1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
54 struct ext4_inode_info
*ei
)
56 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
59 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
60 unsigned int csum_size
= sizeof(dummy_csum
);
62 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
63 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
65 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
66 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
68 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
69 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
70 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
71 EXT4_GOOD_OLD_INODE_SIZE
,
72 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
73 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
74 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
78 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
79 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
85 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
86 struct ext4_inode_info
*ei
)
88 __u32 provided
, calculated
;
90 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
91 cpu_to_le32(EXT4_OS_LINUX
) ||
92 !ext4_has_metadata_csum(inode
->i_sb
))
95 provided
= le16_to_cpu(raw
->i_checksum_lo
);
96 calculated
= ext4_inode_csum(inode
, raw
, ei
);
97 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
98 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
99 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
101 calculated
&= 0xFFFF;
103 return provided
== calculated
;
106 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
107 struct ext4_inode_info
*ei
)
111 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
112 cpu_to_le32(EXT4_OS_LINUX
) ||
113 !ext4_has_metadata_csum(inode
->i_sb
))
116 csum
= ext4_inode_csum(inode
, raw
, ei
);
117 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
118 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
119 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
120 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
126 trace_ext4_begin_ordered_truncate(inode
, new_size
);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode
)->jinode
)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
136 EXT4_I(inode
)->jinode
,
140 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
141 unsigned int length
);
142 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
143 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
144 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode
*inode
)
153 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
154 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
155 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
157 if (ext4_has_inline_data(inode
))
160 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
162 return S_ISLNK(inode
->i_mode
) && inode
->i_size
&&
163 (inode
->i_size
< EXT4_N_BLOCKS
* 4);
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
182 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
183 jbd_debug(2, "restarting handle %p\n", handle
);
184 up_write(&EXT4_I(inode
)->i_data_sem
);
185 ret
= ext4_journal_restart(handle
, nblocks
);
186 down_write(&EXT4_I(inode
)->i_data_sem
);
187 ext4_discard_preallocations(inode
);
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_evict_inode(struct inode
*inode
)
199 int extra_credits
= 3;
200 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
202 trace_ext4_evict_inode(inode
);
204 if (inode
->i_nlink
) {
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
220 * Note that directories do not have this problem because they
221 * don't use page cache.
223 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
224 ext4_should_journal_data(inode
) &&
225 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
226 inode
->i_data
.nrpages
) {
227 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
228 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
230 jbd2_complete_transaction(journal
, commit_tid
);
231 filemap_write_and_wait(&inode
->i_data
);
233 truncate_inode_pages_final(&inode
->i_data
);
238 if (is_bad_inode(inode
))
240 dquot_initialize(inode
);
242 if (ext4_should_order_data(inode
))
243 ext4_begin_ordered_truncate(inode
, 0);
244 truncate_inode_pages_final(&inode
->i_data
);
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
250 sb_start_intwrite(inode
->i_sb
);
252 if (!IS_NOQUOTA(inode
))
253 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
255 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
256 ext4_blocks_for_truncate(inode
)+extra_credits
);
257 if (IS_ERR(handle
)) {
258 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
264 ext4_orphan_del(NULL
, inode
);
265 sb_end_intwrite(inode
->i_sb
);
270 ext4_handle_sync(handle
);
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
279 if (ext4_inode_is_fast_symlink(inode
))
280 memset(EXT4_I(inode
)->i_data
, 0, sizeof(EXT4_I(inode
)->i_data
));
282 err
= ext4_mark_inode_dirty(handle
, inode
);
284 ext4_warning(inode
->i_sb
,
285 "couldn't mark inode dirty (err %d)", err
);
288 if (inode
->i_blocks
) {
289 err
= ext4_truncate(inode
);
291 ext4_error(inode
->i_sb
,
292 "couldn't truncate inode %lu (err %d)",
298 /* Remove xattr references. */
299 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
302 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
304 ext4_journal_stop(handle
);
305 ext4_orphan_del(NULL
, inode
);
306 sb_end_intwrite(inode
->i_sb
);
307 ext4_xattr_inode_array_free(ea_inode_array
);
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
319 ext4_orphan_del(handle
, inode
);
320 EXT4_I(inode
)->i_dtime
= (__u32
)ktime_get_real_seconds();
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
329 if (ext4_mark_inode_dirty(handle
, inode
))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode
);
333 ext4_free_inode(handle
, inode
);
334 ext4_journal_stop(handle
);
335 sb_end_intwrite(inode
->i_sb
);
336 ext4_xattr_inode_array_free(ea_inode_array
);
339 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
343 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
345 return &EXT4_I(inode
)->i_reserved_quota
;
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
353 void ext4_da_update_reserve_space(struct inode
*inode
,
354 int used
, int quota_claim
)
356 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
357 struct ext4_inode_info
*ei
= EXT4_I(inode
);
359 spin_lock(&ei
->i_block_reservation_lock
);
360 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
361 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
362 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__
, inode
->i_ino
, used
,
365 ei
->i_reserved_data_blocks
);
367 used
= ei
->i_reserved_data_blocks
;
370 /* Update per-inode reservations */
371 ei
->i_reserved_data_blocks
-= used
;
372 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
374 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
376 /* Update quota subsystem for data blocks */
378 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
385 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
393 if ((ei
->i_reserved_data_blocks
== 0) &&
394 !inode_is_open_for_write(inode
))
395 ext4_discard_preallocations(inode
);
398 static int __check_block_validity(struct inode
*inode
, const char *func
,
400 struct ext4_map_blocks
*map
)
402 if (ext4_has_feature_journal(inode
->i_sb
) &&
404 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_journal_inum
)))
406 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
408 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
409 "lblock %lu mapped to illegal pblock %llu "
410 "(length %d)", (unsigned long) map
->m_lblk
,
411 map
->m_pblk
, map
->m_len
);
412 return -EFSCORRUPTED
;
417 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
422 if (IS_ENCRYPTED(inode
))
423 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
425 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
432 #define check_block_validity(inode, map) \
433 __check_block_validity((inode), __func__, __LINE__, (map))
435 #ifdef ES_AGGRESSIVE_TEST
436 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
438 struct ext4_map_blocks
*es_map
,
439 struct ext4_map_blocks
*map
,
446 * There is a race window that the result is not the same.
447 * e.g. xfstests #223 when dioread_nolock enables. The reason
448 * is that we lookup a block mapping in extent status tree with
449 * out taking i_data_sem. So at the time the unwritten extent
450 * could be converted.
452 down_read(&EXT4_I(inode
)->i_data_sem
);
453 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
454 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
455 EXT4_GET_BLOCKS_KEEP_SIZE
);
457 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
458 EXT4_GET_BLOCKS_KEEP_SIZE
);
460 up_read((&EXT4_I(inode
)->i_data_sem
));
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
466 if (es_map
->m_lblk
!= map
->m_lblk
||
467 es_map
->m_flags
!= map
->m_flags
||
468 es_map
->m_pblk
!= map
->m_pblk
) {
469 printk("ES cache assertion failed for inode: %lu "
470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
473 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
474 map
->m_len
, map
->m_pblk
, map
->m_flags
,
478 #endif /* ES_AGGRESSIVE_TEST */
481 * The ext4_map_blocks() function tries to look up the requested blocks,
482 * and returns if the blocks are already mapped.
484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
492 * On success, it returns the number of blocks being mapped or allocated. if
493 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
496 * It returns 0 if plain look up failed (blocks have not been allocated), in
497 * that case, @map is returned as unmapped but we still do fill map->m_len to
498 * indicate the length of a hole starting at map->m_lblk.
500 * It returns the error in case of allocation failure.
502 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
503 struct ext4_map_blocks
*map
, int flags
)
505 struct extent_status es
;
508 #ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map
;
511 memcpy(&orig_map
, map
, sizeof(*map
));
515 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
517 (unsigned long) map
->m_lblk
);
520 * ext4_map_blocks returns an int, and m_len is an unsigned int
522 if (unlikely(map
->m_len
> INT_MAX
))
523 map
->m_len
= INT_MAX
;
525 /* We can handle the block number less than EXT_MAX_BLOCKS */
526 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
527 return -EFSCORRUPTED
;
529 /* Lookup extent status tree firstly */
530 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
531 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
532 map
->m_pblk
= ext4_es_pblock(&es
) +
533 map
->m_lblk
- es
.es_lblk
;
534 map
->m_flags
|= ext4_es_is_written(&es
) ?
535 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
536 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
537 if (retval
> map
->m_len
)
540 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
542 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
543 if (retval
> map
->m_len
)
550 #ifdef ES_AGGRESSIVE_TEST
551 ext4_map_blocks_es_recheck(handle
, inode
, map
,
558 * Try to see if we can get the block without requesting a new
561 down_read(&EXT4_I(inode
)->i_data_sem
);
562 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
563 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
564 EXT4_GET_BLOCKS_KEEP_SIZE
);
566 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
567 EXT4_GET_BLOCKS_KEEP_SIZE
);
572 if (unlikely(retval
!= map
->m_len
)) {
573 ext4_warning(inode
->i_sb
,
574 "ES len assertion failed for inode "
575 "%lu: retval %d != map->m_len %d",
576 inode
->i_ino
, retval
, map
->m_len
);
580 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
581 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
582 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
583 !(status
& EXTENT_STATUS_WRITTEN
) &&
584 ext4_es_scan_range(inode
, &ext4_es_is_delayed
, map
->m_lblk
,
585 map
->m_lblk
+ map
->m_len
- 1))
586 status
|= EXTENT_STATUS_DELAYED
;
587 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
588 map
->m_len
, map
->m_pblk
, status
);
592 up_read((&EXT4_I(inode
)->i_data_sem
));
595 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
596 ret
= check_block_validity(inode
, map
);
601 /* If it is only a block(s) look up */
602 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
606 * Returns if the blocks have already allocated
608 * Note that if blocks have been preallocated
609 * ext4_ext_get_block() returns the create = 0
610 * with buffer head unmapped.
612 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
614 * If we need to convert extent to unwritten
615 * we continue and do the actual work in
616 * ext4_ext_map_blocks()
618 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
622 * Here we clear m_flags because after allocating an new extent,
623 * it will be set again.
625 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
628 * New blocks allocate and/or writing to unwritten extent
629 * will possibly result in updating i_data, so we take
630 * the write lock of i_data_sem, and call get_block()
631 * with create == 1 flag.
633 down_write(&EXT4_I(inode
)->i_data_sem
);
636 * We need to check for EXT4 here because migrate
637 * could have changed the inode type in between
639 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
640 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
642 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
644 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
646 * We allocated new blocks which will result in
647 * i_data's format changing. Force the migrate
648 * to fail by clearing migrate flags
650 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
654 * Update reserved blocks/metadata blocks after successful
655 * block allocation which had been deferred till now. We don't
656 * support fallocate for non extent files. So we can update
657 * reserve space here.
660 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
661 ext4_da_update_reserve_space(inode
, retval
, 1);
667 if (unlikely(retval
!= map
->m_len
)) {
668 ext4_warning(inode
->i_sb
,
669 "ES len assertion failed for inode "
670 "%lu: retval %d != map->m_len %d",
671 inode
->i_ino
, retval
, map
->m_len
);
676 * We have to zeroout blocks before inserting them into extent
677 * status tree. Otherwise someone could look them up there and
678 * use them before they are really zeroed. We also have to
679 * unmap metadata before zeroing as otherwise writeback can
680 * overwrite zeros with stale data from block device.
682 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
683 map
->m_flags
& EXT4_MAP_MAPPED
&&
684 map
->m_flags
& EXT4_MAP_NEW
) {
685 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
686 map
->m_pblk
, map
->m_len
);
694 * If the extent has been zeroed out, we don't need to update
695 * extent status tree.
697 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
698 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
699 if (ext4_es_is_written(&es
))
702 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
703 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
704 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
705 !(status
& EXTENT_STATUS_WRITTEN
) &&
706 ext4_es_scan_range(inode
, &ext4_es_is_delayed
, map
->m_lblk
,
707 map
->m_lblk
+ map
->m_len
- 1))
708 status
|= EXTENT_STATUS_DELAYED
;
709 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
710 map
->m_pblk
, status
);
718 up_write((&EXT4_I(inode
)->i_data_sem
));
719 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
720 ret
= check_block_validity(inode
, map
);
725 * Inodes with freshly allocated blocks where contents will be
726 * visible after transaction commit must be on transaction's
729 if (map
->m_flags
& EXT4_MAP_NEW
&&
730 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
731 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
732 !ext4_is_quota_file(inode
) &&
733 ext4_should_order_data(inode
)) {
734 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
735 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
737 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
746 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
747 * we have to be careful as someone else may be manipulating b_state as well.
749 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
751 unsigned long old_state
;
752 unsigned long new_state
;
754 flags
&= EXT4_MAP_FLAGS
;
756 /* Dummy buffer_head? Set non-atomically. */
758 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
762 * Someone else may be modifying b_state. Be careful! This is ugly but
763 * once we get rid of using bh as a container for mapping information
764 * to pass to / from get_block functions, this can go away.
767 old_state
= READ_ONCE(bh
->b_state
);
768 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
770 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
773 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
774 struct buffer_head
*bh
, int flags
)
776 struct ext4_map_blocks map
;
779 if (ext4_has_inline_data(inode
))
783 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
785 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
788 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
789 ext4_update_bh_state(bh
, map
.m_flags
);
790 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
792 } else if (ret
== 0) {
793 /* hole case, need to fill in bh->b_size */
794 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
799 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
800 struct buffer_head
*bh
, int create
)
802 return _ext4_get_block(inode
, iblock
, bh
,
803 create
? EXT4_GET_BLOCKS_CREATE
: 0);
807 * Get block function used when preparing for buffered write if we require
808 * creating an unwritten extent if blocks haven't been allocated. The extent
809 * will be converted to written after the IO is complete.
811 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
812 struct buffer_head
*bh_result
, int create
)
814 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
815 inode
->i_ino
, create
);
816 return _ext4_get_block(inode
, iblock
, bh_result
,
817 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
820 /* Maximum number of blocks we map for direct IO at once. */
821 #define DIO_MAX_BLOCKS 4096
824 * Get blocks function for the cases that need to start a transaction -
825 * generally difference cases of direct IO and DAX IO. It also handles retries
828 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
829 struct buffer_head
*bh_result
, int flags
)
836 /* Trim mapping request to maximum we can map at once for DIO */
837 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
838 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
839 dio_credits
= ext4_chunk_trans_blocks(inode
,
840 bh_result
->b_size
>> inode
->i_blkbits
);
842 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
844 return PTR_ERR(handle
);
846 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
847 ext4_journal_stop(handle
);
849 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
854 /* Get block function for DIO reads and writes to inodes without extents */
855 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
856 struct buffer_head
*bh
, int create
)
858 /* We don't expect handle for direct IO */
859 WARN_ON_ONCE(ext4_journal_current_handle());
862 return _ext4_get_block(inode
, iblock
, bh
, 0);
863 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
867 * Get block function for AIO DIO writes when we create unwritten extent if
868 * blocks are not allocated yet. The extent will be converted to written
869 * after IO is complete.
871 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
872 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
876 /* We don't expect handle for direct IO */
877 WARN_ON_ONCE(ext4_journal_current_handle());
879 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
880 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
883 * When doing DIO using unwritten extents, we need io_end to convert
884 * unwritten extents to written on IO completion. We allocate io_end
885 * once we spot unwritten extent and store it in b_private. Generic
886 * DIO code keeps b_private set and furthermore passes the value to
887 * our completion callback in 'private' argument.
889 if (!ret
&& buffer_unwritten(bh_result
)) {
890 if (!bh_result
->b_private
) {
891 ext4_io_end_t
*io_end
;
893 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
896 bh_result
->b_private
= io_end
;
897 ext4_set_io_unwritten_flag(inode
, io_end
);
899 set_buffer_defer_completion(bh_result
);
906 * Get block function for non-AIO DIO writes when we create unwritten extent if
907 * blocks are not allocated yet. The extent will be converted to written
908 * after IO is complete by ext4_direct_IO_write().
910 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
911 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
915 /* We don't expect handle for direct IO */
916 WARN_ON_ONCE(ext4_journal_current_handle());
918 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
919 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
922 * Mark inode as having pending DIO writes to unwritten extents.
923 * ext4_direct_IO_write() checks this flag and converts extents to
926 if (!ret
&& buffer_unwritten(bh_result
))
927 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
932 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
933 struct buffer_head
*bh_result
, int create
)
937 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
938 inode
->i_ino
, create
);
939 /* We don't expect handle for direct IO */
940 WARN_ON_ONCE(ext4_journal_current_handle());
942 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
944 * Blocks should have been preallocated! ext4_file_write_iter() checks
947 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
954 * `handle' can be NULL if create is zero
956 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
957 ext4_lblk_t block
, int map_flags
)
959 struct ext4_map_blocks map
;
960 struct buffer_head
*bh
;
961 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
964 J_ASSERT(handle
!= NULL
|| create
== 0);
968 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
971 return create
? ERR_PTR(-ENOSPC
) : NULL
;
975 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
977 return ERR_PTR(-ENOMEM
);
978 if (map
.m_flags
& EXT4_MAP_NEW
) {
979 J_ASSERT(create
!= 0);
980 J_ASSERT(handle
!= NULL
);
983 * Now that we do not always journal data, we should
984 * keep in mind whether this should always journal the
985 * new buffer as metadata. For now, regular file
986 * writes use ext4_get_block instead, so it's not a
990 BUFFER_TRACE(bh
, "call get_create_access");
991 err
= ext4_journal_get_create_access(handle
, bh
);
996 if (!buffer_uptodate(bh
)) {
997 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
998 set_buffer_uptodate(bh
);
1001 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1002 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1006 BUFFER_TRACE(bh
, "not a new buffer");
1010 return ERR_PTR(err
);
1013 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1014 ext4_lblk_t block
, int map_flags
)
1016 struct buffer_head
*bh
;
1018 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1021 if (!bh
|| buffer_uptodate(bh
))
1023 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1025 if (buffer_uptodate(bh
))
1028 return ERR_PTR(-EIO
);
1031 /* Read a contiguous batch of blocks. */
1032 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
1033 bool wait
, struct buffer_head
**bhs
)
1037 for (i
= 0; i
< bh_count
; i
++) {
1038 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
1039 if (IS_ERR(bhs
[i
])) {
1040 err
= PTR_ERR(bhs
[i
]);
1046 for (i
= 0; i
< bh_count
; i
++)
1047 /* Note that NULL bhs[i] is valid because of holes. */
1048 if (bhs
[i
] && !buffer_uptodate(bhs
[i
]))
1049 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1,
1055 for (i
= 0; i
< bh_count
; i
++)
1057 wait_on_buffer(bhs
[i
]);
1059 for (i
= 0; i
< bh_count
; i
++) {
1060 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
1068 for (i
= 0; i
< bh_count
; i
++) {
1075 int ext4_walk_page_buffers(handle_t
*handle
,
1076 struct buffer_head
*head
,
1080 int (*fn
)(handle_t
*handle
,
1081 struct buffer_head
*bh
))
1083 struct buffer_head
*bh
;
1084 unsigned block_start
, block_end
;
1085 unsigned blocksize
= head
->b_size
;
1087 struct buffer_head
*next
;
1089 for (bh
= head
, block_start
= 0;
1090 ret
== 0 && (bh
!= head
|| !block_start
);
1091 block_start
= block_end
, bh
= next
) {
1092 next
= bh
->b_this_page
;
1093 block_end
= block_start
+ blocksize
;
1094 if (block_end
<= from
|| block_start
>= to
) {
1095 if (partial
&& !buffer_uptodate(bh
))
1099 err
= (*fn
)(handle
, bh
);
1107 * To preserve ordering, it is essential that the hole instantiation and
1108 * the data write be encapsulated in a single transaction. We cannot
1109 * close off a transaction and start a new one between the ext4_get_block()
1110 * and the commit_write(). So doing the jbd2_journal_start at the start of
1111 * prepare_write() is the right place.
1113 * Also, this function can nest inside ext4_writepage(). In that case, we
1114 * *know* that ext4_writepage() has generated enough buffer credits to do the
1115 * whole page. So we won't block on the journal in that case, which is good,
1116 * because the caller may be PF_MEMALLOC.
1118 * By accident, ext4 can be reentered when a transaction is open via
1119 * quota file writes. If we were to commit the transaction while thus
1120 * reentered, there can be a deadlock - we would be holding a quota
1121 * lock, and the commit would never complete if another thread had a
1122 * transaction open and was blocking on the quota lock - a ranking
1125 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1126 * will _not_ run commit under these circumstances because handle->h_ref
1127 * is elevated. We'll still have enough credits for the tiny quotafile
1130 int do_journal_get_write_access(handle_t
*handle
,
1131 struct buffer_head
*bh
)
1133 int dirty
= buffer_dirty(bh
);
1136 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1139 * __block_write_begin() could have dirtied some buffers. Clean
1140 * the dirty bit as jbd2_journal_get_write_access() could complain
1141 * otherwise about fs integrity issues. Setting of the dirty bit
1142 * by __block_write_begin() isn't a real problem here as we clear
1143 * the bit before releasing a page lock and thus writeback cannot
1144 * ever write the buffer.
1147 clear_buffer_dirty(bh
);
1148 BUFFER_TRACE(bh
, "get write access");
1149 ret
= ext4_journal_get_write_access(handle
, bh
);
1151 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1155 #ifdef CONFIG_FS_ENCRYPTION
1156 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1157 get_block_t
*get_block
)
1159 unsigned from
= pos
& (PAGE_SIZE
- 1);
1160 unsigned to
= from
+ len
;
1161 struct inode
*inode
= page
->mapping
->host
;
1162 unsigned block_start
, block_end
;
1165 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1167 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1168 bool decrypt
= false;
1170 BUG_ON(!PageLocked(page
));
1171 BUG_ON(from
> PAGE_SIZE
);
1172 BUG_ON(to
> PAGE_SIZE
);
1175 if (!page_has_buffers(page
))
1176 create_empty_buffers(page
, blocksize
, 0);
1177 head
= page_buffers(page
);
1178 bbits
= ilog2(blocksize
);
1179 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1181 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1182 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1183 block_end
= block_start
+ blocksize
;
1184 if (block_end
<= from
|| block_start
>= to
) {
1185 if (PageUptodate(page
)) {
1186 if (!buffer_uptodate(bh
))
1187 set_buffer_uptodate(bh
);
1192 clear_buffer_new(bh
);
1193 if (!buffer_mapped(bh
)) {
1194 WARN_ON(bh
->b_size
!= blocksize
);
1195 err
= get_block(inode
, block
, bh
, 1);
1198 if (buffer_new(bh
)) {
1199 if (PageUptodate(page
)) {
1200 clear_buffer_new(bh
);
1201 set_buffer_uptodate(bh
);
1202 mark_buffer_dirty(bh
);
1205 if (block_end
> to
|| block_start
< from
)
1206 zero_user_segments(page
, to
, block_end
,
1211 if (PageUptodate(page
)) {
1212 if (!buffer_uptodate(bh
))
1213 set_buffer_uptodate(bh
);
1216 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1217 !buffer_unwritten(bh
) &&
1218 (block_start
< from
|| block_end
> to
)) {
1219 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1221 decrypt
= IS_ENCRYPTED(inode
) && S_ISREG(inode
->i_mode
);
1225 * If we issued read requests, let them complete.
1227 while (wait_bh
> wait
) {
1228 wait_on_buffer(*--wait_bh
);
1229 if (!buffer_uptodate(*wait_bh
))
1233 page_zero_new_buffers(page
, from
, to
);
1235 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1236 PAGE_SIZE
, 0, page
->index
);
1241 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1242 loff_t pos
, unsigned len
, unsigned flags
,
1243 struct page
**pagep
, void **fsdata
)
1245 struct inode
*inode
= mapping
->host
;
1246 int ret
, needed_blocks
;
1253 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1256 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1258 * Reserve one block more for addition to orphan list in case
1259 * we allocate blocks but write fails for some reason
1261 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1262 index
= pos
>> PAGE_SHIFT
;
1263 from
= pos
& (PAGE_SIZE
- 1);
1266 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1267 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1276 * grab_cache_page_write_begin() can take a long time if the
1277 * system is thrashing due to memory pressure, or if the page
1278 * is being written back. So grab it first before we start
1279 * the transaction handle. This also allows us to allocate
1280 * the page (if needed) without using GFP_NOFS.
1283 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1289 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1290 if (IS_ERR(handle
)) {
1292 return PTR_ERR(handle
);
1296 if (page
->mapping
!= mapping
) {
1297 /* The page got truncated from under us */
1300 ext4_journal_stop(handle
);
1303 /* In case writeback began while the page was unlocked */
1304 wait_for_stable_page(page
);
1306 #ifdef CONFIG_FS_ENCRYPTION
1307 if (ext4_should_dioread_nolock(inode
))
1308 ret
= ext4_block_write_begin(page
, pos
, len
,
1309 ext4_get_block_unwritten
);
1311 ret
= ext4_block_write_begin(page
, pos
, len
,
1314 if (ext4_should_dioread_nolock(inode
))
1315 ret
= __block_write_begin(page
, pos
, len
,
1316 ext4_get_block_unwritten
);
1318 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1320 if (!ret
&& ext4_should_journal_data(inode
)) {
1321 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1323 do_journal_get_write_access
);
1329 * __block_write_begin may have instantiated a few blocks
1330 * outside i_size. Trim these off again. Don't need
1331 * i_size_read because we hold i_mutex.
1333 * Add inode to orphan list in case we crash before
1336 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1337 ext4_orphan_add(handle
, inode
);
1339 ext4_journal_stop(handle
);
1340 if (pos
+ len
> inode
->i_size
) {
1341 ext4_truncate_failed_write(inode
);
1343 * If truncate failed early the inode might
1344 * still be on the orphan list; we need to
1345 * make sure the inode is removed from the
1346 * orphan list in that case.
1349 ext4_orphan_del(NULL
, inode
);
1352 if (ret
== -ENOSPC
&&
1353 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1362 /* For write_end() in data=journal mode */
1363 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1366 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1368 set_buffer_uptodate(bh
);
1369 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1370 clear_buffer_meta(bh
);
1371 clear_buffer_prio(bh
);
1376 * We need to pick up the new inode size which generic_commit_write gave us
1377 * `file' can be NULL - eg, when called from page_symlink().
1379 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1380 * buffers are managed internally.
1382 static int ext4_write_end(struct file
*file
,
1383 struct address_space
*mapping
,
1384 loff_t pos
, unsigned len
, unsigned copied
,
1385 struct page
*page
, void *fsdata
)
1387 handle_t
*handle
= ext4_journal_current_handle();
1388 struct inode
*inode
= mapping
->host
;
1389 loff_t old_size
= inode
->i_size
;
1391 int i_size_changed
= 0;
1392 int inline_data
= ext4_has_inline_data(inode
);
1394 trace_ext4_write_end(inode
, pos
, len
, copied
);
1396 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1405 copied
= block_write_end(file
, mapping
, pos
,
1406 len
, copied
, page
, fsdata
);
1408 * it's important to update i_size while still holding page lock:
1409 * page writeout could otherwise come in and zero beyond i_size.
1411 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1416 pagecache_isize_extended(inode
, old_size
, pos
);
1418 * Don't mark the inode dirty under page lock. First, it unnecessarily
1419 * makes the holding time of page lock longer. Second, it forces lock
1420 * ordering of page lock and transaction start for journaling
1423 if (i_size_changed
|| inline_data
)
1424 ext4_mark_inode_dirty(handle
, inode
);
1426 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1427 /* if we have allocated more blocks and copied
1428 * less. We will have blocks allocated outside
1429 * inode->i_size. So truncate them
1431 ext4_orphan_add(handle
, inode
);
1433 ret2
= ext4_journal_stop(handle
);
1437 if (pos
+ len
> inode
->i_size
) {
1438 ext4_truncate_failed_write(inode
);
1440 * If truncate failed early the inode might still be
1441 * on the orphan list; we need to make sure the inode
1442 * is removed from the orphan list in that case.
1445 ext4_orphan_del(NULL
, inode
);
1448 return ret
? ret
: copied
;
1452 * This is a private version of page_zero_new_buffers() which doesn't
1453 * set the buffer to be dirty, since in data=journalled mode we need
1454 * to call ext4_handle_dirty_metadata() instead.
1456 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1458 unsigned from
, unsigned to
)
1460 unsigned int block_start
= 0, block_end
;
1461 struct buffer_head
*head
, *bh
;
1463 bh
= head
= page_buffers(page
);
1465 block_end
= block_start
+ bh
->b_size
;
1466 if (buffer_new(bh
)) {
1467 if (block_end
> from
&& block_start
< to
) {
1468 if (!PageUptodate(page
)) {
1469 unsigned start
, size
;
1471 start
= max(from
, block_start
);
1472 size
= min(to
, block_end
) - start
;
1474 zero_user(page
, start
, size
);
1475 write_end_fn(handle
, bh
);
1477 clear_buffer_new(bh
);
1480 block_start
= block_end
;
1481 bh
= bh
->b_this_page
;
1482 } while (bh
!= head
);
1485 static int ext4_journalled_write_end(struct file
*file
,
1486 struct address_space
*mapping
,
1487 loff_t pos
, unsigned len
, unsigned copied
,
1488 struct page
*page
, void *fsdata
)
1490 handle_t
*handle
= ext4_journal_current_handle();
1491 struct inode
*inode
= mapping
->host
;
1492 loff_t old_size
= inode
->i_size
;
1496 int size_changed
= 0;
1497 int inline_data
= ext4_has_inline_data(inode
);
1499 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1500 from
= pos
& (PAGE_SIZE
- 1);
1503 BUG_ON(!ext4_handle_valid(handle
));
1506 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1514 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1516 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1518 if (unlikely(copied
< len
))
1519 ext4_journalled_zero_new_buffers(handle
, page
,
1521 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1522 from
+ copied
, &partial
,
1525 SetPageUptodate(page
);
1527 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1528 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1529 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1534 pagecache_isize_extended(inode
, old_size
, pos
);
1536 if (size_changed
|| inline_data
) {
1537 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1542 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1543 /* if we have allocated more blocks and copied
1544 * less. We will have blocks allocated outside
1545 * inode->i_size. So truncate them
1547 ext4_orphan_add(handle
, inode
);
1550 ret2
= ext4_journal_stop(handle
);
1553 if (pos
+ len
> inode
->i_size
) {
1554 ext4_truncate_failed_write(inode
);
1556 * If truncate failed early the inode might still be
1557 * on the orphan list; we need to make sure the inode
1558 * is removed from the orphan list in that case.
1561 ext4_orphan_del(NULL
, inode
);
1564 return ret
? ret
: copied
;
1568 * Reserve space for a single cluster
1570 static int ext4_da_reserve_space(struct inode
*inode
)
1572 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1573 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1577 * We will charge metadata quota at writeout time; this saves
1578 * us from metadata over-estimation, though we may go over by
1579 * a small amount in the end. Here we just reserve for data.
1581 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1585 spin_lock(&ei
->i_block_reservation_lock
);
1586 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1587 spin_unlock(&ei
->i_block_reservation_lock
);
1588 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1591 ei
->i_reserved_data_blocks
++;
1592 trace_ext4_da_reserve_space(inode
);
1593 spin_unlock(&ei
->i_block_reservation_lock
);
1595 return 0; /* success */
1598 void ext4_da_release_space(struct inode
*inode
, int to_free
)
1600 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1601 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1604 return; /* Nothing to release, exit */
1606 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1608 trace_ext4_da_release_space(inode
, to_free
);
1609 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1611 * if there aren't enough reserved blocks, then the
1612 * counter is messed up somewhere. Since this
1613 * function is called from invalidate page, it's
1614 * harmless to return without any action.
1616 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1617 "ino %lu, to_free %d with only %d reserved "
1618 "data blocks", inode
->i_ino
, to_free
,
1619 ei
->i_reserved_data_blocks
);
1621 to_free
= ei
->i_reserved_data_blocks
;
1623 ei
->i_reserved_data_blocks
-= to_free
;
1625 /* update fs dirty data blocks counter */
1626 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1628 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1630 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1633 static void ext4_da_page_release_reservation(struct page
*page
,
1634 unsigned int offset
,
1635 unsigned int length
)
1637 int contiguous_blks
= 0;
1638 struct buffer_head
*head
, *bh
;
1639 unsigned int curr_off
= 0;
1640 struct inode
*inode
= page
->mapping
->host
;
1641 unsigned int stop
= offset
+ length
;
1644 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1646 head
= page_buffers(page
);
1649 unsigned int next_off
= curr_off
+ bh
->b_size
;
1651 if (next_off
> stop
)
1654 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1656 clear_buffer_delay(bh
);
1657 } else if (contiguous_blks
) {
1658 lblk
= page
->index
<<
1659 (PAGE_SHIFT
- inode
->i_blkbits
);
1660 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1662 ext4_es_remove_blks(inode
, lblk
, contiguous_blks
);
1663 contiguous_blks
= 0;
1665 curr_off
= next_off
;
1666 } while ((bh
= bh
->b_this_page
) != head
);
1668 if (contiguous_blks
) {
1669 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1670 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1671 ext4_es_remove_blks(inode
, lblk
, contiguous_blks
);
1677 * Delayed allocation stuff
1680 struct mpage_da_data
{
1681 struct inode
*inode
;
1682 struct writeback_control
*wbc
;
1684 pgoff_t first_page
; /* The first page to write */
1685 pgoff_t next_page
; /* Current page to examine */
1686 pgoff_t last_page
; /* Last page to examine */
1688 * Extent to map - this can be after first_page because that can be
1689 * fully mapped. We somewhat abuse m_flags to store whether the extent
1690 * is delalloc or unwritten.
1692 struct ext4_map_blocks map
;
1693 struct ext4_io_submit io_submit
; /* IO submission data */
1694 unsigned int do_map
:1;
1697 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1702 struct pagevec pvec
;
1703 struct inode
*inode
= mpd
->inode
;
1704 struct address_space
*mapping
= inode
->i_mapping
;
1706 /* This is necessary when next_page == 0. */
1707 if (mpd
->first_page
>= mpd
->next_page
)
1710 index
= mpd
->first_page
;
1711 end
= mpd
->next_page
- 1;
1713 ext4_lblk_t start
, last
;
1714 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1715 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1716 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1719 pagevec_init(&pvec
);
1720 while (index
<= end
) {
1721 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1724 for (i
= 0; i
< nr_pages
; i
++) {
1725 struct page
*page
= pvec
.pages
[i
];
1727 BUG_ON(!PageLocked(page
));
1728 BUG_ON(PageWriteback(page
));
1730 if (page_mapped(page
))
1731 clear_page_dirty_for_io(page
);
1732 block_invalidatepage(page
, 0, PAGE_SIZE
);
1733 ClearPageUptodate(page
);
1737 pagevec_release(&pvec
);
1741 static void ext4_print_free_blocks(struct inode
*inode
)
1743 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1744 struct super_block
*sb
= inode
->i_sb
;
1745 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1747 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1748 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1749 ext4_count_free_clusters(sb
)));
1750 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1751 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1752 (long long) EXT4_C2B(EXT4_SB(sb
),
1753 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1754 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1755 (long long) EXT4_C2B(EXT4_SB(sb
),
1756 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1757 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1758 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1759 ei
->i_reserved_data_blocks
);
1763 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1765 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1769 * ext4_insert_delayed_block - adds a delayed block to the extents status
1770 * tree, incrementing the reserved cluster/block
1771 * count or making a pending reservation
1774 * @inode - file containing the newly added block
1775 * @lblk - logical block to be added
1777 * Returns 0 on success, negative error code on failure.
1779 static int ext4_insert_delayed_block(struct inode
*inode
, ext4_lblk_t lblk
)
1781 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1783 bool allocated
= false;
1786 * If the cluster containing lblk is shared with a delayed,
1787 * written, or unwritten extent in a bigalloc file system, it's
1788 * already been accounted for and does not need to be reserved.
1789 * A pending reservation must be made for the cluster if it's
1790 * shared with a written or unwritten extent and doesn't already
1791 * have one. Written and unwritten extents can be purged from the
1792 * extents status tree if the system is under memory pressure, so
1793 * it's necessary to examine the extent tree if a search of the
1794 * extents status tree doesn't get a match.
1796 if (sbi
->s_cluster_ratio
== 1) {
1797 ret
= ext4_da_reserve_space(inode
);
1798 if (ret
!= 0) /* ENOSPC */
1800 } else { /* bigalloc */
1801 if (!ext4_es_scan_clu(inode
, &ext4_es_is_delonly
, lblk
)) {
1802 if (!ext4_es_scan_clu(inode
,
1803 &ext4_es_is_mapped
, lblk
)) {
1804 ret
= ext4_clu_mapped(inode
,
1805 EXT4_B2C(sbi
, lblk
));
1809 ret
= ext4_da_reserve_space(inode
);
1810 if (ret
!= 0) /* ENOSPC */
1821 ret
= ext4_es_insert_delayed_block(inode
, lblk
, allocated
);
1828 * This function is grabs code from the very beginning of
1829 * ext4_map_blocks, but assumes that the caller is from delayed write
1830 * time. This function looks up the requested blocks and sets the
1831 * buffer delay bit under the protection of i_data_sem.
1833 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1834 struct ext4_map_blocks
*map
,
1835 struct buffer_head
*bh
)
1837 struct extent_status es
;
1839 sector_t invalid_block
= ~((sector_t
) 0xffff);
1840 #ifdef ES_AGGRESSIVE_TEST
1841 struct ext4_map_blocks orig_map
;
1843 memcpy(&orig_map
, map
, sizeof(*map
));
1846 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1850 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1851 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1852 (unsigned long) map
->m_lblk
);
1854 /* Lookup extent status tree firstly */
1855 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1856 if (ext4_es_is_hole(&es
)) {
1858 down_read(&EXT4_I(inode
)->i_data_sem
);
1863 * Delayed extent could be allocated by fallocate.
1864 * So we need to check it.
1866 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1867 map_bh(bh
, inode
->i_sb
, invalid_block
);
1869 set_buffer_delay(bh
);
1873 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1874 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1875 if (retval
> map
->m_len
)
1876 retval
= map
->m_len
;
1877 map
->m_len
= retval
;
1878 if (ext4_es_is_written(&es
))
1879 map
->m_flags
|= EXT4_MAP_MAPPED
;
1880 else if (ext4_es_is_unwritten(&es
))
1881 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1885 #ifdef ES_AGGRESSIVE_TEST
1886 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1892 * Try to see if we can get the block without requesting a new
1893 * file system block.
1895 down_read(&EXT4_I(inode
)->i_data_sem
);
1896 if (ext4_has_inline_data(inode
))
1898 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1899 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1901 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1908 * XXX: __block_prepare_write() unmaps passed block,
1912 ret
= ext4_insert_delayed_block(inode
, map
->m_lblk
);
1918 map_bh(bh
, inode
->i_sb
, invalid_block
);
1920 set_buffer_delay(bh
);
1921 } else if (retval
> 0) {
1923 unsigned int status
;
1925 if (unlikely(retval
!= map
->m_len
)) {
1926 ext4_warning(inode
->i_sb
,
1927 "ES len assertion failed for inode "
1928 "%lu: retval %d != map->m_len %d",
1929 inode
->i_ino
, retval
, map
->m_len
);
1933 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1934 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1935 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1936 map
->m_pblk
, status
);
1942 up_read((&EXT4_I(inode
)->i_data_sem
));
1948 * This is a special get_block_t callback which is used by
1949 * ext4_da_write_begin(). It will either return mapped block or
1950 * reserve space for a single block.
1952 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1953 * We also have b_blocknr = -1 and b_bdev initialized properly
1955 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1956 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1957 * initialized properly.
1959 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1960 struct buffer_head
*bh
, int create
)
1962 struct ext4_map_blocks map
;
1965 BUG_ON(create
== 0);
1966 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1968 map
.m_lblk
= iblock
;
1972 * first, we need to know whether the block is allocated already
1973 * preallocated blocks are unmapped but should treated
1974 * the same as allocated blocks.
1976 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1980 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1981 ext4_update_bh_state(bh
, map
.m_flags
);
1983 if (buffer_unwritten(bh
)) {
1984 /* A delayed write to unwritten bh should be marked
1985 * new and mapped. Mapped ensures that we don't do
1986 * get_block multiple times when we write to the same
1987 * offset and new ensures that we do proper zero out
1988 * for partial write.
1991 set_buffer_mapped(bh
);
1996 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2002 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2008 static int __ext4_journalled_writepage(struct page
*page
,
2011 struct address_space
*mapping
= page
->mapping
;
2012 struct inode
*inode
= mapping
->host
;
2013 struct buffer_head
*page_bufs
= NULL
;
2014 handle_t
*handle
= NULL
;
2015 int ret
= 0, err
= 0;
2016 int inline_data
= ext4_has_inline_data(inode
);
2017 struct buffer_head
*inode_bh
= NULL
;
2019 ClearPageChecked(page
);
2022 BUG_ON(page
->index
!= 0);
2023 BUG_ON(len
> ext4_get_max_inline_size(inode
));
2024 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
2025 if (inode_bh
== NULL
)
2028 page_bufs
= page_buffers(page
);
2033 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2037 * We need to release the page lock before we start the
2038 * journal, so grab a reference so the page won't disappear
2039 * out from under us.
2044 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2045 ext4_writepage_trans_blocks(inode
));
2046 if (IS_ERR(handle
)) {
2047 ret
= PTR_ERR(handle
);
2049 goto out_no_pagelock
;
2051 BUG_ON(!ext4_handle_valid(handle
));
2055 if (page
->mapping
!= mapping
) {
2056 /* The page got truncated from under us */
2057 ext4_journal_stop(handle
);
2063 ret
= ext4_mark_inode_dirty(handle
, inode
);
2065 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2066 do_journal_get_write_access
);
2068 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2073 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2074 err
= ext4_journal_stop(handle
);
2078 if (!ext4_has_inline_data(inode
))
2079 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
2081 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2090 * Note that we don't need to start a transaction unless we're journaling data
2091 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2092 * need to file the inode to the transaction's list in ordered mode because if
2093 * we are writing back data added by write(), the inode is already there and if
2094 * we are writing back data modified via mmap(), no one guarantees in which
2095 * transaction the data will hit the disk. In case we are journaling data, we
2096 * cannot start transaction directly because transaction start ranks above page
2097 * lock so we have to do some magic.
2099 * This function can get called via...
2100 * - ext4_writepages after taking page lock (have journal handle)
2101 * - journal_submit_inode_data_buffers (no journal handle)
2102 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2103 * - grab_page_cache when doing write_begin (have journal handle)
2105 * We don't do any block allocation in this function. If we have page with
2106 * multiple blocks we need to write those buffer_heads that are mapped. This
2107 * is important for mmaped based write. So if we do with blocksize 1K
2108 * truncate(f, 1024);
2109 * a = mmap(f, 0, 4096);
2111 * truncate(f, 4096);
2112 * we have in the page first buffer_head mapped via page_mkwrite call back
2113 * but other buffer_heads would be unmapped but dirty (dirty done via the
2114 * do_wp_page). So writepage should write the first block. If we modify
2115 * the mmap area beyond 1024 we will again get a page_fault and the
2116 * page_mkwrite callback will do the block allocation and mark the
2117 * buffer_heads mapped.
2119 * We redirty the page if we have any buffer_heads that is either delay or
2120 * unwritten in the page.
2122 * We can get recursively called as show below.
2124 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2127 * But since we don't do any block allocation we should not deadlock.
2128 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2130 static int ext4_writepage(struct page
*page
,
2131 struct writeback_control
*wbc
)
2136 struct buffer_head
*page_bufs
= NULL
;
2137 struct inode
*inode
= page
->mapping
->host
;
2138 struct ext4_io_submit io_submit
;
2139 bool keep_towrite
= false;
2141 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
2142 ext4_invalidatepage(page
, 0, PAGE_SIZE
);
2147 trace_ext4_writepage(page
);
2148 size
= i_size_read(inode
);
2149 if (page
->index
== size
>> PAGE_SHIFT
)
2150 len
= size
& ~PAGE_MASK
;
2154 page_bufs
= page_buffers(page
);
2156 * We cannot do block allocation or other extent handling in this
2157 * function. If there are buffers needing that, we have to redirty
2158 * the page. But we may reach here when we do a journal commit via
2159 * journal_submit_inode_data_buffers() and in that case we must write
2160 * allocated buffers to achieve data=ordered mode guarantees.
2162 * Also, if there is only one buffer per page (the fs block
2163 * size == the page size), if one buffer needs block
2164 * allocation or needs to modify the extent tree to clear the
2165 * unwritten flag, we know that the page can't be written at
2166 * all, so we might as well refuse the write immediately.
2167 * Unfortunately if the block size != page size, we can't as
2168 * easily detect this case using ext4_walk_page_buffers(), but
2169 * for the extremely common case, this is an optimization that
2170 * skips a useless round trip through ext4_bio_write_page().
2172 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2173 ext4_bh_delay_or_unwritten
)) {
2174 redirty_page_for_writepage(wbc
, page
);
2175 if ((current
->flags
& PF_MEMALLOC
) ||
2176 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2178 * For memory cleaning there's no point in writing only
2179 * some buffers. So just bail out. Warn if we came here
2180 * from direct reclaim.
2182 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2187 keep_towrite
= true;
2190 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2192 * It's mmapped pagecache. Add buffers and journal it. There
2193 * doesn't seem much point in redirtying the page here.
2195 return __ext4_journalled_writepage(page
, len
);
2197 ext4_io_submit_init(&io_submit
, wbc
);
2198 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2199 if (!io_submit
.io_end
) {
2200 redirty_page_for_writepage(wbc
, page
);
2204 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2205 ext4_io_submit(&io_submit
);
2206 /* Drop io_end reference we got from init */
2207 ext4_put_io_end_defer(io_submit
.io_end
);
2211 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2217 BUG_ON(page
->index
!= mpd
->first_page
);
2218 clear_page_dirty_for_io(page
);
2220 * We have to be very careful here! Nothing protects writeback path
2221 * against i_size changes and the page can be writeably mapped into
2222 * page tables. So an application can be growing i_size and writing
2223 * data through mmap while writeback runs. clear_page_dirty_for_io()
2224 * write-protects our page in page tables and the page cannot get
2225 * written to again until we release page lock. So only after
2226 * clear_page_dirty_for_io() we are safe to sample i_size for
2227 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2228 * on the barrier provided by TestClearPageDirty in
2229 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2230 * after page tables are updated.
2232 size
= i_size_read(mpd
->inode
);
2233 if (page
->index
== size
>> PAGE_SHIFT
)
2234 len
= size
& ~PAGE_MASK
;
2237 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2239 mpd
->wbc
->nr_to_write
--;
2245 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2248 * mballoc gives us at most this number of blocks...
2249 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2250 * The rest of mballoc seems to handle chunks up to full group size.
2252 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2255 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2257 * @mpd - extent of blocks
2258 * @lblk - logical number of the block in the file
2259 * @bh - buffer head we want to add to the extent
2261 * The function is used to collect contig. blocks in the same state. If the
2262 * buffer doesn't require mapping for writeback and we haven't started the
2263 * extent of buffers to map yet, the function returns 'true' immediately - the
2264 * caller can write the buffer right away. Otherwise the function returns true
2265 * if the block has been added to the extent, false if the block couldn't be
2268 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2269 struct buffer_head
*bh
)
2271 struct ext4_map_blocks
*map
= &mpd
->map
;
2273 /* Buffer that doesn't need mapping for writeback? */
2274 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2275 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2276 /* So far no extent to map => we write the buffer right away */
2277 if (map
->m_len
== 0)
2282 /* First block in the extent? */
2283 if (map
->m_len
== 0) {
2284 /* We cannot map unless handle is started... */
2289 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2293 /* Don't go larger than mballoc is willing to allocate */
2294 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2297 /* Can we merge the block to our big extent? */
2298 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2299 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2307 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2309 * @mpd - extent of blocks for mapping
2310 * @head - the first buffer in the page
2311 * @bh - buffer we should start processing from
2312 * @lblk - logical number of the block in the file corresponding to @bh
2314 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2315 * the page for IO if all buffers in this page were mapped and there's no
2316 * accumulated extent of buffers to map or add buffers in the page to the
2317 * extent of buffers to map. The function returns 1 if the caller can continue
2318 * by processing the next page, 0 if it should stop adding buffers to the
2319 * extent to map because we cannot extend it anymore. It can also return value
2320 * < 0 in case of error during IO submission.
2322 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2323 struct buffer_head
*head
,
2324 struct buffer_head
*bh
,
2327 struct inode
*inode
= mpd
->inode
;
2329 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2330 >> inode
->i_blkbits
;
2333 BUG_ON(buffer_locked(bh
));
2335 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2336 /* Found extent to map? */
2339 /* Buffer needs mapping and handle is not started? */
2342 /* Everything mapped so far and we hit EOF */
2345 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2346 /* So far everything mapped? Submit the page for IO. */
2347 if (mpd
->map
.m_len
== 0) {
2348 err
= mpage_submit_page(mpd
, head
->b_page
);
2352 return lblk
< blocks
;
2356 * mpage_map_buffers - update buffers corresponding to changed extent and
2357 * submit fully mapped pages for IO
2359 * @mpd - description of extent to map, on return next extent to map
2361 * Scan buffers corresponding to changed extent (we expect corresponding pages
2362 * to be already locked) and update buffer state according to new extent state.
2363 * We map delalloc buffers to their physical location, clear unwritten bits,
2364 * and mark buffers as uninit when we perform writes to unwritten extents
2365 * and do extent conversion after IO is finished. If the last page is not fully
2366 * mapped, we update @map to the next extent in the last page that needs
2367 * mapping. Otherwise we submit the page for IO.
2369 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2371 struct pagevec pvec
;
2373 struct inode
*inode
= mpd
->inode
;
2374 struct buffer_head
*head
, *bh
;
2375 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2381 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2382 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2383 lblk
= start
<< bpp_bits
;
2384 pblock
= mpd
->map
.m_pblk
;
2386 pagevec_init(&pvec
);
2387 while (start
<= end
) {
2388 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2392 for (i
= 0; i
< nr_pages
; i
++) {
2393 struct page
*page
= pvec
.pages
[i
];
2395 bh
= head
= page_buffers(page
);
2397 if (lblk
< mpd
->map
.m_lblk
)
2399 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2401 * Buffer after end of mapped extent.
2402 * Find next buffer in the page to map.
2405 mpd
->map
.m_flags
= 0;
2407 * FIXME: If dioread_nolock supports
2408 * blocksize < pagesize, we need to make
2409 * sure we add size mapped so far to
2410 * io_end->size as the following call
2411 * can submit the page for IO.
2413 err
= mpage_process_page_bufs(mpd
, head
,
2415 pagevec_release(&pvec
);
2420 if (buffer_delay(bh
)) {
2421 clear_buffer_delay(bh
);
2422 bh
->b_blocknr
= pblock
++;
2424 clear_buffer_unwritten(bh
);
2425 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2428 * FIXME: This is going to break if dioread_nolock
2429 * supports blocksize < pagesize as we will try to
2430 * convert potentially unmapped parts of inode.
2432 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2433 /* Page fully mapped - let IO run! */
2434 err
= mpage_submit_page(mpd
, page
);
2436 pagevec_release(&pvec
);
2440 pagevec_release(&pvec
);
2442 /* Extent fully mapped and matches with page boundary. We are done. */
2444 mpd
->map
.m_flags
= 0;
2448 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2450 struct inode
*inode
= mpd
->inode
;
2451 struct ext4_map_blocks
*map
= &mpd
->map
;
2452 int get_blocks_flags
;
2453 int err
, dioread_nolock
;
2455 trace_ext4_da_write_pages_extent(inode
, map
);
2457 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2458 * to convert an unwritten extent to be initialized (in the case
2459 * where we have written into one or more preallocated blocks). It is
2460 * possible that we're going to need more metadata blocks than
2461 * previously reserved. However we must not fail because we're in
2462 * writeback and there is nothing we can do about it so it might result
2463 * in data loss. So use reserved blocks to allocate metadata if
2466 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2467 * the blocks in question are delalloc blocks. This indicates
2468 * that the blocks and quotas has already been checked when
2469 * the data was copied into the page cache.
2471 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2472 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2473 EXT4_GET_BLOCKS_IO_SUBMIT
;
2474 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2476 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2477 if (map
->m_flags
& (1 << BH_Delay
))
2478 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2480 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2483 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2484 if (!mpd
->io_submit
.io_end
->handle
&&
2485 ext4_handle_valid(handle
)) {
2486 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2487 handle
->h_rsv_handle
= NULL
;
2489 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2492 BUG_ON(map
->m_len
== 0);
2497 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2498 * mpd->len and submit pages underlying it for IO
2500 * @handle - handle for journal operations
2501 * @mpd - extent to map
2502 * @give_up_on_write - we set this to true iff there is a fatal error and there
2503 * is no hope of writing the data. The caller should discard
2504 * dirty pages to avoid infinite loops.
2506 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2507 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2508 * them to initialized or split the described range from larger unwritten
2509 * extent. Note that we need not map all the described range since allocation
2510 * can return less blocks or the range is covered by more unwritten extents. We
2511 * cannot map more because we are limited by reserved transaction credits. On
2512 * the other hand we always make sure that the last touched page is fully
2513 * mapped so that it can be written out (and thus forward progress is
2514 * guaranteed). After mapping we submit all mapped pages for IO.
2516 static int mpage_map_and_submit_extent(handle_t
*handle
,
2517 struct mpage_da_data
*mpd
,
2518 bool *give_up_on_write
)
2520 struct inode
*inode
= mpd
->inode
;
2521 struct ext4_map_blocks
*map
= &mpd
->map
;
2526 mpd
->io_submit
.io_end
->offset
=
2527 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2529 err
= mpage_map_one_extent(handle
, mpd
);
2531 struct super_block
*sb
= inode
->i_sb
;
2533 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2534 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2535 goto invalidate_dirty_pages
;
2537 * Let the uper layers retry transient errors.
2538 * In the case of ENOSPC, if ext4_count_free_blocks()
2539 * is non-zero, a commit should free up blocks.
2541 if ((err
== -ENOMEM
) ||
2542 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2544 goto update_disksize
;
2547 ext4_msg(sb
, KERN_CRIT
,
2548 "Delayed block allocation failed for "
2549 "inode %lu at logical offset %llu with"
2550 " max blocks %u with error %d",
2552 (unsigned long long)map
->m_lblk
,
2553 (unsigned)map
->m_len
, -err
);
2554 ext4_msg(sb
, KERN_CRIT
,
2555 "This should not happen!! Data will "
2558 ext4_print_free_blocks(inode
);
2559 invalidate_dirty_pages
:
2560 *give_up_on_write
= true;
2565 * Update buffer state, submit mapped pages, and get us new
2568 err
= mpage_map_and_submit_buffers(mpd
);
2570 goto update_disksize
;
2571 } while (map
->m_len
);
2575 * Update on-disk size after IO is submitted. Races with
2576 * truncate are avoided by checking i_size under i_data_sem.
2578 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2579 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2583 down_write(&EXT4_I(inode
)->i_data_sem
);
2584 i_size
= i_size_read(inode
);
2585 if (disksize
> i_size
)
2587 if (disksize
> EXT4_I(inode
)->i_disksize
)
2588 EXT4_I(inode
)->i_disksize
= disksize
;
2589 up_write(&EXT4_I(inode
)->i_data_sem
);
2590 err2
= ext4_mark_inode_dirty(handle
, inode
);
2592 ext4_error(inode
->i_sb
,
2593 "Failed to mark inode %lu dirty",
2602 * Calculate the total number of credits to reserve for one writepages
2603 * iteration. This is called from ext4_writepages(). We map an extent of
2604 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2605 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2606 * bpp - 1 blocks in bpp different extents.
2608 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2610 int bpp
= ext4_journal_blocks_per_page(inode
);
2612 return ext4_meta_trans_blocks(inode
,
2613 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2617 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2618 * and underlying extent to map
2620 * @mpd - where to look for pages
2622 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2623 * IO immediately. When we find a page which isn't mapped we start accumulating
2624 * extent of buffers underlying these pages that needs mapping (formed by
2625 * either delayed or unwritten buffers). We also lock the pages containing
2626 * these buffers. The extent found is returned in @mpd structure (starting at
2627 * mpd->lblk with length mpd->len blocks).
2629 * Note that this function can attach bios to one io_end structure which are
2630 * neither logically nor physically contiguous. Although it may seem as an
2631 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2632 * case as we need to track IO to all buffers underlying a page in one io_end.
2634 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2636 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2637 struct pagevec pvec
;
2638 unsigned int nr_pages
;
2639 long left
= mpd
->wbc
->nr_to_write
;
2640 pgoff_t index
= mpd
->first_page
;
2641 pgoff_t end
= mpd
->last_page
;
2644 int blkbits
= mpd
->inode
->i_blkbits
;
2646 struct buffer_head
*head
;
2648 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2649 tag
= PAGECACHE_TAG_TOWRITE
;
2651 tag
= PAGECACHE_TAG_DIRTY
;
2653 pagevec_init(&pvec
);
2655 mpd
->next_page
= index
;
2656 while (index
<= end
) {
2657 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
, end
,
2662 for (i
= 0; i
< nr_pages
; i
++) {
2663 struct page
*page
= pvec
.pages
[i
];
2666 * Accumulated enough dirty pages? This doesn't apply
2667 * to WB_SYNC_ALL mode. For integrity sync we have to
2668 * keep going because someone may be concurrently
2669 * dirtying pages, and we might have synced a lot of
2670 * newly appeared dirty pages, but have not synced all
2671 * of the old dirty pages.
2673 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2676 /* If we can't merge this page, we are done. */
2677 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2682 * If the page is no longer dirty, or its mapping no
2683 * longer corresponds to inode we are writing (which
2684 * means it has been truncated or invalidated), or the
2685 * page is already under writeback and we are not doing
2686 * a data integrity writeback, skip the page
2688 if (!PageDirty(page
) ||
2689 (PageWriteback(page
) &&
2690 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2691 unlikely(page
->mapping
!= mapping
)) {
2696 wait_on_page_writeback(page
);
2697 BUG_ON(PageWriteback(page
));
2699 if (mpd
->map
.m_len
== 0)
2700 mpd
->first_page
= page
->index
;
2701 mpd
->next_page
= page
->index
+ 1;
2702 /* Add all dirty buffers to mpd */
2703 lblk
= ((ext4_lblk_t
)page
->index
) <<
2704 (PAGE_SHIFT
- blkbits
);
2705 head
= page_buffers(page
);
2706 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2712 pagevec_release(&pvec
);
2717 pagevec_release(&pvec
);
2721 static int ext4_writepages(struct address_space
*mapping
,
2722 struct writeback_control
*wbc
)
2724 pgoff_t writeback_index
= 0;
2725 long nr_to_write
= wbc
->nr_to_write
;
2726 int range_whole
= 0;
2728 handle_t
*handle
= NULL
;
2729 struct mpage_da_data mpd
;
2730 struct inode
*inode
= mapping
->host
;
2731 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2732 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2734 struct blk_plug plug
;
2735 bool give_up_on_write
= false;
2737 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2740 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2741 trace_ext4_writepages(inode
, wbc
);
2744 * No pages to write? This is mainly a kludge to avoid starting
2745 * a transaction for special inodes like journal inode on last iput()
2746 * because that could violate lock ordering on umount
2748 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2749 goto out_writepages
;
2751 if (ext4_should_journal_data(inode
)) {
2752 ret
= generic_writepages(mapping
, wbc
);
2753 goto out_writepages
;
2757 * If the filesystem has aborted, it is read-only, so return
2758 * right away instead of dumping stack traces later on that
2759 * will obscure the real source of the problem. We test
2760 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2761 * the latter could be true if the filesystem is mounted
2762 * read-only, and in that case, ext4_writepages should
2763 * *never* be called, so if that ever happens, we would want
2766 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2767 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2769 goto out_writepages
;
2772 if (ext4_should_dioread_nolock(inode
)) {
2774 * We may need to convert up to one extent per block in
2775 * the page and we may dirty the inode.
2777 rsv_blocks
= 1 + ext4_chunk_trans_blocks(inode
,
2778 PAGE_SIZE
>> inode
->i_blkbits
);
2782 * If we have inline data and arrive here, it means that
2783 * we will soon create the block for the 1st page, so
2784 * we'd better clear the inline data here.
2786 if (ext4_has_inline_data(inode
)) {
2787 /* Just inode will be modified... */
2788 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2789 if (IS_ERR(handle
)) {
2790 ret
= PTR_ERR(handle
);
2791 goto out_writepages
;
2793 BUG_ON(ext4_test_inode_state(inode
,
2794 EXT4_STATE_MAY_INLINE_DATA
));
2795 ext4_destroy_inline_data(handle
, inode
);
2796 ext4_journal_stop(handle
);
2799 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2802 if (wbc
->range_cyclic
) {
2803 writeback_index
= mapping
->writeback_index
;
2804 if (writeback_index
)
2806 mpd
.first_page
= writeback_index
;
2809 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2810 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2815 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2817 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2818 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2820 blk_start_plug(&plug
);
2823 * First writeback pages that don't need mapping - we can avoid
2824 * starting a transaction unnecessarily and also avoid being blocked
2825 * in the block layer on device congestion while having transaction
2829 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2830 if (!mpd
.io_submit
.io_end
) {
2834 ret
= mpage_prepare_extent_to_map(&mpd
);
2835 /* Unlock pages we didn't use */
2836 mpage_release_unused_pages(&mpd
, false);
2837 /* Submit prepared bio */
2838 ext4_io_submit(&mpd
.io_submit
);
2839 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2840 mpd
.io_submit
.io_end
= NULL
;
2844 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2845 /* For each extent of pages we use new io_end */
2846 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2847 if (!mpd
.io_submit
.io_end
) {
2853 * We have two constraints: We find one extent to map and we
2854 * must always write out whole page (makes a difference when
2855 * blocksize < pagesize) so that we don't block on IO when we
2856 * try to write out the rest of the page. Journalled mode is
2857 * not supported by delalloc.
2859 BUG_ON(ext4_should_journal_data(inode
));
2860 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2862 /* start a new transaction */
2863 handle
= ext4_journal_start_with_reserve(inode
,
2864 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2865 if (IS_ERR(handle
)) {
2866 ret
= PTR_ERR(handle
);
2867 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2868 "%ld pages, ino %lu; err %d", __func__
,
2869 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2870 /* Release allocated io_end */
2871 ext4_put_io_end(mpd
.io_submit
.io_end
);
2872 mpd
.io_submit
.io_end
= NULL
;
2877 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2878 ret
= mpage_prepare_extent_to_map(&mpd
);
2881 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2885 * We scanned the whole range (or exhausted
2886 * nr_to_write), submitted what was mapped and
2887 * didn't find anything needing mapping. We are
2894 * Caution: If the handle is synchronous,
2895 * ext4_journal_stop() can wait for transaction commit
2896 * to finish which may depend on writeback of pages to
2897 * complete or on page lock to be released. In that
2898 * case, we have to wait until after after we have
2899 * submitted all the IO, released page locks we hold,
2900 * and dropped io_end reference (for extent conversion
2901 * to be able to complete) before stopping the handle.
2903 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2904 ext4_journal_stop(handle
);
2908 /* Unlock pages we didn't use */
2909 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2910 /* Submit prepared bio */
2911 ext4_io_submit(&mpd
.io_submit
);
2914 * Drop our io_end reference we got from init. We have
2915 * to be careful and use deferred io_end finishing if
2916 * we are still holding the transaction as we can
2917 * release the last reference to io_end which may end
2918 * up doing unwritten extent conversion.
2921 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2922 ext4_journal_stop(handle
);
2924 ext4_put_io_end(mpd
.io_submit
.io_end
);
2925 mpd
.io_submit
.io_end
= NULL
;
2927 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2929 * Commit the transaction which would
2930 * free blocks released in the transaction
2933 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2937 /* Fatal error - ENOMEM, EIO... */
2942 blk_finish_plug(&plug
);
2943 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2945 mpd
.last_page
= writeback_index
- 1;
2951 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2953 * Set the writeback_index so that range_cyclic
2954 * mode will write it back later
2956 mapping
->writeback_index
= mpd
.first_page
;
2959 trace_ext4_writepages_result(inode
, wbc
, ret
,
2960 nr_to_write
- wbc
->nr_to_write
);
2961 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2965 static int ext4_dax_writepages(struct address_space
*mapping
,
2966 struct writeback_control
*wbc
)
2969 long nr_to_write
= wbc
->nr_to_write
;
2970 struct inode
*inode
= mapping
->host
;
2971 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2973 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2976 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2977 trace_ext4_writepages(inode
, wbc
);
2979 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
, wbc
);
2980 trace_ext4_writepages_result(inode
, wbc
, ret
,
2981 nr_to_write
- wbc
->nr_to_write
);
2982 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2986 static int ext4_nonda_switch(struct super_block
*sb
)
2988 s64 free_clusters
, dirty_clusters
;
2989 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2992 * switch to non delalloc mode if we are running low
2993 * on free block. The free block accounting via percpu
2994 * counters can get slightly wrong with percpu_counter_batch getting
2995 * accumulated on each CPU without updating global counters
2996 * Delalloc need an accurate free block accounting. So switch
2997 * to non delalloc when we are near to error range.
3000 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
3002 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
3004 * Start pushing delalloc when 1/2 of free blocks are dirty.
3006 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
3007 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
3009 if (2 * free_clusters
< 3 * dirty_clusters
||
3010 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
3012 * free block count is less than 150% of dirty blocks
3013 * or free blocks is less than watermark
3020 /* We always reserve for an inode update; the superblock could be there too */
3021 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
3023 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
3026 if (pos
+ len
<= 0x7fffffffULL
)
3029 /* We might need to update the superblock to set LARGE_FILE */
3033 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3034 loff_t pos
, unsigned len
, unsigned flags
,
3035 struct page
**pagep
, void **fsdata
)
3037 int ret
, retries
= 0;
3040 struct inode
*inode
= mapping
->host
;
3043 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
3046 index
= pos
>> PAGE_SHIFT
;
3048 if (ext4_nonda_switch(inode
->i_sb
) ||
3049 S_ISLNK(inode
->i_mode
)) {
3050 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3051 return ext4_write_begin(file
, mapping
, pos
,
3052 len
, flags
, pagep
, fsdata
);
3054 *fsdata
= (void *)0;
3055 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3057 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
3058 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
3068 * grab_cache_page_write_begin() can take a long time if the
3069 * system is thrashing due to memory pressure, or if the page
3070 * is being written back. So grab it first before we start
3071 * the transaction handle. This also allows us to allocate
3072 * the page (if needed) without using GFP_NOFS.
3075 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3081 * With delayed allocation, we don't log the i_disksize update
3082 * if there is delayed block allocation. But we still need
3083 * to journalling the i_disksize update if writes to the end
3084 * of file which has an already mapped buffer.
3087 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
3088 ext4_da_write_credits(inode
, pos
, len
));
3089 if (IS_ERR(handle
)) {
3091 return PTR_ERR(handle
);
3095 if (page
->mapping
!= mapping
) {
3096 /* The page got truncated from under us */
3099 ext4_journal_stop(handle
);
3102 /* In case writeback began while the page was unlocked */
3103 wait_for_stable_page(page
);
3105 #ifdef CONFIG_FS_ENCRYPTION
3106 ret
= ext4_block_write_begin(page
, pos
, len
,
3107 ext4_da_get_block_prep
);
3109 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3113 ext4_journal_stop(handle
);
3115 * block_write_begin may have instantiated a few blocks
3116 * outside i_size. Trim these off again. Don't need
3117 * i_size_read because we hold i_mutex.
3119 if (pos
+ len
> inode
->i_size
)
3120 ext4_truncate_failed_write(inode
);
3122 if (ret
== -ENOSPC
&&
3123 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3135 * Check if we should update i_disksize
3136 * when write to the end of file but not require block allocation
3138 static int ext4_da_should_update_i_disksize(struct page
*page
,
3139 unsigned long offset
)
3141 struct buffer_head
*bh
;
3142 struct inode
*inode
= page
->mapping
->host
;
3146 bh
= page_buffers(page
);
3147 idx
= offset
>> inode
->i_blkbits
;
3149 for (i
= 0; i
< idx
; i
++)
3150 bh
= bh
->b_this_page
;
3152 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3157 static int ext4_da_write_end(struct file
*file
,
3158 struct address_space
*mapping
,
3159 loff_t pos
, unsigned len
, unsigned copied
,
3160 struct page
*page
, void *fsdata
)
3162 struct inode
*inode
= mapping
->host
;
3164 handle_t
*handle
= ext4_journal_current_handle();
3166 unsigned long start
, end
;
3167 int write_mode
= (int)(unsigned long)fsdata
;
3169 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3170 return ext4_write_end(file
, mapping
, pos
,
3171 len
, copied
, page
, fsdata
);
3173 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3174 start
= pos
& (PAGE_SIZE
- 1);
3175 end
= start
+ copied
- 1;
3178 * generic_write_end() will run mark_inode_dirty() if i_size
3179 * changes. So let's piggyback the i_disksize mark_inode_dirty
3182 new_i_size
= pos
+ copied
;
3183 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3184 if (ext4_has_inline_data(inode
) ||
3185 ext4_da_should_update_i_disksize(page
, end
)) {
3186 ext4_update_i_disksize(inode
, new_i_size
);
3187 /* We need to mark inode dirty even if
3188 * new_i_size is less that inode->i_size
3189 * bu greater than i_disksize.(hint delalloc)
3191 ext4_mark_inode_dirty(handle
, inode
);
3195 if (write_mode
!= CONVERT_INLINE_DATA
&&
3196 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3197 ext4_has_inline_data(inode
))
3198 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3201 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3207 ret2
= ext4_journal_stop(handle
);
3211 return ret
? ret
: copied
;
3214 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3215 unsigned int length
)
3218 * Drop reserved blocks
3220 BUG_ON(!PageLocked(page
));
3221 if (!page_has_buffers(page
))
3224 ext4_da_page_release_reservation(page
, offset
, length
);
3227 ext4_invalidatepage(page
, offset
, length
);
3233 * Force all delayed allocation blocks to be allocated for a given inode.
3235 int ext4_alloc_da_blocks(struct inode
*inode
)
3237 trace_ext4_alloc_da_blocks(inode
);
3239 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3243 * We do something simple for now. The filemap_flush() will
3244 * also start triggering a write of the data blocks, which is
3245 * not strictly speaking necessary (and for users of
3246 * laptop_mode, not even desirable). However, to do otherwise
3247 * would require replicating code paths in:
3249 * ext4_writepages() ->
3250 * write_cache_pages() ---> (via passed in callback function)
3251 * __mpage_da_writepage() -->
3252 * mpage_add_bh_to_extent()
3253 * mpage_da_map_blocks()
3255 * The problem is that write_cache_pages(), located in
3256 * mm/page-writeback.c, marks pages clean in preparation for
3257 * doing I/O, which is not desirable if we're not planning on
3260 * We could call write_cache_pages(), and then redirty all of
3261 * the pages by calling redirty_page_for_writepage() but that
3262 * would be ugly in the extreme. So instead we would need to
3263 * replicate parts of the code in the above functions,
3264 * simplifying them because we wouldn't actually intend to
3265 * write out the pages, but rather only collect contiguous
3266 * logical block extents, call the multi-block allocator, and
3267 * then update the buffer heads with the block allocations.
3269 * For now, though, we'll cheat by calling filemap_flush(),
3270 * which will map the blocks, and start the I/O, but not
3271 * actually wait for the I/O to complete.
3273 return filemap_flush(inode
->i_mapping
);
3277 * bmap() is special. It gets used by applications such as lilo and by
3278 * the swapper to find the on-disk block of a specific piece of data.
3280 * Naturally, this is dangerous if the block concerned is still in the
3281 * journal. If somebody makes a swapfile on an ext4 data-journaling
3282 * filesystem and enables swap, then they may get a nasty shock when the
3283 * data getting swapped to that swapfile suddenly gets overwritten by
3284 * the original zero's written out previously to the journal and
3285 * awaiting writeback in the kernel's buffer cache.
3287 * So, if we see any bmap calls here on a modified, data-journaled file,
3288 * take extra steps to flush any blocks which might be in the cache.
3290 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3292 struct inode
*inode
= mapping
->host
;
3297 * We can get here for an inline file via the FIBMAP ioctl
3299 if (ext4_has_inline_data(inode
))
3302 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3303 test_opt(inode
->i_sb
, DELALLOC
)) {
3305 * With delalloc we want to sync the file
3306 * so that we can make sure we allocate
3309 filemap_write_and_wait(mapping
);
3312 if (EXT4_JOURNAL(inode
) &&
3313 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3315 * This is a REALLY heavyweight approach, but the use of
3316 * bmap on dirty files is expected to be extremely rare:
3317 * only if we run lilo or swapon on a freshly made file
3318 * do we expect this to happen.
3320 * (bmap requires CAP_SYS_RAWIO so this does not
3321 * represent an unprivileged user DOS attack --- we'd be
3322 * in trouble if mortal users could trigger this path at
3325 * NB. EXT4_STATE_JDATA is not set on files other than
3326 * regular files. If somebody wants to bmap a directory
3327 * or symlink and gets confused because the buffer
3328 * hasn't yet been flushed to disk, they deserve
3329 * everything they get.
3332 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3333 journal
= EXT4_JOURNAL(inode
);
3334 jbd2_journal_lock_updates(journal
);
3335 err
= jbd2_journal_flush(journal
);
3336 jbd2_journal_unlock_updates(journal
);
3342 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3345 static int ext4_readpage(struct file
*file
, struct page
*page
)
3348 struct inode
*inode
= page
->mapping
->host
;
3350 trace_ext4_readpage(page
);
3352 if (ext4_has_inline_data(inode
))
3353 ret
= ext4_readpage_inline(inode
, page
);
3356 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1,
3363 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3364 struct list_head
*pages
, unsigned nr_pages
)
3366 struct inode
*inode
= mapping
->host
;
3368 /* If the file has inline data, no need to do readpages. */
3369 if (ext4_has_inline_data(inode
))
3372 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
, true);
3375 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3376 unsigned int length
)
3378 trace_ext4_invalidatepage(page
, offset
, length
);
3380 /* No journalling happens on data buffers when this function is used */
3381 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3383 block_invalidatepage(page
, offset
, length
);
3386 static int __ext4_journalled_invalidatepage(struct page
*page
,
3387 unsigned int offset
,
3388 unsigned int length
)
3390 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3392 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3395 * If it's a full truncate we just forget about the pending dirtying
3397 if (offset
== 0 && length
== PAGE_SIZE
)
3398 ClearPageChecked(page
);
3400 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3403 /* Wrapper for aops... */
3404 static void ext4_journalled_invalidatepage(struct page
*page
,
3405 unsigned int offset
,
3406 unsigned int length
)
3408 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3411 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3413 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3415 trace_ext4_releasepage(page
);
3417 /* Page has dirty journalled data -> cannot release */
3418 if (PageChecked(page
))
3421 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3423 return try_to_free_buffers(page
);
3426 static bool ext4_inode_datasync_dirty(struct inode
*inode
)
3428 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
3431 return !jbd2_transaction_committed(journal
,
3432 EXT4_I(inode
)->i_datasync_tid
);
3433 /* Any metadata buffers to write? */
3434 if (!list_empty(&inode
->i_mapping
->private_list
))
3436 return inode
->i_state
& I_DIRTY_DATASYNC
;
3439 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3440 unsigned flags
, struct iomap
*iomap
)
3442 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3443 unsigned int blkbits
= inode
->i_blkbits
;
3444 unsigned long first_block
, last_block
;
3445 struct ext4_map_blocks map
;
3446 bool delalloc
= false;
3449 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3451 first_block
= offset
>> blkbits
;
3452 last_block
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3453 EXT4_MAX_LOGICAL_BLOCK
);
3455 if (flags
& IOMAP_REPORT
) {
3456 if (ext4_has_inline_data(inode
)) {
3457 ret
= ext4_inline_data_iomap(inode
, iomap
);
3458 if (ret
!= -EAGAIN
) {
3459 if (ret
== 0 && offset
>= iomap
->length
)
3465 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3469 map
.m_lblk
= first_block
;
3470 map
.m_len
= last_block
- first_block
+ 1;
3472 if (flags
& IOMAP_REPORT
) {
3473 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3478 ext4_lblk_t end
= map
.m_lblk
+ map
.m_len
- 1;
3479 struct extent_status es
;
3481 ext4_es_find_extent_range(inode
, &ext4_es_is_delayed
,
3482 map
.m_lblk
, end
, &es
);
3484 if (!es
.es_len
|| es
.es_lblk
> end
) {
3485 /* entire range is a hole */
3486 } else if (es
.es_lblk
> map
.m_lblk
) {
3487 /* range starts with a hole */
3488 map
.m_len
= es
.es_lblk
- map
.m_lblk
;
3490 ext4_lblk_t offs
= 0;
3492 if (es
.es_lblk
< map
.m_lblk
)
3493 offs
= map
.m_lblk
- es
.es_lblk
;
3494 map
.m_lblk
= es
.es_lblk
+ offs
;
3495 map
.m_len
= es
.es_len
- offs
;
3499 } else if (flags
& IOMAP_WRITE
) {
3504 /* Trim mapping request to maximum we can map at once for DIO */
3505 if (map
.m_len
> DIO_MAX_BLOCKS
)
3506 map
.m_len
= DIO_MAX_BLOCKS
;
3507 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3510 * Either we allocate blocks and then we don't get unwritten
3511 * extent so we have reserved enough credits, or the blocks
3512 * are already allocated and unwritten and in that case
3513 * extent conversion fits in the credits as well.
3515 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3518 return PTR_ERR(handle
);
3520 ret
= ext4_map_blocks(handle
, inode
, &map
,
3521 EXT4_GET_BLOCKS_CREATE_ZERO
);
3523 ext4_journal_stop(handle
);
3524 if (ret
== -ENOSPC
&&
3525 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3531 * If we added blocks beyond i_size, we need to make sure they
3532 * will get truncated if we crash before updating i_size in
3533 * ext4_iomap_end(). For faults we don't need to do that (and
3534 * even cannot because for orphan list operations inode_lock is
3535 * required) - if we happen to instantiate block beyond i_size,
3536 * it is because we race with truncate which has already added
3537 * the inode to the orphan list.
3539 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3540 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3543 err
= ext4_orphan_add(handle
, inode
);
3545 ext4_journal_stop(handle
);
3549 ext4_journal_stop(handle
);
3551 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3557 if (ext4_inode_datasync_dirty(inode
))
3558 iomap
->flags
|= IOMAP_F_DIRTY
;
3559 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3560 iomap
->dax_dev
= sbi
->s_daxdev
;
3561 iomap
->offset
= (u64
)first_block
<< blkbits
;
3562 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3565 iomap
->type
= delalloc
? IOMAP_DELALLOC
: IOMAP_HOLE
;
3566 iomap
->addr
= IOMAP_NULL_ADDR
;
3568 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3569 iomap
->type
= IOMAP_MAPPED
;
3570 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3571 iomap
->type
= IOMAP_UNWRITTEN
;
3576 iomap
->addr
= (u64
)map
.m_pblk
<< blkbits
;
3579 if (map
.m_flags
& EXT4_MAP_NEW
)
3580 iomap
->flags
|= IOMAP_F_NEW
;
3585 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3586 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3590 int blkbits
= inode
->i_blkbits
;
3591 bool truncate
= false;
3593 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3596 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3597 if (IS_ERR(handle
)) {
3598 ret
= PTR_ERR(handle
);
3601 if (ext4_update_inode_size(inode
, offset
+ written
))
3602 ext4_mark_inode_dirty(handle
, inode
);
3604 * We may need to truncate allocated but not written blocks beyond EOF.
3606 if (iomap
->offset
+ iomap
->length
>
3607 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3608 ext4_lblk_t written_blk
, end_blk
;
3610 written_blk
= (offset
+ written
) >> blkbits
;
3611 end_blk
= (offset
+ length
) >> blkbits
;
3612 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3616 * Remove inode from orphan list if we were extending a inode and
3617 * everything went fine.
3619 if (!truncate
&& inode
->i_nlink
&&
3620 !list_empty(&EXT4_I(inode
)->i_orphan
))
3621 ext4_orphan_del(handle
, inode
);
3622 ext4_journal_stop(handle
);
3624 ext4_truncate_failed_write(inode
);
3627 * If truncate failed early the inode might still be on the
3628 * orphan list; we need to make sure the inode is removed from
3629 * the orphan list in that case.
3632 ext4_orphan_del(NULL
, inode
);
3637 const struct iomap_ops ext4_iomap_ops
= {
3638 .iomap_begin
= ext4_iomap_begin
,
3639 .iomap_end
= ext4_iomap_end
,
3642 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3643 ssize_t size
, void *private)
3645 ext4_io_end_t
*io_end
= private;
3647 /* if not async direct IO just return */
3651 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3652 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3653 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3656 * Error during AIO DIO. We cannot convert unwritten extents as the
3657 * data was not written. Just clear the unwritten flag and drop io_end.
3660 ext4_clear_io_unwritten_flag(io_end
);
3663 io_end
->offset
= offset
;
3664 io_end
->size
= size
;
3665 ext4_put_io_end(io_end
);
3671 * Handling of direct IO writes.
3673 * For ext4 extent files, ext4 will do direct-io write even to holes,
3674 * preallocated extents, and those write extend the file, no need to
3675 * fall back to buffered IO.
3677 * For holes, we fallocate those blocks, mark them as unwritten
3678 * If those blocks were preallocated, we mark sure they are split, but
3679 * still keep the range to write as unwritten.
3681 * The unwritten extents will be converted to written when DIO is completed.
3682 * For async direct IO, since the IO may still pending when return, we
3683 * set up an end_io call back function, which will do the conversion
3684 * when async direct IO completed.
3686 * If the O_DIRECT write will extend the file then add this inode to the
3687 * orphan list. So recovery will truncate it back to the original size
3688 * if the machine crashes during the write.
3691 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3693 struct file
*file
= iocb
->ki_filp
;
3694 struct inode
*inode
= file
->f_mapping
->host
;
3695 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3697 loff_t offset
= iocb
->ki_pos
;
3698 size_t count
= iov_iter_count(iter
);
3700 get_block_t
*get_block_func
= NULL
;
3702 loff_t final_size
= offset
+ count
;
3706 if (final_size
> inode
->i_size
|| final_size
> ei
->i_disksize
) {
3707 /* Credits for sb + inode write */
3708 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3709 if (IS_ERR(handle
)) {
3710 ret
= PTR_ERR(handle
);
3713 ret
= ext4_orphan_add(handle
, inode
);
3715 ext4_journal_stop(handle
);
3719 ext4_update_i_disksize(inode
, inode
->i_size
);
3720 ext4_journal_stop(handle
);
3723 BUG_ON(iocb
->private == NULL
);
3726 * Make all waiters for direct IO properly wait also for extent
3727 * conversion. This also disallows race between truncate() and
3728 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3730 inode_dio_begin(inode
);
3732 /* If we do a overwrite dio, i_mutex locking can be released */
3733 overwrite
= *((int *)iocb
->private);
3736 inode_unlock(inode
);
3739 * For extent mapped files we could direct write to holes and fallocate.
3741 * Allocated blocks to fill the hole are marked as unwritten to prevent
3742 * parallel buffered read to expose the stale data before DIO complete
3745 * As to previously fallocated extents, ext4 get_block will just simply
3746 * mark the buffer mapped but still keep the extents unwritten.
3748 * For non AIO case, we will convert those unwritten extents to written
3749 * after return back from blockdev_direct_IO. That way we save us from
3750 * allocating io_end structure and also the overhead of offloading
3751 * the extent convertion to a workqueue.
3753 * For async DIO, the conversion needs to be deferred when the
3754 * IO is completed. The ext4 end_io callback function will be
3755 * called to take care of the conversion work. Here for async
3756 * case, we allocate an io_end structure to hook to the iocb.
3758 iocb
->private = NULL
;
3760 get_block_func
= ext4_dio_get_block_overwrite
;
3761 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3762 round_down(offset
, i_blocksize(inode
)) >= inode
->i_size
) {
3763 get_block_func
= ext4_dio_get_block
;
3764 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3765 } else if (is_sync_kiocb(iocb
)) {
3766 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3767 dio_flags
= DIO_LOCKING
;
3769 get_block_func
= ext4_dio_get_block_unwritten_async
;
3770 dio_flags
= DIO_LOCKING
;
3772 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3773 get_block_func
, ext4_end_io_dio
, NULL
,
3776 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3777 EXT4_STATE_DIO_UNWRITTEN
)) {
3780 * for non AIO case, since the IO is already
3781 * completed, we could do the conversion right here
3783 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3787 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3790 inode_dio_end(inode
);
3791 /* take i_mutex locking again if we do a ovewrite dio */
3795 if (ret
< 0 && final_size
> inode
->i_size
)
3796 ext4_truncate_failed_write(inode
);
3798 /* Handle extending of i_size after direct IO write */
3802 /* Credits for sb + inode write */
3803 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3804 if (IS_ERR(handle
)) {
3806 * We wrote the data but cannot extend
3807 * i_size. Bail out. In async io case, we do
3808 * not return error here because we have
3809 * already submmitted the corresponding
3810 * bio. Returning error here makes the caller
3811 * think that this IO is done and failed
3812 * resulting in race with bio's completion
3816 ret
= PTR_ERR(handle
);
3818 ext4_orphan_del(NULL
, inode
);
3823 ext4_orphan_del(handle
, inode
);
3825 loff_t end
= offset
+ ret
;
3826 if (end
> inode
->i_size
|| end
> ei
->i_disksize
) {
3827 ext4_update_i_disksize(inode
, end
);
3828 if (end
> inode
->i_size
)
3829 i_size_write(inode
, end
);
3831 * We're going to return a positive `ret'
3832 * here due to non-zero-length I/O, so there's
3833 * no way of reporting error returns from
3834 * ext4_mark_inode_dirty() to userspace. So
3837 ext4_mark_inode_dirty(handle
, inode
);
3840 err
= ext4_journal_stop(handle
);
3848 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3850 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3851 struct inode
*inode
= mapping
->host
;
3852 size_t count
= iov_iter_count(iter
);
3856 * Shared inode_lock is enough for us - it protects against concurrent
3857 * writes & truncates and since we take care of writing back page cache,
3858 * we are protected against page writeback as well.
3860 inode_lock_shared(inode
);
3861 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3862 iocb
->ki_pos
+ count
- 1);
3865 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3866 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3868 inode_unlock_shared(inode
);
3872 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3874 struct file
*file
= iocb
->ki_filp
;
3875 struct inode
*inode
= file
->f_mapping
->host
;
3876 size_t count
= iov_iter_count(iter
);
3877 loff_t offset
= iocb
->ki_pos
;
3880 #ifdef CONFIG_FS_ENCRYPTION
3881 if (IS_ENCRYPTED(inode
) && S_ISREG(inode
->i_mode
))
3886 * If we are doing data journalling we don't support O_DIRECT
3888 if (ext4_should_journal_data(inode
))
3891 /* Let buffer I/O handle the inline data case. */
3892 if (ext4_has_inline_data(inode
))
3895 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3896 if (iov_iter_rw(iter
) == READ
)
3897 ret
= ext4_direct_IO_read(iocb
, iter
);
3899 ret
= ext4_direct_IO_write(iocb
, iter
);
3900 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3905 * Pages can be marked dirty completely asynchronously from ext4's journalling
3906 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3907 * much here because ->set_page_dirty is called under VFS locks. The page is
3908 * not necessarily locked.
3910 * We cannot just dirty the page and leave attached buffers clean, because the
3911 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3912 * or jbddirty because all the journalling code will explode.
3914 * So what we do is to mark the page "pending dirty" and next time writepage
3915 * is called, propagate that into the buffers appropriately.
3917 static int ext4_journalled_set_page_dirty(struct page
*page
)
3919 SetPageChecked(page
);
3920 return __set_page_dirty_nobuffers(page
);
3923 static int ext4_set_page_dirty(struct page
*page
)
3925 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3926 WARN_ON_ONCE(!page_has_buffers(page
));
3927 return __set_page_dirty_buffers(page
);
3930 static const struct address_space_operations ext4_aops
= {
3931 .readpage
= ext4_readpage
,
3932 .readpages
= ext4_readpages
,
3933 .writepage
= ext4_writepage
,
3934 .writepages
= ext4_writepages
,
3935 .write_begin
= ext4_write_begin
,
3936 .write_end
= ext4_write_end
,
3937 .set_page_dirty
= ext4_set_page_dirty
,
3939 .invalidatepage
= ext4_invalidatepage
,
3940 .releasepage
= ext4_releasepage
,
3941 .direct_IO
= ext4_direct_IO
,
3942 .migratepage
= buffer_migrate_page
,
3943 .is_partially_uptodate
= block_is_partially_uptodate
,
3944 .error_remove_page
= generic_error_remove_page
,
3947 static const struct address_space_operations ext4_journalled_aops
= {
3948 .readpage
= ext4_readpage
,
3949 .readpages
= ext4_readpages
,
3950 .writepage
= ext4_writepage
,
3951 .writepages
= ext4_writepages
,
3952 .write_begin
= ext4_write_begin
,
3953 .write_end
= ext4_journalled_write_end
,
3954 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3956 .invalidatepage
= ext4_journalled_invalidatepage
,
3957 .releasepage
= ext4_releasepage
,
3958 .direct_IO
= ext4_direct_IO
,
3959 .is_partially_uptodate
= block_is_partially_uptodate
,
3960 .error_remove_page
= generic_error_remove_page
,
3963 static const struct address_space_operations ext4_da_aops
= {
3964 .readpage
= ext4_readpage
,
3965 .readpages
= ext4_readpages
,
3966 .writepage
= ext4_writepage
,
3967 .writepages
= ext4_writepages
,
3968 .write_begin
= ext4_da_write_begin
,
3969 .write_end
= ext4_da_write_end
,
3970 .set_page_dirty
= ext4_set_page_dirty
,
3972 .invalidatepage
= ext4_da_invalidatepage
,
3973 .releasepage
= ext4_releasepage
,
3974 .direct_IO
= ext4_direct_IO
,
3975 .migratepage
= buffer_migrate_page
,
3976 .is_partially_uptodate
= block_is_partially_uptodate
,
3977 .error_remove_page
= generic_error_remove_page
,
3980 static const struct address_space_operations ext4_dax_aops
= {
3981 .writepages
= ext4_dax_writepages
,
3982 .direct_IO
= noop_direct_IO
,
3983 .set_page_dirty
= noop_set_page_dirty
,
3985 .invalidatepage
= noop_invalidatepage
,
3988 void ext4_set_aops(struct inode
*inode
)
3990 switch (ext4_inode_journal_mode(inode
)) {
3991 case EXT4_INODE_ORDERED_DATA_MODE
:
3992 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3994 case EXT4_INODE_JOURNAL_DATA_MODE
:
3995 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
4001 inode
->i_mapping
->a_ops
= &ext4_dax_aops
;
4002 else if (test_opt(inode
->i_sb
, DELALLOC
))
4003 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4005 inode
->i_mapping
->a_ops
= &ext4_aops
;
4008 static int __ext4_block_zero_page_range(handle_t
*handle
,
4009 struct address_space
*mapping
, loff_t from
, loff_t length
)
4011 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
4012 unsigned offset
= from
& (PAGE_SIZE
-1);
4013 unsigned blocksize
, pos
;
4015 struct inode
*inode
= mapping
->host
;
4016 struct buffer_head
*bh
;
4020 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
4021 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
4025 blocksize
= inode
->i_sb
->s_blocksize
;
4027 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4029 if (!page_has_buffers(page
))
4030 create_empty_buffers(page
, blocksize
, 0);
4032 /* Find the buffer that contains "offset" */
4033 bh
= page_buffers(page
);
4035 while (offset
>= pos
) {
4036 bh
= bh
->b_this_page
;
4040 if (buffer_freed(bh
)) {
4041 BUFFER_TRACE(bh
, "freed: skip");
4044 if (!buffer_mapped(bh
)) {
4045 BUFFER_TRACE(bh
, "unmapped");
4046 ext4_get_block(inode
, iblock
, bh
, 0);
4047 /* unmapped? It's a hole - nothing to do */
4048 if (!buffer_mapped(bh
)) {
4049 BUFFER_TRACE(bh
, "still unmapped");
4054 /* Ok, it's mapped. Make sure it's up-to-date */
4055 if (PageUptodate(page
))
4056 set_buffer_uptodate(bh
);
4058 if (!buffer_uptodate(bh
)) {
4060 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
4062 /* Uhhuh. Read error. Complain and punt. */
4063 if (!buffer_uptodate(bh
))
4065 if (S_ISREG(inode
->i_mode
) && IS_ENCRYPTED(inode
)) {
4066 /* We expect the key to be set. */
4067 BUG_ON(!fscrypt_has_encryption_key(inode
));
4068 BUG_ON(blocksize
!= PAGE_SIZE
);
4069 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
4070 page
, PAGE_SIZE
, 0, page
->index
));
4073 if (ext4_should_journal_data(inode
)) {
4074 BUFFER_TRACE(bh
, "get write access");
4075 err
= ext4_journal_get_write_access(handle
, bh
);
4079 zero_user(page
, offset
, length
);
4080 BUFFER_TRACE(bh
, "zeroed end of block");
4082 if (ext4_should_journal_data(inode
)) {
4083 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4086 mark_buffer_dirty(bh
);
4087 if (ext4_should_order_data(inode
))
4088 err
= ext4_jbd2_inode_add_write(handle
, inode
);
4098 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4099 * starting from file offset 'from'. The range to be zero'd must
4100 * be contained with in one block. If the specified range exceeds
4101 * the end of the block it will be shortened to end of the block
4102 * that cooresponds to 'from'
4104 static int ext4_block_zero_page_range(handle_t
*handle
,
4105 struct address_space
*mapping
, loff_t from
, loff_t length
)
4107 struct inode
*inode
= mapping
->host
;
4108 unsigned offset
= from
& (PAGE_SIZE
-1);
4109 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4110 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
4113 * correct length if it does not fall between
4114 * 'from' and the end of the block
4116 if (length
> max
|| length
< 0)
4119 if (IS_DAX(inode
)) {
4120 return iomap_zero_range(inode
, from
, length
, NULL
,
4123 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4127 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4128 * up to the end of the block which corresponds to `from'.
4129 * This required during truncate. We need to physically zero the tail end
4130 * of that block so it doesn't yield old data if the file is later grown.
4132 static int ext4_block_truncate_page(handle_t
*handle
,
4133 struct address_space
*mapping
, loff_t from
)
4135 unsigned offset
= from
& (PAGE_SIZE
-1);
4138 struct inode
*inode
= mapping
->host
;
4140 /* If we are processing an encrypted inode during orphan list handling */
4141 if (IS_ENCRYPTED(inode
) && !fscrypt_has_encryption_key(inode
))
4144 blocksize
= inode
->i_sb
->s_blocksize
;
4145 length
= blocksize
- (offset
& (blocksize
- 1));
4147 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4150 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
4151 loff_t lstart
, loff_t length
)
4153 struct super_block
*sb
= inode
->i_sb
;
4154 struct address_space
*mapping
= inode
->i_mapping
;
4155 unsigned partial_start
, partial_end
;
4156 ext4_fsblk_t start
, end
;
4157 loff_t byte_end
= (lstart
+ length
- 1);
4160 partial_start
= lstart
& (sb
->s_blocksize
- 1);
4161 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
4163 start
= lstart
>> sb
->s_blocksize_bits
;
4164 end
= byte_end
>> sb
->s_blocksize_bits
;
4166 /* Handle partial zero within the single block */
4168 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
4169 err
= ext4_block_zero_page_range(handle
, mapping
,
4173 /* Handle partial zero out on the start of the range */
4174 if (partial_start
) {
4175 err
= ext4_block_zero_page_range(handle
, mapping
,
4176 lstart
, sb
->s_blocksize
);
4180 /* Handle partial zero out on the end of the range */
4181 if (partial_end
!= sb
->s_blocksize
- 1)
4182 err
= ext4_block_zero_page_range(handle
, mapping
,
4183 byte_end
- partial_end
,
4188 int ext4_can_truncate(struct inode
*inode
)
4190 if (S_ISREG(inode
->i_mode
))
4192 if (S_ISDIR(inode
->i_mode
))
4194 if (S_ISLNK(inode
->i_mode
))
4195 return !ext4_inode_is_fast_symlink(inode
);
4200 * We have to make sure i_disksize gets properly updated before we truncate
4201 * page cache due to hole punching or zero range. Otherwise i_disksize update
4202 * can get lost as it may have been postponed to submission of writeback but
4203 * that will never happen after we truncate page cache.
4205 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
4209 loff_t size
= i_size_read(inode
);
4211 WARN_ON(!inode_is_locked(inode
));
4212 if (offset
> size
|| offset
+ len
< size
)
4215 if (EXT4_I(inode
)->i_disksize
>= size
)
4218 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4220 return PTR_ERR(handle
);
4221 ext4_update_i_disksize(inode
, size
);
4222 ext4_mark_inode_dirty(handle
, inode
);
4223 ext4_journal_stop(handle
);
4228 static void ext4_wait_dax_page(struct ext4_inode_info
*ei
)
4230 up_write(&ei
->i_mmap_sem
);
4232 down_write(&ei
->i_mmap_sem
);
4235 int ext4_break_layouts(struct inode
*inode
)
4237 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4241 if (WARN_ON_ONCE(!rwsem_is_locked(&ei
->i_mmap_sem
)))
4245 page
= dax_layout_busy_page(inode
->i_mapping
);
4249 error
= ___wait_var_event(&page
->_refcount
,
4250 atomic_read(&page
->_refcount
) == 1,
4251 TASK_INTERRUPTIBLE
, 0, 0,
4252 ext4_wait_dax_page(ei
));
4253 } while (error
== 0);
4259 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4260 * associated with the given offset and length
4262 * @inode: File inode
4263 * @offset: The offset where the hole will begin
4264 * @len: The length of the hole
4266 * Returns: 0 on success or negative on failure
4269 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4271 struct super_block
*sb
= inode
->i_sb
;
4272 ext4_lblk_t first_block
, stop_block
;
4273 struct address_space
*mapping
= inode
->i_mapping
;
4274 loff_t first_block_offset
, last_block_offset
;
4276 unsigned int credits
;
4279 if (!S_ISREG(inode
->i_mode
))
4282 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4285 * Write out all dirty pages to avoid race conditions
4286 * Then release them.
4288 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4289 ret
= filemap_write_and_wait_range(mapping
, offset
,
4290 offset
+ length
- 1);
4297 /* No need to punch hole beyond i_size */
4298 if (offset
>= inode
->i_size
)
4302 * If the hole extends beyond i_size, set the hole
4303 * to end after the page that contains i_size
4305 if (offset
+ length
> inode
->i_size
) {
4306 length
= inode
->i_size
+
4307 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4311 if (offset
& (sb
->s_blocksize
- 1) ||
4312 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4314 * Attach jinode to inode for jbd2 if we do any zeroing of
4317 ret
= ext4_inode_attach_jinode(inode
);
4323 /* Wait all existing dio workers, newcomers will block on i_mutex */
4324 inode_dio_wait(inode
);
4327 * Prevent page faults from reinstantiating pages we have released from
4330 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4332 ret
= ext4_break_layouts(inode
);
4336 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4337 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4339 /* Now release the pages and zero block aligned part of pages*/
4340 if (last_block_offset
> first_block_offset
) {
4341 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4344 truncate_pagecache_range(inode
, first_block_offset
,
4348 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4349 credits
= ext4_writepage_trans_blocks(inode
);
4351 credits
= ext4_blocks_for_truncate(inode
);
4352 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4353 if (IS_ERR(handle
)) {
4354 ret
= PTR_ERR(handle
);
4355 ext4_std_error(sb
, ret
);
4359 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4364 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4365 EXT4_BLOCK_SIZE_BITS(sb
);
4366 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4368 /* If there are blocks to remove, do it */
4369 if (stop_block
> first_block
) {
4371 down_write(&EXT4_I(inode
)->i_data_sem
);
4372 ext4_discard_preallocations(inode
);
4374 ret
= ext4_es_remove_extent(inode
, first_block
,
4375 stop_block
- first_block
);
4377 up_write(&EXT4_I(inode
)->i_data_sem
);
4381 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4382 ret
= ext4_ext_remove_space(inode
, first_block
,
4385 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4388 up_write(&EXT4_I(inode
)->i_data_sem
);
4391 ext4_handle_sync(handle
);
4393 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4394 ext4_mark_inode_dirty(handle
, inode
);
4396 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4398 ext4_journal_stop(handle
);
4400 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4402 inode_unlock(inode
);
4406 int ext4_inode_attach_jinode(struct inode
*inode
)
4408 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4409 struct jbd2_inode
*jinode
;
4411 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4414 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4415 spin_lock(&inode
->i_lock
);
4418 spin_unlock(&inode
->i_lock
);
4421 ei
->jinode
= jinode
;
4422 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4425 spin_unlock(&inode
->i_lock
);
4426 if (unlikely(jinode
!= NULL
))
4427 jbd2_free_inode(jinode
);
4434 * We block out ext4_get_block() block instantiations across the entire
4435 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4436 * simultaneously on behalf of the same inode.
4438 * As we work through the truncate and commit bits of it to the journal there
4439 * is one core, guiding principle: the file's tree must always be consistent on
4440 * disk. We must be able to restart the truncate after a crash.
4442 * The file's tree may be transiently inconsistent in memory (although it
4443 * probably isn't), but whenever we close off and commit a journal transaction,
4444 * the contents of (the filesystem + the journal) must be consistent and
4445 * restartable. It's pretty simple, really: bottom up, right to left (although
4446 * left-to-right works OK too).
4448 * Note that at recovery time, journal replay occurs *before* the restart of
4449 * truncate against the orphan inode list.
4451 * The committed inode has the new, desired i_size (which is the same as
4452 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4453 * that this inode's truncate did not complete and it will again call
4454 * ext4_truncate() to have another go. So there will be instantiated blocks
4455 * to the right of the truncation point in a crashed ext4 filesystem. But
4456 * that's fine - as long as they are linked from the inode, the post-crash
4457 * ext4_truncate() run will find them and release them.
4459 int ext4_truncate(struct inode
*inode
)
4461 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4462 unsigned int credits
;
4465 struct address_space
*mapping
= inode
->i_mapping
;
4468 * There is a possibility that we're either freeing the inode
4469 * or it's a completely new inode. In those cases we might not
4470 * have i_mutex locked because it's not necessary.
4472 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4473 WARN_ON(!inode_is_locked(inode
));
4474 trace_ext4_truncate_enter(inode
);
4476 if (!ext4_can_truncate(inode
))
4479 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4481 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4482 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4484 if (ext4_has_inline_data(inode
)) {
4487 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4494 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4495 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4496 if (ext4_inode_attach_jinode(inode
) < 0)
4500 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4501 credits
= ext4_writepage_trans_blocks(inode
);
4503 credits
= ext4_blocks_for_truncate(inode
);
4505 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4507 return PTR_ERR(handle
);
4509 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4510 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4513 * We add the inode to the orphan list, so that if this
4514 * truncate spans multiple transactions, and we crash, we will
4515 * resume the truncate when the filesystem recovers. It also
4516 * marks the inode dirty, to catch the new size.
4518 * Implication: the file must always be in a sane, consistent
4519 * truncatable state while each transaction commits.
4521 err
= ext4_orphan_add(handle
, inode
);
4525 down_write(&EXT4_I(inode
)->i_data_sem
);
4527 ext4_discard_preallocations(inode
);
4529 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4530 err
= ext4_ext_truncate(handle
, inode
);
4532 ext4_ind_truncate(handle
, inode
);
4534 up_write(&ei
->i_data_sem
);
4539 ext4_handle_sync(handle
);
4543 * If this was a simple ftruncate() and the file will remain alive,
4544 * then we need to clear up the orphan record which we created above.
4545 * However, if this was a real unlink then we were called by
4546 * ext4_evict_inode(), and we allow that function to clean up the
4547 * orphan info for us.
4550 ext4_orphan_del(handle
, inode
);
4552 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4553 ext4_mark_inode_dirty(handle
, inode
);
4554 ext4_journal_stop(handle
);
4556 trace_ext4_truncate_exit(inode
);
4561 * ext4_get_inode_loc returns with an extra refcount against the inode's
4562 * underlying buffer_head on success. If 'in_mem' is true, we have all
4563 * data in memory that is needed to recreate the on-disk version of this
4566 static int __ext4_get_inode_loc(struct inode
*inode
,
4567 struct ext4_iloc
*iloc
, int in_mem
)
4569 struct ext4_group_desc
*gdp
;
4570 struct buffer_head
*bh
;
4571 struct super_block
*sb
= inode
->i_sb
;
4573 int inodes_per_block
, inode_offset
;
4576 if (inode
->i_ino
< EXT4_ROOT_INO
||
4577 inode
->i_ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
4578 return -EFSCORRUPTED
;
4580 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4581 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4586 * Figure out the offset within the block group inode table
4588 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4589 inode_offset
= ((inode
->i_ino
- 1) %
4590 EXT4_INODES_PER_GROUP(sb
));
4591 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4592 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4594 bh
= sb_getblk(sb
, block
);
4597 if (!buffer_uptodate(bh
)) {
4601 * If the buffer has the write error flag, we have failed
4602 * to write out another inode in the same block. In this
4603 * case, we don't have to read the block because we may
4604 * read the old inode data successfully.
4606 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4607 set_buffer_uptodate(bh
);
4609 if (buffer_uptodate(bh
)) {
4610 /* someone brought it uptodate while we waited */
4616 * If we have all information of the inode in memory and this
4617 * is the only valid inode in the block, we need not read the
4621 struct buffer_head
*bitmap_bh
;
4624 start
= inode_offset
& ~(inodes_per_block
- 1);
4626 /* Is the inode bitmap in cache? */
4627 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4628 if (unlikely(!bitmap_bh
))
4632 * If the inode bitmap isn't in cache then the
4633 * optimisation may end up performing two reads instead
4634 * of one, so skip it.
4636 if (!buffer_uptodate(bitmap_bh
)) {
4640 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4641 if (i
== inode_offset
)
4643 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4647 if (i
== start
+ inodes_per_block
) {
4648 /* all other inodes are free, so skip I/O */
4649 memset(bh
->b_data
, 0, bh
->b_size
);
4650 set_buffer_uptodate(bh
);
4658 * If we need to do any I/O, try to pre-readahead extra
4659 * blocks from the inode table.
4661 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4662 ext4_fsblk_t b
, end
, table
;
4664 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4666 table
= ext4_inode_table(sb
, gdp
);
4667 /* s_inode_readahead_blks is always a power of 2 */
4668 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4672 num
= EXT4_INODES_PER_GROUP(sb
);
4673 if (ext4_has_group_desc_csum(sb
))
4674 num
-= ext4_itable_unused_count(sb
, gdp
);
4675 table
+= num
/ inodes_per_block
;
4679 sb_breadahead(sb
, b
++);
4683 * There are other valid inodes in the buffer, this inode
4684 * has in-inode xattrs, or we don't have this inode in memory.
4685 * Read the block from disk.
4687 trace_ext4_load_inode(inode
);
4689 bh
->b_end_io
= end_buffer_read_sync
;
4690 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4692 if (!buffer_uptodate(bh
)) {
4693 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4694 "unable to read itable block");
4704 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4706 /* We have all inode data except xattrs in memory here. */
4707 return __ext4_get_inode_loc(inode
, iloc
,
4708 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4711 static bool ext4_should_use_dax(struct inode
*inode
)
4713 if (!test_opt(inode
->i_sb
, DAX
))
4715 if (!S_ISREG(inode
->i_mode
))
4717 if (ext4_should_journal_data(inode
))
4719 if (ext4_has_inline_data(inode
))
4721 if (ext4_test_inode_flag(inode
, EXT4_INODE_ENCRYPT
))
4726 void ext4_set_inode_flags(struct inode
*inode
)
4728 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4729 unsigned int new_fl
= 0;
4731 if (flags
& EXT4_SYNC_FL
)
4733 if (flags
& EXT4_APPEND_FL
)
4735 if (flags
& EXT4_IMMUTABLE_FL
)
4736 new_fl
|= S_IMMUTABLE
;
4737 if (flags
& EXT4_NOATIME_FL
)
4738 new_fl
|= S_NOATIME
;
4739 if (flags
& EXT4_DIRSYNC_FL
)
4740 new_fl
|= S_DIRSYNC
;
4741 if (ext4_should_use_dax(inode
))
4743 if (flags
& EXT4_ENCRYPT_FL
)
4744 new_fl
|= S_ENCRYPTED
;
4745 if (flags
& EXT4_CASEFOLD_FL
)
4746 new_fl
|= S_CASEFOLD
;
4747 inode_set_flags(inode
, new_fl
,
4748 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
|
4749 S_ENCRYPTED
|S_CASEFOLD
);
4752 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4753 struct ext4_inode_info
*ei
)
4756 struct inode
*inode
= &(ei
->vfs_inode
);
4757 struct super_block
*sb
= inode
->i_sb
;
4759 if (ext4_has_feature_huge_file(sb
)) {
4760 /* we are using combined 48 bit field */
4761 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4762 le32_to_cpu(raw_inode
->i_blocks_lo
);
4763 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4764 /* i_blocks represent file system block size */
4765 return i_blocks
<< (inode
->i_blkbits
- 9);
4770 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4774 static inline int ext4_iget_extra_inode(struct inode
*inode
,
4775 struct ext4_inode
*raw_inode
,
4776 struct ext4_inode_info
*ei
)
4778 __le32
*magic
= (void *)raw_inode
+
4779 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4781 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4782 EXT4_INODE_SIZE(inode
->i_sb
) &&
4783 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4784 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4785 return ext4_find_inline_data_nolock(inode
);
4787 EXT4_I(inode
)->i_inline_off
= 0;
4791 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4793 if (!ext4_has_feature_project(inode
->i_sb
))
4795 *projid
= EXT4_I(inode
)->i_projid
;
4800 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4801 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4804 static inline void ext4_inode_set_iversion_queried(struct inode
*inode
, u64 val
)
4806 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4807 inode_set_iversion_raw(inode
, val
);
4809 inode_set_iversion_queried(inode
, val
);
4811 static inline u64
ext4_inode_peek_iversion(const struct inode
*inode
)
4813 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4814 return inode_peek_iversion_raw(inode
);
4816 return inode_peek_iversion(inode
);
4819 struct inode
*__ext4_iget(struct super_block
*sb
, unsigned long ino
,
4820 ext4_iget_flags flags
, const char *function
,
4823 struct ext4_iloc iloc
;
4824 struct ext4_inode
*raw_inode
;
4825 struct ext4_inode_info
*ei
;
4826 struct inode
*inode
;
4827 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4835 if ((!(flags
& EXT4_IGET_SPECIAL
) &&
4836 (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)) ||
4837 (ino
< EXT4_ROOT_INO
) ||
4838 (ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))) {
4839 if (flags
& EXT4_IGET_HANDLE
)
4840 return ERR_PTR(-ESTALE
);
4841 __ext4_error(sb
, function
, line
,
4842 "inode #%lu: comm %s: iget: illegal inode #",
4843 ino
, current
->comm
);
4844 return ERR_PTR(-EFSCORRUPTED
);
4847 inode
= iget_locked(sb
, ino
);
4849 return ERR_PTR(-ENOMEM
);
4850 if (!(inode
->i_state
& I_NEW
))
4856 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4859 raw_inode
= ext4_raw_inode(&iloc
);
4861 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4862 ext4_error_inode(inode
, function
, line
, 0,
4863 "iget: root inode unallocated");
4864 ret
= -EFSCORRUPTED
;
4868 if ((flags
& EXT4_IGET_HANDLE
) &&
4869 (raw_inode
->i_links_count
== 0) && (raw_inode
->i_mode
== 0)) {
4874 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4875 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4876 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4877 EXT4_INODE_SIZE(inode
->i_sb
) ||
4878 (ei
->i_extra_isize
& 3)) {
4879 ext4_error_inode(inode
, function
, line
, 0,
4880 "iget: bad extra_isize %u "
4883 EXT4_INODE_SIZE(inode
->i_sb
));
4884 ret
= -EFSCORRUPTED
;
4888 ei
->i_extra_isize
= 0;
4890 /* Precompute checksum seed for inode metadata */
4891 if (ext4_has_metadata_csum(sb
)) {
4892 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4894 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4895 __le32 gen
= raw_inode
->i_generation
;
4896 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4898 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4902 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4903 ext4_error_inode(inode
, function
, line
, 0,
4904 "iget: checksum invalid");
4909 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4910 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4911 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4912 if (ext4_has_feature_project(sb
) &&
4913 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4914 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4915 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4917 i_projid
= EXT4_DEF_PROJID
;
4919 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4920 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4921 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4923 i_uid_write(inode
, i_uid
);
4924 i_gid_write(inode
, i_gid
);
4925 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4926 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4928 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4929 ei
->i_inline_off
= 0;
4930 ei
->i_dir_start_lookup
= 0;
4931 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4932 /* We now have enough fields to check if the inode was active or not.
4933 * This is needed because nfsd might try to access dead inodes
4934 * the test is that same one that e2fsck uses
4935 * NeilBrown 1999oct15
4937 if (inode
->i_nlink
== 0) {
4938 if ((inode
->i_mode
== 0 ||
4939 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4940 ino
!= EXT4_BOOT_LOADER_INO
) {
4941 /* this inode is deleted */
4945 /* The only unlinked inodes we let through here have
4946 * valid i_mode and are being read by the orphan
4947 * recovery code: that's fine, we're about to complete
4948 * the process of deleting those.
4949 * OR it is the EXT4_BOOT_LOADER_INO which is
4950 * not initialized on a new filesystem. */
4952 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4953 ext4_set_inode_flags(inode
);
4954 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4955 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4956 if (ext4_has_feature_64bit(sb
))
4958 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4959 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4960 if ((size
= i_size_read(inode
)) < 0) {
4961 ext4_error_inode(inode
, function
, line
, 0,
4962 "iget: bad i_size value: %lld", size
);
4963 ret
= -EFSCORRUPTED
;
4966 ei
->i_disksize
= inode
->i_size
;
4968 ei
->i_reserved_quota
= 0;
4970 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4971 ei
->i_block_group
= iloc
.block_group
;
4972 ei
->i_last_alloc_group
= ~0;
4974 * NOTE! The in-memory inode i_data array is in little-endian order
4975 * even on big-endian machines: we do NOT byteswap the block numbers!
4977 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4978 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4979 INIT_LIST_HEAD(&ei
->i_orphan
);
4982 * Set transaction id's of transactions that have to be committed
4983 * to finish f[data]sync. We set them to currently running transaction
4984 * as we cannot be sure that the inode or some of its metadata isn't
4985 * part of the transaction - the inode could have been reclaimed and
4986 * now it is reread from disk.
4989 transaction_t
*transaction
;
4992 read_lock(&journal
->j_state_lock
);
4993 if (journal
->j_running_transaction
)
4994 transaction
= journal
->j_running_transaction
;
4996 transaction
= journal
->j_committing_transaction
;
4998 tid
= transaction
->t_tid
;
5000 tid
= journal
->j_commit_sequence
;
5001 read_unlock(&journal
->j_state_lock
);
5002 ei
->i_sync_tid
= tid
;
5003 ei
->i_datasync_tid
= tid
;
5006 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5007 if (ei
->i_extra_isize
== 0) {
5008 /* The extra space is currently unused. Use it. */
5009 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
5010 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
5011 EXT4_GOOD_OLD_INODE_SIZE
;
5013 ret
= ext4_iget_extra_inode(inode
, raw_inode
, ei
);
5019 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
5020 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
5021 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
5022 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
5024 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5025 u64 ivers
= le32_to_cpu(raw_inode
->i_disk_version
);
5027 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5028 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5030 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
5032 ext4_inode_set_iversion_queried(inode
, ivers
);
5036 if (ei
->i_file_acl
&&
5037 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
5038 ext4_error_inode(inode
, function
, line
, 0,
5039 "iget: bad extended attribute block %llu",
5041 ret
= -EFSCORRUPTED
;
5043 } else if (!ext4_has_inline_data(inode
)) {
5044 /* validate the block references in the inode */
5045 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5046 (S_ISLNK(inode
->i_mode
) &&
5047 !ext4_inode_is_fast_symlink(inode
))) {
5048 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5049 ret
= ext4_ext_check_inode(inode
);
5051 ret
= ext4_ind_check_inode(inode
);
5057 if (S_ISREG(inode
->i_mode
)) {
5058 inode
->i_op
= &ext4_file_inode_operations
;
5059 inode
->i_fop
= &ext4_file_operations
;
5060 ext4_set_aops(inode
);
5061 } else if (S_ISDIR(inode
->i_mode
)) {
5062 inode
->i_op
= &ext4_dir_inode_operations
;
5063 inode
->i_fop
= &ext4_dir_operations
;
5064 } else if (S_ISLNK(inode
->i_mode
)) {
5065 /* VFS does not allow setting these so must be corruption */
5066 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
5067 ext4_error_inode(inode
, function
, line
, 0,
5068 "iget: immutable or append flags "
5069 "not allowed on symlinks");
5070 ret
= -EFSCORRUPTED
;
5073 if (IS_ENCRYPTED(inode
)) {
5074 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
5075 ext4_set_aops(inode
);
5076 } else if (ext4_inode_is_fast_symlink(inode
)) {
5077 inode
->i_link
= (char *)ei
->i_data
;
5078 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5079 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5080 sizeof(ei
->i_data
) - 1);
5082 inode
->i_op
= &ext4_symlink_inode_operations
;
5083 ext4_set_aops(inode
);
5085 inode_nohighmem(inode
);
5086 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5087 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5088 inode
->i_op
= &ext4_special_inode_operations
;
5089 if (raw_inode
->i_block
[0])
5090 init_special_inode(inode
, inode
->i_mode
,
5091 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5093 init_special_inode(inode
, inode
->i_mode
,
5094 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5095 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
5096 make_bad_inode(inode
);
5098 ret
= -EFSCORRUPTED
;
5099 ext4_error_inode(inode
, function
, line
, 0,
5100 "iget: bogus i_mode (%o)", inode
->i_mode
);
5105 unlock_new_inode(inode
);
5111 return ERR_PTR(ret
);
5114 static int ext4_inode_blocks_set(handle_t
*handle
,
5115 struct ext4_inode
*raw_inode
,
5116 struct ext4_inode_info
*ei
)
5118 struct inode
*inode
= &(ei
->vfs_inode
);
5119 u64 i_blocks
= inode
->i_blocks
;
5120 struct super_block
*sb
= inode
->i_sb
;
5122 if (i_blocks
<= ~0U) {
5124 * i_blocks can be represented in a 32 bit variable
5125 * as multiple of 512 bytes
5127 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5128 raw_inode
->i_blocks_high
= 0;
5129 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5132 if (!ext4_has_feature_huge_file(sb
))
5135 if (i_blocks
<= 0xffffffffffffULL
) {
5137 * i_blocks can be represented in a 48 bit variable
5138 * as multiple of 512 bytes
5140 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5141 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5142 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5144 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5145 /* i_block is stored in file system block size */
5146 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5147 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5148 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5153 struct other_inode
{
5154 unsigned long orig_ino
;
5155 struct ext4_inode
*raw_inode
;
5158 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
5161 struct other_inode
*oi
= (struct other_inode
*) data
;
5163 if ((inode
->i_ino
!= ino
) ||
5164 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
5166 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
5168 spin_lock(&inode
->i_lock
);
5169 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
5170 I_DIRTY_INODE
)) == 0) &&
5171 (inode
->i_state
& I_DIRTY_TIME
)) {
5172 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5174 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
5175 spin_unlock(&inode
->i_lock
);
5177 spin_lock(&ei
->i_raw_lock
);
5178 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
5179 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
5180 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
5181 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
5182 spin_unlock(&ei
->i_raw_lock
);
5183 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
5186 spin_unlock(&inode
->i_lock
);
5191 * Opportunistically update the other time fields for other inodes in
5192 * the same inode table block.
5194 static void ext4_update_other_inodes_time(struct super_block
*sb
,
5195 unsigned long orig_ino
, char *buf
)
5197 struct other_inode oi
;
5199 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
5200 int inode_size
= EXT4_INODE_SIZE(sb
);
5202 oi
.orig_ino
= orig_ino
;
5204 * Calculate the first inode in the inode table block. Inode
5205 * numbers are one-based. That is, the first inode in a block
5206 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5208 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
5209 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
5210 if (ino
== orig_ino
)
5212 oi
.raw_inode
= (struct ext4_inode
*) buf
;
5213 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
5218 * Post the struct inode info into an on-disk inode location in the
5219 * buffer-cache. This gobbles the caller's reference to the
5220 * buffer_head in the inode location struct.
5222 * The caller must have write access to iloc->bh.
5224 static int ext4_do_update_inode(handle_t
*handle
,
5225 struct inode
*inode
,
5226 struct ext4_iloc
*iloc
)
5228 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5229 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5230 struct buffer_head
*bh
= iloc
->bh
;
5231 struct super_block
*sb
= inode
->i_sb
;
5232 int err
= 0, rc
, block
;
5233 int need_datasync
= 0, set_large_file
= 0;
5238 spin_lock(&ei
->i_raw_lock
);
5240 /* For fields not tracked in the in-memory inode,
5241 * initialise them to zero for new inodes. */
5242 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5243 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5245 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5246 i_uid
= i_uid_read(inode
);
5247 i_gid
= i_gid_read(inode
);
5248 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
5249 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5250 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
5251 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
5253 * Fix up interoperability with old kernels. Otherwise, old inodes get
5254 * re-used with the upper 16 bits of the uid/gid intact
5256 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
5257 raw_inode
->i_uid_high
= 0;
5258 raw_inode
->i_gid_high
= 0;
5260 raw_inode
->i_uid_high
=
5261 cpu_to_le16(high_16_bits(i_uid
));
5262 raw_inode
->i_gid_high
=
5263 cpu_to_le16(high_16_bits(i_gid
));
5266 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
5267 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
5268 raw_inode
->i_uid_high
= 0;
5269 raw_inode
->i_gid_high
= 0;
5271 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5273 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5274 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5275 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5276 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5278 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5280 spin_unlock(&ei
->i_raw_lock
);
5283 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5284 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5285 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5286 raw_inode
->i_file_acl_high
=
5287 cpu_to_le16(ei
->i_file_acl
>> 32);
5288 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5289 if (ei
->i_disksize
!= ext4_isize(inode
->i_sb
, raw_inode
)) {
5290 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5293 if (ei
->i_disksize
> 0x7fffffffULL
) {
5294 if (!ext4_has_feature_large_file(sb
) ||
5295 EXT4_SB(sb
)->s_es
->s_rev_level
==
5296 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5299 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5300 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5301 if (old_valid_dev(inode
->i_rdev
)) {
5302 raw_inode
->i_block
[0] =
5303 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5304 raw_inode
->i_block
[1] = 0;
5306 raw_inode
->i_block
[0] = 0;
5307 raw_inode
->i_block
[1] =
5308 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5309 raw_inode
->i_block
[2] = 0;
5311 } else if (!ext4_has_inline_data(inode
)) {
5312 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5313 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5316 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5317 u64 ivers
= ext4_inode_peek_iversion(inode
);
5319 raw_inode
->i_disk_version
= cpu_to_le32(ivers
);
5320 if (ei
->i_extra_isize
) {
5321 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5322 raw_inode
->i_version_hi
=
5323 cpu_to_le32(ivers
>> 32);
5324 raw_inode
->i_extra_isize
=
5325 cpu_to_le16(ei
->i_extra_isize
);
5329 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5330 i_projid
!= EXT4_DEF_PROJID
);
5332 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5333 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5334 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5336 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5337 spin_unlock(&ei
->i_raw_lock
);
5338 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
)
5339 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5342 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5343 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5346 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5347 if (set_large_file
) {
5348 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5349 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5352 ext4_set_feature_large_file(sb
);
5353 ext4_handle_sync(handle
);
5354 err
= ext4_handle_dirty_super(handle
, sb
);
5356 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5359 ext4_std_error(inode
->i_sb
, err
);
5364 * ext4_write_inode()
5366 * We are called from a few places:
5368 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5369 * Here, there will be no transaction running. We wait for any running
5370 * transaction to commit.
5372 * - Within flush work (sys_sync(), kupdate and such).
5373 * We wait on commit, if told to.
5375 * - Within iput_final() -> write_inode_now()
5376 * We wait on commit, if told to.
5378 * In all cases it is actually safe for us to return without doing anything,
5379 * because the inode has been copied into a raw inode buffer in
5380 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5383 * Note that we are absolutely dependent upon all inode dirtiers doing the
5384 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5385 * which we are interested.
5387 * It would be a bug for them to not do this. The code:
5389 * mark_inode_dirty(inode)
5391 * inode->i_size = expr;
5393 * is in error because write_inode() could occur while `stuff()' is running,
5394 * and the new i_size will be lost. Plus the inode will no longer be on the
5395 * superblock's dirty inode list.
5397 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5401 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
) ||
5402 sb_rdonly(inode
->i_sb
))
5405 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5408 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5409 if (ext4_journal_current_handle()) {
5410 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5416 * No need to force transaction in WB_SYNC_NONE mode. Also
5417 * ext4_sync_fs() will force the commit after everything is
5420 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5423 err
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5424 EXT4_I(inode
)->i_sync_tid
);
5426 struct ext4_iloc iloc
;
5428 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5432 * sync(2) will flush the whole buffer cache. No need to do
5433 * it here separately for each inode.
5435 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5436 sync_dirty_buffer(iloc
.bh
);
5437 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5438 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5439 "IO error syncing inode");
5448 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5449 * buffers that are attached to a page stradding i_size and are undergoing
5450 * commit. In that case we have to wait for commit to finish and try again.
5452 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5456 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5457 tid_t commit_tid
= 0;
5460 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5462 * All buffers in the last page remain valid? Then there's nothing to
5463 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5466 if (offset
> PAGE_SIZE
- i_blocksize(inode
))
5469 page
= find_lock_page(inode
->i_mapping
,
5470 inode
->i_size
>> PAGE_SHIFT
);
5473 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5474 PAGE_SIZE
- offset
);
5480 read_lock(&journal
->j_state_lock
);
5481 if (journal
->j_committing_transaction
)
5482 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5483 read_unlock(&journal
->j_state_lock
);
5485 jbd2_log_wait_commit(journal
, commit_tid
);
5492 * Called from notify_change.
5494 * We want to trap VFS attempts to truncate the file as soon as
5495 * possible. In particular, we want to make sure that when the VFS
5496 * shrinks i_size, we put the inode on the orphan list and modify
5497 * i_disksize immediately, so that during the subsequent flushing of
5498 * dirty pages and freeing of disk blocks, we can guarantee that any
5499 * commit will leave the blocks being flushed in an unused state on
5500 * disk. (On recovery, the inode will get truncated and the blocks will
5501 * be freed, so we have a strong guarantee that no future commit will
5502 * leave these blocks visible to the user.)
5504 * Another thing we have to assure is that if we are in ordered mode
5505 * and inode is still attached to the committing transaction, we must
5506 * we start writeout of all the dirty pages which are being truncated.
5507 * This way we are sure that all the data written in the previous
5508 * transaction are already on disk (truncate waits for pages under
5511 * Called with inode->i_mutex down.
5513 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5515 struct inode
*inode
= d_inode(dentry
);
5518 const unsigned int ia_valid
= attr
->ia_valid
;
5520 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5523 error
= setattr_prepare(dentry
, attr
);
5527 error
= fscrypt_prepare_setattr(dentry
, attr
);
5531 if (is_quota_modification(inode
, attr
)) {
5532 error
= dquot_initialize(inode
);
5536 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5537 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5540 /* (user+group)*(old+new) structure, inode write (sb,
5541 * inode block, ? - but truncate inode update has it) */
5542 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5543 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5544 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5545 if (IS_ERR(handle
)) {
5546 error
= PTR_ERR(handle
);
5550 /* dquot_transfer() calls back ext4_get_inode_usage() which
5551 * counts xattr inode references.
5553 down_read(&EXT4_I(inode
)->xattr_sem
);
5554 error
= dquot_transfer(inode
, attr
);
5555 up_read(&EXT4_I(inode
)->xattr_sem
);
5558 ext4_journal_stop(handle
);
5561 /* Update corresponding info in inode so that everything is in
5562 * one transaction */
5563 if (attr
->ia_valid
& ATTR_UID
)
5564 inode
->i_uid
= attr
->ia_uid
;
5565 if (attr
->ia_valid
& ATTR_GID
)
5566 inode
->i_gid
= attr
->ia_gid
;
5567 error
= ext4_mark_inode_dirty(handle
, inode
);
5568 ext4_journal_stop(handle
);
5571 if (attr
->ia_valid
& ATTR_SIZE
) {
5573 loff_t oldsize
= inode
->i_size
;
5574 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5576 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5577 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5579 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5582 if (!S_ISREG(inode
->i_mode
))
5585 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5586 inode_inc_iversion(inode
);
5588 if (ext4_should_order_data(inode
) &&
5589 (attr
->ia_size
< inode
->i_size
)) {
5590 error
= ext4_begin_ordered_truncate(inode
,
5595 if (attr
->ia_size
!= inode
->i_size
) {
5596 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5597 if (IS_ERR(handle
)) {
5598 error
= PTR_ERR(handle
);
5601 if (ext4_handle_valid(handle
) && shrink
) {
5602 error
= ext4_orphan_add(handle
, inode
);
5606 * Update c/mtime on truncate up, ext4_truncate() will
5607 * update c/mtime in shrink case below
5610 inode
->i_mtime
= current_time(inode
);
5611 inode
->i_ctime
= inode
->i_mtime
;
5613 down_write(&EXT4_I(inode
)->i_data_sem
);
5614 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5615 rc
= ext4_mark_inode_dirty(handle
, inode
);
5619 * We have to update i_size under i_data_sem together
5620 * with i_disksize to avoid races with writeback code
5621 * running ext4_wb_update_i_disksize().
5624 i_size_write(inode
, attr
->ia_size
);
5625 up_write(&EXT4_I(inode
)->i_data_sem
);
5626 ext4_journal_stop(handle
);
5629 ext4_orphan_del(NULL
, inode
);
5634 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5637 * Blocks are going to be removed from the inode. Wait
5638 * for dio in flight. Temporarily disable
5639 * dioread_nolock to prevent livelock.
5642 if (!ext4_should_journal_data(inode
)) {
5643 inode_dio_wait(inode
);
5645 ext4_wait_for_tail_page_commit(inode
);
5647 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5649 rc
= ext4_break_layouts(inode
);
5651 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5657 * Truncate pagecache after we've waited for commit
5658 * in data=journal mode to make pages freeable.
5660 truncate_pagecache(inode
, inode
->i_size
);
5662 rc
= ext4_truncate(inode
);
5666 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5670 setattr_copy(inode
, attr
);
5671 mark_inode_dirty(inode
);
5675 * If the call to ext4_truncate failed to get a transaction handle at
5676 * all, we need to clean up the in-core orphan list manually.
5678 if (orphan
&& inode
->i_nlink
)
5679 ext4_orphan_del(NULL
, inode
);
5681 if (!error
&& (ia_valid
& ATTR_MODE
))
5682 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5685 ext4_std_error(inode
->i_sb
, error
);
5691 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5692 u32 request_mask
, unsigned int query_flags
)
5694 struct inode
*inode
= d_inode(path
->dentry
);
5695 struct ext4_inode
*raw_inode
;
5696 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5699 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5700 stat
->result_mask
|= STATX_BTIME
;
5701 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5702 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5705 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5706 if (flags
& EXT4_APPEND_FL
)
5707 stat
->attributes
|= STATX_ATTR_APPEND
;
5708 if (flags
& EXT4_COMPR_FL
)
5709 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5710 if (flags
& EXT4_ENCRYPT_FL
)
5711 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5712 if (flags
& EXT4_IMMUTABLE_FL
)
5713 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5714 if (flags
& EXT4_NODUMP_FL
)
5715 stat
->attributes
|= STATX_ATTR_NODUMP
;
5717 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5718 STATX_ATTR_COMPRESSED
|
5719 STATX_ATTR_ENCRYPTED
|
5720 STATX_ATTR_IMMUTABLE
|
5723 generic_fillattr(inode
, stat
);
5727 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5728 u32 request_mask
, unsigned int query_flags
)
5730 struct inode
*inode
= d_inode(path
->dentry
);
5731 u64 delalloc_blocks
;
5733 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5736 * If there is inline data in the inode, the inode will normally not
5737 * have data blocks allocated (it may have an external xattr block).
5738 * Report at least one sector for such files, so tools like tar, rsync,
5739 * others don't incorrectly think the file is completely sparse.
5741 if (unlikely(ext4_has_inline_data(inode
)))
5742 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5745 * We can't update i_blocks if the block allocation is delayed
5746 * otherwise in the case of system crash before the real block
5747 * allocation is done, we will have i_blocks inconsistent with
5748 * on-disk file blocks.
5749 * We always keep i_blocks updated together with real
5750 * allocation. But to not confuse with user, stat
5751 * will return the blocks that include the delayed allocation
5752 * blocks for this file.
5754 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5755 EXT4_I(inode
)->i_reserved_data_blocks
);
5756 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5760 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5763 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5764 return ext4_ind_trans_blocks(inode
, lblocks
);
5765 return ext4_ext_index_trans_blocks(inode
, pextents
);
5769 * Account for index blocks, block groups bitmaps and block group
5770 * descriptor blocks if modify datablocks and index blocks
5771 * worse case, the indexs blocks spread over different block groups
5773 * If datablocks are discontiguous, they are possible to spread over
5774 * different block groups too. If they are contiguous, with flexbg,
5775 * they could still across block group boundary.
5777 * Also account for superblock, inode, quota and xattr blocks
5779 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5782 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5788 * How many index blocks need to touch to map @lblocks logical blocks
5789 * to @pextents physical extents?
5791 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5796 * Now let's see how many group bitmaps and group descriptors need
5799 groups
= idxblocks
+ pextents
;
5801 if (groups
> ngroups
)
5803 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5804 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5806 /* bitmaps and block group descriptor blocks */
5807 ret
+= groups
+ gdpblocks
;
5809 /* Blocks for super block, inode, quota and xattr blocks */
5810 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5816 * Calculate the total number of credits to reserve to fit
5817 * the modification of a single pages into a single transaction,
5818 * which may include multiple chunks of block allocations.
5820 * This could be called via ext4_write_begin()
5822 * We need to consider the worse case, when
5823 * one new block per extent.
5825 int ext4_writepage_trans_blocks(struct inode
*inode
)
5827 int bpp
= ext4_journal_blocks_per_page(inode
);
5830 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5832 /* Account for data blocks for journalled mode */
5833 if (ext4_should_journal_data(inode
))
5839 * Calculate the journal credits for a chunk of data modification.
5841 * This is called from DIO, fallocate or whoever calling
5842 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5844 * journal buffers for data blocks are not included here, as DIO
5845 * and fallocate do no need to journal data buffers.
5847 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5849 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5853 * The caller must have previously called ext4_reserve_inode_write().
5854 * Give this, we know that the caller already has write access to iloc->bh.
5856 int ext4_mark_iloc_dirty(handle_t
*handle
,
5857 struct inode
*inode
, struct ext4_iloc
*iloc
)
5861 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
5865 if (IS_I_VERSION(inode
))
5866 inode_inc_iversion(inode
);
5868 /* the do_update_inode consumes one bh->b_count */
5871 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5872 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5878 * On success, We end up with an outstanding reference count against
5879 * iloc->bh. This _must_ be cleaned up later.
5883 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5884 struct ext4_iloc
*iloc
)
5888 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5891 err
= ext4_get_inode_loc(inode
, iloc
);
5893 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5894 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5900 ext4_std_error(inode
->i_sb
, err
);
5904 static int __ext4_expand_extra_isize(struct inode
*inode
,
5905 unsigned int new_extra_isize
,
5906 struct ext4_iloc
*iloc
,
5907 handle_t
*handle
, int *no_expand
)
5909 struct ext4_inode
*raw_inode
;
5910 struct ext4_xattr_ibody_header
*header
;
5913 raw_inode
= ext4_raw_inode(iloc
);
5915 header
= IHDR(inode
, raw_inode
);
5917 /* No extended attributes present */
5918 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5919 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5920 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5921 EXT4_I(inode
)->i_extra_isize
, 0,
5922 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5923 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5927 /* try to expand with EAs present */
5928 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5932 * Inode size expansion failed; don't try again
5941 * Expand an inode by new_extra_isize bytes.
5942 * Returns 0 on success or negative error number on failure.
5944 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5945 unsigned int new_extra_isize
,
5946 struct ext4_iloc iloc
,
5952 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5956 * In nojournal mode, we can immediately attempt to expand
5957 * the inode. When journaled, we first need to obtain extra
5958 * buffer credits since we may write into the EA block
5959 * with this same handle. If journal_extend fails, then it will
5960 * only result in a minor loss of functionality for that inode.
5961 * If this is felt to be critical, then e2fsck should be run to
5962 * force a large enough s_min_extra_isize.
5964 if (ext4_handle_valid(handle
) &&
5965 jbd2_journal_extend(handle
,
5966 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) != 0)
5969 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5972 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5973 handle
, &no_expand
);
5974 ext4_write_unlock_xattr(inode
, &no_expand
);
5979 int ext4_expand_extra_isize(struct inode
*inode
,
5980 unsigned int new_extra_isize
,
5981 struct ext4_iloc
*iloc
)
5987 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5992 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5993 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5994 if (IS_ERR(handle
)) {
5995 error
= PTR_ERR(handle
);
6000 ext4_write_lock_xattr(inode
, &no_expand
);
6002 BUFFER_TRACE(iloc
->bh
, "get_write_access");
6003 error
= ext4_journal_get_write_access(handle
, iloc
->bh
);
6009 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
6010 handle
, &no_expand
);
6012 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
6016 ext4_write_unlock_xattr(inode
, &no_expand
);
6018 ext4_journal_stop(handle
);
6023 * What we do here is to mark the in-core inode as clean with respect to inode
6024 * dirtiness (it may still be data-dirty).
6025 * This means that the in-core inode may be reaped by prune_icache
6026 * without having to perform any I/O. This is a very good thing,
6027 * because *any* task may call prune_icache - even ones which
6028 * have a transaction open against a different journal.
6030 * Is this cheating? Not really. Sure, we haven't written the
6031 * inode out, but prune_icache isn't a user-visible syncing function.
6032 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6033 * we start and wait on commits.
6035 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
6037 struct ext4_iloc iloc
;
6038 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
6042 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
6043 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
6047 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
6048 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
6051 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
6055 * ext4_dirty_inode() is called from __mark_inode_dirty()
6057 * We're really interested in the case where a file is being extended.
6058 * i_size has been changed by generic_commit_write() and we thus need
6059 * to include the updated inode in the current transaction.
6061 * Also, dquot_alloc_block() will always dirty the inode when blocks
6062 * are allocated to the file.
6064 * If the inode is marked synchronous, we don't honour that here - doing
6065 * so would cause a commit on atime updates, which we don't bother doing.
6066 * We handle synchronous inodes at the highest possible level.
6068 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6069 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6070 * to copy into the on-disk inode structure are the timestamp files.
6072 void ext4_dirty_inode(struct inode
*inode
, int flags
)
6076 if (flags
== I_DIRTY_TIME
)
6078 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
6082 ext4_mark_inode_dirty(handle
, inode
);
6084 ext4_journal_stop(handle
);
6089 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
6094 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
6097 * We have to be very careful here: changing a data block's
6098 * journaling status dynamically is dangerous. If we write a
6099 * data block to the journal, change the status and then delete
6100 * that block, we risk forgetting to revoke the old log record
6101 * from the journal and so a subsequent replay can corrupt data.
6102 * So, first we make sure that the journal is empty and that
6103 * nobody is changing anything.
6106 journal
= EXT4_JOURNAL(inode
);
6109 if (is_journal_aborted(journal
))
6112 /* Wait for all existing dio workers */
6113 inode_dio_wait(inode
);
6116 * Before flushing the journal and switching inode's aops, we have
6117 * to flush all dirty data the inode has. There can be outstanding
6118 * delayed allocations, there can be unwritten extents created by
6119 * fallocate or buffered writes in dioread_nolock mode covered by
6120 * dirty data which can be converted only after flushing the dirty
6121 * data (and journalled aops don't know how to handle these cases).
6124 down_write(&EXT4_I(inode
)->i_mmap_sem
);
6125 err
= filemap_write_and_wait(inode
->i_mapping
);
6127 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6132 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
6133 jbd2_journal_lock_updates(journal
);
6136 * OK, there are no updates running now, and all cached data is
6137 * synced to disk. We are now in a completely consistent state
6138 * which doesn't have anything in the journal, and we know that
6139 * no filesystem updates are running, so it is safe to modify
6140 * the inode's in-core data-journaling state flag now.
6144 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6146 err
= jbd2_journal_flush(journal
);
6148 jbd2_journal_unlock_updates(journal
);
6149 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6152 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6154 ext4_set_aops(inode
);
6156 jbd2_journal_unlock_updates(journal
);
6157 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6160 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6162 /* Finally we can mark the inode as dirty. */
6164 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
6166 return PTR_ERR(handle
);
6168 err
= ext4_mark_inode_dirty(handle
, inode
);
6169 ext4_handle_sync(handle
);
6170 ext4_journal_stop(handle
);
6171 ext4_std_error(inode
->i_sb
, err
);
6176 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
6178 return !buffer_mapped(bh
);
6181 vm_fault_t
ext4_page_mkwrite(struct vm_fault
*vmf
)
6183 struct vm_area_struct
*vma
= vmf
->vma
;
6184 struct page
*page
= vmf
->page
;
6189 struct file
*file
= vma
->vm_file
;
6190 struct inode
*inode
= file_inode(file
);
6191 struct address_space
*mapping
= inode
->i_mapping
;
6193 get_block_t
*get_block
;
6196 sb_start_pagefault(inode
->i_sb
);
6197 file_update_time(vma
->vm_file
);
6199 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6201 err
= ext4_convert_inline_data(inode
);
6205 /* Delalloc case is easy... */
6206 if (test_opt(inode
->i_sb
, DELALLOC
) &&
6207 !ext4_should_journal_data(inode
) &&
6208 !ext4_nonda_switch(inode
->i_sb
)) {
6210 err
= block_page_mkwrite(vma
, vmf
,
6211 ext4_da_get_block_prep
);
6212 } while (err
== -ENOSPC
&&
6213 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
6218 size
= i_size_read(inode
);
6219 /* Page got truncated from under us? */
6220 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6222 ret
= VM_FAULT_NOPAGE
;
6226 if (page
->index
== size
>> PAGE_SHIFT
)
6227 len
= size
& ~PAGE_MASK
;
6231 * Return if we have all the buffers mapped. This avoids the need to do
6232 * journal_start/journal_stop which can block and take a long time
6234 if (page_has_buffers(page
)) {
6235 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
6237 ext4_bh_unmapped
)) {
6238 /* Wait so that we don't change page under IO */
6239 wait_for_stable_page(page
);
6240 ret
= VM_FAULT_LOCKED
;
6245 /* OK, we need to fill the hole... */
6246 if (ext4_should_dioread_nolock(inode
))
6247 get_block
= ext4_get_block_unwritten
;
6249 get_block
= ext4_get_block
;
6251 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6252 ext4_writepage_trans_blocks(inode
));
6253 if (IS_ERR(handle
)) {
6254 ret
= VM_FAULT_SIGBUS
;
6257 err
= block_page_mkwrite(vma
, vmf
, get_block
);
6258 if (!err
&& ext4_should_journal_data(inode
)) {
6259 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
6260 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
6262 ret
= VM_FAULT_SIGBUS
;
6263 ext4_journal_stop(handle
);
6266 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6268 ext4_journal_stop(handle
);
6269 if (err
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6272 ret
= block_page_mkwrite_return(err
);
6274 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6275 sb_end_pagefault(inode
->i_sb
);
6279 vm_fault_t
ext4_filemap_fault(struct vm_fault
*vmf
)
6281 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
6284 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6285 ret
= filemap_fault(vmf
);
6286 up_read(&EXT4_I(inode
)->i_mmap_sem
);