode_table(sb, gdp);
4312 /* s_inode_readahead_blks is always a power of 2 */
4313 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4317 num = EXT4_INODES_PER_GROUP(sb);
4318 if (ext4_has_group_desc_csum(sb))
4319 num -= ext4_itable_unused_count(sb, gdp);
4320 table += num / inodes_per_block;
4324 sb_breadahead(sb, b++);
4328 * There are other valid inodes in the buffer, this inode
4329 * has in-inode xattrs, or we don't have this inode in memory.
4330 * Read the block from disk.
4332 trace_ext4_load_inode(inode);
4334 bh->b_end_io = end_buffer_read_sync;
4335 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4337 if (!buffer_uptodate(bh)) {
4338 EXT4_ERROR_INODE_BLOCK(inode, block,
4339 "unable to read itable block");
4349 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4351 /* We have all inode data except xattrs in memory here. */
4352 return __ext4_get_inode_loc(inode, iloc,
4353 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4356 void ext4_set_inode_flags(struct inode *inode)
4358 unsigned int flags = EXT4_I(inode)->i_flags;
4359 unsigned int new_fl = 0;
4361 if (flags & EXT4_SYNC_FL)
4363 if (flags & EXT4_APPEND_FL)
4365 if (flags & EXT4_IMMUTABLE_FL)
4366 new_fl |= S_IMMUTABLE;
4367 if (flags & EXT4_NOATIME_FL)
4368 new_fl |= S_NOATIME;
4369 if (flags & EXT4_DIRSYNC_FL)
4370 new_fl |= S_DIRSYNC;
4371 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) &&
4372 !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) &&
4373 !ext4_encrypted_inode(inode))
4375 inode_set_flags(inode, new_fl,
4376 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4379 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4380 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4382 unsigned int vfs_fl;
4383 unsigned long old_fl, new_fl;
4386 vfs_fl = ei->vfs_inode.i_flags;
4387 old_fl = ei->i_flags;
4388 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4389 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4391 if (vfs_fl & S_SYNC)
4392 new_fl |= EXT4_SYNC_FL;
4393 if (vfs_fl & S_APPEND)
4394 new_fl |= EXT4_APPEND_FL;
4395 if (vfs_fl & S_IMMUTABLE)
4396 new_fl |= EXT4_IMMUTABLE_FL;
4397 if (vfs_fl & S_NOATIME)
4398 new_fl |= EXT4_NOATIME_FL;
4399 if (vfs_fl & S_DIRSYNC)
4400 new_fl |= EXT4_DIRSYNC_FL;
4401 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4404 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4405 struct ext4_inode_info *ei)
4408 struct inode *inode = &(ei->vfs_inode);
4409 struct super_block *sb = inode->i_sb;
4411 if (ext4_has_feature_huge_file(sb)) {
4412 /* we are using combined 48 bit field */
4413 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4414 le32_to_cpu(raw_inode->i_blocks_lo);
4415 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4416 /* i_blocks represent file system block size */
4417 return i_blocks << (inode->i_blkbits - 9);
4422 return le32_to_cpu(raw_inode->i_blocks_lo);
4426 static inline void ext4_iget_extra_inode(struct inode *inode,
4427 struct ext4_inode *raw_inode,
4428 struct ext4_inode_info *ei)
4430 __le32 *magic = (void *)raw_inode +
4431 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4432 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4433 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4434 ext4_find_inline_data_nolock(inode);
4436 EXT4_I(inode)->i_inline_off = 0;
4439 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4441 if (!ext4_has_feature_project(inode->i_sb))
4443 *projid = EXT4_I(inode)->i_projid;
4447 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4449 struct ext4_iloc iloc;
4450 struct ext4_inode *raw_inode;
4451 struct ext4_inode_info *ei;
4452 struct inode *inode;
4453 journal_t *journal = EXT4_SB(sb)->s_journal;
4460 inode = iget_locked(sb, ino);
4462 return ERR_PTR(-ENOMEM);
4463 if (!(inode->i_state & I_NEW))
4469 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4472 raw_inode = ext4_raw_inode(&iloc);
4474 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4475 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4476 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4477 EXT4_INODE_SIZE(inode->i_sb)) {
4478 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4479 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4480 EXT4_INODE_SIZE(inode->i_sb));
4481 ret = -EFSCORRUPTED;
4485 ei->i_extra_isize = 0;
4487 /* Precompute checksum seed for inode metadata */
4488 if (ext4_has_metadata_csum(sb)) {
4489 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4491 __le32 inum = cpu_to_le32(inode->i_ino);
4492 __le32 gen = raw_inode->i_generation;
4493 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4495 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4499 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4500 EXT4_ERROR_INODE(inode, "checksum invalid");
4505 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4506 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4507 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4508 if (ext4_has_feature_project(sb) &&
4509 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4510 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4511 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4513 i_projid = EXT4_DEF_PROJID;
4515 if (!(test_opt(inode->i_sb, NO_UID32))) {
4516 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4517 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4519 i_uid_write(inode, i_uid);
4520 i_gid_write(inode, i_gid);
4521 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4522 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4524 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4525 ei->i_inline_off = 0;
4526 ei->i_dir_start_lookup = 0;
4527 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4528 /* We now have enough fields to check if the inode was active or not.
4529 * This is needed because nfsd might try to access dead inodes
4530 * the test is that same one that e2fsck uses
4531 * NeilBrown 1999oct15
4533 if (inode->i_nlink == 0) {
4534 if ((inode->i_mode == 0 ||
4535 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4536 ino != EXT4_BOOT_LOADER_INO) {
4537 /* this inode is deleted */
4541 /* The only unlinked inodes we let through here have
4542 * valid i_mode and are being read by the orphan
4543 * recovery code: that's fine, we're about to complete
4544 * the process of deleting those.
4545 * OR it is the EXT4_BOOT_LOADER_INO which is
4546 * not initialized on a new filesystem. */
4548 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4549 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4550 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4551 if (ext4_has_feature_64bit(sb))
4553 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4554 inode->i_size = ext4_isize(raw_inode);
4555 ei->i_disksize = inode->i_size;
4557 ei->i_reserved_quota = 0;
4559 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4560 ei->i_block_group = iloc.block_group;
4561 ei->i_last_alloc_group = ~0;
4563 * NOTE! The in-memory inode i_data array is in little-endian order
4564 * even on big-endian machines: we do NOT byteswap the block numbers!
4566 for (block = 0; block < EXT4_N_BLOCKS; block++)
4567 ei->i_data[block] = raw_inode->i_block[block];
4568 INIT_LIST_HEAD(&ei->i_orphan);
4571 * Set transaction id's of transactions that have to be committed
4572 * to finish f[data]sync. We set them to currently running transaction
4573 * as we cannot be sure that the inode or some of its metadata isn't
4574 * part of the transaction - the inode could have been reclaimed and
4575 * now it is reread from disk.
4578 transaction_t *transaction;
4581 read_lock(&journal->j_state_lock);
4582 if (journal->j_running_transaction)
4583 transaction = journal->j_running_transaction;
4585 transaction = journal->j_committing_transaction;
4587 tid = transaction->t_tid;
4589 tid = journal->j_commit_sequence;
4590 read_unlock(&journal->j_state_lock);
4591 ei->i_sync_tid = tid;
4592 ei->i_datasync_tid = tid;
4595 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4596 if (ei->i_extra_isize == 0) {
4597 /* The extra space is currently unused. Use it. */
4598 ei->i_extra_isize = sizeof(struct ext4_inode) -
4599 EXT4_GOOD_OLD_INODE_SIZE;
4601 ext4_iget_extra_inode(inode, raw_inode, ei);
4605 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4606 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4607 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4608 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4610 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4611 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4612 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4613 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4615 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4620 if (ei->i_file_acl &&
4621 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4622 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4624 ret = -EFSCORRUPTED;
4626 } else if (!ext4_has_inline_data(inode)) {
4627 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4628 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4629 (S_ISLNK(inode->i_mode) &&
4630 !ext4_inode_is_fast_symlink(inode))))
4631 /* Validate extent which is part of inode */
4632 ret = ext4_ext_check_inode(inode);
4633 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4634 (S_ISLNK(inode->i_mode) &&
4635 !ext4_inode_is_fast_symlink(inode))) {
4636 /* Validate block references which are part of inode */
4637 ret = ext4_ind_check_inode(inode);
4643 if (S_ISREG(inode->i_mode)) {
4644 inode->i_op = &ext4_file_inode_operations;
4645 inode->i_fop = &ext4_file_operations;
4646 ext4_set_aops(inode);
4647 } else if (S_ISDIR(inode->i_mode)) {
4648 inode->i_op = &ext4_dir_inode_operations;
4649 inode->i_fop = &ext4_dir_operations;
4650 } else if (S_ISLNK(inode->i_mode)) {
4651 if (ext4_encrypted_inode(inode)) {
4652 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4653 ext4_set_aops(inode);
4654 } else if (ext4_inode_is_fast_symlink(inode)) {
4655 inode->i_link = (char *)ei->i_data;
4656 inode->i_op = &ext4_fast_symlink_inode_operations;
4657 nd_terminate_link(ei->i_data, inode->i_size,
4658 sizeof(ei->i_data) - 1);
4660 inode->i_op = &ext4_symlink_inode_operations;
4661 ext4_set_aops(inode);
4663 inode_nohighmem(inode);
4664 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4665 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4666 inode->i_op = &ext4_special_inode_operations;
4667 if (raw_inode->i_block[0])
4668 init_special_inode(inode, inode->i_mode,
4669 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4671 init_special_inode(inode, inode->i_mode,
4672 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4673 } else if (ino == EXT4_BOOT_LOADER_INO) {
4674 make_bad_inode(inode);
4676 ret = -EFSCORRUPTED;
4677 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4681 ext4_set_inode_flags(inode);
4682 unlock_new_inode(inode);
4688 return ERR_PTR(ret);
4691 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4693 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4694 return ERR_PTR(-EFSCORRUPTED);
4695 return ext4_iget(sb, ino);
4698 static int ext4_inode_blocks_set(handle_t *handle,
4699 struct ext4_inode *raw_inode,
4700 struct ext4_inode_info *ei)
4702 struct inode *inode = &(ei->vfs_inode);
4703 u64 i_blocks = inode->i_blocks;
4704 struct super_block *sb = inode->i_sb;
4706 if (i_blocks <= ~0U) {
4708 * i_blocks can be represented in a 32 bit variable
4709 * as multiple of 512 bytes
4711 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4712 raw_inode->i_blocks_high = 0;
4713 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4716 if (!ext4_has_feature_huge_file(sb))
4719 if (i_blocks <= 0xffffffffffffULL) {
4721 * i_blocks can be represented in a 48 bit variable
4722 * as multiple of 512 bytes
4724 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4725 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4726 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4728 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4729 /* i_block is stored in file system block size */
4730 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4731 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4732 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4737 struct other_inode {
4738 unsigned long orig_ino;
4739 struct ext4_inode *raw_inode;
4742 static int other_inode_match(struct inode * inode, unsigned long ino,
4745 struct other_inode *oi = (struct other_inode *) data;
4747 if ((inode->i_ino != ino) ||
4748 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4749 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4750 ((inode->i_state & I_DIRTY_TIME) == 0))
4752 spin_lock(&inode->i_lock);
4753 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4754 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4755 (inode->i_state & I_DIRTY_TIME)) {
4756 struct ext4_inode_info *ei = EXT4_I(inode);
4758 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4759 spin_unlock(&inode->i_lock);
4761 spin_lock(&ei->i_raw_lock);
4762 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4763 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4764 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4765 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4766 spin_unlock(&ei->i_raw_lock);
4767 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4770 spin_unlock(&inode->i_lock);
4775 * Opportunistically update the other time fields for other inodes in
4776 * the same inode table block.
4778 static void ext4_update_other_inodes_time(struct super_block *sb,
4779 unsigned long orig_ino, char *buf)
4781 struct other_inode oi;
4783 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4784 int inode_size = EXT4_INODE_SIZE(sb);
4786 oi.orig_ino = orig_ino;
4788 * Calculate the first inode in the inode table block. Inode
4789 * numbers are one-based. That is, the first inode in a block
4790 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4792 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4793 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4794 if (ino == orig_ino)
4796 oi.raw_inode = (struct ext4_inode *) buf;
4797 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4802 * Post the struct inode info into an on-disk inode location in the
4803 * buffer-cache. This gobbles the caller's reference to the
4804 * buffer_head in the inode location struct.
4806 * The caller must have write access to iloc->bh.
4808 static int ext4_do_update_inode(handle_t *handle,
4809 struct inode *inode,
4810 struct ext4_iloc *iloc)
4812 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4813 struct ext4_inode_info *ei = EXT4_I(inode);
4814 struct buffer_head *bh = iloc->bh;
4815 struct super_block *sb = inode->i_sb;
4816 int err = 0, rc, block;
4817 int need_datasync = 0, set_large_file = 0;
4822 spin_lock(&ei->i_raw_lock);
4824 /* For fields not tracked in the in-memory inode,
4825 * initialise them to zero for new inodes. */
4826 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4827 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4829 ext4_get_inode_flags(ei);
4830 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4831 i_uid = i_uid_read(inode);
4832 i_gid = i_gid_read(inode);
4833 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4834 if (!(test_opt(inode->i_sb, NO_UID32))) {
4835 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4836 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4838 * Fix up interoperability with old kernels. Otherwise, old inodes get
4839 * re-used with the upper 16 bits of the uid/gid intact
4841 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4842 raw_inode->i_uid_high = 0;
4843 raw_inode->i_gid_high = 0;
4845 raw_inode->i_uid_high =
4846 cpu_to_le16(high_16_bits(i_uid));
4847 raw_inode->i_gid_high =
4848 cpu_to_le16(high_16_bits(i_gid));
4851 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4852 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4853 raw_inode->i_uid_high = 0;
4854 raw_inode->i_gid_high = 0;
4856 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4858 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4859 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4860 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4861 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4863 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4865 spin_unlock(&ei->i_raw_lock);
4868 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4869 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4870 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4871 raw_inode->i_file_acl_high =
4872 cpu_to_le16(ei->i_file_acl >> 32);
4873 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4874 if (ei->i_disksize != ext4_isize(raw_inode)) {
4875 ext4_isize_set(raw_inode, ei->i_disksize);
4878 if (ei->i_disksize > 0x7fffffffULL) {
4879 if (!ext4_has_feature_large_file(sb) ||
4880 EXT4_SB(sb)->s_es->s_rev_level ==
4881 cpu_to_le32(EXT4_GOOD_OLD_REV))
4884 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4885 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4886 if (old_valid_dev(inode->i_rdev)) {
4887 raw_inode->i_block[0] =
4888 cpu_to_le32(old_encode_dev(inode->i_rdev));
4889 raw_inode->i_block[1] = 0;
4891 raw_inode->i_block[0] = 0;
4892 raw_inode->i_block[1] =
4893 cpu_to_le32(new_encode_dev(inode->i_rdev));
4894 raw_inode->i_block[2] = 0;
4896 } else if (!ext4_has_inline_data(inode)) {
4897 for (block = 0; block < EXT4_N_BLOCKS; block++)
4898 raw_inode->i_block[block] = ei->i_data[block];
4901 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4902 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4903 if (ei->i_extra_isize) {
4904 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4905 raw_inode->i_version_hi =
4906 cpu_to_le32(inode->i_version >> 32);
4907 raw_inode->i_extra_isize =
4908 cpu_to_le16(ei->i_extra_isize);
4912 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
4913 i_projid != EXT4_DEF_PROJID);
4915 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4916 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4917 raw_inode->i_projid = cpu_to_le32(i_projid);
4919 ext4_inode_csum_set(inode, raw_inode, ei);
4920 spin_unlock(&ei->i_raw_lock);
4921 if (inode->i_sb->s_flags & MS_LAZYTIME)
4922 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4925 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4926 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4929 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4930 if (set_large_file) {
4931 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4932 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4935 ext4_update_dynamic_rev(sb);
4936 ext4_set_feature_large_file(sb);
4937 ext4_handle_sync(handle);
4938 err = ext4_handle_dirty_super(handle, sb);
4940 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4943 ext4_std_error(inode->i_sb, err);
4948 * ext4_write_inode()
4950 * We are called from a few places:
4952 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4953 * Here, there will be no transaction running. We wait for any running
4954 * transaction to commit.
4956 * - Within flush work (sys_sync(), kupdate and such).
4957 * We wait on commit, if told to.
4959 * - Within iput_final() -> write_inode_now()
4960 * We wait on commit, if told to.
4962 * In all cases it is actually safe for us to return without doing anything,
4963 * because the inode has been copied into a raw inode buffer in
4964 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4967 * Note that we are absolutely dependent upon all inode dirtiers doing the
4968 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4969 * which we are interested.
4971 * It would be a bug for them to not do this. The code:
4973 * mark_inode_dirty(inode)
4975 * inode->i_size = expr;
4977 * is in error because write_inode() could occur while `stuff()' is running,
4978 * and the new i_size will be lost. Plus the inode will no longer be on the
4979 * superblock's dirty inode list.
4981 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4985 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4988 if (EXT4_SB(inode->i_sb)->s_journal) {
4989 if (ext4_journal_current_handle()) {
4990 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4996 * No need to force transaction in WB_SYNC_NONE mode. Also
4997 * ext4_sync_fs() will force the commit after everything is
5000 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5003 err = ext4_force_commit(inode->i_sb);
5005 struct ext4_iloc iloc;
5007 err = __ext4_get_inode_loc(inode, &iloc, 0);
5011 * sync(2) will flush the whole buffer cache. No need to do
5012 * it here separately for each inode.
5014 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5015 sync_dirty_buffer(iloc.bh);
5016 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5017 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5018 "IO error syncing inode");
5027 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5028 * buffers that are attached to a page stradding i_size and are undergoing
5029 * commit. In that case we have to wait for commit to finish and try again.
5031 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5035 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5036 tid_t commit_tid = 0;
5039 offset = inode->i_size & (PAGE_SIZE - 1);
5041 * All buffers in the last page remain valid? Then there's nothing to
5042 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5045 if (offset > PAGE_SIZE - (1 << inode->i_blkbits))
5048 page = find_lock_page(inode->i_mapping,
5049 inode->i_size >> PAGE_SHIFT);
5052 ret = __ext4_journalled_invalidatepage(page, offset,
5053 PAGE_SIZE - offset);
5059 read_lock(&journal->j_state_lock);
5060 if (journal->j_committing_transaction)
5061 commit_tid = journal->j_committing_transaction->t_tid;
5062 read_unlock(&journal->j_state_lock);
5064 jbd2_log_wait_commit(journal, commit_tid);
5071 * Called from notify_change.
5073 * We want to trap VFS attempts to truncate the file as soon as
5074 * possible. In particular, we want to make sure that when the VFS
5075 * shrinks i_size, we put the inode on the orphan list and modify
5076 * i_disksize immediately, so that during the subsequent flushing of
5077 * dirty pages and freeing of disk blocks, we can guarantee that any
5078 * commit will leave the blocks being flushed in an unused state on
5079 * disk. (On recovery, the inode will get truncated and the blocks will
5080 * be freed, so we have a strong guarantee that no future commit will
5081 * leave these blocks visible to the user.)
5083 * Another thing we have to assure is that if we are in ordered mode
5084 * and inode is still attached to the committing transaction, we must
5085 * we start writeout of all the dirty pages which are being truncated.
5086 * This way we are sure that all the data written in the previous
5087 * transaction are already on disk (truncate waits for pages under
5090 * Called with inode->i_mutex down.
5092 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5094 struct inode *inode = d_inode(dentry);
5097 const unsigned int ia_valid = attr->ia_valid;
5099 error = setattr_prepare(dentry, attr);
5103 if (is_quota_modification(inode, attr)) {
5104 error = dquot_initialize(inode);
5108 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5109 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5112 /* (user+group)*(old+new) structure, inode write (sb,
5113 * inode block, ? - but truncate inode update has it) */
5114 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5115 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5116 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5117 if (IS_ERR(handle)) {
5118 error = PTR_ERR(handle);
5121 error = dquot_transfer(inode, attr);
5123 ext4_journal_stop(handle);
5126 /* Update corresponding info in inode so that everything is in
5127 * one transaction */
5128 if (attr->ia_valid & ATTR_UID)
5129 inode->i_uid = attr->ia_uid;
5130 if (attr->ia_valid & ATTR_GID)
5131 inode->i_gid = attr->ia_gid;
5132 error = ext4_mark_inode_dirty(handle, inode);
5133 ext4_journal_stop(handle);
5136 if (attr->ia_valid & ATTR_SIZE) {
5138 loff_t oldsize = inode->i_size;
5139 int shrink = (attr->ia_size <= inode->i_size);
5141 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5142 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5144 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5147 if (!S_ISREG(inode->i_mode))
5150 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5151 inode_inc_iversion(inode);
5153 if (ext4_should_order_data(inode) &&
5154 (attr->ia_size < inode->i_size)) {
5155 error = ext4_begin_ordered_truncate(inode,
5160 if (attr->ia_size != inode->i_size) {
5161 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5162 if (IS_ERR(handle)) {
5163 error = PTR_ERR(handle);
5166 if (ext4_handle_valid(handle) && shrink) {
5167 error = ext4_orphan_add(handle, inode);
5171 * Update c/mtime on truncate up, ext4_truncate() will
5172 * update c/mtime in shrink case below
5175 inode->i_mtime = current_time(inode);
5176 inode->i_ctime = inode->i_mtime;
5178 down_write(&EXT4_I(inode)->i_data_sem);
5179 EXT4_I(inode)->i_disksize = attr->ia_size;
5180 rc = ext4_mark_inode_dirty(handle, inode);
5184 * We have to update i_size under i_data_sem together
5185 * with i_disksize to avoid races with writeback code
5186 * running ext4_wb_update_i_disksize().
5189 i_size_write(inode, attr->ia_size);
5190 up_write(&EXT4_I(inode)->i_data_sem);
5191 ext4_journal_stop(handle);
5194 ext4_orphan_del(NULL, inode);
5199 pagecache_isize_extended(inode, oldsize, inode->i_size);
5202 * Blocks are going to be removed from the inode. Wait
5203 * for dio in flight. Temporarily disable
5204 * dioread_nolock to prevent livelock.
5207 if (!ext4_should_journal_data(inode)) {
5208 ext4_inode_block_unlocked_dio(inode);
5209 inode_dio_wait(inode);
5210 ext4_inode_resume_unlocked_dio(inode);
5212 ext4_wait_for_tail_page_commit(inode);
5214 down_write(&EXT4_I(inode)->i_mmap_sem);
5216 * Truncate pagecache after we've waited for commit
5217 * in data=journal mode to make pages freeable.
5219 truncate_pagecache(inode, inode->i_size);
5221 rc = ext4_truncate(inode);
5225 up_write(&EXT4_I(inode)->i_mmap_sem);
5229 setattr_copy(inode, attr);
5230 mark_inode_dirty(inode);
5234 * If the call to ext4_truncate failed to get a transaction handle at
5235 * all, we need to clean up the in-core orphan list manually.
5237 if (orphan && inode->i_nlink)
5238 ext4_orphan_del(NULL, inode);
5240 if (!error && (ia_valid & ATTR_MODE))
5241 rc = posix_acl_chmod(inode, inode->i_mode);
5244 ext4_std_error(inode->i_sb, error);
5250 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5253 struct inode *inode;
5254 unsigned long long delalloc_blocks;
5256 inode = d_inode(dentry);
5257 generic_fillattr(inode, stat);
5260 * If there is inline data in the inode, the inode will normally not
5261 * have data blocks allocated (it may have an external xattr block).
5262 * Report at least one sector for such files, so tools like tar, rsync,
5263 * others doen't incorrectly think the file is completely sparse.
5265 if (unlikely(ext4_has_inline_data(inode)))
5266 stat->blocks += (stat->size + 511) >> 9;
5269 * We can't update i_blocks if the block allocation is delayed
5270 * otherwise in the case of system crash before the real block
5271 * allocation is done, we will have i_blocks inconsistent with
5272 * on-disk file blocks.
5273 * We always keep i_blocks updated together with real
5274 * allocation. But to not confuse with user, stat
5275 * will return the blocks that include the delayed allocation
5276 * blocks for this file.
5278 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5279 EXT4_I(inode)->i_reserved_data_blocks);
5280 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5284 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5287 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5288 return ext4_ind_trans_blocks(inode, lblocks);
5289 return ext4_ext_index_trans_blocks(inode, pextents);
5293 * Account for index blocks, block groups bitmaps and block group
5294 * descriptor blocks if modify datablocks and index blocks
5295 * worse case, the indexs blocks spread over different block groups
5297 * If datablocks are discontiguous, they are possible to spread over
5298 * different block groups too. If they are contiguous, with flexbg,
5299 * they could still across block group boundary.
5301 * Also account for superblock, inode, quota and xattr blocks
5303 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5306 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5312 * How many index blocks need to touch to map @lblocks logical blocks
5313 * to @pextents physical extents?
5315 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5320 * Now let's see how many group bitmaps and group descriptors need
5323 groups = idxblocks + pextents;
5325 if (groups > ngroups)
5327 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5328 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5330 /* bitmaps and block group descriptor blocks */
5331 ret += groups + gdpblocks;
5333 /* Blocks for super block, inode, quota and xattr blocks */
5334 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5340 * Calculate the total number of credits to reserve to fit
5341 * the modification of a single pages into a single transaction,
5342 * which may include multiple chunks of block allocations.
5344 * This could be called via ext4_write_begin()
5346 * We need to consider the worse case, when
5347 * one new block per extent.
5349 int ext4_writepage_trans_blocks(struct inode *inode)
5351 int bpp = ext4_journal_blocks_per_page(inode);
5354 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5356 /* Account for data blocks for journalled mode */
5357 if (ext4_should_journal_data(inode))
5363 * Calculate the journal credits for a chunk of data modification.
5365 * This is called from DIO, fallocate or whoever calling
5366 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5368 * journal buffers for data blocks are not included here, as DIO
5369 * and fallocate do no need to journal data buffers.
5371 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5373 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5377 * The caller must have previously called ext4_reserve_inode_write().
5378 * Give this, we know that the caller already has write access to iloc->bh.
5380 int ext4_mark_iloc_dirty(handle_t *handle,
5381 struct inode *inode, struct ext4_iloc *iloc)
5385 if (IS_I_VERSION(inode))
5386 inode_inc_iversion(inode);
5388 /* the do_update_inode consumes one bh->b_count */
5391 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5392 err = ext4_do_update_inode(handle, inode, iloc);
5398 * On success, We end up with an outstanding reference count against
5399 * iloc->bh. This _must_ be cleaned up later.
5403 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5404 struct ext4_iloc *iloc)
5408 err = ext4_get_inode_loc(inode, iloc);
5410 BUFFER_TRACE(iloc->bh, "get_write_access");
5411 err = ext4_journal_get_write_access(handle, iloc->bh);
5417 ext4_std_error(inode->i_sb, err);
5422 * Expand an inode by new_extra_isize bytes.
5423 * Returns 0 on success or negative error number on failure.
5425 static int ext4_expand_extra_isize(struct inode *inode,
5426 unsigned int new_extra_isize,
5427 struct ext4_iloc iloc,
5430 struct ext4_inode *raw_inode;
5431 struct ext4_xattr_ibody_header *header;
5433 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5436 raw_inode = ext4_raw_inode(&iloc);
5438 header = IHDR(inode, raw_inode);
5440 /* No extended attributes present */
5441 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5442 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5443 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5445 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5449 /* try to expand with EAs present */
5450 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5455 * What we do here is to mark the in-core inode as clean with respect to inode
5456 * dirtiness (it may still be data-dirty).
5457 * This means that the in-core inode may be reaped by prune_icache
5458 * without having to perform any I/O. This is a very good thing,
5459 * because *any* task may call prune_icache - even ones which
5460 * have a transaction open against a different journal.
5462 * Is this cheating? Not really. Sure, we haven't written the
5463 * inode out, but prune_icache isn't a user-visible syncing function.
5464 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5465 * we start and wait on commits.
5467 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5469 struct ext4_iloc iloc;
5470 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5471 static unsigned int mnt_count;
5475 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5476 err = ext4_reserve_inode_write(handle, inode, &iloc);
5479 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5480 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5482 * In nojournal mode, we can immediately attempt to expand
5483 * the inode. When journaled, we first need to obtain extra
5484 * buffer credits since we may write into the EA block
5485 * with this same handle. If journal_extend fails, then it will
5486 * only result in a minor loss of functionality for that inode.
5487 * If this is felt to be critical, then e2fsck should be run to
5488 * force a large enough s_min_extra_isize.
5490 if (!ext4_handle_valid(handle) ||
5491 jbd2_journal_extend(handle,
5492 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) == 0) {
5493 ret = ext4_expand_extra_isize(inode,
5494 sbi->s_want_extra_isize,
5498 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5499 ext4_warning(inode->i_sb,
5500 "Unable to expand inode %lu. Delete"
5501 " some EAs or run e2fsck.",
5504 le16_to_cpu(sbi->s_es->s_mnt_count);
5509 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5513 * ext4_dirty_inode() is called from __mark_inode_dirty()
5515 * We're really interested in the case where a file is being extended.
5516 * i_size has been changed by generic_commit_write() and we thus need
5517 * to include the updated inode in the current transaction.
5519 * Also, dquot_alloc_block() will always dirty the inode when blocks
5520 * are allocated to the file.
5522 * If the inode is marked synchronous, we don't honour that here - doing
5523 * so would cause a commit on atime updates, which we don't bother doing.
5524 * We handle synchronous inodes at the highest possible level.
5526 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5527 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5528 * to copy into the on-disk inode structure are the timestamp files.
5530 void ext4_dirty_inode(struct inode *inode, int flags)
5534 if (flags == I_DIRTY_TIME)
5536 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5540 ext4_mark_inode_dirty(handle, inode);
5542 ext4_journal_stop(handle);
5549 * Bind an inode's backing buffer_head into this transaction, to prevent
5550 * it from being flushed to disk early. Unlike
5551 * ext4_reserve_inode_write, this leaves behind no bh reference and
5552 * returns no iloc structure, so the caller needs to repeat the iloc
5553 * lookup to mark the inode dirty later.
5555 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5557 struct ext4_iloc iloc;
5561 err = ext4_get_inode_loc(inode, &iloc);
5563 BUFFER_TRACE(iloc.bh, "get_write_access");
5564 err = jbd2_journal_get_write_access(handle, iloc.bh);
5566 err = ext4_handle_dirty_metadata(handle,
5572 ext4_std_error(inode->i_sb, err);
5577 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5582 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5585 * We have to be very careful here: changing a data block's
5586 * journaling status dynamically is dangerous. If we write a
5587 * data block to the journal, change the status and then delete
5588 * that block, we risk forgetting to revoke the old log record
5589 * from the journal and so a subsequent replay can corrupt data.
5590 * So, first we make sure that the journal is empty and that
5591 * nobody is changing anything.
5594 journal = EXT4_JOURNAL(inode);
5597 if (is_journal_aborted(journal))
5600 /* Wait for all existing dio workers */
5601 ext4_inode_block_unlocked_dio(inode);
5602 inode_dio_wait(inode);
5605 * Before flushing the journal and switching inode's aops, we have
5606 * to flush all dirty data the inode has. There can be outstanding
5607 * delayed allocations, there can be unwritten extents created by
5608 * fallocate or buffered writes in dioread_nolock mode covered by
5609 * dirty data which can be converted only after flushing the dirty
5610 * data (and journalled aops don't know how to handle these cases).
5613 down_write(&EXT4_I(inode)->i_mmap_sem);
5614 err = filemap_write_and_wait(inode->i_mapping);
5616 up_write(&EXT4_I(inode)->i_mmap_sem);
5617 ext4_inode_resume_unlocked_dio(inode);
5622 percpu_down_write(&sbi->s_journal_flag_rwsem);
5623 jbd2_journal_lock_updates(journal);
5626 * OK, there are no updates running now, and all cached data is
5627 * synced to disk. We are now in a completely consistent state
5628 * which doesn't have anything in the journal, and we know that
5629 * no filesystem updates are running, so it is safe to modify
5630 * the inode's in-core data-journaling state flag now.
5634 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5636 err = jbd2_journal_flush(journal);
5638 jbd2_journal_unlock_updates(journal);
5639 percpu_up_write(&sbi->s_journal_flag_rwsem);
5640 ext4_inode_resume_unlocked_dio(inode);
5643 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5645 ext4_set_aops(inode);
5647 * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated.
5648 * E.g. S_DAX may get cleared / set.
5650 ext4_set_inode_flags(inode);
5652 jbd2_journal_unlock_updates(journal);
5653 percpu_up_write(&sbi->s_journal_flag_rwsem);
5656 up_write(&EXT4_I(inode)->i_mmap_sem);
5657 ext4_inode_resume_unlocked_dio(inode);
5659 /* Finally we can mark the inode as dirty. */
5661 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5663 return PTR_ERR(handle);
5665 err = ext4_mark_inode_dirty(handle, inode);
5666 ext4_handle_sync(handle);
5667 ext4_journal_stop(handle);
5668 ext4_std_error(inode->i_sb, err);
5673 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5675 return !buffer_mapped(bh);
5678 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5680 struct page *page = vmf->page;
5684 struct file *file = vma->vm_file;
5685 struct inode *inode = file_inode(file);
5686 struct address_space *mapping = inode->i_mapping;
5688 get_block_t *get_block;
5691 sb_start_pagefault(inode->i_sb);
5692 file_update_time(vma->vm_file);
5694 down_read(&EXT4_I(inode)->i_mmap_sem);
5695 /* Delalloc case is easy... */
5696 if (test_opt(inode->i_sb, DELALLOC) &&
5697 !ext4_should_journal_data(inode) &&
5698 !ext4_nonda_switch(inode->i_sb)) {
5700 ret = block_page_mkwrite(vma, vmf,
5701 ext4_da_get_block_prep);
5702 } while (ret == -ENOSPC &&
5703 ext4_should_retry_alloc(inode->i_sb, &retries));
5708 size = i_size_read(inode);
5709 /* Page got truncated from under us? */
5710 if (page->mapping != mapping || page_offset(page) > size) {
5712 ret = VM_FAULT_NOPAGE;
5716 if (page->index == size >> PAGE_SHIFT)
5717 len = size & ~PAGE_MASK;
5721 * Return if we have all the buffers mapped. This avoids the need to do
5722 * journal_start/journal_stop which can block and take a long time
5724 if (page_has_buffers(page)) {
5725 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5727 ext4_bh_unmapped)) {
5728 /* Wait so that we don't change page under IO */
5729 wait_for_stable_page(page);
5730 ret = VM_FAULT_LOCKED;
5735 /* OK, we need to fill the hole... */
5736 if (ext4_should_dioread_nolock(inode))
5737 get_block = ext4_get_block_unwritten;
5739 get_block = ext4_get_block;
5741 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5742 ext4_writepage_trans_blocks(inode));
5743 if (IS_ERR(handle)) {
5744 ret = VM_FAULT_SIGBUS;
5747 ret = block_page_mkwrite(vma, vmf, get_block);
5748 if (!ret && ext4_should_journal_data(inode)) {
5749 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5750 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5752 ret = VM_FAULT_SIGBUS;
5753 ext4_journal_stop(handle);
5756 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5758 ext4_journal_stop(handle);
5759 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5762 ret = block_page_mkwrite_return(ret);
5764 up_read(&EXT4_I(inode)->i_mmap_sem);
5765 sb_end_pagefault(inode->i_sb);
5769 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5771 struct inode *inode = file_inode(vma->vm_file);
5774 down_read(&EXT4_I(inode)->i_mmap_sem);
5775 err = filemap_fault(vma, vmf);
5776 up_read(&EXT4_I(inode)->i_mmap_sem);
5782 * Find the first extent at or after @lblk in an inode that is not a hole.
5783 * Search for @map_len blocks at most. The extent is returned in @result.
5785 * The function returns 1 if we found an extent. The function returns 0 in
5786 * case there is no extent at or after @lblk and in that case also sets
5787 * @result->es_len to 0. In case of error, the error code is returned.
5789 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5790 unsigned int map_len, struct extent_status *result)
5792 struct ext4_map_blocks map;
5793 struct extent_status es = {};
5797 map.m_len = map_len;
5800 * For non-extent based files this loop may iterate several times since
5801 * we do not determine full hole size.
5803 while (map.m_len > 0) {
5804 ret = ext4_map_blocks(NULL, inode, &map, 0);
5807 /* There's extent covering m_lblk? Just return it. */
5811 ext4_es_store_pblock(result, map.m_pblk);
5812 result->es_lblk = map.m_lblk;
5813 result->es_len = map.m_len;
5814 if (map.m_flags & EXT4_MAP_UNWRITTEN)
5815 status = EXTENT_STATUS_UNWRITTEN;
5817 status = EXTENT_STATUS_WRITTEN;
5818 ext4_es_store_status(result, status);
5821 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5822 map.m_lblk + map.m_len - 1,
5824 /* Is delalloc data before next block in extent tree? */
5825 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5826 ext4_lblk_t offset = 0;
5828 if (es.es_lblk < lblk)
5829 offset = lblk - es.es_lblk;
5830 result->es_lblk = es.es_lblk + offset;
5831 ext4_es_store_pblock(result,
5832 ext4_es_pblock(&es) + offset);
5833 result->es_len = es.es_len - offset;
5834 ext4_es_store_status(result, ext4_es_status(&es));
5838 /* There's a hole at m_lblk, advance us after it */
5839 map.m_lblk += map.m_len;
5840 map_len -= map.m_len;
5841 map.m_len = map_len;